| File: | build/llvm-toolchain-snapshot-15~++20220410100727+3c1483609369/clang/lib/CodeGen/CGBuiltin.cpp |
| Warning: | line 15570, column 5 Undefined or garbage value returned to caller |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===// | |||
| 2 | // | |||
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | |||
| 4 | // See https://llvm.org/LICENSE.txt for license information. | |||
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | |||
| 6 | // | |||
| 7 | //===----------------------------------------------------------------------===// | |||
| 8 | // | |||
| 9 | // This contains code to emit Builtin calls as LLVM code. | |||
| 10 | // | |||
| 11 | //===----------------------------------------------------------------------===// | |||
| 12 | ||||
| 13 | #include "CGCUDARuntime.h" | |||
| 14 | #include "CGCXXABI.h" | |||
| 15 | #include "CGObjCRuntime.h" | |||
| 16 | #include "CGOpenCLRuntime.h" | |||
| 17 | #include "CGRecordLayout.h" | |||
| 18 | #include "CodeGenFunction.h" | |||
| 19 | #include "CodeGenModule.h" | |||
| 20 | #include "ConstantEmitter.h" | |||
| 21 | #include "PatternInit.h" | |||
| 22 | #include "TargetInfo.h" | |||
| 23 | #include "clang/AST/ASTContext.h" | |||
| 24 | #include "clang/AST/Attr.h" | |||
| 25 | #include "clang/AST/Decl.h" | |||
| 26 | #include "clang/AST/OSLog.h" | |||
| 27 | #include "clang/Basic/TargetBuiltins.h" | |||
| 28 | #include "clang/Basic/TargetInfo.h" | |||
| 29 | #include "clang/CodeGen/CGFunctionInfo.h" | |||
| 30 | #include "llvm/ADT/APFloat.h" | |||
| 31 | #include "llvm/ADT/APInt.h" | |||
| 32 | #include "llvm/ADT/SmallPtrSet.h" | |||
| 33 | #include "llvm/ADT/StringExtras.h" | |||
| 34 | #include "llvm/Analysis/ValueTracking.h" | |||
| 35 | #include "llvm/IR/DataLayout.h" | |||
| 36 | #include "llvm/IR/InlineAsm.h" | |||
| 37 | #include "llvm/IR/Intrinsics.h" | |||
| 38 | #include "llvm/IR/IntrinsicsAArch64.h" | |||
| 39 | #include "llvm/IR/IntrinsicsAMDGPU.h" | |||
| 40 | #include "llvm/IR/IntrinsicsARM.h" | |||
| 41 | #include "llvm/IR/IntrinsicsBPF.h" | |||
| 42 | #include "llvm/IR/IntrinsicsHexagon.h" | |||
| 43 | #include "llvm/IR/IntrinsicsNVPTX.h" | |||
| 44 | #include "llvm/IR/IntrinsicsPowerPC.h" | |||
| 45 | #include "llvm/IR/IntrinsicsR600.h" | |||
| 46 | #include "llvm/IR/IntrinsicsRISCV.h" | |||
| 47 | #include "llvm/IR/IntrinsicsS390.h" | |||
| 48 | #include "llvm/IR/IntrinsicsVE.h" | |||
| 49 | #include "llvm/IR/IntrinsicsWebAssembly.h" | |||
| 50 | #include "llvm/IR/IntrinsicsX86.h" | |||
| 51 | #include "llvm/IR/MDBuilder.h" | |||
| 52 | #include "llvm/IR/MatrixBuilder.h" | |||
| 53 | #include "llvm/Support/ConvertUTF.h" | |||
| 54 | #include "llvm/Support/ScopedPrinter.h" | |||
| 55 | #include "llvm/Support/X86TargetParser.h" | |||
| 56 | #include <sstream> | |||
| 57 | ||||
| 58 | using namespace clang; | |||
| 59 | using namespace CodeGen; | |||
| 60 | using namespace llvm; | |||
| 61 | ||||
| 62 | static | |||
| 63 | int64_t clamp(int64_t Value, int64_t Low, int64_t High) { | |||
| 64 | return std::min(High, std::max(Low, Value)); | |||
| 65 | } | |||
| 66 | ||||
| 67 | static void initializeAlloca(CodeGenFunction &CGF, AllocaInst *AI, Value *Size, | |||
| 68 | Align AlignmentInBytes) { | |||
| 69 | ConstantInt *Byte; | |||
| 70 | switch (CGF.getLangOpts().getTrivialAutoVarInit()) { | |||
| 71 | case LangOptions::TrivialAutoVarInitKind::Uninitialized: | |||
| 72 | // Nothing to initialize. | |||
| 73 | return; | |||
| 74 | case LangOptions::TrivialAutoVarInitKind::Zero: | |||
| 75 | Byte = CGF.Builder.getInt8(0x00); | |||
| 76 | break; | |||
| 77 | case LangOptions::TrivialAutoVarInitKind::Pattern: { | |||
| 78 | llvm::Type *Int8 = llvm::IntegerType::getInt8Ty(CGF.CGM.getLLVMContext()); | |||
| 79 | Byte = llvm::dyn_cast<llvm::ConstantInt>( | |||
| 80 | initializationPatternFor(CGF.CGM, Int8)); | |||
| 81 | break; | |||
| 82 | } | |||
| 83 | } | |||
| 84 | if (CGF.CGM.stopAutoInit()) | |||
| 85 | return; | |||
| 86 | auto *I = CGF.Builder.CreateMemSet(AI, Byte, Size, AlignmentInBytes); | |||
| 87 | I->addAnnotationMetadata("auto-init"); | |||
| 88 | } | |||
| 89 | ||||
| 90 | /// getBuiltinLibFunction - Given a builtin id for a function like | |||
| 91 | /// "__builtin_fabsf", return a Function* for "fabsf". | |||
| 92 | llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD, | |||
| 93 | unsigned BuiltinID) { | |||
| 94 | assert(Context.BuiltinInfo.isLibFunction(BuiltinID))(static_cast <bool> (Context.BuiltinInfo.isLibFunction( BuiltinID)) ? void (0) : __assert_fail ("Context.BuiltinInfo.isLibFunction(BuiltinID)" , "clang/lib/CodeGen/CGBuiltin.cpp", 94, __extension__ __PRETTY_FUNCTION__ )); | |||
| 95 | ||||
| 96 | // Get the name, skip over the __builtin_ prefix (if necessary). | |||
| 97 | StringRef Name; | |||
| 98 | GlobalDecl D(FD); | |||
| 99 | ||||
| 100 | // TODO: This list should be expanded or refactored after all GCC-compatible | |||
| 101 | // std libcall builtins are implemented. | |||
| 102 | static SmallDenseMap<unsigned, StringRef, 8> F128Builtins{ | |||
| 103 | {Builtin::BI__builtin_printf, "__printfieee128"}, | |||
| 104 | {Builtin::BI__builtin_vsnprintf, "__vsnprintfieee128"}, | |||
| 105 | {Builtin::BI__builtin_vsprintf, "__vsprintfieee128"}, | |||
| 106 | {Builtin::BI__builtin_sprintf, "__sprintfieee128"}, | |||
| 107 | {Builtin::BI__builtin_snprintf, "__snprintfieee128"}, | |||
| 108 | {Builtin::BI__builtin_fprintf, "__fprintfieee128"}, | |||
| 109 | {Builtin::BI__builtin_nexttowardf128, "__nexttowardieee128"}, | |||
| 110 | }; | |||
| 111 | ||||
| 112 | // If the builtin has been declared explicitly with an assembler label, | |||
| 113 | // use the mangled name. This differs from the plain label on platforms | |||
| 114 | // that prefix labels. | |||
| 115 | if (FD->hasAttr<AsmLabelAttr>()) | |||
| 116 | Name = getMangledName(D); | |||
| 117 | else { | |||
| 118 | // TODO: This mutation should also be applied to other targets other than | |||
| 119 | // PPC, after backend supports IEEE 128-bit style libcalls. | |||
| 120 | if (getTriple().isPPC64() && | |||
| 121 | &getTarget().getLongDoubleFormat() == &llvm::APFloat::IEEEquad() && | |||
| 122 | F128Builtins.find(BuiltinID) != F128Builtins.end()) | |||
| 123 | Name = F128Builtins[BuiltinID]; | |||
| 124 | else | |||
| 125 | Name = Context.BuiltinInfo.getName(BuiltinID) + 10; | |||
| 126 | } | |||
| 127 | ||||
| 128 | llvm::FunctionType *Ty = | |||
| 129 | cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType())); | |||
| 130 | ||||
| 131 | return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false); | |||
| 132 | } | |||
| 133 | ||||
| 134 | /// Emit the conversions required to turn the given value into an | |||
| 135 | /// integer of the given size. | |||
| 136 | static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V, | |||
| 137 | QualType T, llvm::IntegerType *IntType) { | |||
| 138 | V = CGF.EmitToMemory(V, T); | |||
| 139 | ||||
| 140 | if (V->getType()->isPointerTy()) | |||
| 141 | return CGF.Builder.CreatePtrToInt(V, IntType); | |||
| 142 | ||||
| 143 | assert(V->getType() == IntType)(static_cast <bool> (V->getType() == IntType) ? void (0) : __assert_fail ("V->getType() == IntType", "clang/lib/CodeGen/CGBuiltin.cpp" , 143, __extension__ __PRETTY_FUNCTION__)); | |||
| 144 | return V; | |||
| 145 | } | |||
| 146 | ||||
| 147 | static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V, | |||
| 148 | QualType T, llvm::Type *ResultType) { | |||
| 149 | V = CGF.EmitFromMemory(V, T); | |||
| 150 | ||||
| 151 | if (ResultType->isPointerTy()) | |||
| 152 | return CGF.Builder.CreateIntToPtr(V, ResultType); | |||
| 153 | ||||
| 154 | assert(V->getType() == ResultType)(static_cast <bool> (V->getType() == ResultType) ? void (0) : __assert_fail ("V->getType() == ResultType", "clang/lib/CodeGen/CGBuiltin.cpp" , 154, __extension__ __PRETTY_FUNCTION__)); | |||
| 155 | return V; | |||
| 156 | } | |||
| 157 | ||||
| 158 | /// Utility to insert an atomic instruction based on Intrinsic::ID | |||
| 159 | /// and the expression node. | |||
| 160 | static Value *MakeBinaryAtomicValue( | |||
| 161 | CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E, | |||
| 162 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { | |||
| 163 | ||||
| 164 | QualType T = E->getType(); | |||
| 165 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 165, __extension__ __PRETTY_FUNCTION__ )); | |||
| 166 | assert(CGF.getContext().hasSameUnqualifiedType(T,(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 167, __extension__ __PRETTY_FUNCTION__ )) | |||
| 167 | E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 167, __extension__ __PRETTY_FUNCTION__ )); | |||
| 168 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 168, __extension__ __PRETTY_FUNCTION__ )); | |||
| 169 | ||||
| 170 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 171 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); | |||
| 172 | ||||
| 173 | llvm::IntegerType *IntType = | |||
| 174 | llvm::IntegerType::get(CGF.getLLVMContext(), | |||
| 175 | CGF.getContext().getTypeSize(T)); | |||
| 176 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); | |||
| 177 | ||||
| 178 | llvm::Value *Args[2]; | |||
| 179 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); | |||
| 180 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 181 | llvm::Type *ValueType = Args[1]->getType(); | |||
| 182 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); | |||
| 183 | ||||
| 184 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( | |||
| 185 | Kind, Args[0], Args[1], Ordering); | |||
| 186 | return EmitFromInt(CGF, Result, T, ValueType); | |||
| 187 | } | |||
| 188 | ||||
| 189 | static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) { | |||
| 190 | Value *Val = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 191 | Value *Address = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 192 | ||||
| 193 | // Convert the type of the pointer to a pointer to the stored type. | |||
| 194 | Val = CGF.EmitToMemory(Val, E->getArg(0)->getType()); | |||
| 195 | unsigned SrcAddrSpace = Address->getType()->getPointerAddressSpace(); | |||
| 196 | Value *BC = CGF.Builder.CreateBitCast( | |||
| 197 | Address, llvm::PointerType::get(Val->getType(), SrcAddrSpace), "cast"); | |||
| 198 | LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType()); | |||
| 199 | LV.setNontemporal(true); | |||
| 200 | CGF.EmitStoreOfScalar(Val, LV, false); | |||
| 201 | return nullptr; | |||
| 202 | } | |||
| 203 | ||||
| 204 | static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) { | |||
| 205 | Value *Address = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 206 | ||||
| 207 | LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType()); | |||
| 208 | LV.setNontemporal(true); | |||
| 209 | return CGF.EmitLoadOfScalar(LV, E->getExprLoc()); | |||
| 210 | } | |||
| 211 | ||||
| 212 | static RValue EmitBinaryAtomic(CodeGenFunction &CGF, | |||
| 213 | llvm::AtomicRMWInst::BinOp Kind, | |||
| 214 | const CallExpr *E) { | |||
| 215 | return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E)); | |||
| 216 | } | |||
| 217 | ||||
| 218 | /// Utility to insert an atomic instruction based Intrinsic::ID and | |||
| 219 | /// the expression node, where the return value is the result of the | |||
| 220 | /// operation. | |||
| 221 | static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF, | |||
| 222 | llvm::AtomicRMWInst::BinOp Kind, | |||
| 223 | const CallExpr *E, | |||
| 224 | Instruction::BinaryOps Op, | |||
| 225 | bool Invert = false) { | |||
| 226 | QualType T = E->getType(); | |||
| 227 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 227, __extension__ __PRETTY_FUNCTION__ )); | |||
| 228 | assert(CGF.getContext().hasSameUnqualifiedType(T,(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 229, __extension__ __PRETTY_FUNCTION__ )) | |||
| 229 | E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 229, __extension__ __PRETTY_FUNCTION__ )); | |||
| 230 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 230, __extension__ __PRETTY_FUNCTION__ )); | |||
| 231 | ||||
| 232 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 233 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); | |||
| 234 | ||||
| 235 | llvm::IntegerType *IntType = | |||
| 236 | llvm::IntegerType::get(CGF.getLLVMContext(), | |||
| 237 | CGF.getContext().getTypeSize(T)); | |||
| 238 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); | |||
| 239 | ||||
| 240 | llvm::Value *Args[2]; | |||
| 241 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 242 | llvm::Type *ValueType = Args[1]->getType(); | |||
| 243 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); | |||
| 244 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); | |||
| 245 | ||||
| 246 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( | |||
| 247 | Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent); | |||
| 248 | Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]); | |||
| 249 | if (Invert) | |||
| 250 | Result = | |||
| 251 | CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result, | |||
| 252 | llvm::ConstantInt::getAllOnesValue(IntType)); | |||
| 253 | Result = EmitFromInt(CGF, Result, T, ValueType); | |||
| 254 | return RValue::get(Result); | |||
| 255 | } | |||
| 256 | ||||
| 257 | /// Utility to insert an atomic cmpxchg instruction. | |||
| 258 | /// | |||
| 259 | /// @param CGF The current codegen function. | |||
| 260 | /// @param E Builtin call expression to convert to cmpxchg. | |||
| 261 | /// arg0 - address to operate on | |||
| 262 | /// arg1 - value to compare with | |||
| 263 | /// arg2 - new value | |||
| 264 | /// @param ReturnBool Specifies whether to return success flag of | |||
| 265 | /// cmpxchg result or the old value. | |||
| 266 | /// | |||
| 267 | /// @returns result of cmpxchg, according to ReturnBool | |||
| 268 | /// | |||
| 269 | /// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics | |||
| 270 | /// invoke the function EmitAtomicCmpXchgForMSIntrin. | |||
| 271 | static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E, | |||
| 272 | bool ReturnBool) { | |||
| 273 | QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType(); | |||
| 274 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 275 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); | |||
| 276 | ||||
| 277 | llvm::IntegerType *IntType = llvm::IntegerType::get( | |||
| 278 | CGF.getLLVMContext(), CGF.getContext().getTypeSize(T)); | |||
| 279 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); | |||
| 280 | ||||
| 281 | Value *Args[3]; | |||
| 282 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); | |||
| 283 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 284 | llvm::Type *ValueType = Args[1]->getType(); | |||
| 285 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); | |||
| 286 | Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType); | |||
| 287 | ||||
| 288 | Value *Pair = CGF.Builder.CreateAtomicCmpXchg( | |||
| 289 | Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent, | |||
| 290 | llvm::AtomicOrdering::SequentiallyConsistent); | |||
| 291 | if (ReturnBool) | |||
| 292 | // Extract boolean success flag and zext it to int. | |||
| 293 | return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1), | |||
| 294 | CGF.ConvertType(E->getType())); | |||
| 295 | else | |||
| 296 | // Extract old value and emit it using the same type as compare value. | |||
| 297 | return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T, | |||
| 298 | ValueType); | |||
| 299 | } | |||
| 300 | ||||
| 301 | /// This function should be invoked to emit atomic cmpxchg for Microsoft's | |||
| 302 | /// _InterlockedCompareExchange* intrinsics which have the following signature: | |||
| 303 | /// T _InterlockedCompareExchange(T volatile *Destination, | |||
| 304 | /// T Exchange, | |||
| 305 | /// T Comparand); | |||
| 306 | /// | |||
| 307 | /// Whereas the llvm 'cmpxchg' instruction has the following syntax: | |||
| 308 | /// cmpxchg *Destination, Comparand, Exchange. | |||
| 309 | /// So we need to swap Comparand and Exchange when invoking | |||
| 310 | /// CreateAtomicCmpXchg. That is the reason we could not use the above utility | |||
| 311 | /// function MakeAtomicCmpXchgValue since it expects the arguments to be | |||
| 312 | /// already swapped. | |||
| 313 | ||||
| 314 | static | |||
| 315 | Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E, | |||
| 316 | AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) { | |||
| 317 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 317, __extension__ __PRETTY_FUNCTION__ )); | |||
| 318 | assert(CGF.getContext().hasSameUnqualifiedType((static_cast <bool> (CGF.getContext().hasSameUnqualifiedType ( E->getType(), E->getArg(0)->getType()->getPointeeType ())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 319, __extension__ __PRETTY_FUNCTION__ )) | |||
| 319 | E->getType(), E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType ( E->getType(), E->getArg(0)->getType()->getPointeeType ())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 319, __extension__ __PRETTY_FUNCTION__ )); | |||
| 320 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 321, __extension__ __PRETTY_FUNCTION__ )) | |||
| 321 | E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 321, __extension__ __PRETTY_FUNCTION__ )); | |||
| 322 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(2)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 323, __extension__ __PRETTY_FUNCTION__ )) | |||
| 323 | E->getArg(2)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(2)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 323, __extension__ __PRETTY_FUNCTION__ )); | |||
| 324 | ||||
| 325 | auto *Destination = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 326 | auto *Comparand = CGF.EmitScalarExpr(E->getArg(2)); | |||
| 327 | auto *Exchange = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 328 | ||||
| 329 | // For Release ordering, the failure ordering should be Monotonic. | |||
| 330 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ? | |||
| 331 | AtomicOrdering::Monotonic : | |||
| 332 | SuccessOrdering; | |||
| 333 | ||||
| 334 | // The atomic instruction is marked volatile for consistency with MSVC. This | |||
| 335 | // blocks the few atomics optimizations that LLVM has. If we want to optimize | |||
| 336 | // _Interlocked* operations in the future, we will have to remove the volatile | |||
| 337 | // marker. | |||
| 338 | auto *Result = CGF.Builder.CreateAtomicCmpXchg( | |||
| 339 | Destination, Comparand, Exchange, | |||
| 340 | SuccessOrdering, FailureOrdering); | |||
| 341 | Result->setVolatile(true); | |||
| 342 | return CGF.Builder.CreateExtractValue(Result, 0); | |||
| 343 | } | |||
| 344 | ||||
| 345 | // 64-bit Microsoft platforms support 128 bit cmpxchg operations. They are | |||
| 346 | // prototyped like this: | |||
| 347 | // | |||
| 348 | // unsigned char _InterlockedCompareExchange128...( | |||
| 349 | // __int64 volatile * _Destination, | |||
| 350 | // __int64 _ExchangeHigh, | |||
| 351 | // __int64 _ExchangeLow, | |||
| 352 | // __int64 * _ComparandResult); | |||
| 353 | static Value *EmitAtomicCmpXchg128ForMSIntrin(CodeGenFunction &CGF, | |||
| 354 | const CallExpr *E, | |||
| 355 | AtomicOrdering SuccessOrdering) { | |||
| 356 | assert(E->getNumArgs() == 4)(static_cast <bool> (E->getNumArgs() == 4) ? void (0 ) : __assert_fail ("E->getNumArgs() == 4", "clang/lib/CodeGen/CGBuiltin.cpp" , 356, __extension__ __PRETTY_FUNCTION__)); | |||
| 357 | llvm::Value *Destination = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 358 | llvm::Value *ExchangeHigh = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 359 | llvm::Value *ExchangeLow = CGF.EmitScalarExpr(E->getArg(2)); | |||
| 360 | llvm::Value *ComparandPtr = CGF.EmitScalarExpr(E->getArg(3)); | |||
| 361 | ||||
| 362 | assert(Destination->getType()->isPointerTy())(static_cast <bool> (Destination->getType()->isPointerTy ()) ? void (0) : __assert_fail ("Destination->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 362, __extension__ __PRETTY_FUNCTION__ )); | |||
| 363 | assert(!ExchangeHigh->getType()->isPointerTy())(static_cast <bool> (!ExchangeHigh->getType()->isPointerTy ()) ? void (0) : __assert_fail ("!ExchangeHigh->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 363, __extension__ __PRETTY_FUNCTION__ )); | |||
| 364 | assert(!ExchangeLow->getType()->isPointerTy())(static_cast <bool> (!ExchangeLow->getType()->isPointerTy ()) ? void (0) : __assert_fail ("!ExchangeLow->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 364, __extension__ __PRETTY_FUNCTION__ )); | |||
| 365 | assert(ComparandPtr->getType()->isPointerTy())(static_cast <bool> (ComparandPtr->getType()->isPointerTy ()) ? void (0) : __assert_fail ("ComparandPtr->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 365, __extension__ __PRETTY_FUNCTION__ )); | |||
| 366 | ||||
| 367 | // For Release ordering, the failure ordering should be Monotonic. | |||
| 368 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release | |||
| 369 | ? AtomicOrdering::Monotonic | |||
| 370 | : SuccessOrdering; | |||
| 371 | ||||
| 372 | // Convert to i128 pointers and values. | |||
| 373 | llvm::Type *Int128Ty = llvm::IntegerType::get(CGF.getLLVMContext(), 128); | |||
| 374 | llvm::Type *Int128PtrTy = Int128Ty->getPointerTo(); | |||
| 375 | Destination = CGF.Builder.CreateBitCast(Destination, Int128PtrTy); | |||
| 376 | Address ComparandResult(CGF.Builder.CreateBitCast(ComparandPtr, Int128PtrTy), | |||
| 377 | Int128Ty, CGF.getContext().toCharUnitsFromBits(128)); | |||
| 378 | ||||
| 379 | // (((i128)hi) << 64) | ((i128)lo) | |||
| 380 | ExchangeHigh = CGF.Builder.CreateZExt(ExchangeHigh, Int128Ty); | |||
| 381 | ExchangeLow = CGF.Builder.CreateZExt(ExchangeLow, Int128Ty); | |||
| 382 | ExchangeHigh = | |||
| 383 | CGF.Builder.CreateShl(ExchangeHigh, llvm::ConstantInt::get(Int128Ty, 64)); | |||
| 384 | llvm::Value *Exchange = CGF.Builder.CreateOr(ExchangeHigh, ExchangeLow); | |||
| 385 | ||||
| 386 | // Load the comparand for the instruction. | |||
| 387 | llvm::Value *Comparand = CGF.Builder.CreateLoad(ComparandResult); | |||
| 388 | ||||
| 389 | auto *CXI = CGF.Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, | |||
| 390 | SuccessOrdering, FailureOrdering); | |||
| 391 | ||||
| 392 | // The atomic instruction is marked volatile for consistency with MSVC. This | |||
| 393 | // blocks the few atomics optimizations that LLVM has. If we want to optimize | |||
| 394 | // _Interlocked* operations in the future, we will have to remove the volatile | |||
| 395 | // marker. | |||
| 396 | CXI->setVolatile(true); | |||
| 397 | ||||
| 398 | // Store the result as an outparameter. | |||
| 399 | CGF.Builder.CreateStore(CGF.Builder.CreateExtractValue(CXI, 0), | |||
| 400 | ComparandResult); | |||
| 401 | ||||
| 402 | // Get the success boolean and zero extend it to i8. | |||
| 403 | Value *Success = CGF.Builder.CreateExtractValue(CXI, 1); | |||
| 404 | return CGF.Builder.CreateZExt(Success, CGF.Int8Ty); | |||
| 405 | } | |||
| 406 | ||||
| 407 | static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E, | |||
| 408 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { | |||
| 409 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 409, __extension__ __PRETTY_FUNCTION__ )); | |||
| 410 | ||||
| 411 | auto *IntTy = CGF.ConvertType(E->getType()); | |||
| 412 | auto *Result = CGF.Builder.CreateAtomicRMW( | |||
| 413 | AtomicRMWInst::Add, | |||
| 414 | CGF.EmitScalarExpr(E->getArg(0)), | |||
| 415 | ConstantInt::get(IntTy, 1), | |||
| 416 | Ordering); | |||
| 417 | return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1)); | |||
| 418 | } | |||
| 419 | ||||
| 420 | static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E, | |||
| 421 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { | |||
| 422 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 422, __extension__ __PRETTY_FUNCTION__ )); | |||
| 423 | ||||
| 424 | auto *IntTy = CGF.ConvertType(E->getType()); | |||
| 425 | auto *Result = CGF.Builder.CreateAtomicRMW( | |||
| 426 | AtomicRMWInst::Sub, | |||
| 427 | CGF.EmitScalarExpr(E->getArg(0)), | |||
| 428 | ConstantInt::get(IntTy, 1), | |||
| 429 | Ordering); | |||
| 430 | return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1)); | |||
| 431 | } | |||
| 432 | ||||
| 433 | // Build a plain volatile load. | |||
| 434 | static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) { | |||
| 435 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 436 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); | |||
| 437 | CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy); | |||
| 438 | llvm::Type *ITy = | |||
| 439 | llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8); | |||
| 440 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); | |||
| 441 | llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(ITy, Ptr, LoadSize); | |||
| 442 | Load->setVolatile(true); | |||
| 443 | return Load; | |||
| 444 | } | |||
| 445 | ||||
| 446 | // Build a plain volatile store. | |||
| 447 | static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) { | |||
| 448 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 449 | Value *Value = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 450 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); | |||
| 451 | CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy); | |||
| 452 | llvm::Type *ITy = | |||
| 453 | llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8); | |||
| 454 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); | |||
| 455 | llvm::StoreInst *Store = | |||
| 456 | CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize); | |||
| 457 | Store->setVolatile(true); | |||
| 458 | return Store; | |||
| 459 | } | |||
| 460 | ||||
| 461 | // Emit a simple mangled intrinsic that has 1 argument and a return type | |||
| 462 | // matching the argument type. Depending on mode, this may be a constrained | |||
| 463 | // floating-point intrinsic. | |||
| 464 | static Value *emitUnaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, | |||
| 465 | const CallExpr *E, unsigned IntrinsicID, | |||
| 466 | unsigned ConstrainedIntrinsicID) { | |||
| 467 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 468 | ||||
| 469 | if (CGF.Builder.getIsFPConstrained()) { | |||
| 470 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); | |||
| 471 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); | |||
| 472 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0 }); | |||
| 473 | } else { | |||
| 474 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
| 475 | return CGF.Builder.CreateCall(F, Src0); | |||
| 476 | } | |||
| 477 | } | |||
| 478 | ||||
| 479 | // Emit an intrinsic that has 2 operands of the same type as its result. | |||
| 480 | // Depending on mode, this may be a constrained floating-point intrinsic. | |||
| 481 | static Value *emitBinaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, | |||
| 482 | const CallExpr *E, unsigned IntrinsicID, | |||
| 483 | unsigned ConstrainedIntrinsicID) { | |||
| 484 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 485 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 486 | ||||
| 487 | if (CGF.Builder.getIsFPConstrained()) { | |||
| 488 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); | |||
| 489 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); | |||
| 490 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1 }); | |||
| 491 | } else { | |||
| 492 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
| 493 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); | |||
| 494 | } | |||
| 495 | } | |||
| 496 | ||||
| 497 | // Emit an intrinsic that has 3 operands of the same type as its result. | |||
| 498 | // Depending on mode, this may be a constrained floating-point intrinsic. | |||
| 499 | static Value *emitTernaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, | |||
| 500 | const CallExpr *E, unsigned IntrinsicID, | |||
| 501 | unsigned ConstrainedIntrinsicID) { | |||
| 502 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 503 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 504 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); | |||
| 505 | ||||
| 506 | if (CGF.Builder.getIsFPConstrained()) { | |||
| 507 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); | |||
| 508 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); | |||
| 509 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1, Src2 }); | |||
| 510 | } else { | |||
| 511 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
| 512 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
| 513 | } | |||
| 514 | } | |||
| 515 | ||||
| 516 | // Emit an intrinsic where all operands are of the same type as the result. | |||
| 517 | // Depending on mode, this may be a constrained floating-point intrinsic. | |||
| 518 | static Value *emitCallMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, | |||
| 519 | unsigned IntrinsicID, | |||
| 520 | unsigned ConstrainedIntrinsicID, | |||
| 521 | llvm::Type *Ty, | |||
| 522 | ArrayRef<Value *> Args) { | |||
| 523 | Function *F; | |||
| 524 | if (CGF.Builder.getIsFPConstrained()) | |||
| 525 | F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Ty); | |||
| 526 | else | |||
| 527 | F = CGF.CGM.getIntrinsic(IntrinsicID, Ty); | |||
| 528 | ||||
| 529 | if (CGF.Builder.getIsFPConstrained()) | |||
| 530 | return CGF.Builder.CreateConstrainedFPCall(F, Args); | |||
| 531 | else | |||
| 532 | return CGF.Builder.CreateCall(F, Args); | |||
| 533 | } | |||
| 534 | ||||
| 535 | // Emit a simple mangled intrinsic that has 1 argument and a return type | |||
| 536 | // matching the argument type. | |||
| 537 | static Value *emitUnaryBuiltin(CodeGenFunction &CGF, const CallExpr *E, | |||
| 538 | unsigned IntrinsicID, | |||
| 539 | llvm::StringRef Name = "") { | |||
| 540 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 541 | ||||
| 542 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
| 543 | return CGF.Builder.CreateCall(F, Src0, Name); | |||
| 544 | } | |||
| 545 | ||||
| 546 | // Emit an intrinsic that has 2 operands of the same type as its result. | |||
| 547 | static Value *emitBinaryBuiltin(CodeGenFunction &CGF, | |||
| 548 | const CallExpr *E, | |||
| 549 | unsigned IntrinsicID) { | |||
| 550 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 551 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 552 | ||||
| 553 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
| 554 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); | |||
| 555 | } | |||
| 556 | ||||
| 557 | // Emit an intrinsic that has 3 operands of the same type as its result. | |||
| 558 | static Value *emitTernaryBuiltin(CodeGenFunction &CGF, | |||
| 559 | const CallExpr *E, | |||
| 560 | unsigned IntrinsicID) { | |||
| 561 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 562 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 563 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); | |||
| 564 | ||||
| 565 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
| 566 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
| 567 | } | |||
| 568 | ||||
| 569 | // Emit an intrinsic that has 1 float or double operand, and 1 integer. | |||
| 570 | static Value *emitFPIntBuiltin(CodeGenFunction &CGF, | |||
| 571 | const CallExpr *E, | |||
| 572 | unsigned IntrinsicID) { | |||
| 573 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 574 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 575 | ||||
| 576 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); | |||
| 577 | return CGF.Builder.CreateCall(F, {Src0, Src1}); | |||
| 578 | } | |||
| 579 | ||||
| 580 | // Emit an intrinsic that has overloaded integer result and fp operand. | |||
| 581 | static Value * | |||
| 582 | emitMaybeConstrainedFPToIntRoundBuiltin(CodeGenFunction &CGF, const CallExpr *E, | |||
| 583 | unsigned IntrinsicID, | |||
| 584 | unsigned ConstrainedIntrinsicID) { | |||
| 585 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); | |||
| 586 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 587 | ||||
| 588 | if (CGF.Builder.getIsFPConstrained()) { | |||
| 589 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); | |||
| 590 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, | |||
| 591 | {ResultType, Src0->getType()}); | |||
| 592 | return CGF.Builder.CreateConstrainedFPCall(F, {Src0}); | |||
| 593 | } else { | |||
| 594 | Function *F = | |||
| 595 | CGF.CGM.getIntrinsic(IntrinsicID, {ResultType, Src0->getType()}); | |||
| 596 | return CGF.Builder.CreateCall(F, Src0); | |||
| 597 | } | |||
| 598 | } | |||
| 599 | ||||
| 600 | /// EmitFAbs - Emit a call to @llvm.fabs(). | |||
| 601 | static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) { | |||
| 602 | Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType()); | |||
| 603 | llvm::CallInst *Call = CGF.Builder.CreateCall(F, V); | |||
| 604 | Call->setDoesNotAccessMemory(); | |||
| 605 | return Call; | |||
| 606 | } | |||
| 607 | ||||
| 608 | /// Emit the computation of the sign bit for a floating point value. Returns | |||
| 609 | /// the i1 sign bit value. | |||
| 610 | static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) { | |||
| 611 | LLVMContext &C = CGF.CGM.getLLVMContext(); | |||
| 612 | ||||
| 613 | llvm::Type *Ty = V->getType(); | |||
| 614 | int Width = Ty->getPrimitiveSizeInBits(); | |||
| 615 | llvm::Type *IntTy = llvm::IntegerType::get(C, Width); | |||
| 616 | V = CGF.Builder.CreateBitCast(V, IntTy); | |||
| 617 | if (Ty->isPPC_FP128Ty()) { | |||
| 618 | // We want the sign bit of the higher-order double. The bitcast we just | |||
| 619 | // did works as if the double-double was stored to memory and then | |||
| 620 | // read as an i128. The "store" will put the higher-order double in the | |||
| 621 | // lower address in both little- and big-Endian modes, but the "load" | |||
| 622 | // will treat those bits as a different part of the i128: the low bits in | |||
| 623 | // little-Endian, the high bits in big-Endian. Therefore, on big-Endian | |||
| 624 | // we need to shift the high bits down to the low before truncating. | |||
| 625 | Width >>= 1; | |||
| 626 | if (CGF.getTarget().isBigEndian()) { | |||
| 627 | Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width); | |||
| 628 | V = CGF.Builder.CreateLShr(V, ShiftCst); | |||
| 629 | } | |||
| 630 | // We are truncating value in order to extract the higher-order | |||
| 631 | // double, which we will be using to extract the sign from. | |||
| 632 | IntTy = llvm::IntegerType::get(C, Width); | |||
| 633 | V = CGF.Builder.CreateTrunc(V, IntTy); | |||
| 634 | } | |||
| 635 | Value *Zero = llvm::Constant::getNullValue(IntTy); | |||
| 636 | return CGF.Builder.CreateICmpSLT(V, Zero); | |||
| 637 | } | |||
| 638 | ||||
| 639 | static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD, | |||
| 640 | const CallExpr *E, llvm::Constant *calleeValue) { | |||
| 641 | CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD)); | |||
| 642 | return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot()); | |||
| 643 | } | |||
| 644 | ||||
| 645 | /// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.* | |||
| 646 | /// depending on IntrinsicID. | |||
| 647 | /// | |||
| 648 | /// \arg CGF The current codegen function. | |||
| 649 | /// \arg IntrinsicID The ID for the Intrinsic we wish to generate. | |||
| 650 | /// \arg X The first argument to the llvm.*.with.overflow.*. | |||
| 651 | /// \arg Y The second argument to the llvm.*.with.overflow.*. | |||
| 652 | /// \arg Carry The carry returned by the llvm.*.with.overflow.*. | |||
| 653 | /// \returns The result (i.e. sum/product) returned by the intrinsic. | |||
| 654 | static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF, | |||
| 655 | const llvm::Intrinsic::ID IntrinsicID, | |||
| 656 | llvm::Value *X, llvm::Value *Y, | |||
| 657 | llvm::Value *&Carry) { | |||
| 658 | // Make sure we have integers of the same width. | |||
| 659 | assert(X->getType() == Y->getType() &&(static_cast <bool> (X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? void (0) : __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 661, __extension__ __PRETTY_FUNCTION__ )) | |||
| 660 | "Arguments must be the same type. (Did you forget to make sure both "(static_cast <bool> (X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? void (0) : __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 661, __extension__ __PRETTY_FUNCTION__ )) | |||
| 661 | "arguments have the same integer width?)")(static_cast <bool> (X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? void (0) : __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 661, __extension__ __PRETTY_FUNCTION__ )); | |||
| 662 | ||||
| 663 | Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType()); | |||
| 664 | llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y}); | |||
| 665 | Carry = CGF.Builder.CreateExtractValue(Tmp, 1); | |||
| 666 | return CGF.Builder.CreateExtractValue(Tmp, 0); | |||
| 667 | } | |||
| 668 | ||||
| 669 | static Value *emitRangedBuiltin(CodeGenFunction &CGF, | |||
| 670 | unsigned IntrinsicID, | |||
| 671 | int low, int high) { | |||
| 672 | llvm::MDBuilder MDHelper(CGF.getLLVMContext()); | |||
| 673 | llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high)); | |||
| 674 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {}); | |||
| 675 | llvm::Instruction *Call = CGF.Builder.CreateCall(F); | |||
| 676 | Call->setMetadata(llvm::LLVMContext::MD_range, RNode); | |||
| 677 | return Call; | |||
| 678 | } | |||
| 679 | ||||
| 680 | namespace { | |||
| 681 | struct WidthAndSignedness { | |||
| 682 | unsigned Width; | |||
| 683 | bool Signed; | |||
| 684 | }; | |||
| 685 | } | |||
| 686 | ||||
| 687 | static WidthAndSignedness | |||
| 688 | getIntegerWidthAndSignedness(const clang::ASTContext &context, | |||
| 689 | const clang::QualType Type) { | |||
| 690 | assert(Type->isIntegerType() && "Given type is not an integer.")(static_cast <bool> (Type->isIntegerType() && "Given type is not an integer.") ? void (0) : __assert_fail ( "Type->isIntegerType() && \"Given type is not an integer.\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 690, __extension__ __PRETTY_FUNCTION__ )); | |||
| 691 | unsigned Width = Type->isBooleanType() ? 1 | |||
| 692 | : Type->isBitIntType() ? context.getIntWidth(Type) | |||
| 693 | : context.getTypeInfo(Type).Width; | |||
| 694 | bool Signed = Type->isSignedIntegerType(); | |||
| 695 | return {Width, Signed}; | |||
| 696 | } | |||
| 697 | ||||
| 698 | // Given one or more integer types, this function produces an integer type that | |||
| 699 | // encompasses them: any value in one of the given types could be expressed in | |||
| 700 | // the encompassing type. | |||
| 701 | static struct WidthAndSignedness | |||
| 702 | EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) { | |||
| 703 | assert(Types.size() > 0 && "Empty list of types.")(static_cast <bool> (Types.size() > 0 && "Empty list of types." ) ? void (0) : __assert_fail ("Types.size() > 0 && \"Empty list of types.\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 703, __extension__ __PRETTY_FUNCTION__ )); | |||
| 704 | ||||
| 705 | // If any of the given types is signed, we must return a signed type. | |||
| 706 | bool Signed = false; | |||
| 707 | for (const auto &Type : Types) { | |||
| 708 | Signed |= Type.Signed; | |||
| 709 | } | |||
| 710 | ||||
| 711 | // The encompassing type must have a width greater than or equal to the width | |||
| 712 | // of the specified types. Additionally, if the encompassing type is signed, | |||
| 713 | // its width must be strictly greater than the width of any unsigned types | |||
| 714 | // given. | |||
| 715 | unsigned Width = 0; | |||
| 716 | for (const auto &Type : Types) { | |||
| 717 | unsigned MinWidth = Type.Width + (Signed && !Type.Signed); | |||
| 718 | if (Width < MinWidth) { | |||
| 719 | Width = MinWidth; | |||
| 720 | } | |||
| 721 | } | |||
| 722 | ||||
| 723 | return {Width, Signed}; | |||
| 724 | } | |||
| 725 | ||||
| 726 | Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) { | |||
| 727 | llvm::Type *DestType = Int8PtrTy; | |||
| 728 | if (ArgValue->getType() != DestType) | |||
| 729 | ArgValue = | |||
| 730 | Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data()); | |||
| 731 | ||||
| 732 | Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend; | |||
| 733 | return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue); | |||
| 734 | } | |||
| 735 | ||||
| 736 | /// Checks if using the result of __builtin_object_size(p, @p From) in place of | |||
| 737 | /// __builtin_object_size(p, @p To) is correct | |||
| 738 | static bool areBOSTypesCompatible(int From, int To) { | |||
| 739 | // Note: Our __builtin_object_size implementation currently treats Type=0 and | |||
| 740 | // Type=2 identically. Encoding this implementation detail here may make | |||
| 741 | // improving __builtin_object_size difficult in the future, so it's omitted. | |||
| 742 | return From == To || (From == 0 && To == 1) || (From == 3 && To == 2); | |||
| 743 | } | |||
| 744 | ||||
| 745 | static llvm::Value * | |||
| 746 | getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) { | |||
| 747 | return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true); | |||
| 748 | } | |||
| 749 | ||||
| 750 | llvm::Value * | |||
| 751 | CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type, | |||
| 752 | llvm::IntegerType *ResType, | |||
| 753 | llvm::Value *EmittedE, | |||
| 754 | bool IsDynamic) { | |||
| 755 | uint64_t ObjectSize; | |||
| 756 | if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type)) | |||
| 757 | return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic); | |||
| 758 | return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true); | |||
| 759 | } | |||
| 760 | ||||
| 761 | /// Returns a Value corresponding to the size of the given expression. | |||
| 762 | /// This Value may be either of the following: | |||
| 763 | /// - A llvm::Argument (if E is a param with the pass_object_size attribute on | |||
| 764 | /// it) | |||
| 765 | /// - A call to the @llvm.objectsize intrinsic | |||
| 766 | /// | |||
| 767 | /// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null | |||
| 768 | /// and we wouldn't otherwise try to reference a pass_object_size parameter, | |||
| 769 | /// we'll call @llvm.objectsize on EmittedE, rather than emitting E. | |||
| 770 | llvm::Value * | |||
| 771 | CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type, | |||
| 772 | llvm::IntegerType *ResType, | |||
| 773 | llvm::Value *EmittedE, bool IsDynamic) { | |||
| 774 | // We need to reference an argument if the pointer is a parameter with the | |||
| 775 | // pass_object_size attribute. | |||
| 776 | if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) { | |||
| 777 | auto *Param = dyn_cast<ParmVarDecl>(D->getDecl()); | |||
| 778 | auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>(); | |||
| 779 | if (Param != nullptr && PS != nullptr && | |||
| 780 | areBOSTypesCompatible(PS->getType(), Type)) { | |||
| 781 | auto Iter = SizeArguments.find(Param); | |||
| 782 | assert(Iter != SizeArguments.end())(static_cast <bool> (Iter != SizeArguments.end()) ? void (0) : __assert_fail ("Iter != SizeArguments.end()", "clang/lib/CodeGen/CGBuiltin.cpp" , 782, __extension__ __PRETTY_FUNCTION__)); | |||
| 783 | ||||
| 784 | const ImplicitParamDecl *D = Iter->second; | |||
| 785 | auto DIter = LocalDeclMap.find(D); | |||
| 786 | assert(DIter != LocalDeclMap.end())(static_cast <bool> (DIter != LocalDeclMap.end()) ? void (0) : __assert_fail ("DIter != LocalDeclMap.end()", "clang/lib/CodeGen/CGBuiltin.cpp" , 786, __extension__ __PRETTY_FUNCTION__)); | |||
| 787 | ||||
| 788 | return EmitLoadOfScalar(DIter->second, /*Volatile=*/false, | |||
| 789 | getContext().getSizeType(), E->getBeginLoc()); | |||
| 790 | } | |||
| 791 | } | |||
| 792 | ||||
| 793 | // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't | |||
| 794 | // evaluate E for side-effects. In either case, we shouldn't lower to | |||
| 795 | // @llvm.objectsize. | |||
| 796 | if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext()))) | |||
| 797 | return getDefaultBuiltinObjectSizeResult(Type, ResType); | |||
| 798 | ||||
| 799 | Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E); | |||
| 800 | assert(Ptr->getType()->isPointerTy() &&(static_cast <bool> (Ptr->getType()->isPointerTy( ) && "Non-pointer passed to __builtin_object_size?") ? void (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 801, __extension__ __PRETTY_FUNCTION__ )) | |||
| 801 | "Non-pointer passed to __builtin_object_size?")(static_cast <bool> (Ptr->getType()->isPointerTy( ) && "Non-pointer passed to __builtin_object_size?") ? void (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 801, __extension__ __PRETTY_FUNCTION__ )); | |||
| 802 | ||||
| 803 | Function *F = | |||
| 804 | CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()}); | |||
| 805 | ||||
| 806 | // LLVM only supports 0 and 2, make sure that we pass along that as a boolean. | |||
| 807 | Value *Min = Builder.getInt1((Type & 2) != 0); | |||
| 808 | // For GCC compatibility, __builtin_object_size treat NULL as unknown size. | |||
| 809 | Value *NullIsUnknown = Builder.getTrue(); | |||
| 810 | Value *Dynamic = Builder.getInt1(IsDynamic); | |||
| 811 | return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic}); | |||
| 812 | } | |||
| 813 | ||||
| 814 | namespace { | |||
| 815 | /// A struct to generically describe a bit test intrinsic. | |||
| 816 | struct BitTest { | |||
| 817 | enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set }; | |||
| 818 | enum InterlockingKind : uint8_t { | |||
| 819 | Unlocked, | |||
| 820 | Sequential, | |||
| 821 | Acquire, | |||
| 822 | Release, | |||
| 823 | NoFence | |||
| 824 | }; | |||
| 825 | ||||
| 826 | ActionKind Action; | |||
| 827 | InterlockingKind Interlocking; | |||
| 828 | bool Is64Bit; | |||
| 829 | ||||
| 830 | static BitTest decodeBitTestBuiltin(unsigned BuiltinID); | |||
| 831 | }; | |||
| 832 | } // namespace | |||
| 833 | ||||
| 834 | BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) { | |||
| 835 | switch (BuiltinID) { | |||
| 836 | // Main portable variants. | |||
| 837 | case Builtin::BI_bittest: | |||
| 838 | return {TestOnly, Unlocked, false}; | |||
| 839 | case Builtin::BI_bittestandcomplement: | |||
| 840 | return {Complement, Unlocked, false}; | |||
| 841 | case Builtin::BI_bittestandreset: | |||
| 842 | return {Reset, Unlocked, false}; | |||
| 843 | case Builtin::BI_bittestandset: | |||
| 844 | return {Set, Unlocked, false}; | |||
| 845 | case Builtin::BI_interlockedbittestandreset: | |||
| 846 | return {Reset, Sequential, false}; | |||
| 847 | case Builtin::BI_interlockedbittestandset: | |||
| 848 | return {Set, Sequential, false}; | |||
| 849 | ||||
| 850 | // X86-specific 64-bit variants. | |||
| 851 | case Builtin::BI_bittest64: | |||
| 852 | return {TestOnly, Unlocked, true}; | |||
| 853 | case Builtin::BI_bittestandcomplement64: | |||
| 854 | return {Complement, Unlocked, true}; | |||
| 855 | case Builtin::BI_bittestandreset64: | |||
| 856 | return {Reset, Unlocked, true}; | |||
| 857 | case Builtin::BI_bittestandset64: | |||
| 858 | return {Set, Unlocked, true}; | |||
| 859 | case Builtin::BI_interlockedbittestandreset64: | |||
| 860 | return {Reset, Sequential, true}; | |||
| 861 | case Builtin::BI_interlockedbittestandset64: | |||
| 862 | return {Set, Sequential, true}; | |||
| 863 | ||||
| 864 | // ARM/AArch64-specific ordering variants. | |||
| 865 | case Builtin::BI_interlockedbittestandset_acq: | |||
| 866 | return {Set, Acquire, false}; | |||
| 867 | case Builtin::BI_interlockedbittestandset_rel: | |||
| 868 | return {Set, Release, false}; | |||
| 869 | case Builtin::BI_interlockedbittestandset_nf: | |||
| 870 | return {Set, NoFence, false}; | |||
| 871 | case Builtin::BI_interlockedbittestandreset_acq: | |||
| 872 | return {Reset, Acquire, false}; | |||
| 873 | case Builtin::BI_interlockedbittestandreset_rel: | |||
| 874 | return {Reset, Release, false}; | |||
| 875 | case Builtin::BI_interlockedbittestandreset_nf: | |||
| 876 | return {Reset, NoFence, false}; | |||
| 877 | } | |||
| 878 | llvm_unreachable("expected only bittest intrinsics")::llvm::llvm_unreachable_internal("expected only bittest intrinsics" , "clang/lib/CodeGen/CGBuiltin.cpp", 878); | |||
| 879 | } | |||
| 880 | ||||
| 881 | static char bitActionToX86BTCode(BitTest::ActionKind A) { | |||
| 882 | switch (A) { | |||
| 883 | case BitTest::TestOnly: return '\0'; | |||
| 884 | case BitTest::Complement: return 'c'; | |||
| 885 | case BitTest::Reset: return 'r'; | |||
| 886 | case BitTest::Set: return 's'; | |||
| 887 | } | |||
| 888 | llvm_unreachable("invalid action")::llvm::llvm_unreachable_internal("invalid action", "clang/lib/CodeGen/CGBuiltin.cpp" , 888); | |||
| 889 | } | |||
| 890 | ||||
| 891 | static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF, | |||
| 892 | BitTest BT, | |||
| 893 | const CallExpr *E, Value *BitBase, | |||
| 894 | Value *BitPos) { | |||
| 895 | char Action = bitActionToX86BTCode(BT.Action); | |||
| 896 | char SizeSuffix = BT.Is64Bit ? 'q' : 'l'; | |||
| 897 | ||||
| 898 | // Build the assembly. | |||
| 899 | SmallString<64> Asm; | |||
| 900 | raw_svector_ostream AsmOS(Asm); | |||
| 901 | if (BT.Interlocking != BitTest::Unlocked) | |||
| 902 | AsmOS << "lock "; | |||
| 903 | AsmOS << "bt"; | |||
| 904 | if (Action) | |||
| 905 | AsmOS << Action; | |||
| 906 | AsmOS << SizeSuffix << " $2, ($1)"; | |||
| 907 | ||||
| 908 | // Build the constraints. FIXME: We should support immediates when possible. | |||
| 909 | std::string Constraints = "={@ccc},r,r,~{cc},~{memory}"; | |||
| 910 | std::string MachineClobbers = CGF.getTarget().getClobbers(); | |||
| 911 | if (!MachineClobbers.empty()) { | |||
| 912 | Constraints += ','; | |||
| 913 | Constraints += MachineClobbers; | |||
| 914 | } | |||
| 915 | llvm::IntegerType *IntType = llvm::IntegerType::get( | |||
| 916 | CGF.getLLVMContext(), | |||
| 917 | CGF.getContext().getTypeSize(E->getArg(1)->getType())); | |||
| 918 | llvm::Type *IntPtrType = IntType->getPointerTo(); | |||
| 919 | llvm::FunctionType *FTy = | |||
| 920 | llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false); | |||
| 921 | ||||
| 922 | llvm::InlineAsm *IA = | |||
| 923 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); | |||
| 924 | return CGF.Builder.CreateCall(IA, {BitBase, BitPos}); | |||
| 925 | } | |||
| 926 | ||||
| 927 | static llvm::AtomicOrdering | |||
| 928 | getBitTestAtomicOrdering(BitTest::InterlockingKind I) { | |||
| 929 | switch (I) { | |||
| 930 | case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic; | |||
| 931 | case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent; | |||
| 932 | case BitTest::Acquire: return llvm::AtomicOrdering::Acquire; | |||
| 933 | case BitTest::Release: return llvm::AtomicOrdering::Release; | |||
| 934 | case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic; | |||
| 935 | } | |||
| 936 | llvm_unreachable("invalid interlocking")::llvm::llvm_unreachable_internal("invalid interlocking", "clang/lib/CodeGen/CGBuiltin.cpp" , 936); | |||
| 937 | } | |||
| 938 | ||||
| 939 | /// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of | |||
| 940 | /// bits and a bit position and read and optionally modify the bit at that | |||
| 941 | /// position. The position index can be arbitrarily large, i.e. it can be larger | |||
| 942 | /// than 31 or 63, so we need an indexed load in the general case. | |||
| 943 | static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF, | |||
| 944 | unsigned BuiltinID, | |||
| 945 | const CallExpr *E) { | |||
| 946 | Value *BitBase = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 947 | Value *BitPos = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 948 | ||||
| 949 | BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID); | |||
| 950 | ||||
| 951 | // X86 has special BT, BTC, BTR, and BTS instructions that handle the array | |||
| 952 | // indexing operation internally. Use them if possible. | |||
| 953 | if (CGF.getTarget().getTriple().isX86()) | |||
| 954 | return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos); | |||
| 955 | ||||
| 956 | // Otherwise, use generic code to load one byte and test the bit. Use all but | |||
| 957 | // the bottom three bits as the array index, and the bottom three bits to form | |||
| 958 | // a mask. | |||
| 959 | // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0; | |||
| 960 | Value *ByteIndex = CGF.Builder.CreateAShr( | |||
| 961 | BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx"); | |||
| 962 | Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy); | |||
| 963 | Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8, | |||
| 964 | ByteIndex, "bittest.byteaddr"), | |||
| 965 | CGF.Int8Ty, CharUnits::One()); | |||
| 966 | Value *PosLow = | |||
| 967 | CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty), | |||
| 968 | llvm::ConstantInt::get(CGF.Int8Ty, 0x7)); | |||
| 969 | ||||
| 970 | // The updating instructions will need a mask. | |||
| 971 | Value *Mask = nullptr; | |||
| 972 | if (BT.Action != BitTest::TestOnly) { | |||
| 973 | Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow, | |||
| 974 | "bittest.mask"); | |||
| 975 | } | |||
| 976 | ||||
| 977 | // Check the action and ordering of the interlocked intrinsics. | |||
| 978 | llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking); | |||
| 979 | ||||
| 980 | Value *OldByte = nullptr; | |||
| 981 | if (Ordering != llvm::AtomicOrdering::NotAtomic) { | |||
| 982 | // Emit a combined atomicrmw load/store operation for the interlocked | |||
| 983 | // intrinsics. | |||
| 984 | llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or; | |||
| 985 | if (BT.Action == BitTest::Reset) { | |||
| 986 | Mask = CGF.Builder.CreateNot(Mask); | |||
| 987 | RMWOp = llvm::AtomicRMWInst::And; | |||
| 988 | } | |||
| 989 | OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask, | |||
| 990 | Ordering); | |||
| 991 | } else { | |||
| 992 | // Emit a plain load for the non-interlocked intrinsics. | |||
| 993 | OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte"); | |||
| 994 | Value *NewByte = nullptr; | |||
| 995 | switch (BT.Action) { | |||
| 996 | case BitTest::TestOnly: | |||
| 997 | // Don't store anything. | |||
| 998 | break; | |||
| 999 | case BitTest::Complement: | |||
| 1000 | NewByte = CGF.Builder.CreateXor(OldByte, Mask); | |||
| 1001 | break; | |||
| 1002 | case BitTest::Reset: | |||
| 1003 | NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask)); | |||
| 1004 | break; | |||
| 1005 | case BitTest::Set: | |||
| 1006 | NewByte = CGF.Builder.CreateOr(OldByte, Mask); | |||
| 1007 | break; | |||
| 1008 | } | |||
| 1009 | if (NewByte) | |||
| 1010 | CGF.Builder.CreateStore(NewByte, ByteAddr); | |||
| 1011 | } | |||
| 1012 | ||||
| 1013 | // However we loaded the old byte, either by plain load or atomicrmw, shift | |||
| 1014 | // the bit into the low position and mask it to 0 or 1. | |||
| 1015 | Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr"); | |||
| 1016 | return CGF.Builder.CreateAnd( | |||
| 1017 | ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res"); | |||
| 1018 | } | |||
| 1019 | ||||
| 1020 | static llvm::Value *emitPPCLoadReserveIntrinsic(CodeGenFunction &CGF, | |||
| 1021 | unsigned BuiltinID, | |||
| 1022 | const CallExpr *E) { | |||
| 1023 | Value *Addr = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 1024 | ||||
| 1025 | SmallString<64> Asm; | |||
| 1026 | raw_svector_ostream AsmOS(Asm); | |||
| 1027 | llvm::IntegerType *RetType = CGF.Int32Ty; | |||
| 1028 | ||||
| 1029 | switch (BuiltinID) { | |||
| 1030 | case clang::PPC::BI__builtin_ppc_ldarx: | |||
| 1031 | AsmOS << "ldarx "; | |||
| 1032 | RetType = CGF.Int64Ty; | |||
| 1033 | break; | |||
| 1034 | case clang::PPC::BI__builtin_ppc_lwarx: | |||
| 1035 | AsmOS << "lwarx "; | |||
| 1036 | RetType = CGF.Int32Ty; | |||
| 1037 | break; | |||
| 1038 | case clang::PPC::BI__builtin_ppc_lharx: | |||
| 1039 | AsmOS << "lharx "; | |||
| 1040 | RetType = CGF.Int16Ty; | |||
| 1041 | break; | |||
| 1042 | case clang::PPC::BI__builtin_ppc_lbarx: | |||
| 1043 | AsmOS << "lbarx "; | |||
| 1044 | RetType = CGF.Int8Ty; | |||
| 1045 | break; | |||
| 1046 | default: | |||
| 1047 | llvm_unreachable("Expected only PowerPC load reserve intrinsics")::llvm::llvm_unreachable_internal("Expected only PowerPC load reserve intrinsics" , "clang/lib/CodeGen/CGBuiltin.cpp", 1047); | |||
| 1048 | } | |||
| 1049 | ||||
| 1050 | AsmOS << "$0, ${1:y}"; | |||
| 1051 | ||||
| 1052 | std::string Constraints = "=r,*Z,~{memory}"; | |||
| 1053 | std::string MachineClobbers = CGF.getTarget().getClobbers(); | |||
| 1054 | if (!MachineClobbers.empty()) { | |||
| 1055 | Constraints += ','; | |||
| 1056 | Constraints += MachineClobbers; | |||
| 1057 | } | |||
| 1058 | ||||
| 1059 | llvm::Type *IntPtrType = RetType->getPointerTo(); | |||
| 1060 | llvm::FunctionType *FTy = | |||
| 1061 | llvm::FunctionType::get(RetType, {IntPtrType}, false); | |||
| 1062 | ||||
| 1063 | llvm::InlineAsm *IA = | |||
| 1064 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); | |||
| 1065 | llvm::CallInst *CI = CGF.Builder.CreateCall(IA, {Addr}); | |||
| 1066 | CI->addParamAttr( | |||
| 1067 | 0, Attribute::get(CGF.getLLVMContext(), Attribute::ElementType, RetType)); | |||
| 1068 | return CI; | |||
| 1069 | } | |||
| 1070 | ||||
| 1071 | namespace { | |||
| 1072 | enum class MSVCSetJmpKind { | |||
| 1073 | _setjmpex, | |||
| 1074 | _setjmp3, | |||
| 1075 | _setjmp | |||
| 1076 | }; | |||
| 1077 | } | |||
| 1078 | ||||
| 1079 | /// MSVC handles setjmp a bit differently on different platforms. On every | |||
| 1080 | /// architecture except 32-bit x86, the frame address is passed. On x86, extra | |||
| 1081 | /// parameters can be passed as variadic arguments, but we always pass none. | |||
| 1082 | static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind, | |||
| 1083 | const CallExpr *E) { | |||
| 1084 | llvm::Value *Arg1 = nullptr; | |||
| 1085 | llvm::Type *Arg1Ty = nullptr; | |||
| 1086 | StringRef Name; | |||
| 1087 | bool IsVarArg = false; | |||
| 1088 | if (SJKind == MSVCSetJmpKind::_setjmp3) { | |||
| 1089 | Name = "_setjmp3"; | |||
| 1090 | Arg1Ty = CGF.Int32Ty; | |||
| 1091 | Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0); | |||
| 1092 | IsVarArg = true; | |||
| 1093 | } else { | |||
| 1094 | Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex"; | |||
| 1095 | Arg1Ty = CGF.Int8PtrTy; | |||
| 1096 | if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) { | |||
| 1097 | Arg1 = CGF.Builder.CreateCall( | |||
| 1098 | CGF.CGM.getIntrinsic(Intrinsic::sponentry, CGF.AllocaInt8PtrTy)); | |||
| 1099 | } else | |||
| 1100 | Arg1 = CGF.Builder.CreateCall( | |||
| 1101 | CGF.CGM.getIntrinsic(Intrinsic::frameaddress, CGF.AllocaInt8PtrTy), | |||
| 1102 | llvm::ConstantInt::get(CGF.Int32Ty, 0)); | |||
| 1103 | } | |||
| 1104 | ||||
| 1105 | // Mark the call site and declaration with ReturnsTwice. | |||
| 1106 | llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty}; | |||
| 1107 | llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get( | |||
| 1108 | CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex, | |||
| 1109 | llvm::Attribute::ReturnsTwice); | |||
| 1110 | llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction( | |||
| 1111 | llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name, | |||
| 1112 | ReturnsTwiceAttr, /*Local=*/true); | |||
| 1113 | ||||
| 1114 | llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast( | |||
| 1115 | CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy); | |||
| 1116 | llvm::Value *Args[] = {Buf, Arg1}; | |||
| 1117 | llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args); | |||
| 1118 | CB->setAttributes(ReturnsTwiceAttr); | |||
| 1119 | return RValue::get(CB); | |||
| 1120 | } | |||
| 1121 | ||||
| 1122 | // Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code, | |||
| 1123 | // we handle them here. | |||
| 1124 | enum class CodeGenFunction::MSVCIntrin { | |||
| 1125 | _BitScanForward, | |||
| 1126 | _BitScanReverse, | |||
| 1127 | _InterlockedAnd, | |||
| 1128 | _InterlockedDecrement, | |||
| 1129 | _InterlockedExchange, | |||
| 1130 | _InterlockedExchangeAdd, | |||
| 1131 | _InterlockedExchangeSub, | |||
| 1132 | _InterlockedIncrement, | |||
| 1133 | _InterlockedOr, | |||
| 1134 | _InterlockedXor, | |||
| 1135 | _InterlockedExchangeAdd_acq, | |||
| 1136 | _InterlockedExchangeAdd_rel, | |||
| 1137 | _InterlockedExchangeAdd_nf, | |||
| 1138 | _InterlockedExchange_acq, | |||
| 1139 | _InterlockedExchange_rel, | |||
| 1140 | _InterlockedExchange_nf, | |||
| 1141 | _InterlockedCompareExchange_acq, | |||
| 1142 | _InterlockedCompareExchange_rel, | |||
| 1143 | _InterlockedCompareExchange_nf, | |||
| 1144 | _InterlockedCompareExchange128, | |||
| 1145 | _InterlockedCompareExchange128_acq, | |||
| 1146 | _InterlockedCompareExchange128_rel, | |||
| 1147 | _InterlockedCompareExchange128_nf, | |||
| 1148 | _InterlockedOr_acq, | |||
| 1149 | _InterlockedOr_rel, | |||
| 1150 | _InterlockedOr_nf, | |||
| 1151 | _InterlockedXor_acq, | |||
| 1152 | _InterlockedXor_rel, | |||
| 1153 | _InterlockedXor_nf, | |||
| 1154 | _InterlockedAnd_acq, | |||
| 1155 | _InterlockedAnd_rel, | |||
| 1156 | _InterlockedAnd_nf, | |||
| 1157 | _InterlockedIncrement_acq, | |||
| 1158 | _InterlockedIncrement_rel, | |||
| 1159 | _InterlockedIncrement_nf, | |||
| 1160 | _InterlockedDecrement_acq, | |||
| 1161 | _InterlockedDecrement_rel, | |||
| 1162 | _InterlockedDecrement_nf, | |||
| 1163 | __fastfail, | |||
| 1164 | }; | |||
| 1165 | ||||
| 1166 | static Optional<CodeGenFunction::MSVCIntrin> | |||
| 1167 | translateArmToMsvcIntrin(unsigned BuiltinID) { | |||
| 1168 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; | |||
| 1169 | switch (BuiltinID) { | |||
| 1170 | default: | |||
| 1171 | return None; | |||
| 1172 | case ARM::BI_BitScanForward: | |||
| 1173 | case ARM::BI_BitScanForward64: | |||
| 1174 | return MSVCIntrin::_BitScanForward; | |||
| 1175 | case ARM::BI_BitScanReverse: | |||
| 1176 | case ARM::BI_BitScanReverse64: | |||
| 1177 | return MSVCIntrin::_BitScanReverse; | |||
| 1178 | case ARM::BI_InterlockedAnd64: | |||
| 1179 | return MSVCIntrin::_InterlockedAnd; | |||
| 1180 | case ARM::BI_InterlockedExchange64: | |||
| 1181 | return MSVCIntrin::_InterlockedExchange; | |||
| 1182 | case ARM::BI_InterlockedExchangeAdd64: | |||
| 1183 | return MSVCIntrin::_InterlockedExchangeAdd; | |||
| 1184 | case ARM::BI_InterlockedExchangeSub64: | |||
| 1185 | return MSVCIntrin::_InterlockedExchangeSub; | |||
| 1186 | case ARM::BI_InterlockedOr64: | |||
| 1187 | return MSVCIntrin::_InterlockedOr; | |||
| 1188 | case ARM::BI_InterlockedXor64: | |||
| 1189 | return MSVCIntrin::_InterlockedXor; | |||
| 1190 | case ARM::BI_InterlockedDecrement64: | |||
| 1191 | return MSVCIntrin::_InterlockedDecrement; | |||
| 1192 | case ARM::BI_InterlockedIncrement64: | |||
| 1193 | return MSVCIntrin::_InterlockedIncrement; | |||
| 1194 | case ARM::BI_InterlockedExchangeAdd8_acq: | |||
| 1195 | case ARM::BI_InterlockedExchangeAdd16_acq: | |||
| 1196 | case ARM::BI_InterlockedExchangeAdd_acq: | |||
| 1197 | case ARM::BI_InterlockedExchangeAdd64_acq: | |||
| 1198 | return MSVCIntrin::_InterlockedExchangeAdd_acq; | |||
| 1199 | case ARM::BI_InterlockedExchangeAdd8_rel: | |||
| 1200 | case ARM::BI_InterlockedExchangeAdd16_rel: | |||
| 1201 | case ARM::BI_InterlockedExchangeAdd_rel: | |||
| 1202 | case ARM::BI_InterlockedExchangeAdd64_rel: | |||
| 1203 | return MSVCIntrin::_InterlockedExchangeAdd_rel; | |||
| 1204 | case ARM::BI_InterlockedExchangeAdd8_nf: | |||
| 1205 | case ARM::BI_InterlockedExchangeAdd16_nf: | |||
| 1206 | case ARM::BI_InterlockedExchangeAdd_nf: | |||
| 1207 | case ARM::BI_InterlockedExchangeAdd64_nf: | |||
| 1208 | return MSVCIntrin::_InterlockedExchangeAdd_nf; | |||
| 1209 | case ARM::BI_InterlockedExchange8_acq: | |||
| 1210 | case ARM::BI_InterlockedExchange16_acq: | |||
| 1211 | case ARM::BI_InterlockedExchange_acq: | |||
| 1212 | case ARM::BI_InterlockedExchange64_acq: | |||
| 1213 | return MSVCIntrin::_InterlockedExchange_acq; | |||
| 1214 | case ARM::BI_InterlockedExchange8_rel: | |||
| 1215 | case ARM::BI_InterlockedExchange16_rel: | |||
| 1216 | case ARM::BI_InterlockedExchange_rel: | |||
| 1217 | case ARM::BI_InterlockedExchange64_rel: | |||
| 1218 | return MSVCIntrin::_InterlockedExchange_rel; | |||
| 1219 | case ARM::BI_InterlockedExchange8_nf: | |||
| 1220 | case ARM::BI_InterlockedExchange16_nf: | |||
| 1221 | case ARM::BI_InterlockedExchange_nf: | |||
| 1222 | case ARM::BI_InterlockedExchange64_nf: | |||
| 1223 | return MSVCIntrin::_InterlockedExchange_nf; | |||
| 1224 | case ARM::BI_InterlockedCompareExchange8_acq: | |||
| 1225 | case ARM::BI_InterlockedCompareExchange16_acq: | |||
| 1226 | case ARM::BI_InterlockedCompareExchange_acq: | |||
| 1227 | case ARM::BI_InterlockedCompareExchange64_acq: | |||
| 1228 | return MSVCIntrin::_InterlockedCompareExchange_acq; | |||
| 1229 | case ARM::BI_InterlockedCompareExchange8_rel: | |||
| 1230 | case ARM::BI_InterlockedCompareExchange16_rel: | |||
| 1231 | case ARM::BI_InterlockedCompareExchange_rel: | |||
| 1232 | case ARM::BI_InterlockedCompareExchange64_rel: | |||
| 1233 | return MSVCIntrin::_InterlockedCompareExchange_rel; | |||
| 1234 | case ARM::BI_InterlockedCompareExchange8_nf: | |||
| 1235 | case ARM::BI_InterlockedCompareExchange16_nf: | |||
| 1236 | case ARM::BI_InterlockedCompareExchange_nf: | |||
| 1237 | case ARM::BI_InterlockedCompareExchange64_nf: | |||
| 1238 | return MSVCIntrin::_InterlockedCompareExchange_nf; | |||
| 1239 | case ARM::BI_InterlockedOr8_acq: | |||
| 1240 | case ARM::BI_InterlockedOr16_acq: | |||
| 1241 | case ARM::BI_InterlockedOr_acq: | |||
| 1242 | case ARM::BI_InterlockedOr64_acq: | |||
| 1243 | return MSVCIntrin::_InterlockedOr_acq; | |||
| 1244 | case ARM::BI_InterlockedOr8_rel: | |||
| 1245 | case ARM::BI_InterlockedOr16_rel: | |||
| 1246 | case ARM::BI_InterlockedOr_rel: | |||
| 1247 | case ARM::BI_InterlockedOr64_rel: | |||
| 1248 | return MSVCIntrin::_InterlockedOr_rel; | |||
| 1249 | case ARM::BI_InterlockedOr8_nf: | |||
| 1250 | case ARM::BI_InterlockedOr16_nf: | |||
| 1251 | case ARM::BI_InterlockedOr_nf: | |||
| 1252 | case ARM::BI_InterlockedOr64_nf: | |||
| 1253 | return MSVCIntrin::_InterlockedOr_nf; | |||
| 1254 | case ARM::BI_InterlockedXor8_acq: | |||
| 1255 | case ARM::BI_InterlockedXor16_acq: | |||
| 1256 | case ARM::BI_InterlockedXor_acq: | |||
| 1257 | case ARM::BI_InterlockedXor64_acq: | |||
| 1258 | return MSVCIntrin::_InterlockedXor_acq; | |||
| 1259 | case ARM::BI_InterlockedXor8_rel: | |||
| 1260 | case ARM::BI_InterlockedXor16_rel: | |||
| 1261 | case ARM::BI_InterlockedXor_rel: | |||
| 1262 | case ARM::BI_InterlockedXor64_rel: | |||
| 1263 | return MSVCIntrin::_InterlockedXor_rel; | |||
| 1264 | case ARM::BI_InterlockedXor8_nf: | |||
| 1265 | case ARM::BI_InterlockedXor16_nf: | |||
| 1266 | case ARM::BI_InterlockedXor_nf: | |||
| 1267 | case ARM::BI_InterlockedXor64_nf: | |||
| 1268 | return MSVCIntrin::_InterlockedXor_nf; | |||
| 1269 | case ARM::BI_InterlockedAnd8_acq: | |||
| 1270 | case ARM::BI_InterlockedAnd16_acq: | |||
| 1271 | case ARM::BI_InterlockedAnd_acq: | |||
| 1272 | case ARM::BI_InterlockedAnd64_acq: | |||
| 1273 | return MSVCIntrin::_InterlockedAnd_acq; | |||
| 1274 | case ARM::BI_InterlockedAnd8_rel: | |||
| 1275 | case ARM::BI_InterlockedAnd16_rel: | |||
| 1276 | case ARM::BI_InterlockedAnd_rel: | |||
| 1277 | case ARM::BI_InterlockedAnd64_rel: | |||
| 1278 | return MSVCIntrin::_InterlockedAnd_rel; | |||
| 1279 | case ARM::BI_InterlockedAnd8_nf: | |||
| 1280 | case ARM::BI_InterlockedAnd16_nf: | |||
| 1281 | case ARM::BI_InterlockedAnd_nf: | |||
| 1282 | case ARM::BI_InterlockedAnd64_nf: | |||
| 1283 | return MSVCIntrin::_InterlockedAnd_nf; | |||
| 1284 | case ARM::BI_InterlockedIncrement16_acq: | |||
| 1285 | case ARM::BI_InterlockedIncrement_acq: | |||
| 1286 | case ARM::BI_InterlockedIncrement64_acq: | |||
| 1287 | return MSVCIntrin::_InterlockedIncrement_acq; | |||
| 1288 | case ARM::BI_InterlockedIncrement16_rel: | |||
| 1289 | case ARM::BI_InterlockedIncrement_rel: | |||
| 1290 | case ARM::BI_InterlockedIncrement64_rel: | |||
| 1291 | return MSVCIntrin::_InterlockedIncrement_rel; | |||
| 1292 | case ARM::BI_InterlockedIncrement16_nf: | |||
| 1293 | case ARM::BI_InterlockedIncrement_nf: | |||
| 1294 | case ARM::BI_InterlockedIncrement64_nf: | |||
| 1295 | return MSVCIntrin::_InterlockedIncrement_nf; | |||
| 1296 | case ARM::BI_InterlockedDecrement16_acq: | |||
| 1297 | case ARM::BI_InterlockedDecrement_acq: | |||
| 1298 | case ARM::BI_InterlockedDecrement64_acq: | |||
| 1299 | return MSVCIntrin::_InterlockedDecrement_acq; | |||
| 1300 | case ARM::BI_InterlockedDecrement16_rel: | |||
| 1301 | case ARM::BI_InterlockedDecrement_rel: | |||
| 1302 | case ARM::BI_InterlockedDecrement64_rel: | |||
| 1303 | return MSVCIntrin::_InterlockedDecrement_rel; | |||
| 1304 | case ARM::BI_InterlockedDecrement16_nf: | |||
| 1305 | case ARM::BI_InterlockedDecrement_nf: | |||
| 1306 | case ARM::BI_InterlockedDecrement64_nf: | |||
| 1307 | return MSVCIntrin::_InterlockedDecrement_nf; | |||
| 1308 | } | |||
| 1309 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "clang/lib/CodeGen/CGBuiltin.cpp" , 1309); | |||
| 1310 | } | |||
| 1311 | ||||
| 1312 | static Optional<CodeGenFunction::MSVCIntrin> | |||
| 1313 | translateAarch64ToMsvcIntrin(unsigned BuiltinID) { | |||
| 1314 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; | |||
| 1315 | switch (BuiltinID) { | |||
| 1316 | default: | |||
| 1317 | return None; | |||
| 1318 | case AArch64::BI_BitScanForward: | |||
| 1319 | case AArch64::BI_BitScanForward64: | |||
| 1320 | return MSVCIntrin::_BitScanForward; | |||
| 1321 | case AArch64::BI_BitScanReverse: | |||
| 1322 | case AArch64::BI_BitScanReverse64: | |||
| 1323 | return MSVCIntrin::_BitScanReverse; | |||
| 1324 | case AArch64::BI_InterlockedAnd64: | |||
| 1325 | return MSVCIntrin::_InterlockedAnd; | |||
| 1326 | case AArch64::BI_InterlockedExchange64: | |||
| 1327 | return MSVCIntrin::_InterlockedExchange; | |||
| 1328 | case AArch64::BI_InterlockedExchangeAdd64: | |||
| 1329 | return MSVCIntrin::_InterlockedExchangeAdd; | |||
| 1330 | case AArch64::BI_InterlockedExchangeSub64: | |||
| 1331 | return MSVCIntrin::_InterlockedExchangeSub; | |||
| 1332 | case AArch64::BI_InterlockedOr64: | |||
| 1333 | return MSVCIntrin::_InterlockedOr; | |||
| 1334 | case AArch64::BI_InterlockedXor64: | |||
| 1335 | return MSVCIntrin::_InterlockedXor; | |||
| 1336 | case AArch64::BI_InterlockedDecrement64: | |||
| 1337 | return MSVCIntrin::_InterlockedDecrement; | |||
| 1338 | case AArch64::BI_InterlockedIncrement64: | |||
| 1339 | return MSVCIntrin::_InterlockedIncrement; | |||
| 1340 | case AArch64::BI_InterlockedExchangeAdd8_acq: | |||
| 1341 | case AArch64::BI_InterlockedExchangeAdd16_acq: | |||
| 1342 | case AArch64::BI_InterlockedExchangeAdd_acq: | |||
| 1343 | case AArch64::BI_InterlockedExchangeAdd64_acq: | |||
| 1344 | return MSVCIntrin::_InterlockedExchangeAdd_acq; | |||
| 1345 | case AArch64::BI_InterlockedExchangeAdd8_rel: | |||
| 1346 | case AArch64::BI_InterlockedExchangeAdd16_rel: | |||
| 1347 | case AArch64::BI_InterlockedExchangeAdd_rel: | |||
| 1348 | case AArch64::BI_InterlockedExchangeAdd64_rel: | |||
| 1349 | return MSVCIntrin::_InterlockedExchangeAdd_rel; | |||
| 1350 | case AArch64::BI_InterlockedExchangeAdd8_nf: | |||
| 1351 | case AArch64::BI_InterlockedExchangeAdd16_nf: | |||
| 1352 | case AArch64::BI_InterlockedExchangeAdd_nf: | |||
| 1353 | case AArch64::BI_InterlockedExchangeAdd64_nf: | |||
| 1354 | return MSVCIntrin::_InterlockedExchangeAdd_nf; | |||
| 1355 | case AArch64::BI_InterlockedExchange8_acq: | |||
| 1356 | case AArch64::BI_InterlockedExchange16_acq: | |||
| 1357 | case AArch64::BI_InterlockedExchange_acq: | |||
| 1358 | case AArch64::BI_InterlockedExchange64_acq: | |||
| 1359 | return MSVCIntrin::_InterlockedExchange_acq; | |||
| 1360 | case AArch64::BI_InterlockedExchange8_rel: | |||
| 1361 | case AArch64::BI_InterlockedExchange16_rel: | |||
| 1362 | case AArch64::BI_InterlockedExchange_rel: | |||
| 1363 | case AArch64::BI_InterlockedExchange64_rel: | |||
| 1364 | return MSVCIntrin::_InterlockedExchange_rel; | |||
| 1365 | case AArch64::BI_InterlockedExchange8_nf: | |||
| 1366 | case AArch64::BI_InterlockedExchange16_nf: | |||
| 1367 | case AArch64::BI_InterlockedExchange_nf: | |||
| 1368 | case AArch64::BI_InterlockedExchange64_nf: | |||
| 1369 | return MSVCIntrin::_InterlockedExchange_nf; | |||
| 1370 | case AArch64::BI_InterlockedCompareExchange8_acq: | |||
| 1371 | case AArch64::BI_InterlockedCompareExchange16_acq: | |||
| 1372 | case AArch64::BI_InterlockedCompareExchange_acq: | |||
| 1373 | case AArch64::BI_InterlockedCompareExchange64_acq: | |||
| 1374 | return MSVCIntrin::_InterlockedCompareExchange_acq; | |||
| 1375 | case AArch64::BI_InterlockedCompareExchange8_rel: | |||
| 1376 | case AArch64::BI_InterlockedCompareExchange16_rel: | |||
| 1377 | case AArch64::BI_InterlockedCompareExchange_rel: | |||
| 1378 | case AArch64::BI_InterlockedCompareExchange64_rel: | |||
| 1379 | return MSVCIntrin::_InterlockedCompareExchange_rel; | |||
| 1380 | case AArch64::BI_InterlockedCompareExchange8_nf: | |||
| 1381 | case AArch64::BI_InterlockedCompareExchange16_nf: | |||
| 1382 | case AArch64::BI_InterlockedCompareExchange_nf: | |||
| 1383 | case AArch64::BI_InterlockedCompareExchange64_nf: | |||
| 1384 | return MSVCIntrin::_InterlockedCompareExchange_nf; | |||
| 1385 | case AArch64::BI_InterlockedCompareExchange128: | |||
| 1386 | return MSVCIntrin::_InterlockedCompareExchange128; | |||
| 1387 | case AArch64::BI_InterlockedCompareExchange128_acq: | |||
| 1388 | return MSVCIntrin::_InterlockedCompareExchange128_acq; | |||
| 1389 | case AArch64::BI_InterlockedCompareExchange128_nf: | |||
| 1390 | return MSVCIntrin::_InterlockedCompareExchange128_nf; | |||
| 1391 | case AArch64::BI_InterlockedCompareExchange128_rel: | |||
| 1392 | return MSVCIntrin::_InterlockedCompareExchange128_rel; | |||
| 1393 | case AArch64::BI_InterlockedOr8_acq: | |||
| 1394 | case AArch64::BI_InterlockedOr16_acq: | |||
| 1395 | case AArch64::BI_InterlockedOr_acq: | |||
| 1396 | case AArch64::BI_InterlockedOr64_acq: | |||
| 1397 | return MSVCIntrin::_InterlockedOr_acq; | |||
| 1398 | case AArch64::BI_InterlockedOr8_rel: | |||
| 1399 | case AArch64::BI_InterlockedOr16_rel: | |||
| 1400 | case AArch64::BI_InterlockedOr_rel: | |||
| 1401 | case AArch64::BI_InterlockedOr64_rel: | |||
| 1402 | return MSVCIntrin::_InterlockedOr_rel; | |||
| 1403 | case AArch64::BI_InterlockedOr8_nf: | |||
| 1404 | case AArch64::BI_InterlockedOr16_nf: | |||
| 1405 | case AArch64::BI_InterlockedOr_nf: | |||
| 1406 | case AArch64::BI_InterlockedOr64_nf: | |||
| 1407 | return MSVCIntrin::_InterlockedOr_nf; | |||
| 1408 | case AArch64::BI_InterlockedXor8_acq: | |||
| 1409 | case AArch64::BI_InterlockedXor16_acq: | |||
| 1410 | case AArch64::BI_InterlockedXor_acq: | |||
| 1411 | case AArch64::BI_InterlockedXor64_acq: | |||
| 1412 | return MSVCIntrin::_InterlockedXor_acq; | |||
| 1413 | case AArch64::BI_InterlockedXor8_rel: | |||
| 1414 | case AArch64::BI_InterlockedXor16_rel: | |||
| 1415 | case AArch64::BI_InterlockedXor_rel: | |||
| 1416 | case AArch64::BI_InterlockedXor64_rel: | |||
| 1417 | return MSVCIntrin::_InterlockedXor_rel; | |||
| 1418 | case AArch64::BI_InterlockedXor8_nf: | |||
| 1419 | case AArch64::BI_InterlockedXor16_nf: | |||
| 1420 | case AArch64::BI_InterlockedXor_nf: | |||
| 1421 | case AArch64::BI_InterlockedXor64_nf: | |||
| 1422 | return MSVCIntrin::_InterlockedXor_nf; | |||
| 1423 | case AArch64::BI_InterlockedAnd8_acq: | |||
| 1424 | case AArch64::BI_InterlockedAnd16_acq: | |||
| 1425 | case AArch64::BI_InterlockedAnd_acq: | |||
| 1426 | case AArch64::BI_InterlockedAnd64_acq: | |||
| 1427 | return MSVCIntrin::_InterlockedAnd_acq; | |||
| 1428 | case AArch64::BI_InterlockedAnd8_rel: | |||
| 1429 | case AArch64::BI_InterlockedAnd16_rel: | |||
| 1430 | case AArch64::BI_InterlockedAnd_rel: | |||
| 1431 | case AArch64::BI_InterlockedAnd64_rel: | |||
| 1432 | return MSVCIntrin::_InterlockedAnd_rel; | |||
| 1433 | case AArch64::BI_InterlockedAnd8_nf: | |||
| 1434 | case AArch64::BI_InterlockedAnd16_nf: | |||
| 1435 | case AArch64::BI_InterlockedAnd_nf: | |||
| 1436 | case AArch64::BI_InterlockedAnd64_nf: | |||
| 1437 | return MSVCIntrin::_InterlockedAnd_nf; | |||
| 1438 | case AArch64::BI_InterlockedIncrement16_acq: | |||
| 1439 | case AArch64::BI_InterlockedIncrement_acq: | |||
| 1440 | case AArch64::BI_InterlockedIncrement64_acq: | |||
| 1441 | return MSVCIntrin::_InterlockedIncrement_acq; | |||
| 1442 | case AArch64::BI_InterlockedIncrement16_rel: | |||
| 1443 | case AArch64::BI_InterlockedIncrement_rel: | |||
| 1444 | case AArch64::BI_InterlockedIncrement64_rel: | |||
| 1445 | return MSVCIntrin::_InterlockedIncrement_rel; | |||
| 1446 | case AArch64::BI_InterlockedIncrement16_nf: | |||
| 1447 | case AArch64::BI_InterlockedIncrement_nf: | |||
| 1448 | case AArch64::BI_InterlockedIncrement64_nf: | |||
| 1449 | return MSVCIntrin::_InterlockedIncrement_nf; | |||
| 1450 | case AArch64::BI_InterlockedDecrement16_acq: | |||
| 1451 | case AArch64::BI_InterlockedDecrement_acq: | |||
| 1452 | case AArch64::BI_InterlockedDecrement64_acq: | |||
| 1453 | return MSVCIntrin::_InterlockedDecrement_acq; | |||
| 1454 | case AArch64::BI_InterlockedDecrement16_rel: | |||
| 1455 | case AArch64::BI_InterlockedDecrement_rel: | |||
| 1456 | case AArch64::BI_InterlockedDecrement64_rel: | |||
| 1457 | return MSVCIntrin::_InterlockedDecrement_rel; | |||
| 1458 | case AArch64::BI_InterlockedDecrement16_nf: | |||
| 1459 | case AArch64::BI_InterlockedDecrement_nf: | |||
| 1460 | case AArch64::BI_InterlockedDecrement64_nf: | |||
| 1461 | return MSVCIntrin::_InterlockedDecrement_nf; | |||
| 1462 | } | |||
| 1463 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "clang/lib/CodeGen/CGBuiltin.cpp" , 1463); | |||
| 1464 | } | |||
| 1465 | ||||
| 1466 | static Optional<CodeGenFunction::MSVCIntrin> | |||
| 1467 | translateX86ToMsvcIntrin(unsigned BuiltinID) { | |||
| 1468 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; | |||
| 1469 | switch (BuiltinID) { | |||
| 1470 | default: | |||
| 1471 | return None; | |||
| 1472 | case clang::X86::BI_BitScanForward: | |||
| 1473 | case clang::X86::BI_BitScanForward64: | |||
| 1474 | return MSVCIntrin::_BitScanForward; | |||
| 1475 | case clang::X86::BI_BitScanReverse: | |||
| 1476 | case clang::X86::BI_BitScanReverse64: | |||
| 1477 | return MSVCIntrin::_BitScanReverse; | |||
| 1478 | case clang::X86::BI_InterlockedAnd64: | |||
| 1479 | return MSVCIntrin::_InterlockedAnd; | |||
| 1480 | case clang::X86::BI_InterlockedCompareExchange128: | |||
| 1481 | return MSVCIntrin::_InterlockedCompareExchange128; | |||
| 1482 | case clang::X86::BI_InterlockedExchange64: | |||
| 1483 | return MSVCIntrin::_InterlockedExchange; | |||
| 1484 | case clang::X86::BI_InterlockedExchangeAdd64: | |||
| 1485 | return MSVCIntrin::_InterlockedExchangeAdd; | |||
| 1486 | case clang::X86::BI_InterlockedExchangeSub64: | |||
| 1487 | return MSVCIntrin::_InterlockedExchangeSub; | |||
| 1488 | case clang::X86::BI_InterlockedOr64: | |||
| 1489 | return MSVCIntrin::_InterlockedOr; | |||
| 1490 | case clang::X86::BI_InterlockedXor64: | |||
| 1491 | return MSVCIntrin::_InterlockedXor; | |||
| 1492 | case clang::X86::BI_InterlockedDecrement64: | |||
| 1493 | return MSVCIntrin::_InterlockedDecrement; | |||
| 1494 | case clang::X86::BI_InterlockedIncrement64: | |||
| 1495 | return MSVCIntrin::_InterlockedIncrement; | |||
| 1496 | } | |||
| 1497 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "clang/lib/CodeGen/CGBuiltin.cpp" , 1497); | |||
| 1498 | } | |||
| 1499 | ||||
| 1500 | // Emit an MSVC intrinsic. Assumes that arguments have *not* been evaluated. | |||
| 1501 | Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID, | |||
| 1502 | const CallExpr *E) { | |||
| 1503 | switch (BuiltinID) { | |||
| 1504 | case MSVCIntrin::_BitScanForward: | |||
| 1505 | case MSVCIntrin::_BitScanReverse: { | |||
| 1506 | Address IndexAddress(EmitPointerWithAlignment(E->getArg(0))); | |||
| 1507 | Value *ArgValue = EmitScalarExpr(E->getArg(1)); | |||
| 1508 | ||||
| 1509 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 1510 | llvm::Type *IndexType = IndexAddress.getElementType(); | |||
| 1511 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 1512 | ||||
| 1513 | Value *ArgZero = llvm::Constant::getNullValue(ArgType); | |||
| 1514 | Value *ResZero = llvm::Constant::getNullValue(ResultType); | |||
| 1515 | Value *ResOne = llvm::ConstantInt::get(ResultType, 1); | |||
| 1516 | ||||
| 1517 | BasicBlock *Begin = Builder.GetInsertBlock(); | |||
| 1518 | BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn); | |||
| 1519 | Builder.SetInsertPoint(End); | |||
| 1520 | PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result"); | |||
| 1521 | ||||
| 1522 | Builder.SetInsertPoint(Begin); | |||
| 1523 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero); | |||
| 1524 | BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn); | |||
| 1525 | Builder.CreateCondBr(IsZero, End, NotZero); | |||
| 1526 | Result->addIncoming(ResZero, Begin); | |||
| 1527 | ||||
| 1528 | Builder.SetInsertPoint(NotZero); | |||
| 1529 | ||||
| 1530 | if (BuiltinID == MSVCIntrin::_BitScanForward) { | |||
| 1531 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); | |||
| 1532 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); | |||
| 1533 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); | |||
| 1534 | Builder.CreateStore(ZeroCount, IndexAddress, false); | |||
| 1535 | } else { | |||
| 1536 | unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth(); | |||
| 1537 | Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1); | |||
| 1538 | ||||
| 1539 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
| 1540 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); | |||
| 1541 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); | |||
| 1542 | Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount); | |||
| 1543 | Builder.CreateStore(Index, IndexAddress, false); | |||
| 1544 | } | |||
| 1545 | Builder.CreateBr(End); | |||
| 1546 | Result->addIncoming(ResOne, NotZero); | |||
| 1547 | ||||
| 1548 | Builder.SetInsertPoint(End); | |||
| 1549 | return Result; | |||
| 1550 | } | |||
| 1551 | case MSVCIntrin::_InterlockedAnd: | |||
| 1552 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E); | |||
| 1553 | case MSVCIntrin::_InterlockedExchange: | |||
| 1554 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E); | |||
| 1555 | case MSVCIntrin::_InterlockedExchangeAdd: | |||
| 1556 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E); | |||
| 1557 | case MSVCIntrin::_InterlockedExchangeSub: | |||
| 1558 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E); | |||
| 1559 | case MSVCIntrin::_InterlockedOr: | |||
| 1560 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E); | |||
| 1561 | case MSVCIntrin::_InterlockedXor: | |||
| 1562 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E); | |||
| 1563 | case MSVCIntrin::_InterlockedExchangeAdd_acq: | |||
| 1564 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, | |||
| 1565 | AtomicOrdering::Acquire); | |||
| 1566 | case MSVCIntrin::_InterlockedExchangeAdd_rel: | |||
| 1567 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, | |||
| 1568 | AtomicOrdering::Release); | |||
| 1569 | case MSVCIntrin::_InterlockedExchangeAdd_nf: | |||
| 1570 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, | |||
| 1571 | AtomicOrdering::Monotonic); | |||
| 1572 | case MSVCIntrin::_InterlockedExchange_acq: | |||
| 1573 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, | |||
| 1574 | AtomicOrdering::Acquire); | |||
| 1575 | case MSVCIntrin::_InterlockedExchange_rel: | |||
| 1576 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, | |||
| 1577 | AtomicOrdering::Release); | |||
| 1578 | case MSVCIntrin::_InterlockedExchange_nf: | |||
| 1579 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, | |||
| 1580 | AtomicOrdering::Monotonic); | |||
| 1581 | case MSVCIntrin::_InterlockedCompareExchange_acq: | |||
| 1582 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire); | |||
| 1583 | case MSVCIntrin::_InterlockedCompareExchange_rel: | |||
| 1584 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release); | |||
| 1585 | case MSVCIntrin::_InterlockedCompareExchange_nf: | |||
| 1586 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic); | |||
| 1587 | case MSVCIntrin::_InterlockedCompareExchange128: | |||
| 1588 | return EmitAtomicCmpXchg128ForMSIntrin( | |||
| 1589 | *this, E, AtomicOrdering::SequentiallyConsistent); | |||
| 1590 | case MSVCIntrin::_InterlockedCompareExchange128_acq: | |||
| 1591 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Acquire); | |||
| 1592 | case MSVCIntrin::_InterlockedCompareExchange128_rel: | |||
| 1593 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Release); | |||
| 1594 | case MSVCIntrin::_InterlockedCompareExchange128_nf: | |||
| 1595 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Monotonic); | |||
| 1596 | case MSVCIntrin::_InterlockedOr_acq: | |||
| 1597 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, | |||
| 1598 | AtomicOrdering::Acquire); | |||
| 1599 | case MSVCIntrin::_InterlockedOr_rel: | |||
| 1600 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, | |||
| 1601 | AtomicOrdering::Release); | |||
| 1602 | case MSVCIntrin::_InterlockedOr_nf: | |||
| 1603 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, | |||
| 1604 | AtomicOrdering::Monotonic); | |||
| 1605 | case MSVCIntrin::_InterlockedXor_acq: | |||
| 1606 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, | |||
| 1607 | AtomicOrdering::Acquire); | |||
| 1608 | case MSVCIntrin::_InterlockedXor_rel: | |||
| 1609 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, | |||
| 1610 | AtomicOrdering::Release); | |||
| 1611 | case MSVCIntrin::_InterlockedXor_nf: | |||
| 1612 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, | |||
| 1613 | AtomicOrdering::Monotonic); | |||
| 1614 | case MSVCIntrin::_InterlockedAnd_acq: | |||
| 1615 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, | |||
| 1616 | AtomicOrdering::Acquire); | |||
| 1617 | case MSVCIntrin::_InterlockedAnd_rel: | |||
| 1618 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, | |||
| 1619 | AtomicOrdering::Release); | |||
| 1620 | case MSVCIntrin::_InterlockedAnd_nf: | |||
| 1621 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, | |||
| 1622 | AtomicOrdering::Monotonic); | |||
| 1623 | case MSVCIntrin::_InterlockedIncrement_acq: | |||
| 1624 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire); | |||
| 1625 | case MSVCIntrin::_InterlockedIncrement_rel: | |||
| 1626 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release); | |||
| 1627 | case MSVCIntrin::_InterlockedIncrement_nf: | |||
| 1628 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic); | |||
| 1629 | case MSVCIntrin::_InterlockedDecrement_acq: | |||
| 1630 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire); | |||
| 1631 | case MSVCIntrin::_InterlockedDecrement_rel: | |||
| 1632 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release); | |||
| 1633 | case MSVCIntrin::_InterlockedDecrement_nf: | |||
| 1634 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic); | |||
| 1635 | ||||
| 1636 | case MSVCIntrin::_InterlockedDecrement: | |||
| 1637 | return EmitAtomicDecrementValue(*this, E); | |||
| 1638 | case MSVCIntrin::_InterlockedIncrement: | |||
| 1639 | return EmitAtomicIncrementValue(*this, E); | |||
| 1640 | ||||
| 1641 | case MSVCIntrin::__fastfail: { | |||
| 1642 | // Request immediate process termination from the kernel. The instruction | |||
| 1643 | // sequences to do this are documented on MSDN: | |||
| 1644 | // https://msdn.microsoft.com/en-us/library/dn774154.aspx | |||
| 1645 | llvm::Triple::ArchType ISA = getTarget().getTriple().getArch(); | |||
| 1646 | StringRef Asm, Constraints; | |||
| 1647 | switch (ISA) { | |||
| 1648 | default: | |||
| 1649 | ErrorUnsupported(E, "__fastfail call for this architecture"); | |||
| 1650 | break; | |||
| 1651 | case llvm::Triple::x86: | |||
| 1652 | case llvm::Triple::x86_64: | |||
| 1653 | Asm = "int $$0x29"; | |||
| 1654 | Constraints = "{cx}"; | |||
| 1655 | break; | |||
| 1656 | case llvm::Triple::thumb: | |||
| 1657 | Asm = "udf #251"; | |||
| 1658 | Constraints = "{r0}"; | |||
| 1659 | break; | |||
| 1660 | case llvm::Triple::aarch64: | |||
| 1661 | Asm = "brk #0xF003"; | |||
| 1662 | Constraints = "{w0}"; | |||
| 1663 | } | |||
| 1664 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false); | |||
| 1665 | llvm::InlineAsm *IA = | |||
| 1666 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); | |||
| 1667 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( | |||
| 1668 | getLLVMContext(), llvm::AttributeList::FunctionIndex, | |||
| 1669 | llvm::Attribute::NoReturn); | |||
| 1670 | llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0))); | |||
| 1671 | CI->setAttributes(NoReturnAttr); | |||
| 1672 | return CI; | |||
| 1673 | } | |||
| 1674 | } | |||
| 1675 | llvm_unreachable("Incorrect MSVC intrinsic!")::llvm::llvm_unreachable_internal("Incorrect MSVC intrinsic!" , "clang/lib/CodeGen/CGBuiltin.cpp", 1675); | |||
| 1676 | } | |||
| 1677 | ||||
| 1678 | namespace { | |||
| 1679 | // ARC cleanup for __builtin_os_log_format | |||
| 1680 | struct CallObjCArcUse final : EHScopeStack::Cleanup { | |||
| 1681 | CallObjCArcUse(llvm::Value *object) : object(object) {} | |||
| 1682 | llvm::Value *object; | |||
| 1683 | ||||
| 1684 | void Emit(CodeGenFunction &CGF, Flags flags) override { | |||
| 1685 | CGF.EmitARCIntrinsicUse(object); | |||
| 1686 | } | |||
| 1687 | }; | |||
| 1688 | } | |||
| 1689 | ||||
| 1690 | Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E, | |||
| 1691 | BuiltinCheckKind Kind) { | |||
| 1692 | assert((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero)(static_cast <bool> ((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind" ) ? void (0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1693, __extension__ __PRETTY_FUNCTION__ )) | |||
| 1693 | && "Unsupported builtin check kind")(static_cast <bool> ((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind" ) ? void (0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1693, __extension__ __PRETTY_FUNCTION__ )); | |||
| 1694 | ||||
| 1695 | Value *ArgValue = EmitScalarExpr(E); | |||
| 1696 | if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef()) | |||
| 1697 | return ArgValue; | |||
| 1698 | ||||
| 1699 | SanitizerScope SanScope(this); | |||
| 1700 | Value *Cond = Builder.CreateICmpNE( | |||
| 1701 | ArgValue, llvm::Constant::getNullValue(ArgValue->getType())); | |||
| 1702 | EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin), | |||
| 1703 | SanitizerHandler::InvalidBuiltin, | |||
| 1704 | {EmitCheckSourceLocation(E->getExprLoc()), | |||
| 1705 | llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)}, | |||
| 1706 | None); | |||
| 1707 | return ArgValue; | |||
| 1708 | } | |||
| 1709 | ||||
| 1710 | /// Get the argument type for arguments to os_log_helper. | |||
| 1711 | static CanQualType getOSLogArgType(ASTContext &C, int Size) { | |||
| 1712 | QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false); | |||
| 1713 | return C.getCanonicalType(UnsignedTy); | |||
| 1714 | } | |||
| 1715 | ||||
| 1716 | llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction( | |||
| 1717 | const analyze_os_log::OSLogBufferLayout &Layout, | |||
| 1718 | CharUnits BufferAlignment) { | |||
| 1719 | ASTContext &Ctx = getContext(); | |||
| 1720 | ||||
| 1721 | llvm::SmallString<64> Name; | |||
| 1722 | { | |||
| 1723 | raw_svector_ostream OS(Name); | |||
| 1724 | OS << "__os_log_helper"; | |||
| 1725 | OS << "_" << BufferAlignment.getQuantity(); | |||
| 1726 | OS << "_" << int(Layout.getSummaryByte()); | |||
| 1727 | OS << "_" << int(Layout.getNumArgsByte()); | |||
| 1728 | for (const auto &Item : Layout.Items) | |||
| 1729 | OS << "_" << int(Item.getSizeByte()) << "_" | |||
| 1730 | << int(Item.getDescriptorByte()); | |||
| 1731 | } | |||
| 1732 | ||||
| 1733 | if (llvm::Function *F = CGM.getModule().getFunction(Name)) | |||
| 1734 | return F; | |||
| 1735 | ||||
| 1736 | llvm::SmallVector<QualType, 4> ArgTys; | |||
| 1737 | FunctionArgList Args; | |||
| 1738 | Args.push_back(ImplicitParamDecl::Create( | |||
| 1739 | Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"), Ctx.VoidPtrTy, | |||
| 1740 | ImplicitParamDecl::Other)); | |||
| 1741 | ArgTys.emplace_back(Ctx.VoidPtrTy); | |||
| 1742 | ||||
| 1743 | for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) { | |||
| 1744 | char Size = Layout.Items[I].getSizeByte(); | |||
| 1745 | if (!Size) | |||
| 1746 | continue; | |||
| 1747 | ||||
| 1748 | QualType ArgTy = getOSLogArgType(Ctx, Size); | |||
| 1749 | Args.push_back(ImplicitParamDecl::Create( | |||
| 1750 | Ctx, nullptr, SourceLocation(), | |||
| 1751 | &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy, | |||
| 1752 | ImplicitParamDecl::Other)); | |||
| 1753 | ArgTys.emplace_back(ArgTy); | |||
| 1754 | } | |||
| 1755 | ||||
| 1756 | QualType ReturnTy = Ctx.VoidTy; | |||
| 1757 | ||||
| 1758 | // The helper function has linkonce_odr linkage to enable the linker to merge | |||
| 1759 | // identical functions. To ensure the merging always happens, 'noinline' is | |||
| 1760 | // attached to the function when compiling with -Oz. | |||
| 1761 | const CGFunctionInfo &FI = | |||
| 1762 | CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args); | |||
| 1763 | llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI); | |||
| 1764 | llvm::Function *Fn = llvm::Function::Create( | |||
| 1765 | FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule()); | |||
| 1766 | Fn->setVisibility(llvm::GlobalValue::HiddenVisibility); | |||
| 1767 | CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn, /*IsThunk=*/false); | |||
| 1768 | CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn); | |||
| 1769 | Fn->setDoesNotThrow(); | |||
| 1770 | ||||
| 1771 | // Attach 'noinline' at -Oz. | |||
| 1772 | if (CGM.getCodeGenOpts().OptimizeSize == 2) | |||
| 1773 | Fn->addFnAttr(llvm::Attribute::NoInline); | |||
| 1774 | ||||
| 1775 | auto NL = ApplyDebugLocation::CreateEmpty(*this); | |||
| 1776 | StartFunction(GlobalDecl(), ReturnTy, Fn, FI, Args); | |||
| 1777 | ||||
| 1778 | // Create a scope with an artificial location for the body of this function. | |||
| 1779 | auto AL = ApplyDebugLocation::CreateArtificial(*this); | |||
| 1780 | ||||
| 1781 | CharUnits Offset; | |||
| 1782 | Address BufAddr = | |||
| 1783 | Address(Builder.CreateLoad(GetAddrOfLocalVar(Args[0]), "buf"), Int8Ty, | |||
| 1784 | BufferAlignment); | |||
| 1785 | Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()), | |||
| 1786 | Builder.CreateConstByteGEP(BufAddr, Offset++, "summary")); | |||
| 1787 | Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()), | |||
| 1788 | Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs")); | |||
| 1789 | ||||
| 1790 | unsigned I = 1; | |||
| 1791 | for (const auto &Item : Layout.Items) { | |||
| 1792 | Builder.CreateStore( | |||
| 1793 | Builder.getInt8(Item.getDescriptorByte()), | |||
| 1794 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor")); | |||
| 1795 | Builder.CreateStore( | |||
| 1796 | Builder.getInt8(Item.getSizeByte()), | |||
| 1797 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize")); | |||
| 1798 | ||||
| 1799 | CharUnits Size = Item.size(); | |||
| 1800 | if (!Size.getQuantity()) | |||
| 1801 | continue; | |||
| 1802 | ||||
| 1803 | Address Arg = GetAddrOfLocalVar(Args[I]); | |||
| 1804 | Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData"); | |||
| 1805 | Addr = | |||
| 1806 | Builder.CreateElementBitCast(Addr, Arg.getElementType(), "argDataCast"); | |||
| 1807 | Builder.CreateStore(Builder.CreateLoad(Arg), Addr); | |||
| 1808 | Offset += Size; | |||
| 1809 | ++I; | |||
| 1810 | } | |||
| 1811 | ||||
| 1812 | FinishFunction(); | |||
| 1813 | ||||
| 1814 | return Fn; | |||
| 1815 | } | |||
| 1816 | ||||
| 1817 | RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) { | |||
| 1818 | assert(E.getNumArgs() >= 2 &&(static_cast <bool> (E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? void (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1819, __extension__ __PRETTY_FUNCTION__ )) | |||
| 1819 | "__builtin_os_log_format takes at least 2 arguments")(static_cast <bool> (E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? void (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1819, __extension__ __PRETTY_FUNCTION__ )); | |||
| 1820 | ASTContext &Ctx = getContext(); | |||
| 1821 | analyze_os_log::OSLogBufferLayout Layout; | |||
| 1822 | analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout); | |||
| 1823 | Address BufAddr = EmitPointerWithAlignment(E.getArg(0)); | |||
| 1824 | llvm::SmallVector<llvm::Value *, 4> RetainableOperands; | |||
| 1825 | ||||
| 1826 | // Ignore argument 1, the format string. It is not currently used. | |||
| 1827 | CallArgList Args; | |||
| 1828 | Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy); | |||
| 1829 | ||||
| 1830 | for (const auto &Item : Layout.Items) { | |||
| 1831 | int Size = Item.getSizeByte(); | |||
| 1832 | if (!Size) | |||
| 1833 | continue; | |||
| 1834 | ||||
| 1835 | llvm::Value *ArgVal; | |||
| 1836 | ||||
| 1837 | if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) { | |||
| 1838 | uint64_t Val = 0; | |||
| 1839 | for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I) | |||
| 1840 | Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8; | |||
| 1841 | ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val)); | |||
| 1842 | } else if (const Expr *TheExpr = Item.getExpr()) { | |||
| 1843 | ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false); | |||
| 1844 | ||||
| 1845 | // If a temporary object that requires destruction after the full | |||
| 1846 | // expression is passed, push a lifetime-extended cleanup to extend its | |||
| 1847 | // lifetime to the end of the enclosing block scope. | |||
| 1848 | auto LifetimeExtendObject = [&](const Expr *E) { | |||
| 1849 | E = E->IgnoreParenCasts(); | |||
| 1850 | // Extend lifetimes of objects returned by function calls and message | |||
| 1851 | // sends. | |||
| 1852 | ||||
| 1853 | // FIXME: We should do this in other cases in which temporaries are | |||
| 1854 | // created including arguments of non-ARC types (e.g., C++ | |||
| 1855 | // temporaries). | |||
| 1856 | if (isa<CallExpr>(E) || isa<ObjCMessageExpr>(E)) | |||
| 1857 | return true; | |||
| 1858 | return false; | |||
| 1859 | }; | |||
| 1860 | ||||
| 1861 | if (TheExpr->getType()->isObjCRetainableType() && | |||
| 1862 | getLangOpts().ObjCAutoRefCount && LifetimeExtendObject(TheExpr)) { | |||
| 1863 | assert(getEvaluationKind(TheExpr->getType()) == TEK_Scalar &&(static_cast <bool> (getEvaluationKind(TheExpr->getType ()) == TEK_Scalar && "Only scalar can be a ObjC retainable type" ) ? void (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1864, __extension__ __PRETTY_FUNCTION__ )) | |||
| 1864 | "Only scalar can be a ObjC retainable type")(static_cast <bool> (getEvaluationKind(TheExpr->getType ()) == TEK_Scalar && "Only scalar can be a ObjC retainable type" ) ? void (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1864, __extension__ __PRETTY_FUNCTION__ )); | |||
| 1865 | if (!isa<Constant>(ArgVal)) { | |||
| 1866 | CleanupKind Cleanup = getARCCleanupKind(); | |||
| 1867 | QualType Ty = TheExpr->getType(); | |||
| 1868 | Address Alloca = Address::invalid(); | |||
| 1869 | Address Addr = CreateMemTemp(Ty, "os.log.arg", &Alloca); | |||
| 1870 | ArgVal = EmitARCRetain(Ty, ArgVal); | |||
| 1871 | Builder.CreateStore(ArgVal, Addr); | |||
| 1872 | pushLifetimeExtendedDestroy(Cleanup, Alloca, Ty, | |||
| 1873 | CodeGenFunction::destroyARCStrongPrecise, | |||
| 1874 | Cleanup & EHCleanup); | |||
| 1875 | ||||
| 1876 | // Push a clang.arc.use call to ensure ARC optimizer knows that the | |||
| 1877 | // argument has to be alive. | |||
| 1878 | if (CGM.getCodeGenOpts().OptimizationLevel != 0) | |||
| 1879 | pushCleanupAfterFullExpr<CallObjCArcUse>(Cleanup, ArgVal); | |||
| 1880 | } | |||
| 1881 | } | |||
| 1882 | } else { | |||
| 1883 | ArgVal = Builder.getInt32(Item.getConstValue().getQuantity()); | |||
| 1884 | } | |||
| 1885 | ||||
| 1886 | unsigned ArgValSize = | |||
| 1887 | CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType()); | |||
| 1888 | llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(), | |||
| 1889 | ArgValSize); | |||
| 1890 | ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy); | |||
| 1891 | CanQualType ArgTy = getOSLogArgType(Ctx, Size); | |||
| 1892 | // If ArgVal has type x86_fp80, zero-extend ArgVal. | |||
| 1893 | ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy)); | |||
| 1894 | Args.add(RValue::get(ArgVal), ArgTy); | |||
| 1895 | } | |||
| 1896 | ||||
| 1897 | const CGFunctionInfo &FI = | |||
| 1898 | CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args); | |||
| 1899 | llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction( | |||
| 1900 | Layout, BufAddr.getAlignment()); | |||
| 1901 | EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args); | |||
| 1902 | return RValue::get(BufAddr.getPointer()); | |||
| 1903 | } | |||
| 1904 | ||||
| 1905 | static bool isSpecialUnsignedMultiplySignedResult( | |||
| 1906 | unsigned BuiltinID, WidthAndSignedness Op1Info, WidthAndSignedness Op2Info, | |||
| 1907 | WidthAndSignedness ResultInfo) { | |||
| 1908 | return BuiltinID == Builtin::BI__builtin_mul_overflow && | |||
| 1909 | Op1Info.Width == Op2Info.Width && Op2Info.Width == ResultInfo.Width && | |||
| 1910 | !Op1Info.Signed && !Op2Info.Signed && ResultInfo.Signed; | |||
| 1911 | } | |||
| 1912 | ||||
| 1913 | static RValue EmitCheckedUnsignedMultiplySignedResult( | |||
| 1914 | CodeGenFunction &CGF, const clang::Expr *Op1, WidthAndSignedness Op1Info, | |||
| 1915 | const clang::Expr *Op2, WidthAndSignedness Op2Info, | |||
| 1916 | const clang::Expr *ResultArg, QualType ResultQTy, | |||
| 1917 | WidthAndSignedness ResultInfo) { | |||
| 1918 | assert(isSpecialUnsignedMultiplySignedResult((static_cast <bool> (isSpecialUnsignedMultiplySignedResult ( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo ) && "Cannot specialize this multiply") ? void (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1920, __extension__ __PRETTY_FUNCTION__ )) | |||
| 1919 | Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&(static_cast <bool> (isSpecialUnsignedMultiplySignedResult ( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo ) && "Cannot specialize this multiply") ? void (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1920, __extension__ __PRETTY_FUNCTION__ )) | |||
| 1920 | "Cannot specialize this multiply")(static_cast <bool> (isSpecialUnsignedMultiplySignedResult ( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo ) && "Cannot specialize this multiply") ? void (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1920, __extension__ __PRETTY_FUNCTION__ )); | |||
| 1921 | ||||
| 1922 | llvm::Value *V1 = CGF.EmitScalarExpr(Op1); | |||
| 1923 | llvm::Value *V2 = CGF.EmitScalarExpr(Op2); | |||
| 1924 | ||||
| 1925 | llvm::Value *HasOverflow; | |||
| 1926 | llvm::Value *Result = EmitOverflowIntrinsic( | |||
| 1927 | CGF, llvm::Intrinsic::umul_with_overflow, V1, V2, HasOverflow); | |||
| 1928 | ||||
| 1929 | // The intrinsic call will detect overflow when the value is > UINT_MAX, | |||
| 1930 | // however, since the original builtin had a signed result, we need to report | |||
| 1931 | // an overflow when the result is greater than INT_MAX. | |||
| 1932 | auto IntMax = llvm::APInt::getSignedMaxValue(ResultInfo.Width); | |||
| 1933 | llvm::Value *IntMaxValue = llvm::ConstantInt::get(Result->getType(), IntMax); | |||
| 1934 | ||||
| 1935 | llvm::Value *IntMaxOverflow = CGF.Builder.CreateICmpUGT(Result, IntMaxValue); | |||
| 1936 | HasOverflow = CGF.Builder.CreateOr(HasOverflow, IntMaxOverflow); | |||
| 1937 | ||||
| 1938 | bool isVolatile = | |||
| 1939 | ResultArg->getType()->getPointeeType().isVolatileQualified(); | |||
| 1940 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); | |||
| 1941 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, | |||
| 1942 | isVolatile); | |||
| 1943 | return RValue::get(HasOverflow); | |||
| 1944 | } | |||
| 1945 | ||||
| 1946 | /// Determine if a binop is a checked mixed-sign multiply we can specialize. | |||
| 1947 | static bool isSpecialMixedSignMultiply(unsigned BuiltinID, | |||
| 1948 | WidthAndSignedness Op1Info, | |||
| 1949 | WidthAndSignedness Op2Info, | |||
| 1950 | WidthAndSignedness ResultInfo) { | |||
| 1951 | return BuiltinID == Builtin::BI__builtin_mul_overflow && | |||
| 1952 | std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width && | |||
| 1953 | Op1Info.Signed != Op2Info.Signed; | |||
| 1954 | } | |||
| 1955 | ||||
| 1956 | /// Emit a checked mixed-sign multiply. This is a cheaper specialization of | |||
| 1957 | /// the generic checked-binop irgen. | |||
| 1958 | static RValue | |||
| 1959 | EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1, | |||
| 1960 | WidthAndSignedness Op1Info, const clang::Expr *Op2, | |||
| 1961 | WidthAndSignedness Op2Info, | |||
| 1962 | const clang::Expr *ResultArg, QualType ResultQTy, | |||
| 1963 | WidthAndSignedness ResultInfo) { | |||
| 1964 | assert(isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info,(static_cast <bool> (isSpecialMixedSignMultiply(Builtin ::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize") ? void ( 0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1966, __extension__ __PRETTY_FUNCTION__ )) | |||
| 1965 | Op2Info, ResultInfo) &&(static_cast <bool> (isSpecialMixedSignMultiply(Builtin ::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize") ? void ( 0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1966, __extension__ __PRETTY_FUNCTION__ )) | |||
| 1966 | "Not a mixed-sign multipliction we can specialize")(static_cast <bool> (isSpecialMixedSignMultiply(Builtin ::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize") ? void ( 0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1966, __extension__ __PRETTY_FUNCTION__ )); | |||
| 1967 | ||||
| 1968 | // Emit the signed and unsigned operands. | |||
| 1969 | const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2; | |||
| 1970 | const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1; | |||
| 1971 | llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp); | |||
| 1972 | llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp); | |||
| 1973 | unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width; | |||
| 1974 | unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width; | |||
| 1975 | ||||
| 1976 | // One of the operands may be smaller than the other. If so, [s|z]ext it. | |||
| 1977 | if (SignedOpWidth < UnsignedOpWidth) | |||
| 1978 | Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext"); | |||
| 1979 | if (UnsignedOpWidth < SignedOpWidth) | |||
| 1980 | Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext"); | |||
| 1981 | ||||
| 1982 | llvm::Type *OpTy = Signed->getType(); | |||
| 1983 | llvm::Value *Zero = llvm::Constant::getNullValue(OpTy); | |||
| 1984 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); | |||
| 1985 | llvm::Type *ResTy = ResultPtr.getElementType(); | |||
| 1986 | unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width); | |||
| 1987 | ||||
| 1988 | // Take the absolute value of the signed operand. | |||
| 1989 | llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero); | |||
| 1990 | llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed); | |||
| 1991 | llvm::Value *AbsSigned = | |||
| 1992 | CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed); | |||
| 1993 | ||||
| 1994 | // Perform a checked unsigned multiplication. | |||
| 1995 | llvm::Value *UnsignedOverflow; | |||
| 1996 | llvm::Value *UnsignedResult = | |||
| 1997 | EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned, | |||
| 1998 | Unsigned, UnsignedOverflow); | |||
| 1999 | ||||
| 2000 | llvm::Value *Overflow, *Result; | |||
| 2001 | if (ResultInfo.Signed) { | |||
| 2002 | // Signed overflow occurs if the result is greater than INT_MAX or lesser | |||
| 2003 | // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative). | |||
| 2004 | auto IntMax = | |||
| 2005 | llvm::APInt::getSignedMaxValue(ResultInfo.Width).zextOrSelf(OpWidth); | |||
| 2006 | llvm::Value *MaxResult = | |||
| 2007 | CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax), | |||
| 2008 | CGF.Builder.CreateZExt(IsNegative, OpTy)); | |||
| 2009 | llvm::Value *SignedOverflow = | |||
| 2010 | CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult); | |||
| 2011 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow); | |||
| 2012 | ||||
| 2013 | // Prepare the signed result (possibly by negating it). | |||
| 2014 | llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult); | |||
| 2015 | llvm::Value *SignedResult = | |||
| 2016 | CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult); | |||
| 2017 | Result = CGF.Builder.CreateTrunc(SignedResult, ResTy); | |||
| 2018 | } else { | |||
| 2019 | // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX. | |||
| 2020 | llvm::Value *Underflow = CGF.Builder.CreateAnd( | |||
| 2021 | IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult)); | |||
| 2022 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow); | |||
| 2023 | if (ResultInfo.Width < OpWidth) { | |||
| 2024 | auto IntMax = | |||
| 2025 | llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth); | |||
| 2026 | llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT( | |||
| 2027 | UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax)); | |||
| 2028 | Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow); | |||
| 2029 | } | |||
| 2030 | ||||
| 2031 | // Negate the product if it would be negative in infinite precision. | |||
| 2032 | Result = CGF.Builder.CreateSelect( | |||
| 2033 | IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult); | |||
| 2034 | ||||
| 2035 | Result = CGF.Builder.CreateTrunc(Result, ResTy); | |||
| 2036 | } | |||
| 2037 | assert(Overflow && Result && "Missing overflow or result")(static_cast <bool> (Overflow && Result && "Missing overflow or result") ? void (0) : __assert_fail ("Overflow && Result && \"Missing overflow or result\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2037, __extension__ __PRETTY_FUNCTION__ )); | |||
| 2038 | ||||
| 2039 | bool isVolatile = | |||
| 2040 | ResultArg->getType()->getPointeeType().isVolatileQualified(); | |||
| 2041 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, | |||
| 2042 | isVolatile); | |||
| 2043 | return RValue::get(Overflow); | |||
| 2044 | } | |||
| 2045 | ||||
| 2046 | static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType, | |||
| 2047 | LValue RecordLV, CharUnits Align, | |||
| 2048 | llvm::FunctionCallee Func, int Lvl) { | |||
| 2049 | ASTContext &Context = CGF.getContext(); | |||
| 2050 | RecordDecl *RD = RType->castAs<RecordType>()->getDecl()->getDefinition(); | |||
| 2051 | std::string Pad = std::string(Lvl * 4, ' '); | |||
| 2052 | std::string ElementPad = std::string((Lvl + 1) * 4, ' '); | |||
| 2053 | ||||
| 2054 | PrintingPolicy Policy(Context.getLangOpts()); | |||
| 2055 | Policy.AnonymousTagLocations = false; | |||
| 2056 | Value *GString = CGF.Builder.CreateGlobalStringPtr( | |||
| 2057 | llvm::Twine(Pad).concat(RType.getAsString(Policy)).concat(" {\n").str()); | |||
| 2058 | Value *Res = CGF.Builder.CreateCall(Func, {GString}); | |||
| 2059 | ||||
| 2060 | static llvm::DenseMap<QualType, const char *> Types; | |||
| 2061 | if (Types.empty()) { | |||
| 2062 | Types[Context.CharTy] = "%c"; | |||
| 2063 | Types[Context.BoolTy] = "%d"; | |||
| 2064 | Types[Context.SignedCharTy] = "%hhd"; | |||
| 2065 | Types[Context.UnsignedCharTy] = "%hhu"; | |||
| 2066 | Types[Context.IntTy] = "%d"; | |||
| 2067 | Types[Context.UnsignedIntTy] = "%u"; | |||
| 2068 | Types[Context.LongTy] = "%ld"; | |||
| 2069 | Types[Context.UnsignedLongTy] = "%lu"; | |||
| 2070 | Types[Context.LongLongTy] = "%lld"; | |||
| 2071 | Types[Context.UnsignedLongLongTy] = "%llu"; | |||
| 2072 | Types[Context.ShortTy] = "%hd"; | |||
| 2073 | Types[Context.UnsignedShortTy] = "%hu"; | |||
| 2074 | Types[Context.VoidPtrTy] = "%p"; | |||
| 2075 | Types[Context.FloatTy] = "%f"; | |||
| 2076 | Types[Context.DoubleTy] = "%f"; | |||
| 2077 | Types[Context.LongDoubleTy] = "%Lf"; | |||
| 2078 | Types[Context.getPointerType(Context.CharTy)] = "%s"; | |||
| 2079 | Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s"; | |||
| 2080 | } | |||
| 2081 | ||||
| 2082 | for (const auto *FD : RD->fields()) { | |||
| 2083 | Value *TmpRes = nullptr; | |||
| 2084 | ||||
| 2085 | std::string Format = llvm::Twine(ElementPad) | |||
| 2086 | .concat(FD->getType().getAsString()) | |||
| 2087 | .concat(llvm::Twine(' ')) | |||
| 2088 | .concat(FD->getNameAsString()) | |||
| 2089 | .str(); | |||
| 2090 | ||||
| 2091 | if (FD->isBitField()) { | |||
| 2092 | unsigned BitfieldWidth = FD->getBitWidthValue(CGF.getContext()); | |||
| 2093 | ||||
| 2094 | // If current field is a unnamed bitfield, we should dump only one ' ' | |||
| 2095 | // between type-name and ':' | |||
| 2096 | if (!FD->getDeclName().isEmpty()) | |||
| 2097 | Format += ' '; | |||
| 2098 | Format += llvm::Twine(": ").concat(llvm::Twine(BitfieldWidth)).str(); | |||
| 2099 | ||||
| 2100 | // If current field is a zero-width bitfield, we just dump a string like | |||
| 2101 | // 'type-name : 0' | |||
| 2102 | if (FD->isZeroSize(CGF.getContext())) { | |||
| 2103 | Format += "\n"; | |||
| 2104 | GString = CGF.Builder.CreateGlobalStringPtr(Format); | |||
| 2105 | TmpRes = CGF.Builder.CreateCall(Func, {GString}); | |||
| 2106 | Res = CGF.Builder.CreateAdd(Res, TmpRes); | |||
| 2107 | continue; | |||
| 2108 | } | |||
| 2109 | } | |||
| 2110 | ||||
| 2111 | LValue FieldLV = CGF.EmitLValueForField(RecordLV, FD); | |||
| 2112 | QualType CanonicalType = | |||
| 2113 | FD->getType().getUnqualifiedType().getCanonicalType(); | |||
| 2114 | ||||
| 2115 | // We check whether we are in a recursive type | |||
| 2116 | if (CanonicalType->isRecordType()) { | |||
| 2117 | TmpRes = dumpRecord(CGF, CanonicalType, FieldLV, Align, Func, Lvl + 1); | |||
| 2118 | Res = CGF.Builder.CreateAdd(TmpRes, Res); | |||
| 2119 | continue; | |||
| 2120 | } | |||
| 2121 | ||||
| 2122 | // We try to determine the best format to print the current field | |||
| 2123 | const char *TypeFormat = Types.find(CanonicalType) == Types.end() | |||
| 2124 | ? Types[Context.VoidPtrTy] | |||
| 2125 | : Types[CanonicalType]; | |||
| 2126 | ||||
| 2127 | GString = CGF.Builder.CreateGlobalStringPtr(llvm::Twine(Format) | |||
| 2128 | .concat(" = ") | |||
| 2129 | .concat(TypeFormat) | |||
| 2130 | .concat(llvm::Twine('\n')) | |||
| 2131 | .str()); | |||
| 2132 | ||||
| 2133 | RValue RV = FD->isBitField() | |||
| 2134 | ? CGF.EmitLoadOfBitfieldLValue(FieldLV, FD->getLocation()) | |||
| 2135 | : CGF.EmitLoadOfLValue(FieldLV, FD->getLocation()); | |||
| 2136 | TmpRes = CGF.Builder.CreateCall(Func, {GString, RV.getScalarVal()}); | |||
| 2137 | Res = CGF.Builder.CreateAdd(Res, TmpRes); | |||
| 2138 | } | |||
| 2139 | ||||
| 2140 | GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n"); | |||
| 2141 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); | |||
| 2142 | Res = CGF.Builder.CreateAdd(Res, TmpRes); | |||
| 2143 | return Res; | |||
| 2144 | } | |||
| 2145 | ||||
| 2146 | static bool | |||
| 2147 | TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty, | |||
| 2148 | llvm::SmallPtrSetImpl<const Decl *> &Seen) { | |||
| 2149 | if (const auto *Arr = Ctx.getAsArrayType(Ty)) | |||
| 2150 | Ty = Ctx.getBaseElementType(Arr); | |||
| 2151 | ||||
| 2152 | const auto *Record = Ty->getAsCXXRecordDecl(); | |||
| 2153 | if (!Record) | |||
| 2154 | return false; | |||
| 2155 | ||||
| 2156 | // We've already checked this type, or are in the process of checking it. | |||
| 2157 | if (!Seen.insert(Record).second) | |||
| 2158 | return false; | |||
| 2159 | ||||
| 2160 | assert(Record->hasDefinition() &&(static_cast <bool> (Record->hasDefinition() && "Incomplete types should already be diagnosed") ? void (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2161, __extension__ __PRETTY_FUNCTION__ )) | |||
| 2161 | "Incomplete types should already be diagnosed")(static_cast <bool> (Record->hasDefinition() && "Incomplete types should already be diagnosed") ? void (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2161, __extension__ __PRETTY_FUNCTION__ )); | |||
| 2162 | ||||
| 2163 | if (Record->isDynamicClass()) | |||
| 2164 | return true; | |||
| 2165 | ||||
| 2166 | for (FieldDecl *F : Record->fields()) { | |||
| 2167 | if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen)) | |||
| 2168 | return true; | |||
| 2169 | } | |||
| 2170 | return false; | |||
| 2171 | } | |||
| 2172 | ||||
| 2173 | /// Determine if the specified type requires laundering by checking if it is a | |||
| 2174 | /// dynamic class type or contains a subobject which is a dynamic class type. | |||
| 2175 | static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) { | |||
| 2176 | if (!CGM.getCodeGenOpts().StrictVTablePointers) | |||
| 2177 | return false; | |||
| 2178 | llvm::SmallPtrSet<const Decl *, 16> Seen; | |||
| 2179 | return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen); | |||
| 2180 | } | |||
| 2181 | ||||
| 2182 | RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) { | |||
| 2183 | llvm::Value *Src = EmitScalarExpr(E->getArg(0)); | |||
| 2184 | llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1)); | |||
| 2185 | ||||
| 2186 | // The builtin's shift arg may have a different type than the source arg and | |||
| 2187 | // result, but the LLVM intrinsic uses the same type for all values. | |||
| 2188 | llvm::Type *Ty = Src->getType(); | |||
| 2189 | ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false); | |||
| 2190 | ||||
| 2191 | // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same. | |||
| 2192 | unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl; | |||
| 2193 | Function *F = CGM.getIntrinsic(IID, Ty); | |||
| 2194 | return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt })); | |||
| 2195 | } | |||
| 2196 | ||||
| 2197 | // Map math builtins for long-double to f128 version. | |||
| 2198 | static unsigned mutateLongDoubleBuiltin(unsigned BuiltinID) { | |||
| 2199 | switch (BuiltinID) { | |||
| 2200 | #define MUTATE_LDBL(func) \ | |||
| 2201 | case Builtin::BI__builtin_##func##l: \ | |||
| 2202 | return Builtin::BI__builtin_##func##f128; | |||
| 2203 | MUTATE_LDBL(sqrt) | |||
| 2204 | MUTATE_LDBL(cbrt) | |||
| 2205 | MUTATE_LDBL(fabs) | |||
| 2206 | MUTATE_LDBL(log) | |||
| 2207 | MUTATE_LDBL(log2) | |||
| 2208 | MUTATE_LDBL(log10) | |||
| 2209 | MUTATE_LDBL(log1p) | |||
| 2210 | MUTATE_LDBL(logb) | |||
| 2211 | MUTATE_LDBL(exp) | |||
| 2212 | MUTATE_LDBL(exp2) | |||
| 2213 | MUTATE_LDBL(expm1) | |||
| 2214 | MUTATE_LDBL(fdim) | |||
| 2215 | MUTATE_LDBL(hypot) | |||
| 2216 | MUTATE_LDBL(ilogb) | |||
| 2217 | MUTATE_LDBL(pow) | |||
| 2218 | MUTATE_LDBL(fmin) | |||
| 2219 | MUTATE_LDBL(fmax) | |||
| 2220 | MUTATE_LDBL(ceil) | |||
| 2221 | MUTATE_LDBL(trunc) | |||
| 2222 | MUTATE_LDBL(rint) | |||
| 2223 | MUTATE_LDBL(nearbyint) | |||
| 2224 | MUTATE_LDBL(round) | |||
| 2225 | MUTATE_LDBL(floor) | |||
| 2226 | MUTATE_LDBL(lround) | |||
| 2227 | MUTATE_LDBL(llround) | |||
| 2228 | MUTATE_LDBL(lrint) | |||
| 2229 | MUTATE_LDBL(llrint) | |||
| 2230 | MUTATE_LDBL(fmod) | |||
| 2231 | MUTATE_LDBL(modf) | |||
| 2232 | MUTATE_LDBL(nan) | |||
| 2233 | MUTATE_LDBL(nans) | |||
| 2234 | MUTATE_LDBL(inf) | |||
| 2235 | MUTATE_LDBL(fma) | |||
| 2236 | MUTATE_LDBL(sin) | |||
| 2237 | MUTATE_LDBL(cos) | |||
| 2238 | MUTATE_LDBL(tan) | |||
| 2239 | MUTATE_LDBL(sinh) | |||
| 2240 | MUTATE_LDBL(cosh) | |||
| 2241 | MUTATE_LDBL(tanh) | |||
| 2242 | MUTATE_LDBL(asin) | |||
| 2243 | MUTATE_LDBL(acos) | |||
| 2244 | MUTATE_LDBL(atan) | |||
| 2245 | MUTATE_LDBL(asinh) | |||
| 2246 | MUTATE_LDBL(acosh) | |||
| 2247 | MUTATE_LDBL(atanh) | |||
| 2248 | MUTATE_LDBL(atan2) | |||
| 2249 | MUTATE_LDBL(erf) | |||
| 2250 | MUTATE_LDBL(erfc) | |||
| 2251 | MUTATE_LDBL(ldexp) | |||
| 2252 | MUTATE_LDBL(frexp) | |||
| 2253 | MUTATE_LDBL(huge_val) | |||
| 2254 | MUTATE_LDBL(copysign) | |||
| 2255 | MUTATE_LDBL(nextafter) | |||
| 2256 | MUTATE_LDBL(nexttoward) | |||
| 2257 | MUTATE_LDBL(remainder) | |||
| 2258 | MUTATE_LDBL(remquo) | |||
| 2259 | MUTATE_LDBL(scalbln) | |||
| 2260 | MUTATE_LDBL(scalbn) | |||
| 2261 | MUTATE_LDBL(tgamma) | |||
| 2262 | MUTATE_LDBL(lgamma) | |||
| 2263 | #undef MUTATE_LDBL | |||
| 2264 | default: | |||
| 2265 | return BuiltinID; | |||
| 2266 | } | |||
| 2267 | } | |||
| 2268 | ||||
| 2269 | RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, | |||
| 2270 | const CallExpr *E, | |||
| 2271 | ReturnValueSlot ReturnValue) { | |||
| 2272 | const FunctionDecl *FD = GD.getDecl()->getAsFunction(); | |||
| 2273 | // See if we can constant fold this builtin. If so, don't emit it at all. | |||
| 2274 | Expr::EvalResult Result; | |||
| 2275 | if (E->EvaluateAsRValue(Result, CGM.getContext()) && | |||
| 2276 | !Result.hasSideEffects()) { | |||
| 2277 | if (Result.Val.isInt()) | |||
| 2278 | return RValue::get(llvm::ConstantInt::get(getLLVMContext(), | |||
| 2279 | Result.Val.getInt())); | |||
| 2280 | if (Result.Val.isFloat()) | |||
| 2281 | return RValue::get(llvm::ConstantFP::get(getLLVMContext(), | |||
| 2282 | Result.Val.getFloat())); | |||
| 2283 | } | |||
| 2284 | ||||
| 2285 | // If current long-double semantics is IEEE 128-bit, replace math builtins | |||
| 2286 | // of long-double with f128 equivalent. | |||
| 2287 | // TODO: This mutation should also be applied to other targets other than PPC, | |||
| 2288 | // after backend supports IEEE 128-bit style libcalls. | |||
| 2289 | if (getTarget().getTriple().isPPC64() && | |||
| 2290 | &getTarget().getLongDoubleFormat() == &llvm::APFloat::IEEEquad()) | |||
| 2291 | BuiltinID = mutateLongDoubleBuiltin(BuiltinID); | |||
| 2292 | ||||
| 2293 | // If the builtin has been declared explicitly with an assembler label, | |||
| 2294 | // disable the specialized emitting below. Ideally we should communicate the | |||
| 2295 | // rename in IR, or at least avoid generating the intrinsic calls that are | |||
| 2296 | // likely to get lowered to the renamed library functions. | |||
| 2297 | const unsigned BuiltinIDIfNoAsmLabel = | |||
| 2298 | FD->hasAttr<AsmLabelAttr>() ? 0 : BuiltinID; | |||
| 2299 | ||||
| 2300 | // There are LLVM math intrinsics/instructions corresponding to math library | |||
| 2301 | // functions except the LLVM op will never set errno while the math library | |||
| 2302 | // might. Also, math builtins have the same semantics as their math library | |||
| 2303 | // twins. Thus, we can transform math library and builtin calls to their | |||
| 2304 | // LLVM counterparts if the call is marked 'const' (known to never set errno). | |||
| 2305 | if (FD->hasAttr<ConstAttr>()) { | |||
| 2306 | switch (BuiltinIDIfNoAsmLabel) { | |||
| 2307 | case Builtin::BIceil: | |||
| 2308 | case Builtin::BIceilf: | |||
| 2309 | case Builtin::BIceill: | |||
| 2310 | case Builtin::BI__builtin_ceil: | |||
| 2311 | case Builtin::BI__builtin_ceilf: | |||
| 2312 | case Builtin::BI__builtin_ceilf16: | |||
| 2313 | case Builtin::BI__builtin_ceill: | |||
| 2314 | case Builtin::BI__builtin_ceilf128: | |||
| 2315 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2316 | Intrinsic::ceil, | |||
| 2317 | Intrinsic::experimental_constrained_ceil)); | |||
| 2318 | ||||
| 2319 | case Builtin::BIcopysign: | |||
| 2320 | case Builtin::BIcopysignf: | |||
| 2321 | case Builtin::BIcopysignl: | |||
| 2322 | case Builtin::BI__builtin_copysign: | |||
| 2323 | case Builtin::BI__builtin_copysignf: | |||
| 2324 | case Builtin::BI__builtin_copysignf16: | |||
| 2325 | case Builtin::BI__builtin_copysignl: | |||
| 2326 | case Builtin::BI__builtin_copysignf128: | |||
| 2327 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign)); | |||
| 2328 | ||||
| 2329 | case Builtin::BIcos: | |||
| 2330 | case Builtin::BIcosf: | |||
| 2331 | case Builtin::BIcosl: | |||
| 2332 | case Builtin::BI__builtin_cos: | |||
| 2333 | case Builtin::BI__builtin_cosf: | |||
| 2334 | case Builtin::BI__builtin_cosf16: | |||
| 2335 | case Builtin::BI__builtin_cosl: | |||
| 2336 | case Builtin::BI__builtin_cosf128: | |||
| 2337 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2338 | Intrinsic::cos, | |||
| 2339 | Intrinsic::experimental_constrained_cos)); | |||
| 2340 | ||||
| 2341 | case Builtin::BIexp: | |||
| 2342 | case Builtin::BIexpf: | |||
| 2343 | case Builtin::BIexpl: | |||
| 2344 | case Builtin::BI__builtin_exp: | |||
| 2345 | case Builtin::BI__builtin_expf: | |||
| 2346 | case Builtin::BI__builtin_expf16: | |||
| 2347 | case Builtin::BI__builtin_expl: | |||
| 2348 | case Builtin::BI__builtin_expf128: | |||
| 2349 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2350 | Intrinsic::exp, | |||
| 2351 | Intrinsic::experimental_constrained_exp)); | |||
| 2352 | ||||
| 2353 | case Builtin::BIexp2: | |||
| 2354 | case Builtin::BIexp2f: | |||
| 2355 | case Builtin::BIexp2l: | |||
| 2356 | case Builtin::BI__builtin_exp2: | |||
| 2357 | case Builtin::BI__builtin_exp2f: | |||
| 2358 | case Builtin::BI__builtin_exp2f16: | |||
| 2359 | case Builtin::BI__builtin_exp2l: | |||
| 2360 | case Builtin::BI__builtin_exp2f128: | |||
| 2361 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2362 | Intrinsic::exp2, | |||
| 2363 | Intrinsic::experimental_constrained_exp2)); | |||
| 2364 | ||||
| 2365 | case Builtin::BIfabs: | |||
| 2366 | case Builtin::BIfabsf: | |||
| 2367 | case Builtin::BIfabsl: | |||
| 2368 | case Builtin::BI__builtin_fabs: | |||
| 2369 | case Builtin::BI__builtin_fabsf: | |||
| 2370 | case Builtin::BI__builtin_fabsf16: | |||
| 2371 | case Builtin::BI__builtin_fabsl: | |||
| 2372 | case Builtin::BI__builtin_fabsf128: | |||
| 2373 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs)); | |||
| 2374 | ||||
| 2375 | case Builtin::BIfloor: | |||
| 2376 | case Builtin::BIfloorf: | |||
| 2377 | case Builtin::BIfloorl: | |||
| 2378 | case Builtin::BI__builtin_floor: | |||
| 2379 | case Builtin::BI__builtin_floorf: | |||
| 2380 | case Builtin::BI__builtin_floorf16: | |||
| 2381 | case Builtin::BI__builtin_floorl: | |||
| 2382 | case Builtin::BI__builtin_floorf128: | |||
| 2383 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2384 | Intrinsic::floor, | |||
| 2385 | Intrinsic::experimental_constrained_floor)); | |||
| 2386 | ||||
| 2387 | case Builtin::BIfma: | |||
| 2388 | case Builtin::BIfmaf: | |||
| 2389 | case Builtin::BIfmal: | |||
| 2390 | case Builtin::BI__builtin_fma: | |||
| 2391 | case Builtin::BI__builtin_fmaf: | |||
| 2392 | case Builtin::BI__builtin_fmaf16: | |||
| 2393 | case Builtin::BI__builtin_fmal: | |||
| 2394 | case Builtin::BI__builtin_fmaf128: | |||
| 2395 | return RValue::get(emitTernaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2396 | Intrinsic::fma, | |||
| 2397 | Intrinsic::experimental_constrained_fma)); | |||
| 2398 | ||||
| 2399 | case Builtin::BIfmax: | |||
| 2400 | case Builtin::BIfmaxf: | |||
| 2401 | case Builtin::BIfmaxl: | |||
| 2402 | case Builtin::BI__builtin_fmax: | |||
| 2403 | case Builtin::BI__builtin_fmaxf: | |||
| 2404 | case Builtin::BI__builtin_fmaxf16: | |||
| 2405 | case Builtin::BI__builtin_fmaxl: | |||
| 2406 | case Builtin::BI__builtin_fmaxf128: | |||
| 2407 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2408 | Intrinsic::maxnum, | |||
| 2409 | Intrinsic::experimental_constrained_maxnum)); | |||
| 2410 | ||||
| 2411 | case Builtin::BIfmin: | |||
| 2412 | case Builtin::BIfminf: | |||
| 2413 | case Builtin::BIfminl: | |||
| 2414 | case Builtin::BI__builtin_fmin: | |||
| 2415 | case Builtin::BI__builtin_fminf: | |||
| 2416 | case Builtin::BI__builtin_fminf16: | |||
| 2417 | case Builtin::BI__builtin_fminl: | |||
| 2418 | case Builtin::BI__builtin_fminf128: | |||
| 2419 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2420 | Intrinsic::minnum, | |||
| 2421 | Intrinsic::experimental_constrained_minnum)); | |||
| 2422 | ||||
| 2423 | // fmod() is a special-case. It maps to the frem instruction rather than an | |||
| 2424 | // LLVM intrinsic. | |||
| 2425 | case Builtin::BIfmod: | |||
| 2426 | case Builtin::BIfmodf: | |||
| 2427 | case Builtin::BIfmodl: | |||
| 2428 | case Builtin::BI__builtin_fmod: | |||
| 2429 | case Builtin::BI__builtin_fmodf: | |||
| 2430 | case Builtin::BI__builtin_fmodf16: | |||
| 2431 | case Builtin::BI__builtin_fmodl: | |||
| 2432 | case Builtin::BI__builtin_fmodf128: { | |||
| 2433 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 2434 | Value *Arg1 = EmitScalarExpr(E->getArg(0)); | |||
| 2435 | Value *Arg2 = EmitScalarExpr(E->getArg(1)); | |||
| 2436 | return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod")); | |||
| 2437 | } | |||
| 2438 | ||||
| 2439 | case Builtin::BIlog: | |||
| 2440 | case Builtin::BIlogf: | |||
| 2441 | case Builtin::BIlogl: | |||
| 2442 | case Builtin::BI__builtin_log: | |||
| 2443 | case Builtin::BI__builtin_logf: | |||
| 2444 | case Builtin::BI__builtin_logf16: | |||
| 2445 | case Builtin::BI__builtin_logl: | |||
| 2446 | case Builtin::BI__builtin_logf128: | |||
| 2447 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2448 | Intrinsic::log, | |||
| 2449 | Intrinsic::experimental_constrained_log)); | |||
| 2450 | ||||
| 2451 | case Builtin::BIlog10: | |||
| 2452 | case Builtin::BIlog10f: | |||
| 2453 | case Builtin::BIlog10l: | |||
| 2454 | case Builtin::BI__builtin_log10: | |||
| 2455 | case Builtin::BI__builtin_log10f: | |||
| 2456 | case Builtin::BI__builtin_log10f16: | |||
| 2457 | case Builtin::BI__builtin_log10l: | |||
| 2458 | case Builtin::BI__builtin_log10f128: | |||
| 2459 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2460 | Intrinsic::log10, | |||
| 2461 | Intrinsic::experimental_constrained_log10)); | |||
| 2462 | ||||
| 2463 | case Builtin::BIlog2: | |||
| 2464 | case Builtin::BIlog2f: | |||
| 2465 | case Builtin::BIlog2l: | |||
| 2466 | case Builtin::BI__builtin_log2: | |||
| 2467 | case Builtin::BI__builtin_log2f: | |||
| 2468 | case Builtin::BI__builtin_log2f16: | |||
| 2469 | case Builtin::BI__builtin_log2l: | |||
| 2470 | case Builtin::BI__builtin_log2f128: | |||
| 2471 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2472 | Intrinsic::log2, | |||
| 2473 | Intrinsic::experimental_constrained_log2)); | |||
| 2474 | ||||
| 2475 | case Builtin::BInearbyint: | |||
| 2476 | case Builtin::BInearbyintf: | |||
| 2477 | case Builtin::BInearbyintl: | |||
| 2478 | case Builtin::BI__builtin_nearbyint: | |||
| 2479 | case Builtin::BI__builtin_nearbyintf: | |||
| 2480 | case Builtin::BI__builtin_nearbyintl: | |||
| 2481 | case Builtin::BI__builtin_nearbyintf128: | |||
| 2482 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2483 | Intrinsic::nearbyint, | |||
| 2484 | Intrinsic::experimental_constrained_nearbyint)); | |||
| 2485 | ||||
| 2486 | case Builtin::BIpow: | |||
| 2487 | case Builtin::BIpowf: | |||
| 2488 | case Builtin::BIpowl: | |||
| 2489 | case Builtin::BI__builtin_pow: | |||
| 2490 | case Builtin::BI__builtin_powf: | |||
| 2491 | case Builtin::BI__builtin_powf16: | |||
| 2492 | case Builtin::BI__builtin_powl: | |||
| 2493 | case Builtin::BI__builtin_powf128: | |||
| 2494 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2495 | Intrinsic::pow, | |||
| 2496 | Intrinsic::experimental_constrained_pow)); | |||
| 2497 | ||||
| 2498 | case Builtin::BIrint: | |||
| 2499 | case Builtin::BIrintf: | |||
| 2500 | case Builtin::BIrintl: | |||
| 2501 | case Builtin::BI__builtin_rint: | |||
| 2502 | case Builtin::BI__builtin_rintf: | |||
| 2503 | case Builtin::BI__builtin_rintf16: | |||
| 2504 | case Builtin::BI__builtin_rintl: | |||
| 2505 | case Builtin::BI__builtin_rintf128: | |||
| 2506 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2507 | Intrinsic::rint, | |||
| 2508 | Intrinsic::experimental_constrained_rint)); | |||
| 2509 | ||||
| 2510 | case Builtin::BIround: | |||
| 2511 | case Builtin::BIroundf: | |||
| 2512 | case Builtin::BIroundl: | |||
| 2513 | case Builtin::BI__builtin_round: | |||
| 2514 | case Builtin::BI__builtin_roundf: | |||
| 2515 | case Builtin::BI__builtin_roundf16: | |||
| 2516 | case Builtin::BI__builtin_roundl: | |||
| 2517 | case Builtin::BI__builtin_roundf128: | |||
| 2518 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2519 | Intrinsic::round, | |||
| 2520 | Intrinsic::experimental_constrained_round)); | |||
| 2521 | ||||
| 2522 | case Builtin::BIsin: | |||
| 2523 | case Builtin::BIsinf: | |||
| 2524 | case Builtin::BIsinl: | |||
| 2525 | case Builtin::BI__builtin_sin: | |||
| 2526 | case Builtin::BI__builtin_sinf: | |||
| 2527 | case Builtin::BI__builtin_sinf16: | |||
| 2528 | case Builtin::BI__builtin_sinl: | |||
| 2529 | case Builtin::BI__builtin_sinf128: | |||
| 2530 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2531 | Intrinsic::sin, | |||
| 2532 | Intrinsic::experimental_constrained_sin)); | |||
| 2533 | ||||
| 2534 | case Builtin::BIsqrt: | |||
| 2535 | case Builtin::BIsqrtf: | |||
| 2536 | case Builtin::BIsqrtl: | |||
| 2537 | case Builtin::BI__builtin_sqrt: | |||
| 2538 | case Builtin::BI__builtin_sqrtf: | |||
| 2539 | case Builtin::BI__builtin_sqrtf16: | |||
| 2540 | case Builtin::BI__builtin_sqrtl: | |||
| 2541 | case Builtin::BI__builtin_sqrtf128: | |||
| 2542 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2543 | Intrinsic::sqrt, | |||
| 2544 | Intrinsic::experimental_constrained_sqrt)); | |||
| 2545 | ||||
| 2546 | case Builtin::BItrunc: | |||
| 2547 | case Builtin::BItruncf: | |||
| 2548 | case Builtin::BItruncl: | |||
| 2549 | case Builtin::BI__builtin_trunc: | |||
| 2550 | case Builtin::BI__builtin_truncf: | |||
| 2551 | case Builtin::BI__builtin_truncf16: | |||
| 2552 | case Builtin::BI__builtin_truncl: | |||
| 2553 | case Builtin::BI__builtin_truncf128: | |||
| 2554 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, | |||
| 2555 | Intrinsic::trunc, | |||
| 2556 | Intrinsic::experimental_constrained_trunc)); | |||
| 2557 | ||||
| 2558 | case Builtin::BIlround: | |||
| 2559 | case Builtin::BIlroundf: | |||
| 2560 | case Builtin::BIlroundl: | |||
| 2561 | case Builtin::BI__builtin_lround: | |||
| 2562 | case Builtin::BI__builtin_lroundf: | |||
| 2563 | case Builtin::BI__builtin_lroundl: | |||
| 2564 | case Builtin::BI__builtin_lroundf128: | |||
| 2565 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( | |||
| 2566 | *this, E, Intrinsic::lround, | |||
| 2567 | Intrinsic::experimental_constrained_lround)); | |||
| 2568 | ||||
| 2569 | case Builtin::BIllround: | |||
| 2570 | case Builtin::BIllroundf: | |||
| 2571 | case Builtin::BIllroundl: | |||
| 2572 | case Builtin::BI__builtin_llround: | |||
| 2573 | case Builtin::BI__builtin_llroundf: | |||
| 2574 | case Builtin::BI__builtin_llroundl: | |||
| 2575 | case Builtin::BI__builtin_llroundf128: | |||
| 2576 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( | |||
| 2577 | *this, E, Intrinsic::llround, | |||
| 2578 | Intrinsic::experimental_constrained_llround)); | |||
| 2579 | ||||
| 2580 | case Builtin::BIlrint: | |||
| 2581 | case Builtin::BIlrintf: | |||
| 2582 | case Builtin::BIlrintl: | |||
| 2583 | case Builtin::BI__builtin_lrint: | |||
| 2584 | case Builtin::BI__builtin_lrintf: | |||
| 2585 | case Builtin::BI__builtin_lrintl: | |||
| 2586 | case Builtin::BI__builtin_lrintf128: | |||
| 2587 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( | |||
| 2588 | *this, E, Intrinsic::lrint, | |||
| 2589 | Intrinsic::experimental_constrained_lrint)); | |||
| 2590 | ||||
| 2591 | case Builtin::BIllrint: | |||
| 2592 | case Builtin::BIllrintf: | |||
| 2593 | case Builtin::BIllrintl: | |||
| 2594 | case Builtin::BI__builtin_llrint: | |||
| 2595 | case Builtin::BI__builtin_llrintf: | |||
| 2596 | case Builtin::BI__builtin_llrintl: | |||
| 2597 | case Builtin::BI__builtin_llrintf128: | |||
| 2598 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( | |||
| 2599 | *this, E, Intrinsic::llrint, | |||
| 2600 | Intrinsic::experimental_constrained_llrint)); | |||
| 2601 | ||||
| 2602 | default: | |||
| 2603 | break; | |||
| 2604 | } | |||
| 2605 | } | |||
| 2606 | ||||
| 2607 | switch (BuiltinIDIfNoAsmLabel) { | |||
| 2608 | default: break; | |||
| 2609 | case Builtin::BI__builtin___CFStringMakeConstantString: | |||
| 2610 | case Builtin::BI__builtin___NSStringMakeConstantString: | |||
| 2611 | return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType())); | |||
| 2612 | case Builtin::BI__builtin_stdarg_start: | |||
| 2613 | case Builtin::BI__builtin_va_start: | |||
| 2614 | case Builtin::BI__va_start: | |||
| 2615 | case Builtin::BI__builtin_va_end: | |||
| 2616 | return RValue::get( | |||
| 2617 | EmitVAStartEnd(BuiltinID == Builtin::BI__va_start | |||
| 2618 | ? EmitScalarExpr(E->getArg(0)) | |||
| 2619 | : EmitVAListRef(E->getArg(0)).getPointer(), | |||
| 2620 | BuiltinID != Builtin::BI__builtin_va_end)); | |||
| 2621 | case Builtin::BI__builtin_va_copy: { | |||
| 2622 | Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer(); | |||
| 2623 | Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer(); | |||
| 2624 | ||||
| 2625 | llvm::Type *Type = Int8PtrTy; | |||
| 2626 | ||||
| 2627 | DstPtr = Builder.CreateBitCast(DstPtr, Type); | |||
| 2628 | SrcPtr = Builder.CreateBitCast(SrcPtr, Type); | |||
| 2629 | return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy), | |||
| 2630 | {DstPtr, SrcPtr})); | |||
| 2631 | } | |||
| 2632 | case Builtin::BI__builtin_abs: | |||
| 2633 | case Builtin::BI__builtin_labs: | |||
| 2634 | case Builtin::BI__builtin_llabs: { | |||
| 2635 | // X < 0 ? -X : X | |||
| 2636 | // The negation has 'nsw' because abs of INT_MIN is undefined. | |||
| 2637 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2638 | Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg"); | |||
| 2639 | Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType()); | |||
| 2640 | Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond"); | |||
| 2641 | Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs"); | |||
| 2642 | return RValue::get(Result); | |||
| 2643 | } | |||
| 2644 | case Builtin::BI__builtin_complex: { | |||
| 2645 | Value *Real = EmitScalarExpr(E->getArg(0)); | |||
| 2646 | Value *Imag = EmitScalarExpr(E->getArg(1)); | |||
| 2647 | return RValue::getComplex({Real, Imag}); | |||
| 2648 | } | |||
| 2649 | case Builtin::BI__builtin_conj: | |||
| 2650 | case Builtin::BI__builtin_conjf: | |||
| 2651 | case Builtin::BI__builtin_conjl: | |||
| 2652 | case Builtin::BIconj: | |||
| 2653 | case Builtin::BIconjf: | |||
| 2654 | case Builtin::BIconjl: { | |||
| 2655 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
| 2656 | Value *Real = ComplexVal.first; | |||
| 2657 | Value *Imag = ComplexVal.second; | |||
| 2658 | Imag = Builder.CreateFNeg(Imag, "neg"); | |||
| 2659 | return RValue::getComplex(std::make_pair(Real, Imag)); | |||
| 2660 | } | |||
| 2661 | case Builtin::BI__builtin_creal: | |||
| 2662 | case Builtin::BI__builtin_crealf: | |||
| 2663 | case Builtin::BI__builtin_creall: | |||
| 2664 | case Builtin::BIcreal: | |||
| 2665 | case Builtin::BIcrealf: | |||
| 2666 | case Builtin::BIcreall: { | |||
| 2667 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
| 2668 | return RValue::get(ComplexVal.first); | |||
| 2669 | } | |||
| 2670 | ||||
| 2671 | case Builtin::BI__builtin_dump_struct: { | |||
| 2672 | llvm::Type *LLVMIntTy = getTypes().ConvertType(getContext().IntTy); | |||
| 2673 | llvm::FunctionType *LLVMFuncType = llvm::FunctionType::get( | |||
| 2674 | LLVMIntTy, {llvm::Type::getInt8PtrTy(getLLVMContext())}, true); | |||
| 2675 | ||||
| 2676 | Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts()); | |||
| 2677 | CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment(); | |||
| 2678 | ||||
| 2679 | const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts(); | |||
| 2680 | QualType Arg0Type = Arg0->getType()->getPointeeType(); | |||
| 2681 | ||||
| 2682 | Value *RecordPtr = EmitScalarExpr(Arg0); | |||
| 2683 | LValue RecordLV = MakeAddrLValue(RecordPtr, Arg0Type, Arg0Align); | |||
| 2684 | Value *Res = dumpRecord(*this, Arg0Type, RecordLV, Arg0Align, | |||
| 2685 | {LLVMFuncType, Func}, 0); | |||
| 2686 | return RValue::get(Res); | |||
| 2687 | } | |||
| 2688 | ||||
| 2689 | case Builtin::BI__builtin_preserve_access_index: { | |||
| 2690 | // Only enabled preserved access index region when debuginfo | |||
| 2691 | // is available as debuginfo is needed to preserve user-level | |||
| 2692 | // access pattern. | |||
| 2693 | if (!getDebugInfo()) { | |||
| 2694 | CGM.Error(E->getExprLoc(), "using builtin_preserve_access_index() without -g"); | |||
| 2695 | return RValue::get(EmitScalarExpr(E->getArg(0))); | |||
| 2696 | } | |||
| 2697 | ||||
| 2698 | // Nested builtin_preserve_access_index() not supported | |||
| 2699 | if (IsInPreservedAIRegion) { | |||
| 2700 | CGM.Error(E->getExprLoc(), "nested builtin_preserve_access_index() not supported"); | |||
| 2701 | return RValue::get(EmitScalarExpr(E->getArg(0))); | |||
| 2702 | } | |||
| 2703 | ||||
| 2704 | IsInPreservedAIRegion = true; | |||
| 2705 | Value *Res = EmitScalarExpr(E->getArg(0)); | |||
| 2706 | IsInPreservedAIRegion = false; | |||
| 2707 | return RValue::get(Res); | |||
| 2708 | } | |||
| 2709 | ||||
| 2710 | case Builtin::BI__builtin_cimag: | |||
| 2711 | case Builtin::BI__builtin_cimagf: | |||
| 2712 | case Builtin::BI__builtin_cimagl: | |||
| 2713 | case Builtin::BIcimag: | |||
| 2714 | case Builtin::BIcimagf: | |||
| 2715 | case Builtin::BIcimagl: { | |||
| 2716 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
| 2717 | return RValue::get(ComplexVal.second); | |||
| 2718 | } | |||
| 2719 | ||||
| 2720 | case Builtin::BI__builtin_clrsb: | |||
| 2721 | case Builtin::BI__builtin_clrsbl: | |||
| 2722 | case Builtin::BI__builtin_clrsbll: { | |||
| 2723 | // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or | |||
| 2724 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2725 | ||||
| 2726 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2727 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
| 2728 | ||||
| 2729 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2730 | Value *Zero = llvm::Constant::getNullValue(ArgType); | |||
| 2731 | Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg"); | |||
| 2732 | Value *Inverse = Builder.CreateNot(ArgValue, "not"); | |||
| 2733 | Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue); | |||
| 2734 | Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()}); | |||
| 2735 | Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1)); | |||
| 2736 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 2737 | "cast"); | |||
| 2738 | return RValue::get(Result); | |||
| 2739 | } | |||
| 2740 | case Builtin::BI__builtin_ctzs: | |||
| 2741 | case Builtin::BI__builtin_ctz: | |||
| 2742 | case Builtin::BI__builtin_ctzl: | |||
| 2743 | case Builtin::BI__builtin_ctzll: { | |||
| 2744 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero); | |||
| 2745 | ||||
| 2746 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2747 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); | |||
| 2748 | ||||
| 2749 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2750 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); | |||
| 2751 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); | |||
| 2752 | if (Result->getType() != ResultType) | |||
| 2753 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 2754 | "cast"); | |||
| 2755 | return RValue::get(Result); | |||
| 2756 | } | |||
| 2757 | case Builtin::BI__builtin_clzs: | |||
| 2758 | case Builtin::BI__builtin_clz: | |||
| 2759 | case Builtin::BI__builtin_clzl: | |||
| 2760 | case Builtin::BI__builtin_clzll: { | |||
| 2761 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero); | |||
| 2762 | ||||
| 2763 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2764 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
| 2765 | ||||
| 2766 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2767 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); | |||
| 2768 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); | |||
| 2769 | if (Result->getType() != ResultType) | |||
| 2770 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 2771 | "cast"); | |||
| 2772 | return RValue::get(Result); | |||
| 2773 | } | |||
| 2774 | case Builtin::BI__builtin_ffs: | |||
| 2775 | case Builtin::BI__builtin_ffsl: | |||
| 2776 | case Builtin::BI__builtin_ffsll: { | |||
| 2777 | // ffs(x) -> x ? cttz(x) + 1 : 0 | |||
| 2778 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2779 | ||||
| 2780 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2781 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); | |||
| 2782 | ||||
| 2783 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2784 | Value *Tmp = | |||
| 2785 | Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}), | |||
| 2786 | llvm::ConstantInt::get(ArgType, 1)); | |||
| 2787 | Value *Zero = llvm::Constant::getNullValue(ArgType); | |||
| 2788 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero"); | |||
| 2789 | Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs"); | |||
| 2790 | if (Result->getType() != ResultType) | |||
| 2791 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 2792 | "cast"); | |||
| 2793 | return RValue::get(Result); | |||
| 2794 | } | |||
| 2795 | case Builtin::BI__builtin_parity: | |||
| 2796 | case Builtin::BI__builtin_parityl: | |||
| 2797 | case Builtin::BI__builtin_parityll: { | |||
| 2798 | // parity(x) -> ctpop(x) & 1 | |||
| 2799 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2800 | ||||
| 2801 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2802 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); | |||
| 2803 | ||||
| 2804 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2805 | Value *Tmp = Builder.CreateCall(F, ArgValue); | |||
| 2806 | Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1)); | |||
| 2807 | if (Result->getType() != ResultType) | |||
| 2808 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 2809 | "cast"); | |||
| 2810 | return RValue::get(Result); | |||
| 2811 | } | |||
| 2812 | case Builtin::BI__lzcnt16: | |||
| 2813 | case Builtin::BI__lzcnt: | |||
| 2814 | case Builtin::BI__lzcnt64: { | |||
| 2815 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2816 | ||||
| 2817 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2818 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); | |||
| 2819 | ||||
| 2820 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2821 | Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()}); | |||
| 2822 | if (Result->getType() != ResultType) | |||
| 2823 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 2824 | "cast"); | |||
| 2825 | return RValue::get(Result); | |||
| 2826 | } | |||
| 2827 | case Builtin::BI__popcnt16: | |||
| 2828 | case Builtin::BI__popcnt: | |||
| 2829 | case Builtin::BI__popcnt64: | |||
| 2830 | case Builtin::BI__builtin_popcount: | |||
| 2831 | case Builtin::BI__builtin_popcountl: | |||
| 2832 | case Builtin::BI__builtin_popcountll: { | |||
| 2833 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2834 | ||||
| 2835 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2836 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); | |||
| 2837 | ||||
| 2838 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2839 | Value *Result = Builder.CreateCall(F, ArgValue); | |||
| 2840 | if (Result->getType() != ResultType) | |||
| 2841 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 2842 | "cast"); | |||
| 2843 | return RValue::get(Result); | |||
| 2844 | } | |||
| 2845 | case Builtin::BI__builtin_unpredictable: { | |||
| 2846 | // Always return the argument of __builtin_unpredictable. LLVM does not | |||
| 2847 | // handle this builtin. Metadata for this builtin should be added directly | |||
| 2848 | // to instructions such as branches or switches that use it. | |||
| 2849 | return RValue::get(EmitScalarExpr(E->getArg(0))); | |||
| 2850 | } | |||
| 2851 | case Builtin::BI__builtin_expect: { | |||
| 2852 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2853 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2854 | ||||
| 2855 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); | |||
| 2856 | // Don't generate llvm.expect on -O0 as the backend won't use it for | |||
| 2857 | // anything. | |||
| 2858 | // Note, we still IRGen ExpectedValue because it could have side-effects. | |||
| 2859 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) | |||
| 2860 | return RValue::get(ArgValue); | |||
| 2861 | ||||
| 2862 | Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType); | |||
| 2863 | Value *Result = | |||
| 2864 | Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval"); | |||
| 2865 | return RValue::get(Result); | |||
| 2866 | } | |||
| 2867 | case Builtin::BI__builtin_expect_with_probability: { | |||
| 2868 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2869 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 2870 | ||||
| 2871 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); | |||
| 2872 | llvm::APFloat Probability(0.0); | |||
| 2873 | const Expr *ProbArg = E->getArg(2); | |||
| 2874 | bool EvalSucceed = ProbArg->EvaluateAsFloat(Probability, CGM.getContext()); | |||
| 2875 | assert(EvalSucceed && "probability should be able to evaluate as float")(static_cast <bool> (EvalSucceed && "probability should be able to evaluate as float" ) ? void (0) : __assert_fail ("EvalSucceed && \"probability should be able to evaluate as float\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2875, __extension__ __PRETTY_FUNCTION__ )); | |||
| 2876 | (void)EvalSucceed; | |||
| 2877 | bool LoseInfo = false; | |||
| 2878 | Probability.convert(llvm::APFloat::IEEEdouble(), | |||
| 2879 | llvm::RoundingMode::Dynamic, &LoseInfo); | |||
| 2880 | llvm::Type *Ty = ConvertType(ProbArg->getType()); | |||
| 2881 | Constant *Confidence = ConstantFP::get(Ty, Probability); | |||
| 2882 | // Don't generate llvm.expect.with.probability on -O0 as the backend | |||
| 2883 | // won't use it for anything. | |||
| 2884 | // Note, we still IRGen ExpectedValue because it could have side-effects. | |||
| 2885 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) | |||
| 2886 | return RValue::get(ArgValue); | |||
| 2887 | ||||
| 2888 | Function *FnExpect = | |||
| 2889 | CGM.getIntrinsic(Intrinsic::expect_with_probability, ArgType); | |||
| 2890 | Value *Result = Builder.CreateCall( | |||
| 2891 | FnExpect, {ArgValue, ExpectedValue, Confidence}, "expval"); | |||
| 2892 | return RValue::get(Result); | |||
| 2893 | } | |||
| 2894 | case Builtin::BI__builtin_assume_aligned: { | |||
| 2895 | const Expr *Ptr = E->getArg(0); | |||
| 2896 | Value *PtrValue = EmitScalarExpr(Ptr); | |||
| 2897 | Value *OffsetValue = | |||
| 2898 | (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr; | |||
| 2899 | ||||
| 2900 | Value *AlignmentValue = EmitScalarExpr(E->getArg(1)); | |||
| 2901 | ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue); | |||
| 2902 | if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment)) | |||
| 2903 | AlignmentCI = ConstantInt::get(AlignmentCI->getType(), | |||
| 2904 | llvm::Value::MaximumAlignment); | |||
| 2905 | ||||
| 2906 | emitAlignmentAssumption(PtrValue, Ptr, | |||
| 2907 | /*The expr loc is sufficient.*/ SourceLocation(), | |||
| 2908 | AlignmentCI, OffsetValue); | |||
| 2909 | return RValue::get(PtrValue); | |||
| 2910 | } | |||
| 2911 | case Builtin::BI__assume: | |||
| 2912 | case Builtin::BI__builtin_assume: { | |||
| 2913 | if (E->getArg(0)->HasSideEffects(getContext())) | |||
| 2914 | return RValue::get(nullptr); | |||
| 2915 | ||||
| 2916 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2917 | Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume); | |||
| 2918 | return RValue::get(Builder.CreateCall(FnAssume, ArgValue)); | |||
| 2919 | } | |||
| 2920 | case Builtin::BI__arithmetic_fence: { | |||
| 2921 | // Create the builtin call if FastMath is selected, and the target | |||
| 2922 | // supports the builtin, otherwise just return the argument. | |||
| 2923 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 2924 | llvm::FastMathFlags FMF = Builder.getFastMathFlags(); | |||
| 2925 | bool isArithmeticFenceEnabled = | |||
| 2926 | FMF.allowReassoc() && | |||
| 2927 | getContext().getTargetInfo().checkArithmeticFenceSupported(); | |||
| 2928 | QualType ArgType = E->getArg(0)->getType(); | |||
| 2929 | if (ArgType->isComplexType()) { | |||
| 2930 | if (isArithmeticFenceEnabled) { | |||
| 2931 | QualType ElementType = ArgType->castAs<ComplexType>()->getElementType(); | |||
| 2932 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
| 2933 | Value *Real = Builder.CreateArithmeticFence(ComplexVal.first, | |||
| 2934 | ConvertType(ElementType)); | |||
| 2935 | Value *Imag = Builder.CreateArithmeticFence(ComplexVal.second, | |||
| 2936 | ConvertType(ElementType)); | |||
| 2937 | return RValue::getComplex(std::make_pair(Real, Imag)); | |||
| 2938 | } | |||
| 2939 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); | |||
| 2940 | Value *Real = ComplexVal.first; | |||
| 2941 | Value *Imag = ComplexVal.second; | |||
| 2942 | return RValue::getComplex(std::make_pair(Real, Imag)); | |||
| 2943 | } | |||
| 2944 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 2945 | if (isArithmeticFenceEnabled) | |||
| 2946 | return RValue::get( | |||
| 2947 | Builder.CreateArithmeticFence(ArgValue, ConvertType(ArgType))); | |||
| 2948 | return RValue::get(ArgValue); | |||
| 2949 | } | |||
| 2950 | case Builtin::BI__builtin_bswap16: | |||
| 2951 | case Builtin::BI__builtin_bswap32: | |||
| 2952 | case Builtin::BI__builtin_bswap64: | |||
| 2953 | case Builtin::BI_byteswap_ushort: | |||
| 2954 | case Builtin::BI_byteswap_ulong: | |||
| 2955 | case Builtin::BI_byteswap_uint64: { | |||
| 2956 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap)); | |||
| 2957 | } | |||
| 2958 | case Builtin::BI__builtin_bitreverse8: | |||
| 2959 | case Builtin::BI__builtin_bitreverse16: | |||
| 2960 | case Builtin::BI__builtin_bitreverse32: | |||
| 2961 | case Builtin::BI__builtin_bitreverse64: { | |||
| 2962 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse)); | |||
| 2963 | } | |||
| 2964 | case Builtin::BI__builtin_rotateleft8: | |||
| 2965 | case Builtin::BI__builtin_rotateleft16: | |||
| 2966 | case Builtin::BI__builtin_rotateleft32: | |||
| 2967 | case Builtin::BI__builtin_rotateleft64: | |||
| 2968 | case Builtin::BI_rotl8: // Microsoft variants of rotate left | |||
| 2969 | case Builtin::BI_rotl16: | |||
| 2970 | case Builtin::BI_rotl: | |||
| 2971 | case Builtin::BI_lrotl: | |||
| 2972 | case Builtin::BI_rotl64: | |||
| 2973 | return emitRotate(E, false); | |||
| 2974 | ||||
| 2975 | case Builtin::BI__builtin_rotateright8: | |||
| 2976 | case Builtin::BI__builtin_rotateright16: | |||
| 2977 | case Builtin::BI__builtin_rotateright32: | |||
| 2978 | case Builtin::BI__builtin_rotateright64: | |||
| 2979 | case Builtin::BI_rotr8: // Microsoft variants of rotate right | |||
| 2980 | case Builtin::BI_rotr16: | |||
| 2981 | case Builtin::BI_rotr: | |||
| 2982 | case Builtin::BI_lrotr: | |||
| 2983 | case Builtin::BI_rotr64: | |||
| 2984 | return emitRotate(E, true); | |||
| 2985 | ||||
| 2986 | case Builtin::BI__builtin_constant_p: { | |||
| 2987 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 2988 | ||||
| 2989 | const Expr *Arg = E->getArg(0); | |||
| 2990 | QualType ArgType = Arg->getType(); | |||
| 2991 | // FIXME: The allowance for Obj-C pointers and block pointers is historical | |||
| 2992 | // and likely a mistake. | |||
| 2993 | if (!ArgType->isIntegralOrEnumerationType() && !ArgType->isFloatingType() && | |||
| 2994 | !ArgType->isObjCObjectPointerType() && !ArgType->isBlockPointerType()) | |||
| 2995 | // Per the GCC documentation, only numeric constants are recognized after | |||
| 2996 | // inlining. | |||
| 2997 | return RValue::get(ConstantInt::get(ResultType, 0)); | |||
| 2998 | ||||
| 2999 | if (Arg->HasSideEffects(getContext())) | |||
| 3000 | // The argument is unevaluated, so be conservative if it might have | |||
| 3001 | // side-effects. | |||
| 3002 | return RValue::get(ConstantInt::get(ResultType, 0)); | |||
| 3003 | ||||
| 3004 | Value *ArgValue = EmitScalarExpr(Arg); | |||
| 3005 | if (ArgType->isObjCObjectPointerType()) { | |||
| 3006 | // Convert Objective-C objects to id because we cannot distinguish between | |||
| 3007 | // LLVM types for Obj-C classes as they are opaque. | |||
| 3008 | ArgType = CGM.getContext().getObjCIdType(); | |||
| 3009 | ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType)); | |||
| 3010 | } | |||
| 3011 | Function *F = | |||
| 3012 | CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType)); | |||
| 3013 | Value *Result = Builder.CreateCall(F, ArgValue); | |||
| 3014 | if (Result->getType() != ResultType) | |||
| 3015 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false); | |||
| 3016 | return RValue::get(Result); | |||
| 3017 | } | |||
| 3018 | case Builtin::BI__builtin_dynamic_object_size: | |||
| 3019 | case Builtin::BI__builtin_object_size: { | |||
| 3020 | unsigned Type = | |||
| 3021 | E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue(); | |||
| 3022 | auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType())); | |||
| 3023 | ||||
| 3024 | // We pass this builtin onto the optimizer so that it can figure out the | |||
| 3025 | // object size in more complex cases. | |||
| 3026 | bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size; | |||
| 3027 | return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType, | |||
| 3028 | /*EmittedE=*/nullptr, IsDynamic)); | |||
| 3029 | } | |||
| 3030 | case Builtin::BI__builtin_prefetch: { | |||
| 3031 | Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0)); | |||
| 3032 | // FIXME: Technically these constants should of type 'int', yes? | |||
| 3033 | RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) : | |||
| 3034 | llvm::ConstantInt::get(Int32Ty, 0); | |||
| 3035 | Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : | |||
| 3036 | llvm::ConstantInt::get(Int32Ty, 3); | |||
| 3037 | Value *Data = llvm::ConstantInt::get(Int32Ty, 1); | |||
| 3038 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
| 3039 | return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data})); | |||
| 3040 | } | |||
| 3041 | case Builtin::BI__builtin_readcyclecounter: { | |||
| 3042 | Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter); | |||
| 3043 | return RValue::get(Builder.CreateCall(F)); | |||
| 3044 | } | |||
| 3045 | case Builtin::BI__builtin___clear_cache: { | |||
| 3046 | Value *Begin = EmitScalarExpr(E->getArg(0)); | |||
| 3047 | Value *End = EmitScalarExpr(E->getArg(1)); | |||
| 3048 | Function *F = CGM.getIntrinsic(Intrinsic::clear_cache); | |||
| 3049 | return RValue::get(Builder.CreateCall(F, {Begin, End})); | |||
| 3050 | } | |||
| 3051 | case Builtin::BI__builtin_trap: | |||
| 3052 | return RValue::get(EmitTrapCall(Intrinsic::trap)); | |||
| 3053 | case Builtin::BI__debugbreak: | |||
| 3054 | return RValue::get(EmitTrapCall(Intrinsic::debugtrap)); | |||
| 3055 | case Builtin::BI__builtin_unreachable: { | |||
| 3056 | EmitUnreachable(E->getExprLoc()); | |||
| 3057 | ||||
| 3058 | // We do need to preserve an insertion point. | |||
| 3059 | EmitBlock(createBasicBlock("unreachable.cont")); | |||
| 3060 | ||||
| 3061 | return RValue::get(nullptr); | |||
| 3062 | } | |||
| 3063 | ||||
| 3064 | case Builtin::BI__builtin_powi: | |||
| 3065 | case Builtin::BI__builtin_powif: | |||
| 3066 | case Builtin::BI__builtin_powil: { | |||
| 3067 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 3068 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
| 3069 | ||||
| 3070 | if (Builder.getIsFPConstrained()) { | |||
| 3071 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 3072 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_powi, | |||
| 3073 | Src0->getType()); | |||
| 3074 | return RValue::get(Builder.CreateConstrainedFPCall(F, { Src0, Src1 })); | |||
| 3075 | } | |||
| 3076 | ||||
| 3077 | Function *F = CGM.getIntrinsic(Intrinsic::powi, | |||
| 3078 | { Src0->getType(), Src1->getType() }); | |||
| 3079 | return RValue::get(Builder.CreateCall(F, { Src0, Src1 })); | |||
| 3080 | } | |||
| 3081 | case Builtin::BI__builtin_isgreater: | |||
| 3082 | case Builtin::BI__builtin_isgreaterequal: | |||
| 3083 | case Builtin::BI__builtin_isless: | |||
| 3084 | case Builtin::BI__builtin_islessequal: | |||
| 3085 | case Builtin::BI__builtin_islessgreater: | |||
| 3086 | case Builtin::BI__builtin_isunordered: { | |||
| 3087 | // Ordered comparisons: we know the arguments to these are matching scalar | |||
| 3088 | // floating point values. | |||
| 3089 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 3090 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. | |||
| 3091 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 3092 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 3093 | ||||
| 3094 | switch (BuiltinID) { | |||
| 3095 | default: llvm_unreachable("Unknown ordered comparison")::llvm::llvm_unreachable_internal("Unknown ordered comparison" , "clang/lib/CodeGen/CGBuiltin.cpp", 3095); | |||
| 3096 | case Builtin::BI__builtin_isgreater: | |||
| 3097 | LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp"); | |||
| 3098 | break; | |||
| 3099 | case Builtin::BI__builtin_isgreaterequal: | |||
| 3100 | LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp"); | |||
| 3101 | break; | |||
| 3102 | case Builtin::BI__builtin_isless: | |||
| 3103 | LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp"); | |||
| 3104 | break; | |||
| 3105 | case Builtin::BI__builtin_islessequal: | |||
| 3106 | LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp"); | |||
| 3107 | break; | |||
| 3108 | case Builtin::BI__builtin_islessgreater: | |||
| 3109 | LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp"); | |||
| 3110 | break; | |||
| 3111 | case Builtin::BI__builtin_isunordered: | |||
| 3112 | LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp"); | |||
| 3113 | break; | |||
| 3114 | } | |||
| 3115 | // ZExt bool to int type. | |||
| 3116 | return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType()))); | |||
| 3117 | } | |||
| 3118 | case Builtin::BI__builtin_isnan: { | |||
| 3119 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 3120 | Value *V = EmitScalarExpr(E->getArg(0)); | |||
| 3121 | llvm::Type *Ty = V->getType(); | |||
| 3122 | const llvm::fltSemantics &Semantics = Ty->getFltSemantics(); | |||
| 3123 | if (!Builder.getIsFPConstrained() || | |||
| 3124 | Builder.getDefaultConstrainedExcept() == fp::ebIgnore || | |||
| 3125 | !Ty->isIEEE()) { | |||
| 3126 | V = Builder.CreateFCmpUNO(V, V, "cmp"); | |||
| 3127 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); | |||
| 3128 | } | |||
| 3129 | ||||
| 3130 | if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM)) | |||
| 3131 | return RValue::get(Result); | |||
| 3132 | ||||
| 3133 | // NaN has all exp bits set and a non zero significand. Therefore: | |||
| 3134 | // isnan(V) == ((exp mask - (abs(V) & exp mask)) < 0) | |||
| 3135 | unsigned bitsize = Ty->getScalarSizeInBits(); | |||
| 3136 | llvm::IntegerType *IntTy = Builder.getIntNTy(bitsize); | |||
| 3137 | Value *IntV = Builder.CreateBitCast(V, IntTy); | |||
| 3138 | APInt AndMask = APInt::getSignedMaxValue(bitsize); | |||
| 3139 | Value *AbsV = | |||
| 3140 | Builder.CreateAnd(IntV, llvm::ConstantInt::get(IntTy, AndMask)); | |||
| 3141 | APInt ExpMask = APFloat::getInf(Semantics).bitcastToAPInt(); | |||
| 3142 | Value *Sub = | |||
| 3143 | Builder.CreateSub(llvm::ConstantInt::get(IntTy, ExpMask), AbsV); | |||
| 3144 | // V = sign bit (Sub) <=> V = (Sub < 0) | |||
| 3145 | V = Builder.CreateLShr(Sub, llvm::ConstantInt::get(IntTy, bitsize - 1)); | |||
| 3146 | if (bitsize > 32) | |||
| 3147 | V = Builder.CreateTrunc(V, ConvertType(E->getType())); | |||
| 3148 | return RValue::get(V); | |||
| 3149 | } | |||
| 3150 | ||||
| 3151 | case Builtin::BI__builtin_elementwise_abs: { | |||
| 3152 | Value *Result; | |||
| 3153 | QualType QT = E->getArg(0)->getType(); | |||
| 3154 | ||||
| 3155 | if (auto *VecTy = QT->getAs<VectorType>()) | |||
| 3156 | QT = VecTy->getElementType(); | |||
| 3157 | if (QT->isIntegerType()) | |||
| 3158 | Result = Builder.CreateBinaryIntrinsic( | |||
| 3159 | llvm::Intrinsic::abs, EmitScalarExpr(E->getArg(0)), | |||
| 3160 | Builder.getFalse(), nullptr, "elt.abs"); | |||
| 3161 | else | |||
| 3162 | Result = emitUnaryBuiltin(*this, E, llvm::Intrinsic::fabs, "elt.abs"); | |||
| 3163 | ||||
| 3164 | return RValue::get(Result); | |||
| 3165 | } | |||
| 3166 | ||||
| 3167 | case Builtin::BI__builtin_elementwise_ceil: | |||
| 3168 | return RValue::get( | |||
| 3169 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::ceil, "elt.ceil")); | |||
| 3170 | case Builtin::BI__builtin_elementwise_floor: | |||
| 3171 | return RValue::get( | |||
| 3172 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::floor, "elt.floor")); | |||
| 3173 | case Builtin::BI__builtin_elementwise_roundeven: | |||
| 3174 | return RValue::get(emitUnaryBuiltin(*this, E, llvm::Intrinsic::roundeven, | |||
| 3175 | "elt.roundeven")); | |||
| 3176 | case Builtin::BI__builtin_elementwise_trunc: | |||
| 3177 | return RValue::get( | |||
| 3178 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::trunc, "elt.trunc")); | |||
| 3179 | ||||
| 3180 | case Builtin::BI__builtin_elementwise_add_sat: | |||
| 3181 | case Builtin::BI__builtin_elementwise_sub_sat: { | |||
| 3182 | Value *Op0 = EmitScalarExpr(E->getArg(0)); | |||
| 3183 | Value *Op1 = EmitScalarExpr(E->getArg(1)); | |||
| 3184 | Value *Result; | |||
| 3185 | assert(Op0->getType()->isIntOrIntVectorTy() && "integer type expected")(static_cast <bool> (Op0->getType()->isIntOrIntVectorTy () && "integer type expected") ? void (0) : __assert_fail ("Op0->getType()->isIntOrIntVectorTy() && \"integer type expected\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3185, __extension__ __PRETTY_FUNCTION__ )); | |||
| 3186 | QualType Ty = E->getArg(0)->getType(); | |||
| 3187 | if (auto *VecTy = Ty->getAs<VectorType>()) | |||
| 3188 | Ty = VecTy->getElementType(); | |||
| 3189 | bool IsSigned = Ty->isSignedIntegerType(); | |||
| 3190 | unsigned Opc; | |||
| 3191 | if (BuiltinIDIfNoAsmLabel == Builtin::BI__builtin_elementwise_add_sat) | |||
| 3192 | Opc = IsSigned ? llvm::Intrinsic::sadd_sat : llvm::Intrinsic::uadd_sat; | |||
| 3193 | else | |||
| 3194 | Opc = IsSigned ? llvm::Intrinsic::ssub_sat : llvm::Intrinsic::usub_sat; | |||
| 3195 | Result = Builder.CreateBinaryIntrinsic(Opc, Op0, Op1, nullptr, "elt.sat"); | |||
| 3196 | return RValue::get(Result); | |||
| 3197 | } | |||
| 3198 | ||||
| 3199 | case Builtin::BI__builtin_elementwise_max: { | |||
| 3200 | Value *Op0 = EmitScalarExpr(E->getArg(0)); | |||
| 3201 | Value *Op1 = EmitScalarExpr(E->getArg(1)); | |||
| 3202 | Value *Result; | |||
| 3203 | if (Op0->getType()->isIntOrIntVectorTy()) { | |||
| 3204 | QualType Ty = E->getArg(0)->getType(); | |||
| 3205 | if (auto *VecTy = Ty->getAs<VectorType>()) | |||
| 3206 | Ty = VecTy->getElementType(); | |||
| 3207 | Result = Builder.CreateBinaryIntrinsic(Ty->isSignedIntegerType() | |||
| 3208 | ? llvm::Intrinsic::smax | |||
| 3209 | : llvm::Intrinsic::umax, | |||
| 3210 | Op0, Op1, nullptr, "elt.max"); | |||
| 3211 | } else | |||
| 3212 | Result = Builder.CreateMaxNum(Op0, Op1, "elt.max"); | |||
| 3213 | return RValue::get(Result); | |||
| 3214 | } | |||
| 3215 | case Builtin::BI__builtin_elementwise_min: { | |||
| 3216 | Value *Op0 = EmitScalarExpr(E->getArg(0)); | |||
| 3217 | Value *Op1 = EmitScalarExpr(E->getArg(1)); | |||
| 3218 | Value *Result; | |||
| 3219 | if (Op0->getType()->isIntOrIntVectorTy()) { | |||
| 3220 | QualType Ty = E->getArg(0)->getType(); | |||
| 3221 | if (auto *VecTy = Ty->getAs<VectorType>()) | |||
| 3222 | Ty = VecTy->getElementType(); | |||
| 3223 | Result = Builder.CreateBinaryIntrinsic(Ty->isSignedIntegerType() | |||
| 3224 | ? llvm::Intrinsic::smin | |||
| 3225 | : llvm::Intrinsic::umin, | |||
| 3226 | Op0, Op1, nullptr, "elt.min"); | |||
| 3227 | } else | |||
| 3228 | Result = Builder.CreateMinNum(Op0, Op1, "elt.min"); | |||
| 3229 | return RValue::get(Result); | |||
| 3230 | } | |||
| 3231 | ||||
| 3232 | case Builtin::BI__builtin_reduce_max: { | |||
| 3233 | auto GetIntrinsicID = [](QualType QT) { | |||
| 3234 | if (auto *VecTy = QT->getAs<VectorType>()) | |||
| 3235 | QT = VecTy->getElementType(); | |||
| 3236 | if (QT->isSignedIntegerType()) | |||
| 3237 | return llvm::Intrinsic::vector_reduce_smax; | |||
| 3238 | if (QT->isUnsignedIntegerType()) | |||
| 3239 | return llvm::Intrinsic::vector_reduce_umax; | |||
| 3240 | assert(QT->isFloatingType() && "must have a float here")(static_cast <bool> (QT->isFloatingType() && "must have a float here") ? void (0) : __assert_fail ("QT->isFloatingType() && \"must have a float here\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3240, __extension__ __PRETTY_FUNCTION__ )); | |||
| 3241 | return llvm::Intrinsic::vector_reduce_fmax; | |||
| 3242 | }; | |||
| 3243 | return RValue::get(emitUnaryBuiltin( | |||
| 3244 | *this, E, GetIntrinsicID(E->getArg(0)->getType()), "rdx.min")); | |||
| 3245 | } | |||
| 3246 | ||||
| 3247 | case Builtin::BI__builtin_reduce_min: { | |||
| 3248 | auto GetIntrinsicID = [](QualType QT) { | |||
| 3249 | if (auto *VecTy = QT->getAs<VectorType>()) | |||
| 3250 | QT = VecTy->getElementType(); | |||
| 3251 | if (QT->isSignedIntegerType()) | |||
| 3252 | return llvm::Intrinsic::vector_reduce_smin; | |||
| 3253 | if (QT->isUnsignedIntegerType()) | |||
| 3254 | return llvm::Intrinsic::vector_reduce_umin; | |||
| 3255 | assert(QT->isFloatingType() && "must have a float here")(static_cast <bool> (QT->isFloatingType() && "must have a float here") ? void (0) : __assert_fail ("QT->isFloatingType() && \"must have a float here\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3255, __extension__ __PRETTY_FUNCTION__ )); | |||
| 3256 | return llvm::Intrinsic::vector_reduce_fmin; | |||
| 3257 | }; | |||
| 3258 | ||||
| 3259 | return RValue::get(emitUnaryBuiltin( | |||
| 3260 | *this, E, GetIntrinsicID(E->getArg(0)->getType()), "rdx.min")); | |||
| 3261 | } | |||
| 3262 | ||||
| 3263 | case Builtin::BI__builtin_reduce_xor: | |||
| 3264 | return RValue::get(emitUnaryBuiltin( | |||
| 3265 | *this, E, llvm::Intrinsic::vector_reduce_xor, "rdx.xor")); | |||
| 3266 | case Builtin::BI__builtin_reduce_or: | |||
| 3267 | return RValue::get(emitUnaryBuiltin( | |||
| 3268 | *this, E, llvm::Intrinsic::vector_reduce_or, "rdx.or")); | |||
| 3269 | case Builtin::BI__builtin_reduce_and: | |||
| 3270 | return RValue::get(emitUnaryBuiltin( | |||
| 3271 | *this, E, llvm::Intrinsic::vector_reduce_and, "rdx.and")); | |||
| 3272 | ||||
| 3273 | case Builtin::BI__builtin_matrix_transpose: { | |||
| 3274 | auto *MatrixTy = E->getArg(0)->getType()->castAs<ConstantMatrixType>(); | |||
| 3275 | Value *MatValue = EmitScalarExpr(E->getArg(0)); | |||
| 3276 | MatrixBuilder MB(Builder); | |||
| 3277 | Value *Result = MB.CreateMatrixTranspose(MatValue, MatrixTy->getNumRows(), | |||
| 3278 | MatrixTy->getNumColumns()); | |||
| 3279 | return RValue::get(Result); | |||
| 3280 | } | |||
| 3281 | ||||
| 3282 | case Builtin::BI__builtin_matrix_column_major_load: { | |||
| 3283 | MatrixBuilder MB(Builder); | |||
| 3284 | // Emit everything that isn't dependent on the first parameter type | |||
| 3285 | Value *Stride = EmitScalarExpr(E->getArg(3)); | |||
| 3286 | const auto *ResultTy = E->getType()->getAs<ConstantMatrixType>(); | |||
| 3287 | auto *PtrTy = E->getArg(0)->getType()->getAs<PointerType>(); | |||
| 3288 | assert(PtrTy && "arg0 must be of pointer type")(static_cast <bool> (PtrTy && "arg0 must be of pointer type" ) ? void (0) : __assert_fail ("PtrTy && \"arg0 must be of pointer type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3288, __extension__ __PRETTY_FUNCTION__ )); | |||
| 3289 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); | |||
| 3290 | ||||
| 3291 | Address Src = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3292 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(0)->getType(), | |||
| 3293 | E->getArg(0)->getExprLoc(), FD, 0); | |||
| 3294 | Value *Result = MB.CreateColumnMajorLoad( | |||
| 3295 | Src.getElementType(), Src.getPointer(), | |||
| 3296 | Align(Src.getAlignment().getQuantity()), Stride, IsVolatile, | |||
| 3297 | ResultTy->getNumRows(), ResultTy->getNumColumns(), | |||
| 3298 | "matrix"); | |||
| 3299 | return RValue::get(Result); | |||
| 3300 | } | |||
| 3301 | ||||
| 3302 | case Builtin::BI__builtin_matrix_column_major_store: { | |||
| 3303 | MatrixBuilder MB(Builder); | |||
| 3304 | Value *Matrix = EmitScalarExpr(E->getArg(0)); | |||
| 3305 | Address Dst = EmitPointerWithAlignment(E->getArg(1)); | |||
| 3306 | Value *Stride = EmitScalarExpr(E->getArg(2)); | |||
| 3307 | ||||
| 3308 | const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>(); | |||
| 3309 | auto *PtrTy = E->getArg(1)->getType()->getAs<PointerType>(); | |||
| 3310 | assert(PtrTy && "arg1 must be of pointer type")(static_cast <bool> (PtrTy && "arg1 must be of pointer type" ) ? void (0) : __assert_fail ("PtrTy && \"arg1 must be of pointer type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3310, __extension__ __PRETTY_FUNCTION__ )); | |||
| 3311 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); | |||
| 3312 | ||||
| 3313 | EmitNonNullArgCheck(RValue::get(Dst.getPointer()), E->getArg(1)->getType(), | |||
| 3314 | E->getArg(1)->getExprLoc(), FD, 0); | |||
| 3315 | Value *Result = MB.CreateColumnMajorStore( | |||
| 3316 | Matrix, Dst.getPointer(), Align(Dst.getAlignment().getQuantity()), | |||
| 3317 | Stride, IsVolatile, MatrixTy->getNumRows(), MatrixTy->getNumColumns()); | |||
| 3318 | return RValue::get(Result); | |||
| 3319 | } | |||
| 3320 | ||||
| 3321 | case Builtin::BIfinite: | |||
| 3322 | case Builtin::BI__finite: | |||
| 3323 | case Builtin::BIfinitef: | |||
| 3324 | case Builtin::BI__finitef: | |||
| 3325 | case Builtin::BIfinitel: | |||
| 3326 | case Builtin::BI__finitel: | |||
| 3327 | case Builtin::BI__builtin_isinf: | |||
| 3328 | case Builtin::BI__builtin_isfinite: { | |||
| 3329 | // isinf(x) --> fabs(x) == infinity | |||
| 3330 | // isfinite(x) --> fabs(x) != infinity | |||
| 3331 | // x != NaN via the ordered compare in either case. | |||
| 3332 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 3333 | Value *V = EmitScalarExpr(E->getArg(0)); | |||
| 3334 | llvm::Type *Ty = V->getType(); | |||
| 3335 | if (!Builder.getIsFPConstrained() || | |||
| 3336 | Builder.getDefaultConstrainedExcept() == fp::ebIgnore || | |||
| 3337 | !Ty->isIEEE()) { | |||
| 3338 | Value *Fabs = EmitFAbs(*this, V); | |||
| 3339 | Constant *Infinity = ConstantFP::getInfinity(V->getType()); | |||
| 3340 | CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf) | |||
| 3341 | ? CmpInst::FCMP_OEQ | |||
| 3342 | : CmpInst::FCMP_ONE; | |||
| 3343 | Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf"); | |||
| 3344 | return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType()))); | |||
| 3345 | } | |||
| 3346 | ||||
| 3347 | if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM)) | |||
| 3348 | return RValue::get(Result); | |||
| 3349 | ||||
| 3350 | // Inf values have all exp bits set and a zero significand. Therefore: | |||
| 3351 | // isinf(V) == ((V << 1) == ((exp mask) << 1)) | |||
| 3352 | // isfinite(V) == ((V << 1) < ((exp mask) << 1)) using unsigned comparison | |||
| 3353 | unsigned bitsize = Ty->getScalarSizeInBits(); | |||
| 3354 | llvm::IntegerType *IntTy = Builder.getIntNTy(bitsize); | |||
| 3355 | Value *IntV = Builder.CreateBitCast(V, IntTy); | |||
| 3356 | Value *Shl1 = Builder.CreateShl(IntV, 1); | |||
| 3357 | const llvm::fltSemantics &Semantics = Ty->getFltSemantics(); | |||
| 3358 | APInt ExpMask = APFloat::getInf(Semantics).bitcastToAPInt(); | |||
| 3359 | Value *ExpMaskShl1 = llvm::ConstantInt::get(IntTy, ExpMask.shl(1)); | |||
| 3360 | if (BuiltinID == Builtin::BI__builtin_isinf) | |||
| 3361 | V = Builder.CreateICmpEQ(Shl1, ExpMaskShl1); | |||
| 3362 | else | |||
| 3363 | V = Builder.CreateICmpULT(Shl1, ExpMaskShl1); | |||
| 3364 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); | |||
| 3365 | } | |||
| 3366 | ||||
| 3367 | case Builtin::BI__builtin_isinf_sign: { | |||
| 3368 | // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0 | |||
| 3369 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 3370 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. | |||
| 3371 | Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 3372 | Value *AbsArg = EmitFAbs(*this, Arg); | |||
| 3373 | Value *IsInf = Builder.CreateFCmpOEQ( | |||
| 3374 | AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf"); | |||
| 3375 | Value *IsNeg = EmitSignBit(*this, Arg); | |||
| 3376 | ||||
| 3377 | llvm::Type *IntTy = ConvertType(E->getType()); | |||
| 3378 | Value *Zero = Constant::getNullValue(IntTy); | |||
| 3379 | Value *One = ConstantInt::get(IntTy, 1); | |||
| 3380 | Value *NegativeOne = ConstantInt::get(IntTy, -1); | |||
| 3381 | Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One); | |||
| 3382 | Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero); | |||
| 3383 | return RValue::get(Result); | |||
| 3384 | } | |||
| 3385 | ||||
| 3386 | case Builtin::BI__builtin_isnormal: { | |||
| 3387 | // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min | |||
| 3388 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 3389 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. | |||
| 3390 | Value *V = EmitScalarExpr(E->getArg(0)); | |||
| 3391 | Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq"); | |||
| 3392 | ||||
| 3393 | Value *Abs = EmitFAbs(*this, V); | |||
| 3394 | Value *IsLessThanInf = | |||
| 3395 | Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf"); | |||
| 3396 | APFloat Smallest = APFloat::getSmallestNormalized( | |||
| 3397 | getContext().getFloatTypeSemantics(E->getArg(0)->getType())); | |||
| 3398 | Value *IsNormal = | |||
| 3399 | Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest), | |||
| 3400 | "isnormal"); | |||
| 3401 | V = Builder.CreateAnd(Eq, IsLessThanInf, "and"); | |||
| 3402 | V = Builder.CreateAnd(V, IsNormal, "and"); | |||
| 3403 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); | |||
| 3404 | } | |||
| 3405 | ||||
| 3406 | case Builtin::BI__builtin_flt_rounds: { | |||
| 3407 | Function *F = CGM.getIntrinsic(Intrinsic::flt_rounds); | |||
| 3408 | ||||
| 3409 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 3410 | Value *Result = Builder.CreateCall(F); | |||
| 3411 | if (Result->getType() != ResultType) | |||
| 3412 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 3413 | "cast"); | |||
| 3414 | return RValue::get(Result); | |||
| 3415 | } | |||
| 3416 | ||||
| 3417 | case Builtin::BI__builtin_fpclassify: { | |||
| 3418 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 3419 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. | |||
| 3420 | Value *V = EmitScalarExpr(E->getArg(5)); | |||
| 3421 | llvm::Type *Ty = ConvertType(E->getArg(5)->getType()); | |||
| 3422 | ||||
| 3423 | // Create Result | |||
| 3424 | BasicBlock *Begin = Builder.GetInsertBlock(); | |||
| 3425 | BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn); | |||
| 3426 | Builder.SetInsertPoint(End); | |||
| 3427 | PHINode *Result = | |||
| 3428 | Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4, | |||
| 3429 | "fpclassify_result"); | |||
| 3430 | ||||
| 3431 | // if (V==0) return FP_ZERO | |||
| 3432 | Builder.SetInsertPoint(Begin); | |||
| 3433 | Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty), | |||
| 3434 | "iszero"); | |||
| 3435 | Value *ZeroLiteral = EmitScalarExpr(E->getArg(4)); | |||
| 3436 | BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn); | |||
| 3437 | Builder.CreateCondBr(IsZero, End, NotZero); | |||
| 3438 | Result->addIncoming(ZeroLiteral, Begin); | |||
| 3439 | ||||
| 3440 | // if (V != V) return FP_NAN | |||
| 3441 | Builder.SetInsertPoint(NotZero); | |||
| 3442 | Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp"); | |||
| 3443 | Value *NanLiteral = EmitScalarExpr(E->getArg(0)); | |||
| 3444 | BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn); | |||
| 3445 | Builder.CreateCondBr(IsNan, End, NotNan); | |||
| 3446 | Result->addIncoming(NanLiteral, NotZero); | |||
| 3447 | ||||
| 3448 | // if (fabs(V) == infinity) return FP_INFINITY | |||
| 3449 | Builder.SetInsertPoint(NotNan); | |||
| 3450 | Value *VAbs = EmitFAbs(*this, V); | |||
| 3451 | Value *IsInf = | |||
| 3452 | Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()), | |||
| 3453 | "isinf"); | |||
| 3454 | Value *InfLiteral = EmitScalarExpr(E->getArg(1)); | |||
| 3455 | BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn); | |||
| 3456 | Builder.CreateCondBr(IsInf, End, NotInf); | |||
| 3457 | Result->addIncoming(InfLiteral, NotNan); | |||
| 3458 | ||||
| 3459 | // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL | |||
| 3460 | Builder.SetInsertPoint(NotInf); | |||
| 3461 | APFloat Smallest = APFloat::getSmallestNormalized( | |||
| 3462 | getContext().getFloatTypeSemantics(E->getArg(5)->getType())); | |||
| 3463 | Value *IsNormal = | |||
| 3464 | Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest), | |||
| 3465 | "isnormal"); | |||
| 3466 | Value *NormalResult = | |||
| 3467 | Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)), | |||
| 3468 | EmitScalarExpr(E->getArg(3))); | |||
| 3469 | Builder.CreateBr(End); | |||
| 3470 | Result->addIncoming(NormalResult, NotInf); | |||
| 3471 | ||||
| 3472 | // return Result | |||
| 3473 | Builder.SetInsertPoint(End); | |||
| 3474 | return RValue::get(Result); | |||
| 3475 | } | |||
| 3476 | ||||
| 3477 | case Builtin::BIalloca: | |||
| 3478 | case Builtin::BI_alloca: | |||
| 3479 | case Builtin::BI__builtin_alloca_uninitialized: | |||
| 3480 | case Builtin::BI__builtin_alloca: { | |||
| 3481 | Value *Size = EmitScalarExpr(E->getArg(0)); | |||
| 3482 | const TargetInfo &TI = getContext().getTargetInfo(); | |||
| 3483 | // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__. | |||
| 3484 | const Align SuitableAlignmentInBytes = | |||
| 3485 | CGM.getContext() | |||
| 3486 | .toCharUnitsFromBits(TI.getSuitableAlign()) | |||
| 3487 | .getAsAlign(); | |||
| 3488 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); | |||
| 3489 | AI->setAlignment(SuitableAlignmentInBytes); | |||
| 3490 | if (BuiltinID != Builtin::BI__builtin_alloca_uninitialized) | |||
| 3491 | initializeAlloca(*this, AI, Size, SuitableAlignmentInBytes); | |||
| 3492 | return RValue::get(AI); | |||
| 3493 | } | |||
| 3494 | ||||
| 3495 | case Builtin::BI__builtin_alloca_with_align_uninitialized: | |||
| 3496 | case Builtin::BI__builtin_alloca_with_align: { | |||
| 3497 | Value *Size = EmitScalarExpr(E->getArg(0)); | |||
| 3498 | Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1)); | |||
| 3499 | auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue); | |||
| 3500 | unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue(); | |||
| 3501 | const Align AlignmentInBytes = | |||
| 3502 | CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getAsAlign(); | |||
| 3503 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); | |||
| 3504 | AI->setAlignment(AlignmentInBytes); | |||
| 3505 | if (BuiltinID != Builtin::BI__builtin_alloca_with_align_uninitialized) | |||
| 3506 | initializeAlloca(*this, AI, Size, AlignmentInBytes); | |||
| 3507 | return RValue::get(AI); | |||
| 3508 | } | |||
| 3509 | ||||
| 3510 | case Builtin::BIbzero: | |||
| 3511 | case Builtin::BI__builtin_bzero: { | |||
| 3512 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3513 | Value *SizeVal = EmitScalarExpr(E->getArg(1)); | |||
| 3514 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
| 3515 | E->getArg(0)->getExprLoc(), FD, 0); | |||
| 3516 | Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false); | |||
| 3517 | return RValue::get(nullptr); | |||
| 3518 | } | |||
| 3519 | case Builtin::BImemcpy: | |||
| 3520 | case Builtin::BI__builtin_memcpy: | |||
| 3521 | case Builtin::BImempcpy: | |||
| 3522 | case Builtin::BI__builtin_mempcpy: { | |||
| 3523 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3524 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
| 3525 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
| 3526 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
| 3527 | E->getArg(0)->getExprLoc(), FD, 0); | |||
| 3528 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), | |||
| 3529 | E->getArg(1)->getExprLoc(), FD, 1); | |||
| 3530 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); | |||
| 3531 | if (BuiltinID == Builtin::BImempcpy || | |||
| 3532 | BuiltinID == Builtin::BI__builtin_mempcpy) | |||
| 3533 | return RValue::get(Builder.CreateInBoundsGEP(Dest.getElementType(), | |||
| 3534 | Dest.getPointer(), SizeVal)); | |||
| 3535 | else | |||
| 3536 | return RValue::get(Dest.getPointer()); | |||
| 3537 | } | |||
| 3538 | ||||
| 3539 | case Builtin::BI__builtin_memcpy_inline: { | |||
| 3540 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3541 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
| 3542 | uint64_t Size = | |||
| 3543 | E->getArg(2)->EvaluateKnownConstInt(getContext()).getZExtValue(); | |||
| 3544 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
| 3545 | E->getArg(0)->getExprLoc(), FD, 0); | |||
| 3546 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), | |||
| 3547 | E->getArg(1)->getExprLoc(), FD, 1); | |||
| 3548 | Builder.CreateMemCpyInline(Dest, Src, Size); | |||
| 3549 | return RValue::get(nullptr); | |||
| 3550 | } | |||
| 3551 | ||||
| 3552 | case Builtin::BI__builtin_char_memchr: | |||
| 3553 | BuiltinID = Builtin::BI__builtin_memchr; | |||
| 3554 | break; | |||
| 3555 | ||||
| 3556 | case Builtin::BI__builtin___memcpy_chk: { | |||
| 3557 | // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2. | |||
| 3558 | Expr::EvalResult SizeResult, DstSizeResult; | |||
| 3559 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || | |||
| 3560 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) | |||
| 3561 | break; | |||
| 3562 | llvm::APSInt Size = SizeResult.Val.getInt(); | |||
| 3563 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); | |||
| 3564 | if (Size.ugt(DstSize)) | |||
| 3565 | break; | |||
| 3566 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3567 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
| 3568 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); | |||
| 3569 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); | |||
| 3570 | return RValue::get(Dest.getPointer()); | |||
| 3571 | } | |||
| 3572 | ||||
| 3573 | case Builtin::BI__builtin_objc_memmove_collectable: { | |||
| 3574 | Address DestAddr = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3575 | Address SrcAddr = EmitPointerWithAlignment(E->getArg(1)); | |||
| 3576 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
| 3577 | CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, | |||
| 3578 | DestAddr, SrcAddr, SizeVal); | |||
| 3579 | return RValue::get(DestAddr.getPointer()); | |||
| 3580 | } | |||
| 3581 | ||||
| 3582 | case Builtin::BI__builtin___memmove_chk: { | |||
| 3583 | // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2. | |||
| 3584 | Expr::EvalResult SizeResult, DstSizeResult; | |||
| 3585 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || | |||
| 3586 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) | |||
| 3587 | break; | |||
| 3588 | llvm::APSInt Size = SizeResult.Val.getInt(); | |||
| 3589 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); | |||
| 3590 | if (Size.ugt(DstSize)) | |||
| 3591 | break; | |||
| 3592 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3593 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
| 3594 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); | |||
| 3595 | Builder.CreateMemMove(Dest, Src, SizeVal, false); | |||
| 3596 | return RValue::get(Dest.getPointer()); | |||
| 3597 | } | |||
| 3598 | ||||
| 3599 | case Builtin::BImemmove: | |||
| 3600 | case Builtin::BI__builtin_memmove: { | |||
| 3601 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3602 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
| 3603 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
| 3604 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
| 3605 | E->getArg(0)->getExprLoc(), FD, 0); | |||
| 3606 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), | |||
| 3607 | E->getArg(1)->getExprLoc(), FD, 1); | |||
| 3608 | Builder.CreateMemMove(Dest, Src, SizeVal, false); | |||
| 3609 | return RValue::get(Dest.getPointer()); | |||
| 3610 | } | |||
| 3611 | case Builtin::BImemset: | |||
| 3612 | case Builtin::BI__builtin_memset: { | |||
| 3613 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3614 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), | |||
| 3615 | Builder.getInt8Ty()); | |||
| 3616 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); | |||
| 3617 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), | |||
| 3618 | E->getArg(0)->getExprLoc(), FD, 0); | |||
| 3619 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); | |||
| 3620 | return RValue::get(Dest.getPointer()); | |||
| 3621 | } | |||
| 3622 | case Builtin::BI__builtin___memset_chk: { | |||
| 3623 | // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2. | |||
| 3624 | Expr::EvalResult SizeResult, DstSizeResult; | |||
| 3625 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || | |||
| 3626 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) | |||
| 3627 | break; | |||
| 3628 | llvm::APSInt Size = SizeResult.Val.getInt(); | |||
| 3629 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); | |||
| 3630 | if (Size.ugt(DstSize)) | |||
| 3631 | break; | |||
| 3632 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3633 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), | |||
| 3634 | Builder.getInt8Ty()); | |||
| 3635 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); | |||
| 3636 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); | |||
| 3637 | return RValue::get(Dest.getPointer()); | |||
| 3638 | } | |||
| 3639 | case Builtin::BI__builtin_wmemchr: { | |||
| 3640 | // The MSVC runtime library does not provide a definition of wmemchr, so we | |||
| 3641 | // need an inline implementation. | |||
| 3642 | if (!getTarget().getTriple().isOSMSVCRT()) | |||
| 3643 | break; | |||
| 3644 | ||||
| 3645 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); | |||
| 3646 | Value *Str = EmitScalarExpr(E->getArg(0)); | |||
| 3647 | Value *Chr = EmitScalarExpr(E->getArg(1)); | |||
| 3648 | Value *Size = EmitScalarExpr(E->getArg(2)); | |||
| 3649 | ||||
| 3650 | BasicBlock *Entry = Builder.GetInsertBlock(); | |||
| 3651 | BasicBlock *CmpEq = createBasicBlock("wmemchr.eq"); | |||
| 3652 | BasicBlock *Next = createBasicBlock("wmemchr.next"); | |||
| 3653 | BasicBlock *Exit = createBasicBlock("wmemchr.exit"); | |||
| 3654 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); | |||
| 3655 | Builder.CreateCondBr(SizeEq0, Exit, CmpEq); | |||
| 3656 | ||||
| 3657 | EmitBlock(CmpEq); | |||
| 3658 | PHINode *StrPhi = Builder.CreatePHI(Str->getType(), 2); | |||
| 3659 | StrPhi->addIncoming(Str, Entry); | |||
| 3660 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); | |||
| 3661 | SizePhi->addIncoming(Size, Entry); | |||
| 3662 | CharUnits WCharAlign = | |||
| 3663 | getContext().getTypeAlignInChars(getContext().WCharTy); | |||
| 3664 | Value *StrCh = Builder.CreateAlignedLoad(WCharTy, StrPhi, WCharAlign); | |||
| 3665 | Value *FoundChr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 0); | |||
| 3666 | Value *StrEqChr = Builder.CreateICmpEQ(StrCh, Chr); | |||
| 3667 | Builder.CreateCondBr(StrEqChr, Exit, Next); | |||
| 3668 | ||||
| 3669 | EmitBlock(Next); | |||
| 3670 | Value *NextStr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 1); | |||
| 3671 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); | |||
| 3672 | Value *NextSizeEq0 = | |||
| 3673 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); | |||
| 3674 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpEq); | |||
| 3675 | StrPhi->addIncoming(NextStr, Next); | |||
| 3676 | SizePhi->addIncoming(NextSize, Next); | |||
| 3677 | ||||
| 3678 | EmitBlock(Exit); | |||
| 3679 | PHINode *Ret = Builder.CreatePHI(Str->getType(), 3); | |||
| 3680 | Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Entry); | |||
| 3681 | Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Next); | |||
| 3682 | Ret->addIncoming(FoundChr, CmpEq); | |||
| 3683 | return RValue::get(Ret); | |||
| 3684 | } | |||
| 3685 | case Builtin::BI__builtin_wmemcmp: { | |||
| 3686 | // The MSVC runtime library does not provide a definition of wmemcmp, so we | |||
| 3687 | // need an inline implementation. | |||
| 3688 | if (!getTarget().getTriple().isOSMSVCRT()) | |||
| 3689 | break; | |||
| 3690 | ||||
| 3691 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); | |||
| 3692 | ||||
| 3693 | Value *Dst = EmitScalarExpr(E->getArg(0)); | |||
| 3694 | Value *Src = EmitScalarExpr(E->getArg(1)); | |||
| 3695 | Value *Size = EmitScalarExpr(E->getArg(2)); | |||
| 3696 | ||||
| 3697 | BasicBlock *Entry = Builder.GetInsertBlock(); | |||
| 3698 | BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt"); | |||
| 3699 | BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt"); | |||
| 3700 | BasicBlock *Next = createBasicBlock("wmemcmp.next"); | |||
| 3701 | BasicBlock *Exit = createBasicBlock("wmemcmp.exit"); | |||
| 3702 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); | |||
| 3703 | Builder.CreateCondBr(SizeEq0, Exit, CmpGT); | |||
| 3704 | ||||
| 3705 | EmitBlock(CmpGT); | |||
| 3706 | PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2); | |||
| 3707 | DstPhi->addIncoming(Dst, Entry); | |||
| 3708 | PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2); | |||
| 3709 | SrcPhi->addIncoming(Src, Entry); | |||
| 3710 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); | |||
| 3711 | SizePhi->addIncoming(Size, Entry); | |||
| 3712 | CharUnits WCharAlign = | |||
| 3713 | getContext().getTypeAlignInChars(getContext().WCharTy); | |||
| 3714 | Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign); | |||
| 3715 | Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign); | |||
| 3716 | Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh); | |||
| 3717 | Builder.CreateCondBr(DstGtSrc, Exit, CmpLT); | |||
| 3718 | ||||
| 3719 | EmitBlock(CmpLT); | |||
| 3720 | Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh); | |||
| 3721 | Builder.CreateCondBr(DstLtSrc, Exit, Next); | |||
| 3722 | ||||
| 3723 | EmitBlock(Next); | |||
| 3724 | Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1); | |||
| 3725 | Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1); | |||
| 3726 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); | |||
| 3727 | Value *NextSizeEq0 = | |||
| 3728 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); | |||
| 3729 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT); | |||
| 3730 | DstPhi->addIncoming(NextDst, Next); | |||
| 3731 | SrcPhi->addIncoming(NextSrc, Next); | |||
| 3732 | SizePhi->addIncoming(NextSize, Next); | |||
| 3733 | ||||
| 3734 | EmitBlock(Exit); | |||
| 3735 | PHINode *Ret = Builder.CreatePHI(IntTy, 4); | |||
| 3736 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry); | |||
| 3737 | Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT); | |||
| 3738 | Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT); | |||
| 3739 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Next); | |||
| 3740 | return RValue::get(Ret); | |||
| 3741 | } | |||
| 3742 | case Builtin::BI__builtin_dwarf_cfa: { | |||
| 3743 | // The offset in bytes from the first argument to the CFA. | |||
| 3744 | // | |||
| 3745 | // Why on earth is this in the frontend? Is there any reason at | |||
| 3746 | // all that the backend can't reasonably determine this while | |||
| 3747 | // lowering llvm.eh.dwarf.cfa()? | |||
| 3748 | // | |||
| 3749 | // TODO: If there's a satisfactory reason, add a target hook for | |||
| 3750 | // this instead of hard-coding 0, which is correct for most targets. | |||
| 3751 | int32_t Offset = 0; | |||
| 3752 | ||||
| 3753 | Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa); | |||
| 3754 | return RValue::get(Builder.CreateCall(F, | |||
| 3755 | llvm::ConstantInt::get(Int32Ty, Offset))); | |||
| 3756 | } | |||
| 3757 | case Builtin::BI__builtin_return_address: { | |||
| 3758 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), | |||
| 3759 | getContext().UnsignedIntTy); | |||
| 3760 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); | |||
| 3761 | return RValue::get(Builder.CreateCall(F, Depth)); | |||
| 3762 | } | |||
| 3763 | case Builtin::BI_ReturnAddress: { | |||
| 3764 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); | |||
| 3765 | return RValue::get(Builder.CreateCall(F, Builder.getInt32(0))); | |||
| 3766 | } | |||
| 3767 | case Builtin::BI__builtin_frame_address: { | |||
| 3768 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), | |||
| 3769 | getContext().UnsignedIntTy); | |||
| 3770 | Function *F = CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy); | |||
| 3771 | return RValue::get(Builder.CreateCall(F, Depth)); | |||
| 3772 | } | |||
| 3773 | case Builtin::BI__builtin_extract_return_addr: { | |||
| 3774 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
| 3775 | Value *Result = getTargetHooks().decodeReturnAddress(*this, Address); | |||
| 3776 | return RValue::get(Result); | |||
| 3777 | } | |||
| 3778 | case Builtin::BI__builtin_frob_return_addr: { | |||
| 3779 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
| 3780 | Value *Result = getTargetHooks().encodeReturnAddress(*this, Address); | |||
| 3781 | return RValue::get(Result); | |||
| 3782 | } | |||
| 3783 | case Builtin::BI__builtin_dwarf_sp_column: { | |||
| 3784 | llvm::IntegerType *Ty | |||
| 3785 | = cast<llvm::IntegerType>(ConvertType(E->getType())); | |||
| 3786 | int Column = getTargetHooks().getDwarfEHStackPointer(CGM); | |||
| 3787 | if (Column == -1) { | |||
| 3788 | CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column"); | |||
| 3789 | return RValue::get(llvm::UndefValue::get(Ty)); | |||
| 3790 | } | |||
| 3791 | return RValue::get(llvm::ConstantInt::get(Ty, Column, true)); | |||
| 3792 | } | |||
| 3793 | case Builtin::BI__builtin_init_dwarf_reg_size_table: { | |||
| 3794 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
| 3795 | if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address)) | |||
| 3796 | CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table"); | |||
| 3797 | return RValue::get(llvm::UndefValue::get(ConvertType(E->getType()))); | |||
| 3798 | } | |||
| 3799 | case Builtin::BI__builtin_eh_return: { | |||
| 3800 | Value *Int = EmitScalarExpr(E->getArg(0)); | |||
| 3801 | Value *Ptr = EmitScalarExpr(E->getArg(1)); | |||
| 3802 | ||||
| 3803 | llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType()); | |||
| 3804 | assert((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) &&(static_cast <bool> ((IntTy->getBitWidth() == 32 || IntTy ->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? void (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3805, __extension__ __PRETTY_FUNCTION__ )) | |||
| 3805 | "LLVM's __builtin_eh_return only supports 32- and 64-bit variants")(static_cast <bool> ((IntTy->getBitWidth() == 32 || IntTy ->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? void (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3805, __extension__ __PRETTY_FUNCTION__ )); | |||
| 3806 | Function *F = | |||
| 3807 | CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32 | |||
| 3808 | : Intrinsic::eh_return_i64); | |||
| 3809 | Builder.CreateCall(F, {Int, Ptr}); | |||
| 3810 | Builder.CreateUnreachable(); | |||
| 3811 | ||||
| 3812 | // We do need to preserve an insertion point. | |||
| 3813 | EmitBlock(createBasicBlock("builtin_eh_return.cont")); | |||
| 3814 | ||||
| 3815 | return RValue::get(nullptr); | |||
| 3816 | } | |||
| 3817 | case Builtin::BI__builtin_unwind_init: { | |||
| 3818 | Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init); | |||
| 3819 | return RValue::get(Builder.CreateCall(F)); | |||
| 3820 | } | |||
| 3821 | case Builtin::BI__builtin_extend_pointer: { | |||
| 3822 | // Extends a pointer to the size of an _Unwind_Word, which is | |||
| 3823 | // uint64_t on all platforms. Generally this gets poked into a | |||
| 3824 | // register and eventually used as an address, so if the | |||
| 3825 | // addressing registers are wider than pointers and the platform | |||
| 3826 | // doesn't implicitly ignore high-order bits when doing | |||
| 3827 | // addressing, we need to make sure we zext / sext based on | |||
| 3828 | // the platform's expectations. | |||
| 3829 | // | |||
| 3830 | // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html | |||
| 3831 | ||||
| 3832 | // Cast the pointer to intptr_t. | |||
| 3833 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 3834 | Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast"); | |||
| 3835 | ||||
| 3836 | // If that's 64 bits, we're done. | |||
| 3837 | if (IntPtrTy->getBitWidth() == 64) | |||
| 3838 | return RValue::get(Result); | |||
| 3839 | ||||
| 3840 | // Otherwise, ask the codegen data what to do. | |||
| 3841 | if (getTargetHooks().extendPointerWithSExt()) | |||
| 3842 | return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext")); | |||
| 3843 | else | |||
| 3844 | return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext")); | |||
| 3845 | } | |||
| 3846 | case Builtin::BI__builtin_setjmp: { | |||
| 3847 | // Buffer is a void**. | |||
| 3848 | Address Buf = EmitPointerWithAlignment(E->getArg(0)); | |||
| 3849 | ||||
| 3850 | // Store the frame pointer to the setjmp buffer. | |||
| 3851 | Value *FrameAddr = Builder.CreateCall( | |||
| 3852 | CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy), | |||
| 3853 | ConstantInt::get(Int32Ty, 0)); | |||
| 3854 | Builder.CreateStore(FrameAddr, Buf); | |||
| 3855 | ||||
| 3856 | // Store the stack pointer to the setjmp buffer. | |||
| 3857 | Value *StackAddr = | |||
| 3858 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave)); | |||
| 3859 | Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2); | |||
| 3860 | Builder.CreateStore(StackAddr, StackSaveSlot); | |||
| 3861 | ||||
| 3862 | // Call LLVM's EH setjmp, which is lightweight. | |||
| 3863 | Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp); | |||
| 3864 | Buf = Builder.CreateElementBitCast(Buf, Int8Ty); | |||
| 3865 | return RValue::get(Builder.CreateCall(F, Buf.getPointer())); | |||
| 3866 | } | |||
| 3867 | case Builtin::BI__builtin_longjmp: { | |||
| 3868 | Value *Buf = EmitScalarExpr(E->getArg(0)); | |||
| 3869 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); | |||
| 3870 | ||||
| 3871 | // Call LLVM's EH longjmp, which is lightweight. | |||
| 3872 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf); | |||
| 3873 | ||||
| 3874 | // longjmp doesn't return; mark this as unreachable. | |||
| 3875 | Builder.CreateUnreachable(); | |||
| 3876 | ||||
| 3877 | // We do need to preserve an insertion point. | |||
| 3878 | EmitBlock(createBasicBlock("longjmp.cont")); | |||
| 3879 | ||||
| 3880 | return RValue::get(nullptr); | |||
| 3881 | } | |||
| 3882 | case Builtin::BI__builtin_launder: { | |||
| 3883 | const Expr *Arg = E->getArg(0); | |||
| 3884 | QualType ArgTy = Arg->getType()->getPointeeType(); | |||
| 3885 | Value *Ptr = EmitScalarExpr(Arg); | |||
| 3886 | if (TypeRequiresBuiltinLaunder(CGM, ArgTy)) | |||
| 3887 | Ptr = Builder.CreateLaunderInvariantGroup(Ptr); | |||
| 3888 | ||||
| 3889 | return RValue::get(Ptr); | |||
| 3890 | } | |||
| 3891 | case Builtin::BI__sync_fetch_and_add: | |||
| 3892 | case Builtin::BI__sync_fetch_and_sub: | |||
| 3893 | case Builtin::BI__sync_fetch_and_or: | |||
| 3894 | case Builtin::BI__sync_fetch_and_and: | |||
| 3895 | case Builtin::BI__sync_fetch_and_xor: | |||
| 3896 | case Builtin::BI__sync_fetch_and_nand: | |||
| 3897 | case Builtin::BI__sync_add_and_fetch: | |||
| 3898 | case Builtin::BI__sync_sub_and_fetch: | |||
| 3899 | case Builtin::BI__sync_and_and_fetch: | |||
| 3900 | case Builtin::BI__sync_or_and_fetch: | |||
| 3901 | case Builtin::BI__sync_xor_and_fetch: | |||
| 3902 | case Builtin::BI__sync_nand_and_fetch: | |||
| 3903 | case Builtin::BI__sync_val_compare_and_swap: | |||
| 3904 | case Builtin::BI__sync_bool_compare_and_swap: | |||
| 3905 | case Builtin::BI__sync_lock_test_and_set: | |||
| 3906 | case Builtin::BI__sync_lock_release: | |||
| 3907 | case Builtin::BI__sync_swap: | |||
| 3908 | llvm_unreachable("Shouldn't make it through sema")::llvm::llvm_unreachable_internal("Shouldn't make it through sema" , "clang/lib/CodeGen/CGBuiltin.cpp", 3908); | |||
| 3909 | case Builtin::BI__sync_fetch_and_add_1: | |||
| 3910 | case Builtin::BI__sync_fetch_and_add_2: | |||
| 3911 | case Builtin::BI__sync_fetch_and_add_4: | |||
| 3912 | case Builtin::BI__sync_fetch_and_add_8: | |||
| 3913 | case Builtin::BI__sync_fetch_and_add_16: | |||
| 3914 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E); | |||
| 3915 | case Builtin::BI__sync_fetch_and_sub_1: | |||
| 3916 | case Builtin::BI__sync_fetch_and_sub_2: | |||
| 3917 | case Builtin::BI__sync_fetch_and_sub_4: | |||
| 3918 | case Builtin::BI__sync_fetch_and_sub_8: | |||
| 3919 | case Builtin::BI__sync_fetch_and_sub_16: | |||
| 3920 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E); | |||
| 3921 | case Builtin::BI__sync_fetch_and_or_1: | |||
| 3922 | case Builtin::BI__sync_fetch_and_or_2: | |||
| 3923 | case Builtin::BI__sync_fetch_and_or_4: | |||
| 3924 | case Builtin::BI__sync_fetch_and_or_8: | |||
| 3925 | case Builtin::BI__sync_fetch_and_or_16: | |||
| 3926 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E); | |||
| 3927 | case Builtin::BI__sync_fetch_and_and_1: | |||
| 3928 | case Builtin::BI__sync_fetch_and_and_2: | |||
| 3929 | case Builtin::BI__sync_fetch_and_and_4: | |||
| 3930 | case Builtin::BI__sync_fetch_and_and_8: | |||
| 3931 | case Builtin::BI__sync_fetch_and_and_16: | |||
| 3932 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E); | |||
| 3933 | case Builtin::BI__sync_fetch_and_xor_1: | |||
| 3934 | case Builtin::BI__sync_fetch_and_xor_2: | |||
| 3935 | case Builtin::BI__sync_fetch_and_xor_4: | |||
| 3936 | case Builtin::BI__sync_fetch_and_xor_8: | |||
| 3937 | case Builtin::BI__sync_fetch_and_xor_16: | |||
| 3938 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E); | |||
| 3939 | case Builtin::BI__sync_fetch_and_nand_1: | |||
| 3940 | case Builtin::BI__sync_fetch_and_nand_2: | |||
| 3941 | case Builtin::BI__sync_fetch_and_nand_4: | |||
| 3942 | case Builtin::BI__sync_fetch_and_nand_8: | |||
| 3943 | case Builtin::BI__sync_fetch_and_nand_16: | |||
| 3944 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E); | |||
| 3945 | ||||
| 3946 | // Clang extensions: not overloaded yet. | |||
| 3947 | case Builtin::BI__sync_fetch_and_min: | |||
| 3948 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E); | |||
| 3949 | case Builtin::BI__sync_fetch_and_max: | |||
| 3950 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E); | |||
| 3951 | case Builtin::BI__sync_fetch_and_umin: | |||
| 3952 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E); | |||
| 3953 | case Builtin::BI__sync_fetch_and_umax: | |||
| 3954 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E); | |||
| 3955 | ||||
| 3956 | case Builtin::BI__sync_add_and_fetch_1: | |||
| 3957 | case Builtin::BI__sync_add_and_fetch_2: | |||
| 3958 | case Builtin::BI__sync_add_and_fetch_4: | |||
| 3959 | case Builtin::BI__sync_add_and_fetch_8: | |||
| 3960 | case Builtin::BI__sync_add_and_fetch_16: | |||
| 3961 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E, | |||
| 3962 | llvm::Instruction::Add); | |||
| 3963 | case Builtin::BI__sync_sub_and_fetch_1: | |||
| 3964 | case Builtin::BI__sync_sub_and_fetch_2: | |||
| 3965 | case Builtin::BI__sync_sub_and_fetch_4: | |||
| 3966 | case Builtin::BI__sync_sub_and_fetch_8: | |||
| 3967 | case Builtin::BI__sync_sub_and_fetch_16: | |||
| 3968 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E, | |||
| 3969 | llvm::Instruction::Sub); | |||
| 3970 | case Builtin::BI__sync_and_and_fetch_1: | |||
| 3971 | case Builtin::BI__sync_and_and_fetch_2: | |||
| 3972 | case Builtin::BI__sync_and_and_fetch_4: | |||
| 3973 | case Builtin::BI__sync_and_and_fetch_8: | |||
| 3974 | case Builtin::BI__sync_and_and_fetch_16: | |||
| 3975 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E, | |||
| 3976 | llvm::Instruction::And); | |||
| 3977 | case Builtin::BI__sync_or_and_fetch_1: | |||
| 3978 | case Builtin::BI__sync_or_and_fetch_2: | |||
| 3979 | case Builtin::BI__sync_or_and_fetch_4: | |||
| 3980 | case Builtin::BI__sync_or_and_fetch_8: | |||
| 3981 | case Builtin::BI__sync_or_and_fetch_16: | |||
| 3982 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E, | |||
| 3983 | llvm::Instruction::Or); | |||
| 3984 | case Builtin::BI__sync_xor_and_fetch_1: | |||
| 3985 | case Builtin::BI__sync_xor_and_fetch_2: | |||
| 3986 | case Builtin::BI__sync_xor_and_fetch_4: | |||
| 3987 | case Builtin::BI__sync_xor_and_fetch_8: | |||
| 3988 | case Builtin::BI__sync_xor_and_fetch_16: | |||
| 3989 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E, | |||
| 3990 | llvm::Instruction::Xor); | |||
| 3991 | case Builtin::BI__sync_nand_and_fetch_1: | |||
| 3992 | case Builtin::BI__sync_nand_and_fetch_2: | |||
| 3993 | case Builtin::BI__sync_nand_and_fetch_4: | |||
| 3994 | case Builtin::BI__sync_nand_and_fetch_8: | |||
| 3995 | case Builtin::BI__sync_nand_and_fetch_16: | |||
| 3996 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E, | |||
| 3997 | llvm::Instruction::And, true); | |||
| 3998 | ||||
| 3999 | case Builtin::BI__sync_val_compare_and_swap_1: | |||
| 4000 | case Builtin::BI__sync_val_compare_and_swap_2: | |||
| 4001 | case Builtin::BI__sync_val_compare_and_swap_4: | |||
| 4002 | case Builtin::BI__sync_val_compare_and_swap_8: | |||
| 4003 | case Builtin::BI__sync_val_compare_and_swap_16: | |||
| 4004 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, false)); | |||
| 4005 | ||||
| 4006 | case Builtin::BI__sync_bool_compare_and_swap_1: | |||
| 4007 | case Builtin::BI__sync_bool_compare_and_swap_2: | |||
| 4008 | case Builtin::BI__sync_bool_compare_and_swap_4: | |||
| 4009 | case Builtin::BI__sync_bool_compare_and_swap_8: | |||
| 4010 | case Builtin::BI__sync_bool_compare_and_swap_16: | |||
| 4011 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, true)); | |||
| 4012 | ||||
| 4013 | case Builtin::BI__sync_swap_1: | |||
| 4014 | case Builtin::BI__sync_swap_2: | |||
| 4015 | case Builtin::BI__sync_swap_4: | |||
| 4016 | case Builtin::BI__sync_swap_8: | |||
| 4017 | case Builtin::BI__sync_swap_16: | |||
| 4018 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); | |||
| 4019 | ||||
| 4020 | case Builtin::BI__sync_lock_test_and_set_1: | |||
| 4021 | case Builtin::BI__sync_lock_test_and_set_2: | |||
| 4022 | case Builtin::BI__sync_lock_test_and_set_4: | |||
| 4023 | case Builtin::BI__sync_lock_test_and_set_8: | |||
| 4024 | case Builtin::BI__sync_lock_test_and_set_16: | |||
| 4025 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); | |||
| 4026 | ||||
| 4027 | case Builtin::BI__sync_lock_release_1: | |||
| 4028 | case Builtin::BI__sync_lock_release_2: | |||
| 4029 | case Builtin::BI__sync_lock_release_4: | |||
| 4030 | case Builtin::BI__sync_lock_release_8: | |||
| 4031 | case Builtin::BI__sync_lock_release_16: { | |||
| 4032 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 4033 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); | |||
| 4034 | CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy); | |||
| 4035 | llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(), | |||
| 4036 | StoreSize.getQuantity() * 8); | |||
| 4037 | Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo()); | |||
| 4038 | llvm::StoreInst *Store = | |||
| 4039 | Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr, | |||
| 4040 | StoreSize); | |||
| 4041 | Store->setAtomic(llvm::AtomicOrdering::Release); | |||
| 4042 | return RValue::get(nullptr); | |||
| 4043 | } | |||
| 4044 | ||||
| 4045 | case Builtin::BI__sync_synchronize: { | |||
| 4046 | // We assume this is supposed to correspond to a C++0x-style | |||
| 4047 | // sequentially-consistent fence (i.e. this is only usable for | |||
| 4048 | // synchronization, not device I/O or anything like that). This intrinsic | |||
| 4049 | // is really badly designed in the sense that in theory, there isn't | |||
| 4050 | // any way to safely use it... but in practice, it mostly works | |||
| 4051 | // to use it with non-atomic loads and stores to get acquire/release | |||
| 4052 | // semantics. | |||
| 4053 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent); | |||
| 4054 | return RValue::get(nullptr); | |||
| 4055 | } | |||
| 4056 | ||||
| 4057 | case Builtin::BI__builtin_nontemporal_load: | |||
| 4058 | return RValue::get(EmitNontemporalLoad(*this, E)); | |||
| 4059 | case Builtin::BI__builtin_nontemporal_store: | |||
| 4060 | return RValue::get(EmitNontemporalStore(*this, E)); | |||
| 4061 | case Builtin::BI__c11_atomic_is_lock_free: | |||
| 4062 | case Builtin::BI__atomic_is_lock_free: { | |||
| 4063 | // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the | |||
| 4064 | // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since | |||
| 4065 | // _Atomic(T) is always properly-aligned. | |||
| 4066 | const char *LibCallName = "__atomic_is_lock_free"; | |||
| 4067 | CallArgList Args; | |||
| 4068 | Args.add(RValue::get(EmitScalarExpr(E->getArg(0))), | |||
| 4069 | getContext().getSizeType()); | |||
| 4070 | if (BuiltinID == Builtin::BI__atomic_is_lock_free) | |||
| 4071 | Args.add(RValue::get(EmitScalarExpr(E->getArg(1))), | |||
| 4072 | getContext().VoidPtrTy); | |||
| 4073 | else | |||
| 4074 | Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)), | |||
| 4075 | getContext().VoidPtrTy); | |||
| 4076 | const CGFunctionInfo &FuncInfo = | |||
| 4077 | CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args); | |||
| 4078 | llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo); | |||
| 4079 | llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName); | |||
| 4080 | return EmitCall(FuncInfo, CGCallee::forDirect(Func), | |||
| 4081 | ReturnValueSlot(), Args); | |||
| 4082 | } | |||
| 4083 | ||||
| 4084 | case Builtin::BI__atomic_test_and_set: { | |||
| 4085 | // Look at the argument type to determine whether this is a volatile | |||
| 4086 | // operation. The parameter type is always volatile. | |||
| 4087 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); | |||
| 4088 | bool Volatile = | |||
| 4089 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); | |||
| 4090 | ||||
| 4091 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 4092 | unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace(); | |||
| 4093 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); | |||
| 4094 | Value *NewVal = Builder.getInt8(1); | |||
| 4095 | Value *Order = EmitScalarExpr(E->getArg(1)); | |||
| 4096 | if (isa<llvm::ConstantInt>(Order)) { | |||
| 4097 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); | |||
| 4098 | AtomicRMWInst *Result = nullptr; | |||
| 4099 | switch (ord) { | |||
| 4100 | case 0: // memory_order_relaxed | |||
| 4101 | default: // invalid order | |||
| 4102 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
| 4103 | llvm::AtomicOrdering::Monotonic); | |||
| 4104 | break; | |||
| 4105 | case 1: // memory_order_consume | |||
| 4106 | case 2: // memory_order_acquire | |||
| 4107 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
| 4108 | llvm::AtomicOrdering::Acquire); | |||
| 4109 | break; | |||
| 4110 | case 3: // memory_order_release | |||
| 4111 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
| 4112 | llvm::AtomicOrdering::Release); | |||
| 4113 | break; | |||
| 4114 | case 4: // memory_order_acq_rel | |||
| 4115 | ||||
| 4116 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
| 4117 | llvm::AtomicOrdering::AcquireRelease); | |||
| 4118 | break; | |||
| 4119 | case 5: // memory_order_seq_cst | |||
| 4120 | Result = Builder.CreateAtomicRMW( | |||
| 4121 | llvm::AtomicRMWInst::Xchg, Ptr, NewVal, | |||
| 4122 | llvm::AtomicOrdering::SequentiallyConsistent); | |||
| 4123 | break; | |||
| 4124 | } | |||
| 4125 | Result->setVolatile(Volatile); | |||
| 4126 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); | |||
| 4127 | } | |||
| 4128 | ||||
| 4129 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); | |||
| 4130 | ||||
| 4131 | llvm::BasicBlock *BBs[5] = { | |||
| 4132 | createBasicBlock("monotonic", CurFn), | |||
| 4133 | createBasicBlock("acquire", CurFn), | |||
| 4134 | createBasicBlock("release", CurFn), | |||
| 4135 | createBasicBlock("acqrel", CurFn), | |||
| 4136 | createBasicBlock("seqcst", CurFn) | |||
| 4137 | }; | |||
| 4138 | llvm::AtomicOrdering Orders[5] = { | |||
| 4139 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire, | |||
| 4140 | llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease, | |||
| 4141 | llvm::AtomicOrdering::SequentiallyConsistent}; | |||
| 4142 | ||||
| 4143 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); | |||
| 4144 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); | |||
| 4145 | ||||
| 4146 | Builder.SetInsertPoint(ContBB); | |||
| 4147 | PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set"); | |||
| 4148 | ||||
| 4149 | for (unsigned i = 0; i < 5; ++i) { | |||
| 4150 | Builder.SetInsertPoint(BBs[i]); | |||
| 4151 | AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, | |||
| 4152 | Ptr, NewVal, Orders[i]); | |||
| 4153 | RMW->setVolatile(Volatile); | |||
| 4154 | Result->addIncoming(RMW, BBs[i]); | |||
| 4155 | Builder.CreateBr(ContBB); | |||
| 4156 | } | |||
| 4157 | ||||
| 4158 | SI->addCase(Builder.getInt32(0), BBs[0]); | |||
| 4159 | SI->addCase(Builder.getInt32(1), BBs[1]); | |||
| 4160 | SI->addCase(Builder.getInt32(2), BBs[1]); | |||
| 4161 | SI->addCase(Builder.getInt32(3), BBs[2]); | |||
| 4162 | SI->addCase(Builder.getInt32(4), BBs[3]); | |||
| 4163 | SI->addCase(Builder.getInt32(5), BBs[4]); | |||
| 4164 | ||||
| 4165 | Builder.SetInsertPoint(ContBB); | |||
| 4166 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); | |||
| 4167 | } | |||
| 4168 | ||||
| 4169 | case Builtin::BI__atomic_clear: { | |||
| 4170 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); | |||
| 4171 | bool Volatile = | |||
| 4172 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); | |||
| 4173 | ||||
| 4174 | Address Ptr = EmitPointerWithAlignment(E->getArg(0)); | |||
| 4175 | Ptr = Builder.CreateElementBitCast(Ptr, Int8Ty); | |||
| 4176 | Value *NewVal = Builder.getInt8(0); | |||
| 4177 | Value *Order = EmitScalarExpr(E->getArg(1)); | |||
| 4178 | if (isa<llvm::ConstantInt>(Order)) { | |||
| 4179 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); | |||
| 4180 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); | |||
| 4181 | switch (ord) { | |||
| 4182 | case 0: // memory_order_relaxed | |||
| 4183 | default: // invalid order | |||
| 4184 | Store->setOrdering(llvm::AtomicOrdering::Monotonic); | |||
| 4185 | break; | |||
| 4186 | case 3: // memory_order_release | |||
| 4187 | Store->setOrdering(llvm::AtomicOrdering::Release); | |||
| 4188 | break; | |||
| 4189 | case 5: // memory_order_seq_cst | |||
| 4190 | Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent); | |||
| 4191 | break; | |||
| 4192 | } | |||
| 4193 | return RValue::get(nullptr); | |||
| 4194 | } | |||
| 4195 | ||||
| 4196 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); | |||
| 4197 | ||||
| 4198 | llvm::BasicBlock *BBs[3] = { | |||
| 4199 | createBasicBlock("monotonic", CurFn), | |||
| 4200 | createBasicBlock("release", CurFn), | |||
| 4201 | createBasicBlock("seqcst", CurFn) | |||
| 4202 | }; | |||
| 4203 | llvm::AtomicOrdering Orders[3] = { | |||
| 4204 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release, | |||
| 4205 | llvm::AtomicOrdering::SequentiallyConsistent}; | |||
| 4206 | ||||
| 4207 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); | |||
| 4208 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); | |||
| 4209 | ||||
| 4210 | for (unsigned i = 0; i < 3; ++i) { | |||
| 4211 | Builder.SetInsertPoint(BBs[i]); | |||
| 4212 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); | |||
| 4213 | Store->setOrdering(Orders[i]); | |||
| 4214 | Builder.CreateBr(ContBB); | |||
| 4215 | } | |||
| 4216 | ||||
| 4217 | SI->addCase(Builder.getInt32(0), BBs[0]); | |||
| 4218 | SI->addCase(Builder.getInt32(3), BBs[1]); | |||
| 4219 | SI->addCase(Builder.getInt32(5), BBs[2]); | |||
| 4220 | ||||
| 4221 | Builder.SetInsertPoint(ContBB); | |||
| 4222 | return RValue::get(nullptr); | |||
| 4223 | } | |||
| 4224 | ||||
| 4225 | case Builtin::BI__atomic_thread_fence: | |||
| 4226 | case Builtin::BI__atomic_signal_fence: | |||
| 4227 | case Builtin::BI__c11_atomic_thread_fence: | |||
| 4228 | case Builtin::BI__c11_atomic_signal_fence: { | |||
| 4229 | llvm::SyncScope::ID SSID; | |||
| 4230 | if (BuiltinID == Builtin::BI__atomic_signal_fence || | |||
| 4231 | BuiltinID == Builtin::BI__c11_atomic_signal_fence) | |||
| 4232 | SSID = llvm::SyncScope::SingleThread; | |||
| 4233 | else | |||
| 4234 | SSID = llvm::SyncScope::System; | |||
| 4235 | Value *Order = EmitScalarExpr(E->getArg(0)); | |||
| 4236 | if (isa<llvm::ConstantInt>(Order)) { | |||
| 4237 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); | |||
| 4238 | switch (ord) { | |||
| 4239 | case 0: // memory_order_relaxed | |||
| 4240 | default: // invalid order | |||
| 4241 | break; | |||
| 4242 | case 1: // memory_order_consume | |||
| 4243 | case 2: // memory_order_acquire | |||
| 4244 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); | |||
| 4245 | break; | |||
| 4246 | case 3: // memory_order_release | |||
| 4247 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); | |||
| 4248 | break; | |||
| 4249 | case 4: // memory_order_acq_rel | |||
| 4250 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); | |||
| 4251 | break; | |||
| 4252 | case 5: // memory_order_seq_cst | |||
| 4253 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); | |||
| 4254 | break; | |||
| 4255 | } | |||
| 4256 | return RValue::get(nullptr); | |||
| 4257 | } | |||
| 4258 | ||||
| 4259 | llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB; | |||
| 4260 | AcquireBB = createBasicBlock("acquire", CurFn); | |||
| 4261 | ReleaseBB = createBasicBlock("release", CurFn); | |||
| 4262 | AcqRelBB = createBasicBlock("acqrel", CurFn); | |||
| 4263 | SeqCstBB = createBasicBlock("seqcst", CurFn); | |||
| 4264 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); | |||
| 4265 | ||||
| 4266 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); | |||
| 4267 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB); | |||
| 4268 | ||||
| 4269 | Builder.SetInsertPoint(AcquireBB); | |||
| 4270 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); | |||
| 4271 | Builder.CreateBr(ContBB); | |||
| 4272 | SI->addCase(Builder.getInt32(1), AcquireBB); | |||
| 4273 | SI->addCase(Builder.getInt32(2), AcquireBB); | |||
| 4274 | ||||
| 4275 | Builder.SetInsertPoint(ReleaseBB); | |||
| 4276 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); | |||
| 4277 | Builder.CreateBr(ContBB); | |||
| 4278 | SI->addCase(Builder.getInt32(3), ReleaseBB); | |||
| 4279 | ||||
| 4280 | Builder.SetInsertPoint(AcqRelBB); | |||
| 4281 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); | |||
| 4282 | Builder.CreateBr(ContBB); | |||
| 4283 | SI->addCase(Builder.getInt32(4), AcqRelBB); | |||
| 4284 | ||||
| 4285 | Builder.SetInsertPoint(SeqCstBB); | |||
| 4286 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); | |||
| 4287 | Builder.CreateBr(ContBB); | |||
| 4288 | SI->addCase(Builder.getInt32(5), SeqCstBB); | |||
| 4289 | ||||
| 4290 | Builder.SetInsertPoint(ContBB); | |||
| 4291 | return RValue::get(nullptr); | |||
| 4292 | } | |||
| 4293 | ||||
| 4294 | case Builtin::BI__builtin_signbit: | |||
| 4295 | case Builtin::BI__builtin_signbitf: | |||
| 4296 | case Builtin::BI__builtin_signbitl: { | |||
| 4297 | return RValue::get( | |||
| 4298 | Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))), | |||
| 4299 | ConvertType(E->getType()))); | |||
| 4300 | } | |||
| 4301 | case Builtin::BI__warn_memset_zero_len: | |||
| 4302 | return RValue::getIgnored(); | |||
| 4303 | case Builtin::BI__annotation: { | |||
| 4304 | // Re-encode each wide string to UTF8 and make an MDString. | |||
| 4305 | SmallVector<Metadata *, 1> Strings; | |||
| 4306 | for (const Expr *Arg : E->arguments()) { | |||
| 4307 | const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts()); | |||
| 4308 | assert(Str->getCharByteWidth() == 2)(static_cast <bool> (Str->getCharByteWidth() == 2) ? void (0) : __assert_fail ("Str->getCharByteWidth() == 2", "clang/lib/CodeGen/CGBuiltin.cpp", 4308, __extension__ __PRETTY_FUNCTION__ )); | |||
| 4309 | StringRef WideBytes = Str->getBytes(); | |||
| 4310 | std::string StrUtf8; | |||
| 4311 | if (!convertUTF16ToUTF8String( | |||
| 4312 | makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) { | |||
| 4313 | CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument"); | |||
| 4314 | continue; | |||
| 4315 | } | |||
| 4316 | Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8)); | |||
| 4317 | } | |||
| 4318 | ||||
| 4319 | // Build and MDTuple of MDStrings and emit the intrinsic call. | |||
| 4320 | llvm::Function *F = | |||
| 4321 | CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {}); | |||
| 4322 | MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings); | |||
| 4323 | Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple)); | |||
| 4324 | return RValue::getIgnored(); | |||
| 4325 | } | |||
| 4326 | case Builtin::BI__builtin_annotation: { | |||
| 4327 | llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0)); | |||
| 4328 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::annotation, | |||
| 4329 | AnnVal->getType()); | |||
| 4330 | ||||
| 4331 | // Get the annotation string, go through casts. Sema requires this to be a | |||
| 4332 | // non-wide string literal, potentially casted, so the cast<> is safe. | |||
| 4333 | const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts(); | |||
| 4334 | StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString(); | |||
| 4335 | return RValue::get( | |||
| 4336 | EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc(), nullptr)); | |||
| 4337 | } | |||
| 4338 | case Builtin::BI__builtin_addcb: | |||
| 4339 | case Builtin::BI__builtin_addcs: | |||
| 4340 | case Builtin::BI__builtin_addc: | |||
| 4341 | case Builtin::BI__builtin_addcl: | |||
| 4342 | case Builtin::BI__builtin_addcll: | |||
| 4343 | case Builtin::BI__builtin_subcb: | |||
| 4344 | case Builtin::BI__builtin_subcs: | |||
| 4345 | case Builtin::BI__builtin_subc: | |||
| 4346 | case Builtin::BI__builtin_subcl: | |||
| 4347 | case Builtin::BI__builtin_subcll: { | |||
| 4348 | ||||
| 4349 | // We translate all of these builtins from expressions of the form: | |||
| 4350 | // int x = ..., y = ..., carryin = ..., carryout, result; | |||
| 4351 | // result = __builtin_addc(x, y, carryin, &carryout); | |||
| 4352 | // | |||
| 4353 | // to LLVM IR of the form: | |||
| 4354 | // | |||
| 4355 | // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y) | |||
| 4356 | // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0 | |||
| 4357 | // %carry1 = extractvalue {i32, i1} %tmp1, 1 | |||
| 4358 | // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1, | |||
| 4359 | // i32 %carryin) | |||
| 4360 | // %result = extractvalue {i32, i1} %tmp2, 0 | |||
| 4361 | // %carry2 = extractvalue {i32, i1} %tmp2, 1 | |||
| 4362 | // %tmp3 = or i1 %carry1, %carry2 | |||
| 4363 | // %tmp4 = zext i1 %tmp3 to i32 | |||
| 4364 | // store i32 %tmp4, i32* %carryout | |||
| 4365 | ||||
| 4366 | // Scalarize our inputs. | |||
| 4367 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 4368 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 4369 | llvm::Value *Carryin = EmitScalarExpr(E->getArg(2)); | |||
| 4370 | Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3)); | |||
| 4371 | ||||
| 4372 | // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow. | |||
| 4373 | llvm::Intrinsic::ID IntrinsicId; | |||
| 4374 | switch (BuiltinID) { | |||
| 4375 | default: llvm_unreachable("Unknown multiprecision builtin id.")::llvm::llvm_unreachable_internal("Unknown multiprecision builtin id." , "clang/lib/CodeGen/CGBuiltin.cpp", 4375); | |||
| 4376 | case Builtin::BI__builtin_addcb: | |||
| 4377 | case Builtin::BI__builtin_addcs: | |||
| 4378 | case Builtin::BI__builtin_addc: | |||
| 4379 | case Builtin::BI__builtin_addcl: | |||
| 4380 | case Builtin::BI__builtin_addcll: | |||
| 4381 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; | |||
| 4382 | break; | |||
| 4383 | case Builtin::BI__builtin_subcb: | |||
| 4384 | case Builtin::BI__builtin_subcs: | |||
| 4385 | case Builtin::BI__builtin_subc: | |||
| 4386 | case Builtin::BI__builtin_subcl: | |||
| 4387 | case Builtin::BI__builtin_subcll: | |||
| 4388 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; | |||
| 4389 | break; | |||
| 4390 | } | |||
| 4391 | ||||
| 4392 | // Construct our resulting LLVM IR expression. | |||
| 4393 | llvm::Value *Carry1; | |||
| 4394 | llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId, | |||
| 4395 | X, Y, Carry1); | |||
| 4396 | llvm::Value *Carry2; | |||
| 4397 | llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId, | |||
| 4398 | Sum1, Carryin, Carry2); | |||
| 4399 | llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2), | |||
| 4400 | X->getType()); | |||
| 4401 | Builder.CreateStore(CarryOut, CarryOutPtr); | |||
| 4402 | return RValue::get(Sum2); | |||
| 4403 | } | |||
| 4404 | ||||
| 4405 | case Builtin::BI__builtin_add_overflow: | |||
| 4406 | case Builtin::BI__builtin_sub_overflow: | |||
| 4407 | case Builtin::BI__builtin_mul_overflow: { | |||
| 4408 | const clang::Expr *LeftArg = E->getArg(0); | |||
| 4409 | const clang::Expr *RightArg = E->getArg(1); | |||
| 4410 | const clang::Expr *ResultArg = E->getArg(2); | |||
| 4411 | ||||
| 4412 | clang::QualType ResultQTy = | |||
| 4413 | ResultArg->getType()->castAs<PointerType>()->getPointeeType(); | |||
| 4414 | ||||
| 4415 | WidthAndSignedness LeftInfo = | |||
| 4416 | getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType()); | |||
| 4417 | WidthAndSignedness RightInfo = | |||
| 4418 | getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType()); | |||
| 4419 | WidthAndSignedness ResultInfo = | |||
| 4420 | getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy); | |||
| 4421 | ||||
| 4422 | // Handle mixed-sign multiplication as a special case, because adding | |||
| 4423 | // runtime or backend support for our generic irgen would be too expensive. | |||
| 4424 | if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo)) | |||
| 4425 | return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg, | |||
| 4426 | RightInfo, ResultArg, ResultQTy, | |||
| 4427 | ResultInfo); | |||
| 4428 | ||||
| 4429 | if (isSpecialUnsignedMultiplySignedResult(BuiltinID, LeftInfo, RightInfo, | |||
| 4430 | ResultInfo)) | |||
| 4431 | return EmitCheckedUnsignedMultiplySignedResult( | |||
| 4432 | *this, LeftArg, LeftInfo, RightArg, RightInfo, ResultArg, ResultQTy, | |||
| 4433 | ResultInfo); | |||
| 4434 | ||||
| 4435 | WidthAndSignedness EncompassingInfo = | |||
| 4436 | EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo}); | |||
| 4437 | ||||
| 4438 | llvm::Type *EncompassingLLVMTy = | |||
| 4439 | llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width); | |||
| 4440 | ||||
| 4441 | llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy); | |||
| 4442 | ||||
| 4443 | llvm::Intrinsic::ID IntrinsicId; | |||
| 4444 | switch (BuiltinID) { | |||
| 4445 | default: | |||
| 4446 | llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "clang/lib/CodeGen/CGBuiltin.cpp", 4446); | |||
| 4447 | case Builtin::BI__builtin_add_overflow: | |||
| 4448 | IntrinsicId = EncompassingInfo.Signed | |||
| 4449 | ? llvm::Intrinsic::sadd_with_overflow | |||
| 4450 | : llvm::Intrinsic::uadd_with_overflow; | |||
| 4451 | break; | |||
| 4452 | case Builtin::BI__builtin_sub_overflow: | |||
| 4453 | IntrinsicId = EncompassingInfo.Signed | |||
| 4454 | ? llvm::Intrinsic::ssub_with_overflow | |||
| 4455 | : llvm::Intrinsic::usub_with_overflow; | |||
| 4456 | break; | |||
| 4457 | case Builtin::BI__builtin_mul_overflow: | |||
| 4458 | IntrinsicId = EncompassingInfo.Signed | |||
| 4459 | ? llvm::Intrinsic::smul_with_overflow | |||
| 4460 | : llvm::Intrinsic::umul_with_overflow; | |||
| 4461 | break; | |||
| 4462 | } | |||
| 4463 | ||||
| 4464 | llvm::Value *Left = EmitScalarExpr(LeftArg); | |||
| 4465 | llvm::Value *Right = EmitScalarExpr(RightArg); | |||
| 4466 | Address ResultPtr = EmitPointerWithAlignment(ResultArg); | |||
| 4467 | ||||
| 4468 | // Extend each operand to the encompassing type. | |||
| 4469 | Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed); | |||
| 4470 | Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed); | |||
| 4471 | ||||
| 4472 | // Perform the operation on the extended values. | |||
| 4473 | llvm::Value *Overflow, *Result; | |||
| 4474 | Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow); | |||
| 4475 | ||||
| 4476 | if (EncompassingInfo.Width > ResultInfo.Width) { | |||
| 4477 | // The encompassing type is wider than the result type, so we need to | |||
| 4478 | // truncate it. | |||
| 4479 | llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy); | |||
| 4480 | ||||
| 4481 | // To see if the truncation caused an overflow, we will extend | |||
| 4482 | // the result and then compare it to the original result. | |||
| 4483 | llvm::Value *ResultTruncExt = Builder.CreateIntCast( | |||
| 4484 | ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed); | |||
| 4485 | llvm::Value *TruncationOverflow = | |||
| 4486 | Builder.CreateICmpNE(Result, ResultTruncExt); | |||
| 4487 | ||||
| 4488 | Overflow = Builder.CreateOr(Overflow, TruncationOverflow); | |||
| 4489 | Result = ResultTrunc; | |||
| 4490 | } | |||
| 4491 | ||||
| 4492 | // Finally, store the result using the pointer. | |||
| 4493 | bool isVolatile = | |||
| 4494 | ResultArg->getType()->getPointeeType().isVolatileQualified(); | |||
| 4495 | Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile); | |||
| 4496 | ||||
| 4497 | return RValue::get(Overflow); | |||
| 4498 | } | |||
| 4499 | ||||
| 4500 | case Builtin::BI__builtin_uadd_overflow: | |||
| 4501 | case Builtin::BI__builtin_uaddl_overflow: | |||
| 4502 | case Builtin::BI__builtin_uaddll_overflow: | |||
| 4503 | case Builtin::BI__builtin_usub_overflow: | |||
| 4504 | case Builtin::BI__builtin_usubl_overflow: | |||
| 4505 | case Builtin::BI__builtin_usubll_overflow: | |||
| 4506 | case Builtin::BI__builtin_umul_overflow: | |||
| 4507 | case Builtin::BI__builtin_umull_overflow: | |||
| 4508 | case Builtin::BI__builtin_umulll_overflow: | |||
| 4509 | case Builtin::BI__builtin_sadd_overflow: | |||
| 4510 | case Builtin::BI__builtin_saddl_overflow: | |||
| 4511 | case Builtin::BI__builtin_saddll_overflow: | |||
| 4512 | case Builtin::BI__builtin_ssub_overflow: | |||
| 4513 | case Builtin::BI__builtin_ssubl_overflow: | |||
| 4514 | case Builtin::BI__builtin_ssubll_overflow: | |||
| 4515 | case Builtin::BI__builtin_smul_overflow: | |||
| 4516 | case Builtin::BI__builtin_smull_overflow: | |||
| 4517 | case Builtin::BI__builtin_smulll_overflow: { | |||
| 4518 | ||||
| 4519 | // We translate all of these builtins directly to the relevant llvm IR node. | |||
| 4520 | ||||
| 4521 | // Scalarize our inputs. | |||
| 4522 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 4523 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 4524 | Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2)); | |||
| 4525 | ||||
| 4526 | // Decide which of the overflow intrinsics we are lowering to: | |||
| 4527 | llvm::Intrinsic::ID IntrinsicId; | |||
| 4528 | switch (BuiltinID) { | |||
| 4529 | default: llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "clang/lib/CodeGen/CGBuiltin.cpp", 4529); | |||
| 4530 | case Builtin::BI__builtin_uadd_overflow: | |||
| 4531 | case Builtin::BI__builtin_uaddl_overflow: | |||
| 4532 | case Builtin::BI__builtin_uaddll_overflow: | |||
| 4533 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; | |||
| 4534 | break; | |||
| 4535 | case Builtin::BI__builtin_usub_overflow: | |||
| 4536 | case Builtin::BI__builtin_usubl_overflow: | |||
| 4537 | case Builtin::BI__builtin_usubll_overflow: | |||
| 4538 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; | |||
| 4539 | break; | |||
| 4540 | case Builtin::BI__builtin_umul_overflow: | |||
| 4541 | case Builtin::BI__builtin_umull_overflow: | |||
| 4542 | case Builtin::BI__builtin_umulll_overflow: | |||
| 4543 | IntrinsicId = llvm::Intrinsic::umul_with_overflow; | |||
| 4544 | break; | |||
| 4545 | case Builtin::BI__builtin_sadd_overflow: | |||
| 4546 | case Builtin::BI__builtin_saddl_overflow: | |||
| 4547 | case Builtin::BI__builtin_saddll_overflow: | |||
| 4548 | IntrinsicId = llvm::Intrinsic::sadd_with_overflow; | |||
| 4549 | break; | |||
| 4550 | case Builtin::BI__builtin_ssub_overflow: | |||
| 4551 | case Builtin::BI__builtin_ssubl_overflow: | |||
| 4552 | case Builtin::BI__builtin_ssubll_overflow: | |||
| 4553 | IntrinsicId = llvm::Intrinsic::ssub_with_overflow; | |||
| 4554 | break; | |||
| 4555 | case Builtin::BI__builtin_smul_overflow: | |||
| 4556 | case Builtin::BI__builtin_smull_overflow: | |||
| 4557 | case Builtin::BI__builtin_smulll_overflow: | |||
| 4558 | IntrinsicId = llvm::Intrinsic::smul_with_overflow; | |||
| 4559 | break; | |||
| 4560 | } | |||
| 4561 | ||||
| 4562 | ||||
| 4563 | llvm::Value *Carry; | |||
| 4564 | llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry); | |||
| 4565 | Builder.CreateStore(Sum, SumOutPtr); | |||
| 4566 | ||||
| 4567 | return RValue::get(Carry); | |||
| 4568 | } | |||
| 4569 | case Builtin::BI__builtin_addressof: | |||
| 4570 | return RValue::get(EmitLValue(E->getArg(0)).getPointer(*this)); | |||
| 4571 | case Builtin::BI__builtin_function_start: | |||
| 4572 | return RValue::get(CGM.GetFunctionStart( | |||
| 4573 | E->getArg(0)->getAsBuiltinConstantDeclRef(CGM.getContext()))); | |||
| 4574 | case Builtin::BI__builtin_operator_new: | |||
| 4575 | return EmitBuiltinNewDeleteCall( | |||
| 4576 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false); | |||
| 4577 | case Builtin::BI__builtin_operator_delete: | |||
| 4578 | return EmitBuiltinNewDeleteCall( | |||
| 4579 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true); | |||
| 4580 | ||||
| 4581 | case Builtin::BI__builtin_is_aligned: | |||
| 4582 | return EmitBuiltinIsAligned(E); | |||
| 4583 | case Builtin::BI__builtin_align_up: | |||
| 4584 | return EmitBuiltinAlignTo(E, true); | |||
| 4585 | case Builtin::BI__builtin_align_down: | |||
| 4586 | return EmitBuiltinAlignTo(E, false); | |||
| 4587 | ||||
| 4588 | case Builtin::BI__noop: | |||
| 4589 | // __noop always evaluates to an integer literal zero. | |||
| 4590 | return RValue::get(ConstantInt::get(IntTy, 0)); | |||
| 4591 | case Builtin::BI__builtin_call_with_static_chain: { | |||
| 4592 | const CallExpr *Call = cast<CallExpr>(E->getArg(0)); | |||
| 4593 | const Expr *Chain = E->getArg(1); | |||
| 4594 | return EmitCall(Call->getCallee()->getType(), | |||
| 4595 | EmitCallee(Call->getCallee()), Call, ReturnValue, | |||
| 4596 | EmitScalarExpr(Chain)); | |||
| 4597 | } | |||
| 4598 | case Builtin::BI_InterlockedExchange8: | |||
| 4599 | case Builtin::BI_InterlockedExchange16: | |||
| 4600 | case Builtin::BI_InterlockedExchange: | |||
| 4601 | case Builtin::BI_InterlockedExchangePointer: | |||
| 4602 | return RValue::get( | |||
| 4603 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E)); | |||
| 4604 | case Builtin::BI_InterlockedCompareExchangePointer: | |||
| 4605 | case Builtin::BI_InterlockedCompareExchangePointer_nf: { | |||
| 4606 | llvm::Type *RTy; | |||
| 4607 | llvm::IntegerType *IntType = | |||
| 4608 | IntegerType::get(getLLVMContext(), | |||
| 4609 | getContext().getTypeSize(E->getType())); | |||
| 4610 | llvm::Type *IntPtrType = IntType->getPointerTo(); | |||
| 4611 | ||||
| 4612 | llvm::Value *Destination = | |||
| 4613 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType); | |||
| 4614 | ||||
| 4615 | llvm::Value *Exchange = EmitScalarExpr(E->getArg(1)); | |||
| 4616 | RTy = Exchange->getType(); | |||
| 4617 | Exchange = Builder.CreatePtrToInt(Exchange, IntType); | |||
| 4618 | ||||
| 4619 | llvm::Value *Comparand = | |||
| 4620 | Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType); | |||
| 4621 | ||||
| 4622 | auto Ordering = | |||
| 4623 | BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ? | |||
| 4624 | AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent; | |||
| 4625 | ||||
| 4626 | auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, | |||
| 4627 | Ordering, Ordering); | |||
| 4628 | Result->setVolatile(true); | |||
| 4629 | ||||
| 4630 | return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result, | |||
| 4631 | 0), | |||
| 4632 | RTy)); | |||
| 4633 | } | |||
| 4634 | case Builtin::BI_InterlockedCompareExchange8: | |||
| 4635 | case Builtin::BI_InterlockedCompareExchange16: | |||
| 4636 | case Builtin::BI_InterlockedCompareExchange: | |||
| 4637 | case Builtin::BI_InterlockedCompareExchange64: | |||
| 4638 | return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E)); | |||
| 4639 | case Builtin::BI_InterlockedIncrement16: | |||
| 4640 | case Builtin::BI_InterlockedIncrement: | |||
| 4641 | return RValue::get( | |||
| 4642 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E)); | |||
| 4643 | case Builtin::BI_InterlockedDecrement16: | |||
| 4644 | case Builtin::BI_InterlockedDecrement: | |||
| 4645 | return RValue::get( | |||
| 4646 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E)); | |||
| 4647 | case Builtin::BI_InterlockedAnd8: | |||
| 4648 | case Builtin::BI_InterlockedAnd16: | |||
| 4649 | case Builtin::BI_InterlockedAnd: | |||
| 4650 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E)); | |||
| 4651 | case Builtin::BI_InterlockedExchangeAdd8: | |||
| 4652 | case Builtin::BI_InterlockedExchangeAdd16: | |||
| 4653 | case Builtin::BI_InterlockedExchangeAdd: | |||
| 4654 | return RValue::get( | |||
| 4655 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E)); | |||
| 4656 | case Builtin::BI_InterlockedExchangeSub8: | |||
| 4657 | case Builtin::BI_InterlockedExchangeSub16: | |||
| 4658 | case Builtin::BI_InterlockedExchangeSub: | |||
| 4659 | return RValue::get( | |||
| 4660 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E)); | |||
| 4661 | case Builtin::BI_InterlockedOr8: | |||
| 4662 | case Builtin::BI_InterlockedOr16: | |||
| 4663 | case Builtin::BI_InterlockedOr: | |||
| 4664 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E)); | |||
| 4665 | case Builtin::BI_InterlockedXor8: | |||
| 4666 | case Builtin::BI_InterlockedXor16: | |||
| 4667 | case Builtin::BI_InterlockedXor: | |||
| 4668 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E)); | |||
| 4669 | ||||
| 4670 | case Builtin::BI_bittest64: | |||
| 4671 | case Builtin::BI_bittest: | |||
| 4672 | case Builtin::BI_bittestandcomplement64: | |||
| 4673 | case Builtin::BI_bittestandcomplement: | |||
| 4674 | case Builtin::BI_bittestandreset64: | |||
| 4675 | case Builtin::BI_bittestandreset: | |||
| 4676 | case Builtin::BI_bittestandset64: | |||
| 4677 | case Builtin::BI_bittestandset: | |||
| 4678 | case Builtin::BI_interlockedbittestandreset: | |||
| 4679 | case Builtin::BI_interlockedbittestandreset64: | |||
| 4680 | case Builtin::BI_interlockedbittestandset64: | |||
| 4681 | case Builtin::BI_interlockedbittestandset: | |||
| 4682 | case Builtin::BI_interlockedbittestandset_acq: | |||
| 4683 | case Builtin::BI_interlockedbittestandset_rel: | |||
| 4684 | case Builtin::BI_interlockedbittestandset_nf: | |||
| 4685 | case Builtin::BI_interlockedbittestandreset_acq: | |||
| 4686 | case Builtin::BI_interlockedbittestandreset_rel: | |||
| 4687 | case Builtin::BI_interlockedbittestandreset_nf: | |||
| 4688 | return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E)); | |||
| 4689 | ||||
| 4690 | // These builtins exist to emit regular volatile loads and stores not | |||
| 4691 | // affected by the -fms-volatile setting. | |||
| 4692 | case Builtin::BI__iso_volatile_load8: | |||
| 4693 | case Builtin::BI__iso_volatile_load16: | |||
| 4694 | case Builtin::BI__iso_volatile_load32: | |||
| 4695 | case Builtin::BI__iso_volatile_load64: | |||
| 4696 | return RValue::get(EmitISOVolatileLoad(*this, E)); | |||
| 4697 | case Builtin::BI__iso_volatile_store8: | |||
| 4698 | case Builtin::BI__iso_volatile_store16: | |||
| 4699 | case Builtin::BI__iso_volatile_store32: | |||
| 4700 | case Builtin::BI__iso_volatile_store64: | |||
| 4701 | return RValue::get(EmitISOVolatileStore(*this, E)); | |||
| 4702 | ||||
| 4703 | case Builtin::BI__exception_code: | |||
| 4704 | case Builtin::BI_exception_code: | |||
| 4705 | return RValue::get(EmitSEHExceptionCode()); | |||
| 4706 | case Builtin::BI__exception_info: | |||
| 4707 | case Builtin::BI_exception_info: | |||
| 4708 | return RValue::get(EmitSEHExceptionInfo()); | |||
| 4709 | case Builtin::BI__abnormal_termination: | |||
| 4710 | case Builtin::BI_abnormal_termination: | |||
| 4711 | return RValue::get(EmitSEHAbnormalTermination()); | |||
| 4712 | case Builtin::BI_setjmpex: | |||
| 4713 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && | |||
| 4714 | E->getArg(0)->getType()->isPointerType()) | |||
| 4715 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); | |||
| 4716 | break; | |||
| 4717 | case Builtin::BI_setjmp: | |||
| 4718 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && | |||
| 4719 | E->getArg(0)->getType()->isPointerType()) { | |||
| 4720 | if (getTarget().getTriple().getArch() == llvm::Triple::x86) | |||
| 4721 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E); | |||
| 4722 | else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64) | |||
| 4723 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); | |||
| 4724 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E); | |||
| 4725 | } | |||
| 4726 | break; | |||
| 4727 | ||||
| 4728 | case Builtin::BI__GetExceptionInfo: { | |||
| 4729 | if (llvm::GlobalVariable *GV = | |||
| 4730 | CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType())) | |||
| 4731 | return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy)); | |||
| 4732 | break; | |||
| 4733 | } | |||
| 4734 | ||||
| 4735 | case Builtin::BI__fastfail: | |||
| 4736 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E)); | |||
| 4737 | ||||
| 4738 | case Builtin::BI__builtin_coro_size: { | |||
| 4739 | auto & Context = getContext(); | |||
| 4740 | auto SizeTy = Context.getSizeType(); | |||
| 4741 | auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy)); | |||
| 4742 | Function *F = CGM.getIntrinsic(Intrinsic::coro_size, T); | |||
| 4743 | return RValue::get(Builder.CreateCall(F)); | |||
| 4744 | } | |||
| 4745 | ||||
| 4746 | case Builtin::BI__builtin_coro_id: | |||
| 4747 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_id); | |||
| 4748 | case Builtin::BI__builtin_coro_promise: | |||
| 4749 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise); | |||
| 4750 | case Builtin::BI__builtin_coro_resume: | |||
| 4751 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume); | |||
| 4752 | case Builtin::BI__builtin_coro_frame: | |||
| 4753 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame); | |||
| 4754 | case Builtin::BI__builtin_coro_noop: | |||
| 4755 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop); | |||
| 4756 | case Builtin::BI__builtin_coro_free: | |||
| 4757 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_free); | |||
| 4758 | case Builtin::BI__builtin_coro_destroy: | |||
| 4759 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy); | |||
| 4760 | case Builtin::BI__builtin_coro_done: | |||
| 4761 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_done); | |||
| 4762 | case Builtin::BI__builtin_coro_alloc: | |||
| 4763 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc); | |||
| 4764 | case Builtin::BI__builtin_coro_begin: | |||
| 4765 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin); | |||
| 4766 | case Builtin::BI__builtin_coro_end: | |||
| 4767 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_end); | |||
| 4768 | case Builtin::BI__builtin_coro_suspend: | |||
| 4769 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend); | |||
| 4770 | ||||
| 4771 | // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions | |||
| 4772 | case Builtin::BIread_pipe: | |||
| 4773 | case Builtin::BIwrite_pipe: { | |||
| 4774 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), | |||
| 4775 | *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
| 4776 | CGOpenCLRuntime OpenCLRT(CGM); | |||
| 4777 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
| 4778 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
| 4779 | ||||
| 4780 | // Type of the generic packet parameter. | |||
| 4781 | unsigned GenericAS = | |||
| 4782 | getContext().getTargetAddressSpace(LangAS::opencl_generic); | |||
| 4783 | llvm::Type *I8PTy = llvm::PointerType::get( | |||
| 4784 | llvm::Type::getInt8Ty(getLLVMContext()), GenericAS); | |||
| 4785 | ||||
| 4786 | // Testing which overloaded version we should generate the call for. | |||
| 4787 | if (2U == E->getNumArgs()) { | |||
| 4788 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2" | |||
| 4789 | : "__write_pipe_2"; | |||
| 4790 | // Creating a generic function type to be able to call with any builtin or | |||
| 4791 | // user defined type. | |||
| 4792 | llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty}; | |||
| 4793 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
| 4794 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 4795 | Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy); | |||
| 4796 | return RValue::get( | |||
| 4797 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
| 4798 | {Arg0, BCast, PacketSize, PacketAlign})); | |||
| 4799 | } else { | |||
| 4800 | assert(4 == E->getNumArgs() &&(static_cast <bool> (4 == E->getNumArgs() && "Illegal number of parameters to pipe function") ? void (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 4801, __extension__ __PRETTY_FUNCTION__ )) | |||
| 4801 | "Illegal number of parameters to pipe function")(static_cast <bool> (4 == E->getNumArgs() && "Illegal number of parameters to pipe function") ? void (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 4801, __extension__ __PRETTY_FUNCTION__ )); | |||
| 4802 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4" | |||
| 4803 | : "__write_pipe_4"; | |||
| 4804 | ||||
| 4805 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy, | |||
| 4806 | Int32Ty, Int32Ty}; | |||
| 4807 | Value *Arg2 = EmitScalarExpr(E->getArg(2)), | |||
| 4808 | *Arg3 = EmitScalarExpr(E->getArg(3)); | |||
| 4809 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
| 4810 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 4811 | Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy); | |||
| 4812 | // We know the third argument is an integer type, but we may need to cast | |||
| 4813 | // it to i32. | |||
| 4814 | if (Arg2->getType() != Int32Ty) | |||
| 4815 | Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty); | |||
| 4816 | return RValue::get( | |||
| 4817 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
| 4818 | {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign})); | |||
| 4819 | } | |||
| 4820 | } | |||
| 4821 | // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write | |||
| 4822 | // functions | |||
| 4823 | case Builtin::BIreserve_read_pipe: | |||
| 4824 | case Builtin::BIreserve_write_pipe: | |||
| 4825 | case Builtin::BIwork_group_reserve_read_pipe: | |||
| 4826 | case Builtin::BIwork_group_reserve_write_pipe: | |||
| 4827 | case Builtin::BIsub_group_reserve_read_pipe: | |||
| 4828 | case Builtin::BIsub_group_reserve_write_pipe: { | |||
| 4829 | // Composing the mangled name for the function. | |||
| 4830 | const char *Name; | |||
| 4831 | if (BuiltinID == Builtin::BIreserve_read_pipe) | |||
| 4832 | Name = "__reserve_read_pipe"; | |||
| 4833 | else if (BuiltinID == Builtin::BIreserve_write_pipe) | |||
| 4834 | Name = "__reserve_write_pipe"; | |||
| 4835 | else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe) | |||
| 4836 | Name = "__work_group_reserve_read_pipe"; | |||
| 4837 | else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe) | |||
| 4838 | Name = "__work_group_reserve_write_pipe"; | |||
| 4839 | else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe) | |||
| 4840 | Name = "__sub_group_reserve_read_pipe"; | |||
| 4841 | else | |||
| 4842 | Name = "__sub_group_reserve_write_pipe"; | |||
| 4843 | ||||
| 4844 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), | |||
| 4845 | *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
| 4846 | llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy); | |||
| 4847 | CGOpenCLRuntime OpenCLRT(CGM); | |||
| 4848 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
| 4849 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
| 4850 | ||||
| 4851 | // Building the generic function prototype. | |||
| 4852 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty}; | |||
| 4853 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
| 4854 | ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 4855 | // We know the second argument is an integer type, but we may need to cast | |||
| 4856 | // it to i32. | |||
| 4857 | if (Arg1->getType() != Int32Ty) | |||
| 4858 | Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty); | |||
| 4859 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
| 4860 | {Arg0, Arg1, PacketSize, PacketAlign})); | |||
| 4861 | } | |||
| 4862 | // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write | |||
| 4863 | // functions | |||
| 4864 | case Builtin::BIcommit_read_pipe: | |||
| 4865 | case Builtin::BIcommit_write_pipe: | |||
| 4866 | case Builtin::BIwork_group_commit_read_pipe: | |||
| 4867 | case Builtin::BIwork_group_commit_write_pipe: | |||
| 4868 | case Builtin::BIsub_group_commit_read_pipe: | |||
| 4869 | case Builtin::BIsub_group_commit_write_pipe: { | |||
| 4870 | const char *Name; | |||
| 4871 | if (BuiltinID == Builtin::BIcommit_read_pipe) | |||
| 4872 | Name = "__commit_read_pipe"; | |||
| 4873 | else if (BuiltinID == Builtin::BIcommit_write_pipe) | |||
| 4874 | Name = "__commit_write_pipe"; | |||
| 4875 | else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe) | |||
| 4876 | Name = "__work_group_commit_read_pipe"; | |||
| 4877 | else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe) | |||
| 4878 | Name = "__work_group_commit_write_pipe"; | |||
| 4879 | else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe) | |||
| 4880 | Name = "__sub_group_commit_read_pipe"; | |||
| 4881 | else | |||
| 4882 | Name = "__sub_group_commit_write_pipe"; | |||
| 4883 | ||||
| 4884 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), | |||
| 4885 | *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
| 4886 | CGOpenCLRuntime OpenCLRT(CGM); | |||
| 4887 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
| 4888 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
| 4889 | ||||
| 4890 | // Building the generic function prototype. | |||
| 4891 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty}; | |||
| 4892 | llvm::FunctionType *FTy = | |||
| 4893 | llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()), | |||
| 4894 | llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 4895 | ||||
| 4896 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
| 4897 | {Arg0, Arg1, PacketSize, PacketAlign})); | |||
| 4898 | } | |||
| 4899 | // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions | |||
| 4900 | case Builtin::BIget_pipe_num_packets: | |||
| 4901 | case Builtin::BIget_pipe_max_packets: { | |||
| 4902 | const char *BaseName; | |||
| 4903 | const auto *PipeTy = E->getArg(0)->getType()->castAs<PipeType>(); | |||
| 4904 | if (BuiltinID == Builtin::BIget_pipe_num_packets) | |||
| 4905 | BaseName = "__get_pipe_num_packets"; | |||
| 4906 | else | |||
| 4907 | BaseName = "__get_pipe_max_packets"; | |||
| 4908 | std::string Name = std::string(BaseName) + | |||
| 4909 | std::string(PipeTy->isReadOnly() ? "_ro" : "_wo"); | |||
| 4910 | ||||
| 4911 | // Building the generic function prototype. | |||
| 4912 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
| 4913 | CGOpenCLRuntime OpenCLRT(CGM); | |||
| 4914 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); | |||
| 4915 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); | |||
| 4916 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty}; | |||
| 4917 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
| 4918 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 4919 | ||||
| 4920 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
| 4921 | {Arg0, PacketSize, PacketAlign})); | |||
| 4922 | } | |||
| 4923 | ||||
| 4924 | // OpenCL v2.0 s6.13.9 - Address space qualifier functions. | |||
| 4925 | case Builtin::BIto_global: | |||
| 4926 | case Builtin::BIto_local: | |||
| 4927 | case Builtin::BIto_private: { | |||
| 4928 | auto Arg0 = EmitScalarExpr(E->getArg(0)); | |||
| 4929 | auto NewArgT = llvm::PointerType::get(Int8Ty, | |||
| 4930 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
| 4931 | auto NewRetT = llvm::PointerType::get(Int8Ty, | |||
| 4932 | CGM.getContext().getTargetAddressSpace( | |||
| 4933 | E->getType()->getPointeeType().getAddressSpace())); | |||
| 4934 | auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false); | |||
| 4935 | llvm::Value *NewArg; | |||
| 4936 | if (Arg0->getType()->getPointerAddressSpace() != | |||
| 4937 | NewArgT->getPointerAddressSpace()) | |||
| 4938 | NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT); | |||
| 4939 | else | |||
| 4940 | NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT); | |||
| 4941 | auto NewName = std::string("__") + E->getDirectCallee()->getName().str(); | |||
| 4942 | auto NewCall = | |||
| 4943 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg}); | |||
| 4944 | return RValue::get(Builder.CreateBitOrPointerCast(NewCall, | |||
| 4945 | ConvertType(E->getType()))); | |||
| 4946 | } | |||
| 4947 | ||||
| 4948 | // OpenCL v2.0, s6.13.17 - Enqueue kernel function. | |||
| 4949 | // It contains four different overload formats specified in Table 6.13.17.1. | |||
| 4950 | case Builtin::BIenqueue_kernel: { | |||
| 4951 | StringRef Name; // Generated function call name | |||
| 4952 | unsigned NumArgs = E->getNumArgs(); | |||
| 4953 | ||||
| 4954 | llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy); | |||
| 4955 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
| 4956 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
| 4957 | ||||
| 4958 | llvm::Value *Queue = EmitScalarExpr(E->getArg(0)); | |||
| 4959 | llvm::Value *Flags = EmitScalarExpr(E->getArg(1)); | |||
| 4960 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(2)); | |||
| 4961 | llvm::Value *Range = NDRangeL.getAddress(*this).getPointer(); | |||
| 4962 | llvm::Type *RangeTy = NDRangeL.getAddress(*this).getType(); | |||
| 4963 | ||||
| 4964 | if (NumArgs == 4) { | |||
| 4965 | // The most basic form of the call with parameters: | |||
| 4966 | // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void) | |||
| 4967 | Name = "__enqueue_kernel_basic"; | |||
| 4968 | llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy, | |||
| 4969 | GenericVoidPtrTy}; | |||
| 4970 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
| 4971 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 4972 | ||||
| 4973 | auto Info = | |||
| 4974 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); | |||
| 4975 | llvm::Value *Kernel = | |||
| 4976 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
| 4977 | llvm::Value *Block = | |||
| 4978 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
| 4979 | ||||
| 4980 | AttrBuilder B(Builder.getContext()); | |||
| 4981 | B.addByValAttr(NDRangeL.getAddress(*this).getElementType()); | |||
| 4982 | llvm::AttributeList ByValAttrSet = | |||
| 4983 | llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B); | |||
| 4984 | ||||
| 4985 | auto RTCall = | |||
| 4986 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet), | |||
| 4987 | {Queue, Flags, Range, Kernel, Block}); | |||
| 4988 | RTCall->setAttributes(ByValAttrSet); | |||
| 4989 | return RValue::get(RTCall); | |||
| 4990 | } | |||
| 4991 | assert(NumArgs >= 5 && "Invalid enqueue_kernel signature")(static_cast <bool> (NumArgs >= 5 && "Invalid enqueue_kernel signature" ) ? void (0) : __assert_fail ("NumArgs >= 5 && \"Invalid enqueue_kernel signature\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 4991, __extension__ __PRETTY_FUNCTION__ )); | |||
| 4992 | ||||
| 4993 | // Create a temporary array to hold the sizes of local pointer arguments | |||
| 4994 | // for the block. \p First is the position of the first size argument. | |||
| 4995 | auto CreateArrayForSizeVar = [=](unsigned First) | |||
| 4996 | -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> { | |||
| 4997 | llvm::APInt ArraySize(32, NumArgs - First); | |||
| 4998 | QualType SizeArrayTy = getContext().getConstantArrayType( | |||
| 4999 | getContext().getSizeType(), ArraySize, nullptr, ArrayType::Normal, | |||
| 5000 | /*IndexTypeQuals=*/0); | |||
| 5001 | auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes"); | |||
| 5002 | llvm::Value *TmpPtr = Tmp.getPointer(); | |||
| 5003 | llvm::Value *TmpSize = EmitLifetimeStart( | |||
| 5004 | CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr); | |||
| 5005 | llvm::Value *ElemPtr; | |||
| 5006 | // Each of the following arguments specifies the size of the corresponding | |||
| 5007 | // argument passed to the enqueued block. | |||
| 5008 | auto *Zero = llvm::ConstantInt::get(IntTy, 0); | |||
| 5009 | for (unsigned I = First; I < NumArgs; ++I) { | |||
| 5010 | auto *Index = llvm::ConstantInt::get(IntTy, I - First); | |||
| 5011 | auto *GEP = Builder.CreateGEP(Tmp.getElementType(), TmpPtr, | |||
| 5012 | {Zero, Index}); | |||
| 5013 | if (I == First) | |||
| 5014 | ElemPtr = GEP; | |||
| 5015 | auto *V = | |||
| 5016 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy); | |||
| 5017 | Builder.CreateAlignedStore( | |||
| 5018 | V, GEP, CGM.getDataLayout().getPrefTypeAlign(SizeTy)); | |||
| 5019 | } | |||
| 5020 | return std::tie(ElemPtr, TmpSize, TmpPtr); | |||
| 5021 | }; | |||
| 5022 | ||||
| 5023 | // Could have events and/or varargs. | |||
| 5024 | if (E->getArg(3)->getType()->isBlockPointerType()) { | |||
| 5025 | // No events passed, but has variadic arguments. | |||
| 5026 | Name = "__enqueue_kernel_varargs"; | |||
| 5027 | auto Info = | |||
| 5028 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); | |||
| 5029 | llvm::Value *Kernel = | |||
| 5030 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
| 5031 | auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
| 5032 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; | |||
| 5033 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4); | |||
| 5034 | ||||
| 5035 | // Create a vector of the arguments, as well as a constant value to | |||
| 5036 | // express to the runtime the number of variadic arguments. | |||
| 5037 | llvm::Value *const Args[] = {Queue, Flags, | |||
| 5038 | Range, Kernel, | |||
| 5039 | Block, ConstantInt::get(IntTy, NumArgs - 4), | |||
| 5040 | ElemPtr}; | |||
| 5041 | llvm::Type *const ArgTys[] = { | |||
| 5042 | QueueTy, IntTy, RangeTy, GenericVoidPtrTy, | |||
| 5043 | GenericVoidPtrTy, IntTy, ElemPtr->getType()}; | |||
| 5044 | ||||
| 5045 | llvm::FunctionType *FTy = llvm::FunctionType::get(Int32Ty, ArgTys, false); | |||
| 5046 | auto Call = RValue::get( | |||
| 5047 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Args)); | |||
| 5048 | if (TmpSize) | |||
| 5049 | EmitLifetimeEnd(TmpSize, TmpPtr); | |||
| 5050 | return Call; | |||
| 5051 | } | |||
| 5052 | // Any calls now have event arguments passed. | |||
| 5053 | if (NumArgs >= 7) { | |||
| 5054 | llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy); | |||
| 5055 | llvm::PointerType *EventPtrTy = EventTy->getPointerTo( | |||
| 5056 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
| 5057 | ||||
| 5058 | llvm::Value *NumEvents = | |||
| 5059 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty); | |||
| 5060 | ||||
| 5061 | // Since SemaOpenCLBuiltinEnqueueKernel allows fifth and sixth arguments | |||
| 5062 | // to be a null pointer constant (including `0` literal), we can take it | |||
| 5063 | // into account and emit null pointer directly. | |||
| 5064 | llvm::Value *EventWaitList = nullptr; | |||
| 5065 | if (E->getArg(4)->isNullPointerConstant( | |||
| 5066 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { | |||
| 5067 | EventWaitList = llvm::ConstantPointerNull::get(EventPtrTy); | |||
| 5068 | } else { | |||
| 5069 | EventWaitList = E->getArg(4)->getType()->isArrayType() | |||
| 5070 | ? EmitArrayToPointerDecay(E->getArg(4)).getPointer() | |||
| 5071 | : EmitScalarExpr(E->getArg(4)); | |||
| 5072 | // Convert to generic address space. | |||
| 5073 | EventWaitList = Builder.CreatePointerCast(EventWaitList, EventPtrTy); | |||
| 5074 | } | |||
| 5075 | llvm::Value *EventRet = nullptr; | |||
| 5076 | if (E->getArg(5)->isNullPointerConstant( | |||
| 5077 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { | |||
| 5078 | EventRet = llvm::ConstantPointerNull::get(EventPtrTy); | |||
| 5079 | } else { | |||
| 5080 | EventRet = | |||
| 5081 | Builder.CreatePointerCast(EmitScalarExpr(E->getArg(5)), EventPtrTy); | |||
| 5082 | } | |||
| 5083 | ||||
| 5084 | auto Info = | |||
| 5085 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6)); | |||
| 5086 | llvm::Value *Kernel = | |||
| 5087 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
| 5088 | llvm::Value *Block = | |||
| 5089 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
| 5090 | ||||
| 5091 | std::vector<llvm::Type *> ArgTys = { | |||
| 5092 | QueueTy, Int32Ty, RangeTy, Int32Ty, | |||
| 5093 | EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy}; | |||
| 5094 | ||||
| 5095 | std::vector<llvm::Value *> Args = {Queue, Flags, Range, | |||
| 5096 | NumEvents, EventWaitList, EventRet, | |||
| 5097 | Kernel, Block}; | |||
| 5098 | ||||
| 5099 | if (NumArgs == 7) { | |||
| 5100 | // Has events but no variadics. | |||
| 5101 | Name = "__enqueue_kernel_basic_events"; | |||
| 5102 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
| 5103 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 5104 | return RValue::get( | |||
| 5105 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
| 5106 | llvm::ArrayRef<llvm::Value *>(Args))); | |||
| 5107 | } | |||
| 5108 | // Has event info and variadics | |||
| 5109 | // Pass the number of variadics to the runtime function too. | |||
| 5110 | Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7)); | |||
| 5111 | ArgTys.push_back(Int32Ty); | |||
| 5112 | Name = "__enqueue_kernel_events_varargs"; | |||
| 5113 | ||||
| 5114 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; | |||
| 5115 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7); | |||
| 5116 | Args.push_back(ElemPtr); | |||
| 5117 | ArgTys.push_back(ElemPtr->getType()); | |||
| 5118 | ||||
| 5119 | llvm::FunctionType *FTy = llvm::FunctionType::get( | |||
| 5120 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); | |||
| 5121 | auto Call = | |||
| 5122 | RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), | |||
| 5123 | llvm::ArrayRef<llvm::Value *>(Args))); | |||
| 5124 | if (TmpSize) | |||
| 5125 | EmitLifetimeEnd(TmpSize, TmpPtr); | |||
| 5126 | return Call; | |||
| 5127 | } | |||
| 5128 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 5129 | } | |||
| 5130 | // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block | |||
| 5131 | // parameter. | |||
| 5132 | case Builtin::BIget_kernel_work_group_size: { | |||
| 5133 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
| 5134 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
| 5135 | auto Info = | |||
| 5136 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); | |||
| 5137 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
| 5138 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
| 5139 | return RValue::get(EmitRuntimeCall( | |||
| 5140 | CGM.CreateRuntimeFunction( | |||
| 5141 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, | |||
| 5142 | false), | |||
| 5143 | "__get_kernel_work_group_size_impl"), | |||
| 5144 | {Kernel, Arg})); | |||
| 5145 | } | |||
| 5146 | case Builtin::BIget_kernel_preferred_work_group_size_multiple: { | |||
| 5147 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
| 5148 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
| 5149 | auto Info = | |||
| 5150 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); | |||
| 5151 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
| 5152 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
| 5153 | return RValue::get(EmitRuntimeCall( | |||
| 5154 | CGM.CreateRuntimeFunction( | |||
| 5155 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, | |||
| 5156 | false), | |||
| 5157 | "__get_kernel_preferred_work_group_size_multiple_impl"), | |||
| 5158 | {Kernel, Arg})); | |||
| 5159 | } | |||
| 5160 | case Builtin::BIget_kernel_max_sub_group_size_for_ndrange: | |||
| 5161 | case Builtin::BIget_kernel_sub_group_count_for_ndrange: { | |||
| 5162 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( | |||
| 5163 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); | |||
| 5164 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(0)); | |||
| 5165 | llvm::Value *NDRange = NDRangeL.getAddress(*this).getPointer(); | |||
| 5166 | auto Info = | |||
| 5167 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1)); | |||
| 5168 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); | |||
| 5169 | Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); | |||
| 5170 | const char *Name = | |||
| 5171 | BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange | |||
| 5172 | ? "__get_kernel_max_sub_group_size_for_ndrange_impl" | |||
| 5173 | : "__get_kernel_sub_group_count_for_ndrange_impl"; | |||
| 5174 | return RValue::get(EmitRuntimeCall( | |||
| 5175 | CGM.CreateRuntimeFunction( | |||
| 5176 | llvm::FunctionType::get( | |||
| 5177 | IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy}, | |||
| 5178 | false), | |||
| 5179 | Name), | |||
| 5180 | {NDRange, Kernel, Block})); | |||
| 5181 | } | |||
| 5182 | ||||
| 5183 | case Builtin::BI__builtin_store_half: | |||
| 5184 | case Builtin::BI__builtin_store_halff: { | |||
| 5185 | Value *Val = EmitScalarExpr(E->getArg(0)); | |||
| 5186 | Address Address = EmitPointerWithAlignment(E->getArg(1)); | |||
| 5187 | Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy()); | |||
| 5188 | return RValue::get(Builder.CreateStore(HalfVal, Address)); | |||
| 5189 | } | |||
| 5190 | case Builtin::BI__builtin_load_half: { | |||
| 5191 | Address Address = EmitPointerWithAlignment(E->getArg(0)); | |||
| 5192 | Value *HalfVal = Builder.CreateLoad(Address); | |||
| 5193 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy())); | |||
| 5194 | } | |||
| 5195 | case Builtin::BI__builtin_load_halff: { | |||
| 5196 | Address Address = EmitPointerWithAlignment(E->getArg(0)); | |||
| 5197 | Value *HalfVal = Builder.CreateLoad(Address); | |||
| 5198 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy())); | |||
| 5199 | } | |||
| 5200 | case Builtin::BIprintf: | |||
| 5201 | if (getTarget().getTriple().isNVPTX() || | |||
| 5202 | getTarget().getTriple().isAMDGCN()) { | |||
| 5203 | if (getLangOpts().OpenMPIsDevice) | |||
| 5204 | return EmitOpenMPDevicePrintfCallExpr(E); | |||
| 5205 | if (getTarget().getTriple().isNVPTX()) | |||
| 5206 | return EmitNVPTXDevicePrintfCallExpr(E); | |||
| 5207 | if (getTarget().getTriple().isAMDGCN() && getLangOpts().HIP) | |||
| 5208 | return EmitAMDGPUDevicePrintfCallExpr(E); | |||
| 5209 | } | |||
| 5210 | ||||
| 5211 | break; | |||
| 5212 | case Builtin::BI__builtin_canonicalize: | |||
| 5213 | case Builtin::BI__builtin_canonicalizef: | |||
| 5214 | case Builtin::BI__builtin_canonicalizef16: | |||
| 5215 | case Builtin::BI__builtin_canonicalizel: | |||
| 5216 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize)); | |||
| 5217 | ||||
| 5218 | case Builtin::BI__builtin_thread_pointer: { | |||
| 5219 | if (!getContext().getTargetInfo().isTLSSupported()) | |||
| 5220 | CGM.ErrorUnsupported(E, "__builtin_thread_pointer"); | |||
| 5221 | // Fall through - it's already mapped to the intrinsic by GCCBuiltin. | |||
| 5222 | break; | |||
| 5223 | } | |||
| 5224 | case Builtin::BI__builtin_os_log_format: | |||
| 5225 | return emitBuiltinOSLogFormat(*E); | |||
| 5226 | ||||
| 5227 | case Builtin::BI__xray_customevent: { | |||
| 5228 | if (!ShouldXRayInstrumentFunction()) | |||
| 5229 | return RValue::getIgnored(); | |||
| 5230 | ||||
| 5231 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( | |||
| 5232 | XRayInstrKind::Custom)) | |||
| 5233 | return RValue::getIgnored(); | |||
| 5234 | ||||
| 5235 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) | |||
| 5236 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents()) | |||
| 5237 | return RValue::getIgnored(); | |||
| 5238 | ||||
| 5239 | Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent); | |||
| 5240 | auto FTy = F->getFunctionType(); | |||
| 5241 | auto Arg0 = E->getArg(0); | |||
| 5242 | auto Arg0Val = EmitScalarExpr(Arg0); | |||
| 5243 | auto Arg0Ty = Arg0->getType(); | |||
| 5244 | auto PTy0 = FTy->getParamType(0); | |||
| 5245 | if (PTy0 != Arg0Val->getType()) { | |||
| 5246 | if (Arg0Ty->isArrayType()) | |||
| 5247 | Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer(); | |||
| 5248 | else | |||
| 5249 | Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0); | |||
| 5250 | } | |||
| 5251 | auto Arg1 = EmitScalarExpr(E->getArg(1)); | |||
| 5252 | auto PTy1 = FTy->getParamType(1); | |||
| 5253 | if (PTy1 != Arg1->getType()) | |||
| 5254 | Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1); | |||
| 5255 | return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1})); | |||
| 5256 | } | |||
| 5257 | ||||
| 5258 | case Builtin::BI__xray_typedevent: { | |||
| 5259 | // TODO: There should be a way to always emit events even if the current | |||
| 5260 | // function is not instrumented. Losing events in a stream can cripple | |||
| 5261 | // a trace. | |||
| 5262 | if (!ShouldXRayInstrumentFunction()) | |||
| 5263 | return RValue::getIgnored(); | |||
| 5264 | ||||
| 5265 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( | |||
| 5266 | XRayInstrKind::Typed)) | |||
| 5267 | return RValue::getIgnored(); | |||
| 5268 | ||||
| 5269 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) | |||
| 5270 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents()) | |||
| 5271 | return RValue::getIgnored(); | |||
| 5272 | ||||
| 5273 | Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent); | |||
| 5274 | auto FTy = F->getFunctionType(); | |||
| 5275 | auto Arg0 = EmitScalarExpr(E->getArg(0)); | |||
| 5276 | auto PTy0 = FTy->getParamType(0); | |||
| 5277 | if (PTy0 != Arg0->getType()) | |||
| 5278 | Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0); | |||
| 5279 | auto Arg1 = E->getArg(1); | |||
| 5280 | auto Arg1Val = EmitScalarExpr(Arg1); | |||
| 5281 | auto Arg1Ty = Arg1->getType(); | |||
| 5282 | auto PTy1 = FTy->getParamType(1); | |||
| 5283 | if (PTy1 != Arg1Val->getType()) { | |||
| 5284 | if (Arg1Ty->isArrayType()) | |||
| 5285 | Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer(); | |||
| 5286 | else | |||
| 5287 | Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1); | |||
| 5288 | } | |||
| 5289 | auto Arg2 = EmitScalarExpr(E->getArg(2)); | |||
| 5290 | auto PTy2 = FTy->getParamType(2); | |||
| 5291 | if (PTy2 != Arg2->getType()) | |||
| 5292 | Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2); | |||
| 5293 | return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2})); | |||
| 5294 | } | |||
| 5295 | ||||
| 5296 | case Builtin::BI__builtin_ms_va_start: | |||
| 5297 | case Builtin::BI__builtin_ms_va_end: | |||
| 5298 | return RValue::get( | |||
| 5299 | EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(), | |||
| 5300 | BuiltinID == Builtin::BI__builtin_ms_va_start)); | |||
| 5301 | ||||
| 5302 | case Builtin::BI__builtin_ms_va_copy: { | |||
| 5303 | // Lower this manually. We can't reliably determine whether or not any | |||
| 5304 | // given va_copy() is for a Win64 va_list from the calling convention | |||
| 5305 | // alone, because it's legal to do this from a System V ABI function. | |||
| 5306 | // With opaque pointer types, we won't have enough information in LLVM | |||
| 5307 | // IR to determine this from the argument types, either. Best to do it | |||
| 5308 | // now, while we have enough information. | |||
| 5309 | Address DestAddr = EmitMSVAListRef(E->getArg(0)); | |||
| 5310 | Address SrcAddr = EmitMSVAListRef(E->getArg(1)); | |||
| 5311 | ||||
| 5312 | llvm::Type *BPP = Int8PtrPtrTy; | |||
| 5313 | ||||
| 5314 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"), | |||
| 5315 | Int8PtrTy, DestAddr.getAlignment()); | |||
| 5316 | SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"), | |||
| 5317 | Int8PtrTy, SrcAddr.getAlignment()); | |||
| 5318 | ||||
| 5319 | Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val"); | |||
| 5320 | return RValue::get(Builder.CreateStore(ArgPtr, DestAddr)); | |||
| 5321 | } | |||
| 5322 | ||||
| 5323 | case Builtin::BI__builtin_get_device_side_mangled_name: { | |||
| 5324 | auto Name = CGM.getCUDARuntime().getDeviceSideName( | |||
| 5325 | cast<DeclRefExpr>(E->getArg(0)->IgnoreImpCasts())->getDecl()); | |||
| 5326 | auto Str = CGM.GetAddrOfConstantCString(Name, ""); | |||
| 5327 | llvm::Constant *Zeros[] = {llvm::ConstantInt::get(SizeTy, 0), | |||
| 5328 | llvm::ConstantInt::get(SizeTy, 0)}; | |||
| 5329 | auto *Ptr = llvm::ConstantExpr::getGetElementPtr(Str.getElementType(), | |||
| 5330 | Str.getPointer(), Zeros); | |||
| 5331 | return RValue::get(Ptr); | |||
| 5332 | } | |||
| 5333 | } | |||
| 5334 | ||||
| 5335 | // If this is an alias for a lib function (e.g. __builtin_sin), emit | |||
| 5336 | // the call using the normal call path, but using the unmangled | |||
| 5337 | // version of the function name. | |||
| 5338 | if (getContext().BuiltinInfo.isLibFunction(BuiltinID)) | |||
| 5339 | return emitLibraryCall(*this, FD, E, | |||
| 5340 | CGM.getBuiltinLibFunction(FD, BuiltinID)); | |||
| 5341 | ||||
| 5342 | // If this is a predefined lib function (e.g. malloc), emit the call | |||
| 5343 | // using exactly the normal call path. | |||
| 5344 | if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID)) | |||
| 5345 | return emitLibraryCall(*this, FD, E, | |||
| 5346 | cast<llvm::Constant>(EmitScalarExpr(E->getCallee()))); | |||
| 5347 | ||||
| 5348 | // Check that a call to a target specific builtin has the correct target | |||
| 5349 | // features. | |||
| 5350 | // This is down here to avoid non-target specific builtins, however, if | |||
| 5351 | // generic builtins start to require generic target features then we | |||
| 5352 | // can move this up to the beginning of the function. | |||
| 5353 | checkTargetFeatures(E, FD); | |||
| 5354 | ||||
| 5355 | if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID)) | |||
| 5356 | LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth); | |||
| 5357 | ||||
| 5358 | // See if we have a target specific intrinsic. | |||
| 5359 | const char *Name = getContext().BuiltinInfo.getName(BuiltinID); | |||
| 5360 | Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic; | |||
| 5361 | StringRef Prefix = | |||
| 5362 | llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch()); | |||
| 5363 | if (!Prefix.empty()) { | |||
| 5364 | IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name); | |||
| 5365 | // NOTE we don't need to perform a compatibility flag check here since the | |||
| 5366 | // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the | |||
| 5367 | // MS builtins via ALL_MS_LANGUAGES and are filtered earlier. | |||
| 5368 | if (IntrinsicID == Intrinsic::not_intrinsic) | |||
| 5369 | IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name); | |||
| 5370 | } | |||
| 5371 | ||||
| 5372 | if (IntrinsicID != Intrinsic::not_intrinsic) { | |||
| 5373 | SmallVector<Value*, 16> Args; | |||
| 5374 | ||||
| 5375 | // Find out if any arguments are required to be integer constant | |||
| 5376 | // expressions. | |||
| 5377 | unsigned ICEArguments = 0; | |||
| 5378 | ASTContext::GetBuiltinTypeError Error; | |||
| 5379 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
| 5380 | assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None && "Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5380, __extension__ __PRETTY_FUNCTION__ )); | |||
| 5381 | ||||
| 5382 | Function *F = CGM.getIntrinsic(IntrinsicID); | |||
| 5383 | llvm::FunctionType *FTy = F->getFunctionType(); | |||
| 5384 | ||||
| 5385 | for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) { | |||
| 5386 | Value *ArgValue; | |||
| 5387 | // If this is a normal argument, just emit it as a scalar. | |||
| 5388 | if ((ICEArguments & (1 << i)) == 0) { | |||
| 5389 | ArgValue = EmitScalarExpr(E->getArg(i)); | |||
| 5390 | } else { | |||
| 5391 | // If this is required to be a constant, constant fold it so that we | |||
| 5392 | // know that the generated intrinsic gets a ConstantInt. | |||
| 5393 | ArgValue = llvm::ConstantInt::get( | |||
| 5394 | getLLVMContext(), | |||
| 5395 | *E->getArg(i)->getIntegerConstantExpr(getContext())); | |||
| 5396 | } | |||
| 5397 | ||||
| 5398 | // If the intrinsic arg type is different from the builtin arg type | |||
| 5399 | // we need to do a bit cast. | |||
| 5400 | llvm::Type *PTy = FTy->getParamType(i); | |||
| 5401 | if (PTy != ArgValue->getType()) { | |||
| 5402 | // XXX - vector of pointers? | |||
| 5403 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) { | |||
| 5404 | if (PtrTy->getAddressSpace() != | |||
| 5405 | ArgValue->getType()->getPointerAddressSpace()) { | |||
| 5406 | ArgValue = Builder.CreateAddrSpaceCast( | |||
| 5407 | ArgValue, | |||
| 5408 | ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace())); | |||
| 5409 | } | |||
| 5410 | } | |||
| 5411 | ||||
| 5412 | assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&(static_cast <bool> (PTy->canLosslesslyBitCastTo(FTy ->getParamType(i)) && "Must be able to losslessly bit cast to param" ) ? void (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5413, __extension__ __PRETTY_FUNCTION__ )) | |||
| 5413 | "Must be able to losslessly bit cast to param")(static_cast <bool> (PTy->canLosslesslyBitCastTo(FTy ->getParamType(i)) && "Must be able to losslessly bit cast to param" ) ? void (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5413, __extension__ __PRETTY_FUNCTION__ )); | |||
| 5414 | // Cast vector type (e.g., v256i32) to x86_amx, this only happen | |||
| 5415 | // in amx intrinsics. | |||
| 5416 | if (PTy->isX86_AMXTy()) | |||
| 5417 | ArgValue = Builder.CreateIntrinsic(Intrinsic::x86_cast_vector_to_tile, | |||
| 5418 | {ArgValue->getType()}, {ArgValue}); | |||
| 5419 | else | |||
| 5420 | ArgValue = Builder.CreateBitCast(ArgValue, PTy); | |||
| 5421 | } | |||
| 5422 | ||||
| 5423 | Args.push_back(ArgValue); | |||
| 5424 | } | |||
| 5425 | ||||
| 5426 | Value *V = Builder.CreateCall(F, Args); | |||
| 5427 | QualType BuiltinRetType = E->getType(); | |||
| 5428 | ||||
| 5429 | llvm::Type *RetTy = VoidTy; | |||
| 5430 | if (!BuiltinRetType->isVoidType()) | |||
| 5431 | RetTy = ConvertType(BuiltinRetType); | |||
| 5432 | ||||
| 5433 | if (RetTy != V->getType()) { | |||
| 5434 | // XXX - vector of pointers? | |||
| 5435 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) { | |||
| 5436 | if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) { | |||
| 5437 | V = Builder.CreateAddrSpaceCast( | |||
| 5438 | V, V->getType()->getPointerTo(PtrTy->getAddressSpace())); | |||
| 5439 | } | |||
| 5440 | } | |||
| 5441 | ||||
| 5442 | assert(V->getType()->canLosslesslyBitCastTo(RetTy) &&(static_cast <bool> (V->getType()->canLosslesslyBitCastTo (RetTy) && "Must be able to losslessly bit cast result type" ) ? void (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5443, __extension__ __PRETTY_FUNCTION__ )) | |||
| 5443 | "Must be able to losslessly bit cast result type")(static_cast <bool> (V->getType()->canLosslesslyBitCastTo (RetTy) && "Must be able to losslessly bit cast result type" ) ? void (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5443, __extension__ __PRETTY_FUNCTION__ )); | |||
| 5444 | // Cast x86_amx to vector type (e.g., v256i32), this only happen | |||
| 5445 | // in amx intrinsics. | |||
| 5446 | if (V->getType()->isX86_AMXTy()) | |||
| 5447 | V = Builder.CreateIntrinsic(Intrinsic::x86_cast_tile_to_vector, {RetTy}, | |||
| 5448 | {V}); | |||
| 5449 | else | |||
| 5450 | V = Builder.CreateBitCast(V, RetTy); | |||
| 5451 | } | |||
| 5452 | ||||
| 5453 | return RValue::get(V); | |||
| 5454 | } | |||
| 5455 | ||||
| 5456 | // Some target-specific builtins can have aggregate return values, e.g. | |||
| 5457 | // __builtin_arm_mve_vld2q_u32. So if the result is an aggregate, force | |||
| 5458 | // ReturnValue to be non-null, so that the target-specific emission code can | |||
| 5459 | // always just emit into it. | |||
| 5460 | TypeEvaluationKind EvalKind = getEvaluationKind(E->getType()); | |||
| 5461 | if (EvalKind == TEK_Aggregate && ReturnValue.isNull()) { | |||
| 5462 | Address DestPtr = CreateMemTemp(E->getType(), "agg.tmp"); | |||
| 5463 | ReturnValue = ReturnValueSlot(DestPtr, false); | |||
| 5464 | } | |||
| 5465 | ||||
| 5466 | // Now see if we can emit a target-specific builtin. | |||
| 5467 | if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E, ReturnValue)) { | |||
| 5468 | switch (EvalKind) { | |||
| 5469 | case TEK_Scalar: | |||
| 5470 | return RValue::get(V); | |||
| 5471 | case TEK_Aggregate: | |||
| 5472 | return RValue::getAggregate(ReturnValue.getValue(), | |||
| 5473 | ReturnValue.isVolatile()); | |||
| 5474 | case TEK_Complex: | |||
| 5475 | llvm_unreachable("No current target builtin returns complex")::llvm::llvm_unreachable_internal("No current target builtin returns complex" , "clang/lib/CodeGen/CGBuiltin.cpp", 5475); | |||
| 5476 | } | |||
| 5477 | llvm_unreachable("Bad evaluation kind in EmitBuiltinExpr")::llvm::llvm_unreachable_internal("Bad evaluation kind in EmitBuiltinExpr" , "clang/lib/CodeGen/CGBuiltin.cpp", 5477); | |||
| 5478 | } | |||
| 5479 | ||||
| 5480 | ErrorUnsupported(E, "builtin function"); | |||
| 5481 | ||||
| 5482 | // Unknown builtin, for now just dump it out and return undef. | |||
| 5483 | return GetUndefRValue(E->getType()); | |||
| 5484 | } | |||
| 5485 | ||||
| 5486 | static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF, | |||
| 5487 | unsigned BuiltinID, const CallExpr *E, | |||
| 5488 | ReturnValueSlot ReturnValue, | |||
| 5489 | llvm::Triple::ArchType Arch) { | |||
| 5490 | switch (Arch) { | |||
| 5491 | case llvm::Triple::arm: | |||
| 5492 | case llvm::Triple::armeb: | |||
| 5493 | case llvm::Triple::thumb: | |||
| 5494 | case llvm::Triple::thumbeb: | |||
| 5495 | return CGF->EmitARMBuiltinExpr(BuiltinID, E, ReturnValue, Arch); | |||
| 5496 | case llvm::Triple::aarch64: | |||
| 5497 | case llvm::Triple::aarch64_32: | |||
| 5498 | case llvm::Triple::aarch64_be: | |||
| 5499 | return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch); | |||
| 5500 | case llvm::Triple::bpfeb: | |||
| 5501 | case llvm::Triple::bpfel: | |||
| 5502 | return CGF->EmitBPFBuiltinExpr(BuiltinID, E); | |||
| 5503 | case llvm::Triple::x86: | |||
| 5504 | case llvm::Triple::x86_64: | |||
| 5505 | return CGF->EmitX86BuiltinExpr(BuiltinID, E); | |||
| 5506 | case llvm::Triple::ppc: | |||
| 5507 | case llvm::Triple::ppcle: | |||
| 5508 | case llvm::Triple::ppc64: | |||
| 5509 | case llvm::Triple::ppc64le: | |||
| 5510 | return CGF->EmitPPCBuiltinExpr(BuiltinID, E); | |||
| 5511 | case llvm::Triple::r600: | |||
| 5512 | case llvm::Triple::amdgcn: | |||
| 5513 | return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E); | |||
| 5514 | case llvm::Triple::systemz: | |||
| 5515 | return CGF->EmitSystemZBuiltinExpr(BuiltinID, E); | |||
| 5516 | case llvm::Triple::nvptx: | |||
| 5517 | case llvm::Triple::nvptx64: | |||
| 5518 | return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E); | |||
| 5519 | case llvm::Triple::wasm32: | |||
| 5520 | case llvm::Triple::wasm64: | |||
| 5521 | return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E); | |||
| 5522 | case llvm::Triple::hexagon: | |||
| 5523 | return CGF->EmitHexagonBuiltinExpr(BuiltinID, E); | |||
| 5524 | case llvm::Triple::riscv32: | |||
| 5525 | case llvm::Triple::riscv64: | |||
| 5526 | return CGF->EmitRISCVBuiltinExpr(BuiltinID, E, ReturnValue); | |||
| 5527 | default: | |||
| 5528 | return nullptr; | |||
| 5529 | } | |||
| 5530 | } | |||
| 5531 | ||||
| 5532 | Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID, | |||
| 5533 | const CallExpr *E, | |||
| 5534 | ReturnValueSlot ReturnValue) { | |||
| 5535 | if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) { | |||
| 5536 | assert(getContext().getAuxTargetInfo() && "Missing aux target info")(static_cast <bool> (getContext().getAuxTargetInfo() && "Missing aux target info") ? void (0) : __assert_fail ("getContext().getAuxTargetInfo() && \"Missing aux target info\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5536, __extension__ __PRETTY_FUNCTION__ )); | |||
| 5537 | return EmitTargetArchBuiltinExpr( | |||
| 5538 | this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E, | |||
| 5539 | ReturnValue, getContext().getAuxTargetInfo()->getTriple().getArch()); | |||
| 5540 | } | |||
| 5541 | ||||
| 5542 | return EmitTargetArchBuiltinExpr(this, BuiltinID, E, ReturnValue, | |||
| 5543 | getTarget().getTriple().getArch()); | |||
| 5544 | } | |||
| 5545 | ||||
| 5546 | static llvm::FixedVectorType *GetNeonType(CodeGenFunction *CGF, | |||
| 5547 | NeonTypeFlags TypeFlags, | |||
| 5548 | bool HasLegalHalfType = true, | |||
| 5549 | bool V1Ty = false, | |||
| 5550 | bool AllowBFloatArgsAndRet = true) { | |||
| 5551 | int IsQuad = TypeFlags.isQuad(); | |||
| 5552 | switch (TypeFlags.getEltType()) { | |||
| 5553 | case NeonTypeFlags::Int8: | |||
| 5554 | case NeonTypeFlags::Poly8: | |||
| 5555 | return llvm::FixedVectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad)); | |||
| 5556 | case NeonTypeFlags::Int16: | |||
| 5557 | case NeonTypeFlags::Poly16: | |||
| 5558 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); | |||
| 5559 | case NeonTypeFlags::BFloat16: | |||
| 5560 | if (AllowBFloatArgsAndRet) | |||
| 5561 | return llvm::FixedVectorType::get(CGF->BFloatTy, V1Ty ? 1 : (4 << IsQuad)); | |||
| 5562 | else | |||
| 5563 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); | |||
| 5564 | case NeonTypeFlags::Float16: | |||
| 5565 | if (HasLegalHalfType) | |||
| 5566 | return llvm::FixedVectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad)); | |||
| 5567 | else | |||
| 5568 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); | |||
| 5569 | case NeonTypeFlags::Int32: | |||
| 5570 | return llvm::FixedVectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad)); | |||
| 5571 | case NeonTypeFlags::Int64: | |||
| 5572 | case NeonTypeFlags::Poly64: | |||
| 5573 | return llvm::FixedVectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad)); | |||
| 5574 | case NeonTypeFlags::Poly128: | |||
| 5575 | // FIXME: i128 and f128 doesn't get fully support in Clang and llvm. | |||
| 5576 | // There is a lot of i128 and f128 API missing. | |||
| 5577 | // so we use v16i8 to represent poly128 and get pattern matched. | |||
| 5578 | return llvm::FixedVectorType::get(CGF->Int8Ty, 16); | |||
| 5579 | case NeonTypeFlags::Float32: | |||
| 5580 | return llvm::FixedVectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad)); | |||
| 5581 | case NeonTypeFlags::Float64: | |||
| 5582 | return llvm::FixedVectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad)); | |||
| 5583 | } | |||
| 5584 | llvm_unreachable("Unknown vector element type!")::llvm::llvm_unreachable_internal("Unknown vector element type!" , "clang/lib/CodeGen/CGBuiltin.cpp", 5584); | |||
| 5585 | } | |||
| 5586 | ||||
| 5587 | static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF, | |||
| 5588 | NeonTypeFlags IntTypeFlags) { | |||
| 5589 | int IsQuad = IntTypeFlags.isQuad(); | |||
| 5590 | switch (IntTypeFlags.getEltType()) { | |||
| 5591 | case NeonTypeFlags::Int16: | |||
| 5592 | return llvm::FixedVectorType::get(CGF->HalfTy, (4 << IsQuad)); | |||
| 5593 | case NeonTypeFlags::Int32: | |||
| 5594 | return llvm::FixedVectorType::get(CGF->FloatTy, (2 << IsQuad)); | |||
| 5595 | case NeonTypeFlags::Int64: | |||
| 5596 | return llvm::FixedVectorType::get(CGF->DoubleTy, (1 << IsQuad)); | |||
| 5597 | default: | |||
| 5598 | llvm_unreachable("Type can't be converted to floating-point!")::llvm::llvm_unreachable_internal("Type can't be converted to floating-point!" , "clang/lib/CodeGen/CGBuiltin.cpp", 5598); | |||
| 5599 | } | |||
| 5600 | } | |||
| 5601 | ||||
| 5602 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C, | |||
| 5603 | const ElementCount &Count) { | |||
| 5604 | Value *SV = llvm::ConstantVector::getSplat(Count, C); | |||
| 5605 | return Builder.CreateShuffleVector(V, V, SV, "lane"); | |||
| 5606 | } | |||
| 5607 | ||||
| 5608 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) { | |||
| 5609 | ElementCount EC = cast<llvm::VectorType>(V->getType())->getElementCount(); | |||
| 5610 | return EmitNeonSplat(V, C, EC); | |||
| 5611 | } | |||
| 5612 | ||||
| 5613 | Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops, | |||
| 5614 | const char *name, | |||
| 5615 | unsigned shift, bool rightshift) { | |||
| 5616 | unsigned j = 0; | |||
| 5617 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); | |||
| 5618 | ai != ae; ++ai, ++j) { | |||
| 5619 | if (F->isConstrainedFPIntrinsic()) | |||
| 5620 | if (ai->getType()->isMetadataTy()) | |||
| 5621 | continue; | |||
| 5622 | if (shift > 0 && shift == j) | |||
| 5623 | Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift); | |||
| 5624 | else | |||
| 5625 | Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name); | |||
| 5626 | } | |||
| 5627 | ||||
| 5628 | if (F->isConstrainedFPIntrinsic()) | |||
| 5629 | return Builder.CreateConstrainedFPCall(F, Ops, name); | |||
| 5630 | else | |||
| 5631 | return Builder.CreateCall(F, Ops, name); | |||
| 5632 | } | |||
| 5633 | ||||
| 5634 | Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty, | |||
| 5635 | bool neg) { | |||
| 5636 | int SV = cast<ConstantInt>(V)->getSExtValue(); | |||
| 5637 | return ConstantInt::get(Ty, neg ? -SV : SV); | |||
| 5638 | } | |||
| 5639 | ||||
| 5640 | // Right-shift a vector by a constant. | |||
| 5641 | Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift, | |||
| 5642 | llvm::Type *Ty, bool usgn, | |||
| 5643 | const char *name) { | |||
| 5644 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); | |||
| 5645 | ||||
| 5646 | int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue(); | |||
| 5647 | int EltSize = VTy->getScalarSizeInBits(); | |||
| 5648 | ||||
| 5649 | Vec = Builder.CreateBitCast(Vec, Ty); | |||
| 5650 | ||||
| 5651 | // lshr/ashr are undefined when the shift amount is equal to the vector | |||
| 5652 | // element size. | |||
| 5653 | if (ShiftAmt == EltSize) { | |||
| 5654 | if (usgn) { | |||
| 5655 | // Right-shifting an unsigned value by its size yields 0. | |||
| 5656 | return llvm::ConstantAggregateZero::get(VTy); | |||
| 5657 | } else { | |||
| 5658 | // Right-shifting a signed value by its size is equivalent | |||
| 5659 | // to a shift of size-1. | |||
| 5660 | --ShiftAmt; | |||
| 5661 | Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt); | |||
| 5662 | } | |||
| 5663 | } | |||
| 5664 | ||||
| 5665 | Shift = EmitNeonShiftVector(Shift, Ty, false); | |||
| 5666 | if (usgn) | |||
| 5667 | return Builder.CreateLShr(Vec, Shift, name); | |||
| 5668 | else | |||
| 5669 | return Builder.CreateAShr(Vec, Shift, name); | |||
| 5670 | } | |||
| 5671 | ||||
| 5672 | enum { | |||
| 5673 | AddRetType = (1 << 0), | |||
| 5674 | Add1ArgType = (1 << 1), | |||
| 5675 | Add2ArgTypes = (1 << 2), | |||
| 5676 | ||||
| 5677 | VectorizeRetType = (1 << 3), | |||
| 5678 | VectorizeArgTypes = (1 << 4), | |||
| 5679 | ||||
| 5680 | InventFloatType = (1 << 5), | |||
| 5681 | UnsignedAlts = (1 << 6), | |||
| 5682 | ||||
| 5683 | Use64BitVectors = (1 << 7), | |||
| 5684 | Use128BitVectors = (1 << 8), | |||
| 5685 | ||||
| 5686 | Vectorize1ArgType = Add1ArgType | VectorizeArgTypes, | |||
| 5687 | VectorRet = AddRetType | VectorizeRetType, | |||
| 5688 | VectorRetGetArgs01 = | |||
| 5689 | AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes, | |||
| 5690 | FpCmpzModifiers = | |||
| 5691 | AddRetType | VectorizeRetType | Add1ArgType | InventFloatType | |||
| 5692 | }; | |||
| 5693 | ||||
| 5694 | namespace { | |||
| 5695 | struct ARMVectorIntrinsicInfo { | |||
| 5696 | const char *NameHint; | |||
| 5697 | unsigned BuiltinID; | |||
| 5698 | unsigned LLVMIntrinsic; | |||
| 5699 | unsigned AltLLVMIntrinsic; | |||
| 5700 | uint64_t TypeModifier; | |||
| 5701 | ||||
| 5702 | bool operator<(unsigned RHSBuiltinID) const { | |||
| 5703 | return BuiltinID < RHSBuiltinID; | |||
| 5704 | } | |||
| 5705 | bool operator<(const ARMVectorIntrinsicInfo &TE) const { | |||
| 5706 | return BuiltinID < TE.BuiltinID; | |||
| 5707 | } | |||
| 5708 | }; | |||
| 5709 | } // end anonymous namespace | |||
| 5710 | ||||
| 5711 | #define NEONMAP0(NameBase) \ | |||
| 5712 | { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 } | |||
| 5713 | ||||
| 5714 | #define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \ | |||
| 5715 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ | |||
| 5716 | Intrinsic::LLVMIntrinsic, 0, TypeModifier } | |||
| 5717 | ||||
| 5718 | #define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \ | |||
| 5719 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ | |||
| 5720 | Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \ | |||
| 5721 | TypeModifier } | |||
| 5722 | ||||
| 5723 | static const ARMVectorIntrinsicInfo ARMSIMDIntrinsicMap [] = { | |||
| 5724 | NEONMAP1(__a32_vcvt_bf16_v, arm_neon_vcvtfp2bf, 0), | |||
| 5725 | NEONMAP0(splat_lane_v), | |||
| 5726 | NEONMAP0(splat_laneq_v), | |||
| 5727 | NEONMAP0(splatq_lane_v), | |||
| 5728 | NEONMAP0(splatq_laneq_v), | |||
| 5729 | NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), | |||
| 5730 | NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), | |||
| 5731 | NEONMAP1(vabs_v, arm_neon_vabs, 0), | |||
| 5732 | NEONMAP1(vabsq_v, arm_neon_vabs, 0), | |||
| 5733 | NEONMAP0(vadd_v), | |||
| 5734 | NEONMAP0(vaddhn_v), | |||
| 5735 | NEONMAP0(vaddq_v), | |||
| 5736 | NEONMAP1(vaesdq_v, arm_neon_aesd, 0), | |||
| 5737 | NEONMAP1(vaeseq_v, arm_neon_aese, 0), | |||
| 5738 | NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0), | |||
| 5739 | NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0), | |||
| 5740 | NEONMAP1(vbfdot_v, arm_neon_bfdot, 0), | |||
| 5741 | NEONMAP1(vbfdotq_v, arm_neon_bfdot, 0), | |||
| 5742 | NEONMAP1(vbfmlalbq_v, arm_neon_bfmlalb, 0), | |||
| 5743 | NEONMAP1(vbfmlaltq_v, arm_neon_bfmlalt, 0), | |||
| 5744 | NEONMAP1(vbfmmlaq_v, arm_neon_bfmmla, 0), | |||
| 5745 | NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType), | |||
| 5746 | NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType), | |||
| 5747 | NEONMAP1(vcadd_rot270_v, arm_neon_vcadd_rot270, Add1ArgType), | |||
| 5748 | NEONMAP1(vcadd_rot90_v, arm_neon_vcadd_rot90, Add1ArgType), | |||
| 5749 | NEONMAP1(vcaddq_rot270_v, arm_neon_vcadd_rot270, Add1ArgType), | |||
| 5750 | NEONMAP1(vcaddq_rot90_v, arm_neon_vcadd_rot90, Add1ArgType), | |||
| 5751 | NEONMAP1(vcage_v, arm_neon_vacge, 0), | |||
| 5752 | NEONMAP1(vcageq_v, arm_neon_vacge, 0), | |||
| 5753 | NEONMAP1(vcagt_v, arm_neon_vacgt, 0), | |||
| 5754 | NEONMAP1(vcagtq_v, arm_neon_vacgt, 0), | |||
| 5755 | NEONMAP1(vcale_v, arm_neon_vacge, 0), | |||
| 5756 | NEONMAP1(vcaleq_v, arm_neon_vacge, 0), | |||
| 5757 | NEONMAP1(vcalt_v, arm_neon_vacgt, 0), | |||
| 5758 | NEONMAP1(vcaltq_v, arm_neon_vacgt, 0), | |||
| 5759 | NEONMAP0(vceqz_v), | |||
| 5760 | NEONMAP0(vceqzq_v), | |||
| 5761 | NEONMAP0(vcgez_v), | |||
| 5762 | NEONMAP0(vcgezq_v), | |||
| 5763 | NEONMAP0(vcgtz_v), | |||
| 5764 | NEONMAP0(vcgtzq_v), | |||
| 5765 | NEONMAP0(vclez_v), | |||
| 5766 | NEONMAP0(vclezq_v), | |||
| 5767 | NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType), | |||
| 5768 | NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType), | |||
| 5769 | NEONMAP0(vcltz_v), | |||
| 5770 | NEONMAP0(vcltzq_v), | |||
| 5771 | NEONMAP1(vclz_v, ctlz, Add1ArgType), | |||
| 5772 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), | |||
| 5773 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), | |||
| 5774 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), | |||
| 5775 | NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0), | |||
| 5776 | NEONMAP0(vcvt_f16_v), | |||
| 5777 | NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0), | |||
| 5778 | NEONMAP0(vcvt_f32_v), | |||
| 5779 | NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
| 5780 | NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
| 5781 | NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0), | |||
| 5782 | NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0), | |||
| 5783 | NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0), | |||
| 5784 | NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0), | |||
| 5785 | NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0), | |||
| 5786 | NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0), | |||
| 5787 | NEONMAP0(vcvt_s16_v), | |||
| 5788 | NEONMAP0(vcvt_s32_v), | |||
| 5789 | NEONMAP0(vcvt_s64_v), | |||
| 5790 | NEONMAP0(vcvt_u16_v), | |||
| 5791 | NEONMAP0(vcvt_u32_v), | |||
| 5792 | NEONMAP0(vcvt_u64_v), | |||
| 5793 | NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0), | |||
| 5794 | NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0), | |||
| 5795 | NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0), | |||
| 5796 | NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0), | |||
| 5797 | NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0), | |||
| 5798 | NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0), | |||
| 5799 | NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0), | |||
| 5800 | NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0), | |||
| 5801 | NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0), | |||
| 5802 | NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0), | |||
| 5803 | NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0), | |||
| 5804 | NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0), | |||
| 5805 | NEONMAP1(vcvth_bf16_f32, arm_neon_vcvtbfp2bf, 0), | |||
| 5806 | NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0), | |||
| 5807 | NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0), | |||
| 5808 | NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0), | |||
| 5809 | NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0), | |||
| 5810 | NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0), | |||
| 5811 | NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0), | |||
| 5812 | NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0), | |||
| 5813 | NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0), | |||
| 5814 | NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0), | |||
| 5815 | NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0), | |||
| 5816 | NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0), | |||
| 5817 | NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0), | |||
| 5818 | NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0), | |||
| 5819 | NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0), | |||
| 5820 | NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0), | |||
| 5821 | NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0), | |||
| 5822 | NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0), | |||
| 5823 | NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0), | |||
| 5824 | NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0), | |||
| 5825 | NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0), | |||
| 5826 | NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0), | |||
| 5827 | NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0), | |||
| 5828 | NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0), | |||
| 5829 | NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0), | |||
| 5830 | NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0), | |||
| 5831 | NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0), | |||
| 5832 | NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0), | |||
| 5833 | NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0), | |||
| 5834 | NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0), | |||
| 5835 | NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0), | |||
| 5836 | NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0), | |||
| 5837 | NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0), | |||
| 5838 | NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0), | |||
| 5839 | NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0), | |||
| 5840 | NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0), | |||
| 5841 | NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0), | |||
| 5842 | NEONMAP0(vcvtq_f16_v), | |||
| 5843 | NEONMAP0(vcvtq_f32_v), | |||
| 5844 | NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
| 5845 | NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), | |||
| 5846 | NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0), | |||
| 5847 | NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0), | |||
| 5848 | NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0), | |||
| 5849 | NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0), | |||
| 5850 | NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0), | |||
| 5851 | NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0), | |||
| 5852 | NEONMAP0(vcvtq_s16_v), | |||
| 5853 | NEONMAP0(vcvtq_s32_v), | |||
| 5854 | NEONMAP0(vcvtq_s64_v), | |||
| 5855 | NEONMAP0(vcvtq_u16_v), | |||
| 5856 | NEONMAP0(vcvtq_u32_v), | |||
| 5857 | NEONMAP0(vcvtq_u64_v), | |||
| 5858 | NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0), | |||
| 5859 | NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0), | |||
| 5860 | NEONMAP0(vext_v), | |||
| 5861 | NEONMAP0(vextq_v), | |||
| 5862 | NEONMAP0(vfma_v), | |||
| 5863 | NEONMAP0(vfmaq_v), | |||
| 5864 | NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), | |||
| 5865 | NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), | |||
| 5866 | NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), | |||
| 5867 | NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), | |||
| 5868 | NEONMAP0(vld1_dup_v), | |||
| 5869 | NEONMAP1(vld1_v, arm_neon_vld1, 0), | |||
| 5870 | NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0), | |||
| 5871 | NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0), | |||
| 5872 | NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0), | |||
| 5873 | NEONMAP0(vld1q_dup_v), | |||
| 5874 | NEONMAP1(vld1q_v, arm_neon_vld1, 0), | |||
| 5875 | NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0), | |||
| 5876 | NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0), | |||
| 5877 | NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0), | |||
| 5878 | NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0), | |||
| 5879 | NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0), | |||
| 5880 | NEONMAP1(vld2_v, arm_neon_vld2, 0), | |||
| 5881 | NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0), | |||
| 5882 | NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0), | |||
| 5883 | NEONMAP1(vld2q_v, arm_neon_vld2, 0), | |||
| 5884 | NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0), | |||
| 5885 | NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0), | |||
| 5886 | NEONMAP1(vld3_v, arm_neon_vld3, 0), | |||
| 5887 | NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0), | |||
| 5888 | NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0), | |||
| 5889 | NEONMAP1(vld3q_v, arm_neon_vld3, 0), | |||
| 5890 | NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0), | |||
| 5891 | NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0), | |||
| 5892 | NEONMAP1(vld4_v, arm_neon_vld4, 0), | |||
| 5893 | NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0), | |||
| 5894 | NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0), | |||
| 5895 | NEONMAP1(vld4q_v, arm_neon_vld4, 0), | |||
| 5896 | NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), | |||
| 5897 | NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType), | |||
| 5898 | NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType), | |||
| 5899 | NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), | |||
| 5900 | NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), | |||
| 5901 | NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType), | |||
| 5902 | NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType), | |||
| 5903 | NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), | |||
| 5904 | NEONMAP2(vmmlaq_v, arm_neon_ummla, arm_neon_smmla, 0), | |||
| 5905 | NEONMAP0(vmovl_v), | |||
| 5906 | NEONMAP0(vmovn_v), | |||
| 5907 | NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType), | |||
| 5908 | NEONMAP0(vmull_v), | |||
| 5909 | NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType), | |||
| 5910 | NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), | |||
| 5911 | NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), | |||
| 5912 | NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType), | |||
| 5913 | NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), | |||
| 5914 | NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), | |||
| 5915 | NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType), | |||
| 5916 | NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts), | |||
| 5917 | NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts), | |||
| 5918 | NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType), | |||
| 5919 | NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType), | |||
| 5920 | NEONMAP2(vqadd_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts), | |||
| 5921 | NEONMAP2(vqaddq_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts), | |||
| 5922 | NEONMAP2(vqdmlal_v, arm_neon_vqdmull, sadd_sat, 0), | |||
| 5923 | NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, ssub_sat, 0), | |||
| 5924 | NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType), | |||
| 5925 | NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType), | |||
| 5926 | NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType), | |||
| 5927 | NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts), | |||
| 5928 | NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType), | |||
| 5929 | NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType), | |||
| 5930 | NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType), | |||
| 5931 | NEONMAP1(vqrdmlah_v, arm_neon_vqrdmlah, Add1ArgType), | |||
| 5932 | NEONMAP1(vqrdmlahq_v, arm_neon_vqrdmlah, Add1ArgType), | |||
| 5933 | NEONMAP1(vqrdmlsh_v, arm_neon_vqrdmlsh, Add1ArgType), | |||
| 5934 | NEONMAP1(vqrdmlshq_v, arm_neon_vqrdmlsh, Add1ArgType), | |||
| 5935 | NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType), | |||
| 5936 | NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType), | |||
| 5937 | NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), | |||
| 5938 | NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), | |||
| 5939 | NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), | |||
| 5940 | NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), | |||
| 5941 | NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), | |||
| 5942 | NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), | |||
| 5943 | NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0), | |||
| 5944 | NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0), | |||
| 5945 | NEONMAP2(vqsub_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts), | |||
| 5946 | NEONMAP2(vqsubq_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts), | |||
| 5947 | NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType), | |||
| 5948 | NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), | |||
| 5949 | NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), | |||
| 5950 | NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType), | |||
| 5951 | NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType), | |||
| 5952 | NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), | |||
| 5953 | NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), | |||
| 5954 | NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType), | |||
| 5955 | NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType), | |||
| 5956 | NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType), | |||
| 5957 | NEONMAP0(vrndi_v), | |||
| 5958 | NEONMAP0(vrndiq_v), | |||
| 5959 | NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType), | |||
| 5960 | NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType), | |||
| 5961 | NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType), | |||
| 5962 | NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType), | |||
| 5963 | NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType), | |||
| 5964 | NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType), | |||
| 5965 | NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType), | |||
| 5966 | NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType), | |||
| 5967 | NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType), | |||
| 5968 | NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), | |||
| 5969 | NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), | |||
| 5970 | NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), | |||
| 5971 | NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), | |||
| 5972 | NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), | |||
| 5973 | NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), | |||
| 5974 | NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType), | |||
| 5975 | NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType), | |||
| 5976 | NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType), | |||
| 5977 | NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0), | |||
| 5978 | NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0), | |||
| 5979 | NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0), | |||
| 5980 | NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0), | |||
| 5981 | NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0), | |||
| 5982 | NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0), | |||
| 5983 | NEONMAP0(vshl_n_v), | |||
| 5984 | NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), | |||
| 5985 | NEONMAP0(vshll_n_v), | |||
| 5986 | NEONMAP0(vshlq_n_v), | |||
| 5987 | NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), | |||
| 5988 | NEONMAP0(vshr_n_v), | |||
| 5989 | NEONMAP0(vshrn_n_v), | |||
| 5990 | NEONMAP0(vshrq_n_v), | |||
| 5991 | NEONMAP1(vst1_v, arm_neon_vst1, 0), | |||
| 5992 | NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0), | |||
| 5993 | NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0), | |||
| 5994 | NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0), | |||
| 5995 | NEONMAP1(vst1q_v, arm_neon_vst1, 0), | |||
| 5996 | NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0), | |||
| 5997 | NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0), | |||
| 5998 | NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0), | |||
| 5999 | NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0), | |||
| 6000 | NEONMAP1(vst2_v, arm_neon_vst2, 0), | |||
| 6001 | NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0), | |||
| 6002 | NEONMAP1(vst2q_v, arm_neon_vst2, 0), | |||
| 6003 | NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0), | |||
| 6004 | NEONMAP1(vst3_v, arm_neon_vst3, 0), | |||
| 6005 | NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0), | |||
| 6006 | NEONMAP1(vst3q_v, arm_neon_vst3, 0), | |||
| 6007 | NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0), | |||
| 6008 | NEONMAP1(vst4_v, arm_neon_vst4, 0), | |||
| 6009 | NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0), | |||
| 6010 | NEONMAP1(vst4q_v, arm_neon_vst4, 0), | |||
| 6011 | NEONMAP0(vsubhn_v), | |||
| 6012 | NEONMAP0(vtrn_v), | |||
| 6013 | NEONMAP0(vtrnq_v), | |||
| 6014 | NEONMAP0(vtst_v), | |||
| 6015 | NEONMAP0(vtstq_v), | |||
| 6016 | NEONMAP1(vusdot_v, arm_neon_usdot, 0), | |||
| 6017 | NEONMAP1(vusdotq_v, arm_neon_usdot, 0), | |||
| 6018 | NEONMAP1(vusmmlaq_v, arm_neon_usmmla, 0), | |||
| 6019 | NEONMAP0(vuzp_v), | |||
| 6020 | NEONMAP0(vuzpq_v), | |||
| 6021 | NEONMAP0(vzip_v), | |||
| 6022 | NEONMAP0(vzipq_v) | |||
| 6023 | }; | |||
| 6024 | ||||
| 6025 | static const ARMVectorIntrinsicInfo AArch64SIMDIntrinsicMap[] = { | |||
| 6026 | NEONMAP1(__a64_vcvtq_low_bf16_v, aarch64_neon_bfcvtn, 0), | |||
| 6027 | NEONMAP0(splat_lane_v), | |||
| 6028 | NEONMAP0(splat_laneq_v), | |||
| 6029 | NEONMAP0(splatq_lane_v), | |||
| 6030 | NEONMAP0(splatq_laneq_v), | |||
| 6031 | NEONMAP1(vabs_v, aarch64_neon_abs, 0), | |||
| 6032 | NEONMAP1(vabsq_v, aarch64_neon_abs, 0), | |||
| 6033 | NEONMAP0(vadd_v), | |||
| 6034 | NEONMAP0(vaddhn_v), | |||
| 6035 | NEONMAP0(vaddq_p128), | |||
| 6036 | NEONMAP0(vaddq_v), | |||
| 6037 | NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0), | |||
| 6038 | NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0), | |||
| 6039 | NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0), | |||
| 6040 | NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0), | |||
| 6041 | NEONMAP2(vbcaxq_v, aarch64_crypto_bcaxu, aarch64_crypto_bcaxs, Add1ArgType | UnsignedAlts), | |||
| 6042 | NEONMAP1(vbfdot_v, aarch64_neon_bfdot, 0), | |||
| 6043 | NEONMAP1(vbfdotq_v, aarch64_neon_bfdot, 0), | |||
| 6044 | NEONMAP1(vbfmlalbq_v, aarch64_neon_bfmlalb, 0), | |||
| 6045 | NEONMAP1(vbfmlaltq_v, aarch64_neon_bfmlalt, 0), | |||
| 6046 | NEONMAP1(vbfmmlaq_v, aarch64_neon_bfmmla, 0), | |||
| 6047 | NEONMAP1(vcadd_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType), | |||
| 6048 | NEONMAP1(vcadd_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType), | |||
| 6049 | NEONMAP1(vcaddq_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType), | |||
| 6050 | NEONMAP1(vcaddq_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType), | |||
| 6051 | NEONMAP1(vcage_v, aarch64_neon_facge, 0), | |||
| 6052 | NEONMAP1(vcageq_v, aarch64_neon_facge, 0), | |||
| 6053 | NEONMAP1(vcagt_v, aarch64_neon_facgt, 0), | |||
| 6054 | NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0), | |||
| 6055 | NEONMAP1(vcale_v, aarch64_neon_facge, 0), | |||
| 6056 | NEONMAP1(vcaleq_v, aarch64_neon_facge, 0), | |||
| 6057 | NEONMAP1(vcalt_v, aarch64_neon_facgt, 0), | |||
| 6058 | NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0), | |||
| 6059 | NEONMAP0(vceqz_v), | |||
| 6060 | NEONMAP0(vceqzq_v), | |||
| 6061 | NEONMAP0(vcgez_v), | |||
| 6062 | NEONMAP0(vcgezq_v), | |||
| 6063 | NEONMAP0(vcgtz_v), | |||
| 6064 | NEONMAP0(vcgtzq_v), | |||
| 6065 | NEONMAP0(vclez_v), | |||
| 6066 | NEONMAP0(vclezq_v), | |||
| 6067 | NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType), | |||
| 6068 | NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType), | |||
| 6069 | NEONMAP0(vcltz_v), | |||
| 6070 | NEONMAP0(vcltzq_v), | |||
| 6071 | NEONMAP1(vclz_v, ctlz, Add1ArgType), | |||
| 6072 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), | |||
| 6073 | NEONMAP1(vcmla_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType), | |||
| 6074 | NEONMAP1(vcmla_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType), | |||
| 6075 | NEONMAP1(vcmla_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType), | |||
| 6076 | NEONMAP1(vcmla_v, aarch64_neon_vcmla_rot0, Add1ArgType), | |||
| 6077 | NEONMAP1(vcmlaq_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType), | |||
| 6078 | NEONMAP1(vcmlaq_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType), | |||
| 6079 | NEONMAP1(vcmlaq_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType), | |||
| 6080 | NEONMAP1(vcmlaq_v, aarch64_neon_vcmla_rot0, Add1ArgType), | |||
| 6081 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), | |||
| 6082 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), | |||
| 6083 | NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0), | |||
| 6084 | NEONMAP0(vcvt_f16_v), | |||
| 6085 | NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0), | |||
| 6086 | NEONMAP0(vcvt_f32_v), | |||
| 6087 | NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
| 6088 | NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
| 6089 | NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
| 6090 | NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), | |||
| 6091 | NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), | |||
| 6092 | NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), | |||
| 6093 | NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), | |||
| 6094 | NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), | |||
| 6095 | NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), | |||
| 6096 | NEONMAP0(vcvtq_f16_v), | |||
| 6097 | NEONMAP0(vcvtq_f32_v), | |||
| 6098 | NEONMAP1(vcvtq_high_bf16_v, aarch64_neon_bfcvtn2, 0), | |||
| 6099 | NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
| 6100 | NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
| 6101 | NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), | |||
| 6102 | NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), | |||
| 6103 | NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), | |||
| 6104 | NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), | |||
| 6105 | NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), | |||
| 6106 | NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), | |||
| 6107 | NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), | |||
| 6108 | NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType), | |||
| 6109 | NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0), | |||
| 6110 | NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0), | |||
| 6111 | NEONMAP2(veor3q_v, aarch64_crypto_eor3u, aarch64_crypto_eor3s, Add1ArgType | UnsignedAlts), | |||
| 6112 | NEONMAP0(vext_v), | |||
| 6113 | NEONMAP0(vextq_v), | |||
| 6114 | NEONMAP0(vfma_v), | |||
| 6115 | NEONMAP0(vfmaq_v), | |||
| 6116 | NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0), | |||
| 6117 | NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0), | |||
| 6118 | NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0), | |||
| 6119 | NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0), | |||
| 6120 | NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0), | |||
| 6121 | NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0), | |||
| 6122 | NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0), | |||
| 6123 | NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0), | |||
| 6124 | NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), | |||
| 6125 | NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), | |||
| 6126 | NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), | |||
| 6127 | NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), | |||
| 6128 | NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0), | |||
| 6129 | NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0), | |||
| 6130 | NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0), | |||
| 6131 | NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0), | |||
| 6132 | NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0), | |||
| 6133 | NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0), | |||
| 6134 | NEONMAP2(vmmlaq_v, aarch64_neon_ummla, aarch64_neon_smmla, 0), | |||
| 6135 | NEONMAP0(vmovl_v), | |||
| 6136 | NEONMAP0(vmovn_v), | |||
| 6137 | NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType), | |||
| 6138 | NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType), | |||
| 6139 | NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType), | |||
| 6140 | NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), | |||
| 6141 | NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), | |||
| 6142 | NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType), | |||
| 6143 | NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType), | |||
| 6144 | NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType), | |||
| 6145 | NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), | |||
| 6146 | NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), | |||
| 6147 | NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0), | |||
| 6148 | NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0), | |||
| 6149 | NEONMAP1(vqdmulh_lane_v, aarch64_neon_sqdmulh_lane, 0), | |||
| 6150 | NEONMAP1(vqdmulh_laneq_v, aarch64_neon_sqdmulh_laneq, 0), | |||
| 6151 | NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType), | |||
| 6152 | NEONMAP1(vqdmulhq_lane_v, aarch64_neon_sqdmulh_lane, 0), | |||
| 6153 | NEONMAP1(vqdmulhq_laneq_v, aarch64_neon_sqdmulh_laneq, 0), | |||
| 6154 | NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType), | |||
| 6155 | NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType), | |||
| 6156 | NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts), | |||
| 6157 | NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType), | |||
| 6158 | NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType), | |||
| 6159 | NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType), | |||
| 6160 | NEONMAP1(vqrdmlah_v, aarch64_neon_sqrdmlah, Add1ArgType), | |||
| 6161 | NEONMAP1(vqrdmlahq_v, aarch64_neon_sqrdmlah, Add1ArgType), | |||
| 6162 | NEONMAP1(vqrdmlsh_v, aarch64_neon_sqrdmlsh, Add1ArgType), | |||
| 6163 | NEONMAP1(vqrdmlshq_v, aarch64_neon_sqrdmlsh, Add1ArgType), | |||
| 6164 | NEONMAP1(vqrdmulh_lane_v, aarch64_neon_sqrdmulh_lane, 0), | |||
| 6165 | NEONMAP1(vqrdmulh_laneq_v, aarch64_neon_sqrdmulh_laneq, 0), | |||
| 6166 | NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType), | |||
| 6167 | NEONMAP1(vqrdmulhq_lane_v, aarch64_neon_sqrdmulh_lane, 0), | |||
| 6168 | NEONMAP1(vqrdmulhq_laneq_v, aarch64_neon_sqrdmulh_laneq, 0), | |||
| 6169 | NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType), | |||
| 6170 | NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), | |||
| 6171 | NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), | |||
| 6172 | NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts), | |||
| 6173 | NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), | |||
| 6174 | NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts), | |||
| 6175 | NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), | |||
| 6176 | NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0), | |||
| 6177 | NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0), | |||
| 6178 | NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), | |||
| 6179 | NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), | |||
| 6180 | NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType), | |||
| 6181 | NEONMAP1(vrax1q_v, aarch64_crypto_rax1, 0), | |||
| 6182 | NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), | |||
| 6183 | NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), | |||
| 6184 | NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType), | |||
| 6185 | NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType), | |||
| 6186 | NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), | |||
| 6187 | NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), | |||
| 6188 | NEONMAP1(vrnd32x_v, aarch64_neon_frint32x, Add1ArgType), | |||
| 6189 | NEONMAP1(vrnd32xq_v, aarch64_neon_frint32x, Add1ArgType), | |||
| 6190 | NEONMAP1(vrnd32z_v, aarch64_neon_frint32z, Add1ArgType), | |||
| 6191 | NEONMAP1(vrnd32zq_v, aarch64_neon_frint32z, Add1ArgType), | |||
| 6192 | NEONMAP1(vrnd64x_v, aarch64_neon_frint64x, Add1ArgType), | |||
| 6193 | NEONMAP1(vrnd64xq_v, aarch64_neon_frint64x, Add1ArgType), | |||
| 6194 | NEONMAP1(vrnd64z_v, aarch64_neon_frint64z, Add1ArgType), | |||
| 6195 | NEONMAP1(vrnd64zq_v, aarch64_neon_frint64z, Add1ArgType), | |||
| 6196 | NEONMAP0(vrndi_v), | |||
| 6197 | NEONMAP0(vrndiq_v), | |||
| 6198 | NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), | |||
| 6199 | NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), | |||
| 6200 | NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), | |||
| 6201 | NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), | |||
| 6202 | NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), | |||
| 6203 | NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), | |||
| 6204 | NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType), | |||
| 6205 | NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType), | |||
| 6206 | NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType), | |||
| 6207 | NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0), | |||
| 6208 | NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0), | |||
| 6209 | NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0), | |||
| 6210 | NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0), | |||
| 6211 | NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0), | |||
| 6212 | NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0), | |||
| 6213 | NEONMAP1(vsha512h2q_v, aarch64_crypto_sha512h2, 0), | |||
| 6214 | NEONMAP1(vsha512hq_v, aarch64_crypto_sha512h, 0), | |||
| 6215 | NEONMAP1(vsha512su0q_v, aarch64_crypto_sha512su0, 0), | |||
| 6216 | NEONMAP1(vsha512su1q_v, aarch64_crypto_sha512su1, 0), | |||
| 6217 | NEONMAP0(vshl_n_v), | |||
| 6218 | NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), | |||
| 6219 | NEONMAP0(vshll_n_v), | |||
| 6220 | NEONMAP0(vshlq_n_v), | |||
| 6221 | NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), | |||
| 6222 | NEONMAP0(vshr_n_v), | |||
| 6223 | NEONMAP0(vshrn_n_v), | |||
| 6224 | NEONMAP0(vshrq_n_v), | |||
| 6225 | NEONMAP1(vsm3partw1q_v, aarch64_crypto_sm3partw1, 0), | |||
| 6226 | NEONMAP1(vsm3partw2q_v, aarch64_crypto_sm3partw2, 0), | |||
| 6227 | NEONMAP1(vsm3ss1q_v, aarch64_crypto_sm3ss1, 0), | |||
| 6228 | NEONMAP1(vsm3tt1aq_v, aarch64_crypto_sm3tt1a, 0), | |||
| 6229 | NEONMAP1(vsm3tt1bq_v, aarch64_crypto_sm3tt1b, 0), | |||
| 6230 | NEONMAP1(vsm3tt2aq_v, aarch64_crypto_sm3tt2a, 0), | |||
| 6231 | NEONMAP1(vsm3tt2bq_v, aarch64_crypto_sm3tt2b, 0), | |||
| 6232 | NEONMAP1(vsm4ekeyq_v, aarch64_crypto_sm4ekey, 0), | |||
| 6233 | NEONMAP1(vsm4eq_v, aarch64_crypto_sm4e, 0), | |||
| 6234 | NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0), | |||
| 6235 | NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0), | |||
| 6236 | NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0), | |||
| 6237 | NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0), | |||
| 6238 | NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0), | |||
| 6239 | NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0), | |||
| 6240 | NEONMAP0(vsubhn_v), | |||
| 6241 | NEONMAP0(vtst_v), | |||
| 6242 | NEONMAP0(vtstq_v), | |||
| 6243 | NEONMAP1(vusdot_v, aarch64_neon_usdot, 0), | |||
| 6244 | NEONMAP1(vusdotq_v, aarch64_neon_usdot, 0), | |||
| 6245 | NEONMAP1(vusmmlaq_v, aarch64_neon_usmmla, 0), | |||
| 6246 | NEONMAP1(vxarq_v, aarch64_crypto_xar, 0), | |||
| 6247 | }; | |||
| 6248 | ||||
| 6249 | static const ARMVectorIntrinsicInfo AArch64SISDIntrinsicMap[] = { | |||
| 6250 | NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType), | |||
| 6251 | NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType), | |||
| 6252 | NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType), | |||
| 6253 | NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), | |||
| 6254 | NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), | |||
| 6255 | NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), | |||
| 6256 | NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), | |||
| 6257 | NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), | |||
| 6258 | NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), | |||
| 6259 | NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
| 6260 | NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), | |||
| 6261 | NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType), | |||
| 6262 | NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), | |||
| 6263 | NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType), | |||
| 6264 | NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
| 6265 | NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
| 6266 | NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
| 6267 | NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
| 6268 | NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
| 6269 | NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
| 6270 | NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
| 6271 | NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType), | |||
| 6272 | NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
| 6273 | NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), | |||
| 6274 | NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
| 6275 | NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
| 6276 | NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
| 6277 | NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
| 6278 | NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
| 6279 | NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
| 6280 | NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
| 6281 | NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
| 6282 | NEONMAP1(vcvtd_s64_f64, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), | |||
| 6283 | NEONMAP1(vcvtd_u64_f64, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), | |||
| 6284 | NEONMAP1(vcvth_bf16_f32, aarch64_neon_bfcvt, 0), | |||
| 6285 | NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
| 6286 | NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
| 6287 | NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
| 6288 | NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
| 6289 | NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
| 6290 | NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
| 6291 | NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
| 6292 | NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
| 6293 | NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
| 6294 | NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
| 6295 | NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
| 6296 | NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
| 6297 | NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
| 6298 | NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
| 6299 | NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
| 6300 | NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
| 6301 | NEONMAP1(vcvts_s32_f32, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), | |||
| 6302 | NEONMAP1(vcvts_u32_f32, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), | |||
| 6303 | NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0), | |||
| 6304 | NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
| 6305 | NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
| 6306 | NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
| 6307 | NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
| 6308 | NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), | |||
| 6309 | NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), | |||
| 6310 | NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
| 6311 | NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
| 6312 | NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), | |||
| 6313 | NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), | |||
| 6314 | NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
| 6315 | NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
| 6316 | NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
| 6317 | NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
| 6318 | NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), | |||
| 6319 | NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), | |||
| 6320 | NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
| 6321 | NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
| 6322 | NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), | |||
| 6323 | NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), | |||
| 6324 | NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0), | |||
| 6325 | NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType), | |||
| 6326 | NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType), | |||
| 6327 | NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
| 6328 | NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), | |||
| 6329 | NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
| 6330 | NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), | |||
| 6331 | NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
| 6332 | NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), | |||
| 6333 | NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
| 6334 | NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), | |||
| 6335 | NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
| 6336 | NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), | |||
| 6337 | NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), | |||
| 6338 | NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType), | |||
| 6339 | NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), | |||
| 6340 | NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType), | |||
| 6341 | NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6342 | NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6343 | NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType), | |||
| 6344 | NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType), | |||
| 6345 | NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6346 | NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6347 | NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType), | |||
| 6348 | NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType), | |||
| 6349 | NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors), | |||
| 6350 | NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType), | |||
| 6351 | NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors), | |||
| 6352 | NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0), | |||
| 6353 | NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType), | |||
| 6354 | NEONMAP1(vqmovnd_u64, aarch64_neon_scalar_uqxtn, AddRetType | Add1ArgType), | |||
| 6355 | NEONMAP1(vqmovnh_s16, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), | |||
| 6356 | NEONMAP1(vqmovnh_u16, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), | |||
| 6357 | NEONMAP1(vqmovns_s32, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), | |||
| 6358 | NEONMAP1(vqmovns_u32, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), | |||
| 6359 | NEONMAP1(vqmovund_s64, aarch64_neon_scalar_sqxtun, AddRetType | Add1ArgType), | |||
| 6360 | NEONMAP1(vqmovunh_s16, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), | |||
| 6361 | NEONMAP1(vqmovuns_s32, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), | |||
| 6362 | NEONMAP1(vqnegb_s8, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), | |||
| 6363 | NEONMAP1(vqnegd_s64, aarch64_neon_sqneg, Add1ArgType), | |||
| 6364 | NEONMAP1(vqnegh_s16, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), | |||
| 6365 | NEONMAP1(vqnegs_s32, aarch64_neon_sqneg, Add1ArgType), | |||
| 6366 | NEONMAP1(vqrdmlahh_s16, aarch64_neon_sqrdmlah, Vectorize1ArgType | Use64BitVectors), | |||
| 6367 | NEONMAP1(vqrdmlahs_s32, aarch64_neon_sqrdmlah, Add1ArgType), | |||
| 6368 | NEONMAP1(vqrdmlshh_s16, aarch64_neon_sqrdmlsh, Vectorize1ArgType | Use64BitVectors), | |||
| 6369 | NEONMAP1(vqrdmlshs_s32, aarch64_neon_sqrdmlsh, Add1ArgType), | |||
| 6370 | NEONMAP1(vqrdmulhh_s16, aarch64_neon_sqrdmulh, Vectorize1ArgType | Use64BitVectors), | |||
| 6371 | NEONMAP1(vqrdmulhs_s32, aarch64_neon_sqrdmulh, Add1ArgType), | |||
| 6372 | NEONMAP1(vqrshlb_s8, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6373 | NEONMAP1(vqrshlb_u8, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6374 | NEONMAP1(vqrshld_s64, aarch64_neon_sqrshl, Add1ArgType), | |||
| 6375 | NEONMAP1(vqrshld_u64, aarch64_neon_uqrshl, Add1ArgType), | |||
| 6376 | NEONMAP1(vqrshlh_s16, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6377 | NEONMAP1(vqrshlh_u16, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6378 | NEONMAP1(vqrshls_s32, aarch64_neon_sqrshl, Add1ArgType), | |||
| 6379 | NEONMAP1(vqrshls_u32, aarch64_neon_uqrshl, Add1ArgType), | |||
| 6380 | NEONMAP1(vqrshrnd_n_s64, aarch64_neon_sqrshrn, AddRetType), | |||
| 6381 | NEONMAP1(vqrshrnd_n_u64, aarch64_neon_uqrshrn, AddRetType), | |||
| 6382 | NEONMAP1(vqrshrnh_n_s16, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), | |||
| 6383 | NEONMAP1(vqrshrnh_n_u16, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), | |||
| 6384 | NEONMAP1(vqrshrns_n_s32, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), | |||
| 6385 | NEONMAP1(vqrshrns_n_u32, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), | |||
| 6386 | NEONMAP1(vqrshrund_n_s64, aarch64_neon_sqrshrun, AddRetType), | |||
| 6387 | NEONMAP1(vqrshrunh_n_s16, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), | |||
| 6388 | NEONMAP1(vqrshruns_n_s32, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), | |||
| 6389 | NEONMAP1(vqshlb_n_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6390 | NEONMAP1(vqshlb_n_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6391 | NEONMAP1(vqshlb_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6392 | NEONMAP1(vqshlb_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6393 | NEONMAP1(vqshld_s64, aarch64_neon_sqshl, Add1ArgType), | |||
| 6394 | NEONMAP1(vqshld_u64, aarch64_neon_uqshl, Add1ArgType), | |||
| 6395 | NEONMAP1(vqshlh_n_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6396 | NEONMAP1(vqshlh_n_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6397 | NEONMAP1(vqshlh_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6398 | NEONMAP1(vqshlh_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), | |||
| 6399 | NEONMAP1(vqshls_n_s32, aarch64_neon_sqshl, Add1ArgType), | |||
| 6400 | NEONMAP1(vqshls_n_u32, aarch64_neon_uqshl, Add1ArgType), | |||
| 6401 | NEONMAP1(vqshls_s32, aarch64_neon_sqshl, Add1ArgType), | |||
| 6402 | NEONMAP1(vqshls_u32, aarch64_neon_uqshl, Add1ArgType), | |||
| 6403 | NEONMAP1(vqshlub_n_s8, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), | |||
| 6404 | NEONMAP1(vqshluh_n_s16, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), | |||
| 6405 | NEONMAP1(vqshlus_n_s32, aarch64_neon_sqshlu, Add1ArgType), | |||
| 6406 | NEONMAP1(vqshrnd_n_s64, aarch64_neon_sqshrn, AddRetType), | |||
| 6407 | NEONMAP1(vqshrnd_n_u64, aarch64_neon_uqshrn, AddRetType), | |||
| 6408 | NEONMAP1(vqshrnh_n_s16, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), | |||
| 6409 | NEONMAP1(vqshrnh_n_u16, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), | |||
| 6410 | NEONMAP1(vqshrns_n_s32, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), | |||
| 6411 | NEONMAP1(vqshrns_n_u32, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), | |||
| 6412 | NEONMAP1(vqshrund_n_s64, aarch64_neon_sqshrun, AddRetType), | |||
| 6413 | NEONMAP1(vqshrunh_n_s16, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), | |||
| 6414 | NEONMAP1(vqshruns_n_s32, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), | |||
| 6415 | NEONMAP1(vqsubb_s8, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors), | |||
| 6416 | NEONMAP1(vqsubb_u8, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors), | |||
| 6417 | NEONMAP1(vqsubd_s64, aarch64_neon_sqsub, Add1ArgType), | |||
| 6418 | NEONMAP1(vqsubd_u64, aarch64_neon_uqsub, Add1ArgType), | |||
| 6419 | NEONMAP1(vqsubh_s16, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors), | |||
| 6420 | NEONMAP1(vqsubh_u16, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors), | |||
| 6421 | NEONMAP1(vqsubs_s32, aarch64_neon_sqsub, Add1ArgType), | |||
| 6422 | NEONMAP1(vqsubs_u32, aarch64_neon_uqsub, Add1ArgType), | |||
| 6423 | NEONMAP1(vrecped_f64, aarch64_neon_frecpe, Add1ArgType), | |||
| 6424 | NEONMAP1(vrecpes_f32, aarch64_neon_frecpe, Add1ArgType), | |||
| 6425 | NEONMAP1(vrecpxd_f64, aarch64_neon_frecpx, Add1ArgType), | |||
| 6426 | NEONMAP1(vrecpxs_f32, aarch64_neon_frecpx, Add1ArgType), | |||
| 6427 | NEONMAP1(vrshld_s64, aarch64_neon_srshl, Add1ArgType), | |||
| 6428 | NEONMAP1(vrshld_u64, aarch64_neon_urshl, Add1ArgType), | |||
| 6429 | NEONMAP1(vrsqrted_f64, aarch64_neon_frsqrte, Add1ArgType), | |||
| 6430 | NEONMAP1(vrsqrtes_f32, aarch64_neon_frsqrte, Add1ArgType), | |||
| 6431 | NEONMAP1(vrsqrtsd_f64, aarch64_neon_frsqrts, Add1ArgType), | |||
| 6432 | NEONMAP1(vrsqrtss_f32, aarch64_neon_frsqrts, Add1ArgType), | |||
| 6433 | NEONMAP1(vsha1cq_u32, aarch64_crypto_sha1c, 0), | |||
| 6434 | NEONMAP1(vsha1h_u32, aarch64_crypto_sha1h, 0), | |||
| 6435 | NEONMAP1(vsha1mq_u32, aarch64_crypto_sha1m, 0), | |||
| 6436 | NEONMAP1(vsha1pq_u32, aarch64_crypto_sha1p, 0), | |||
| 6437 | NEONMAP1(vshld_s64, aarch64_neon_sshl, Add1ArgType), | |||
| 6438 | NEONMAP1(vshld_u64, aarch64_neon_ushl, Add1ArgType), | |||
| 6439 | NEONMAP1(vslid_n_s64, aarch64_neon_vsli, Vectorize1ArgType), | |||
| 6440 | NEONMAP1(vslid_n_u64, aarch64_neon_vsli, Vectorize1ArgType), | |||
| 6441 | NEONMAP1(vsqaddb_u8, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6442 | NEONMAP1(vsqaddd_u64, aarch64_neon_usqadd, Add1ArgType), | |||
| 6443 | NEONMAP1(vsqaddh_u16, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6444 | NEONMAP1(vsqadds_u32, aarch64_neon_usqadd, Add1ArgType), | |||
| 6445 | NEONMAP1(vsrid_n_s64, aarch64_neon_vsri, Vectorize1ArgType), | |||
| 6446 | NEONMAP1(vsrid_n_u64, aarch64_neon_vsri, Vectorize1ArgType), | |||
| 6447 | NEONMAP1(vuqaddb_s8, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6448 | NEONMAP1(vuqaddd_s64, aarch64_neon_suqadd, Add1ArgType), | |||
| 6449 | NEONMAP1(vuqaddh_s16, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors), | |||
| 6450 | NEONMAP1(vuqadds_s32, aarch64_neon_suqadd, Add1ArgType), | |||
| 6451 | // FP16 scalar intrinisics go here. | |||
| 6452 | NEONMAP1(vabdh_f16, aarch64_sisd_fabd, Add1ArgType), | |||
| 6453 | NEONMAP1(vcvtah_s32_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
| 6454 | NEONMAP1(vcvtah_s64_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType), | |||
| 6455 | NEONMAP1(vcvtah_u32_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
| 6456 | NEONMAP1(vcvtah_u64_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType), | |||
| 6457 | NEONMAP1(vcvth_n_f16_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
| 6458 | NEONMAP1(vcvth_n_f16_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), | |||
| 6459 | NEONMAP1(vcvth_n_f16_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
| 6460 | NEONMAP1(vcvth_n_f16_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), | |||
| 6461 | NEONMAP1(vcvth_n_s32_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
| 6462 | NEONMAP1(vcvth_n_s64_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), | |||
| 6463 | NEONMAP1(vcvth_n_u32_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
| 6464 | NEONMAP1(vcvth_n_u64_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), | |||
| 6465 | NEONMAP1(vcvth_s32_f16, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), | |||
| 6466 | NEONMAP1(vcvth_s64_f16, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), | |||
| 6467 | NEONMAP1(vcvth_u32_f16, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), | |||
| 6468 | NEONMAP1(vcvth_u64_f16, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), | |||
| 6469 | NEONMAP1(vcvtmh_s32_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
| 6470 | NEONMAP1(vcvtmh_s64_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType), | |||
| 6471 | NEONMAP1(vcvtmh_u32_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
| 6472 | NEONMAP1(vcvtmh_u64_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), | |||
| 6473 | NEONMAP1(vcvtnh_s32_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
| 6474 | NEONMAP1(vcvtnh_s64_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType), | |||
| 6475 | NEONMAP1(vcvtnh_u32_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
| 6476 | NEONMAP1(vcvtnh_u64_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), | |||
| 6477 | NEONMAP1(vcvtph_s32_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
| 6478 | NEONMAP1(vcvtph_s64_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType), | |||
| 6479 | NEONMAP1(vcvtph_u32_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
| 6480 | NEONMAP1(vcvtph_u64_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), | |||
| 6481 | NEONMAP1(vmulxh_f16, aarch64_neon_fmulx, Add1ArgType), | |||
| 6482 | NEONMAP1(vrecpeh_f16, aarch64_neon_frecpe, Add1ArgType), | |||
| 6483 | NEONMAP1(vrecpxh_f16, aarch64_neon_frecpx, Add1ArgType), | |||
| 6484 | NEONMAP1(vrsqrteh_f16, aarch64_neon_frsqrte, Add1ArgType), | |||
| 6485 | NEONMAP1(vrsqrtsh_f16, aarch64_neon_frsqrts, Add1ArgType), | |||
| 6486 | }; | |||
| 6487 | ||||
| 6488 | #undef NEONMAP0 | |||
| 6489 | #undef NEONMAP1 | |||
| 6490 | #undef NEONMAP2 | |||
| 6491 | ||||
| 6492 | #define SVEMAP1(NameBase, LLVMIntrinsic, TypeModifier) \ | |||
| 6493 | { \ | |||
| 6494 | #NameBase, SVE::BI__builtin_sve_##NameBase, Intrinsic::LLVMIntrinsic, 0, \ | |||
| 6495 | TypeModifier \ | |||
| 6496 | } | |||
| 6497 | ||||
| 6498 | #define SVEMAP2(NameBase, TypeModifier) \ | |||
| 6499 | { #NameBase, SVE::BI__builtin_sve_##NameBase, 0, 0, TypeModifier } | |||
| 6500 | static const ARMVectorIntrinsicInfo AArch64SVEIntrinsicMap[] = { | |||
| 6501 | #define GET_SVE_LLVM_INTRINSIC_MAP | |||
| 6502 | #include "clang/Basic/arm_sve_builtin_cg.inc" | |||
| 6503 | #include "clang/Basic/BuiltinsAArch64NeonSVEBridge_cg.def" | |||
| 6504 | #undef GET_SVE_LLVM_INTRINSIC_MAP | |||
| 6505 | }; | |||
| 6506 | ||||
| 6507 | #undef SVEMAP1 | |||
| 6508 | #undef SVEMAP2 | |||
| 6509 | ||||
| 6510 | static bool NEONSIMDIntrinsicsProvenSorted = false; | |||
| 6511 | ||||
| 6512 | static bool AArch64SIMDIntrinsicsProvenSorted = false; | |||
| 6513 | static bool AArch64SISDIntrinsicsProvenSorted = false; | |||
| 6514 | static bool AArch64SVEIntrinsicsProvenSorted = false; | |||
| 6515 | ||||
| 6516 | static const ARMVectorIntrinsicInfo * | |||
| 6517 | findARMVectorIntrinsicInMap(ArrayRef<ARMVectorIntrinsicInfo> IntrinsicMap, | |||
| 6518 | unsigned BuiltinID, bool &MapProvenSorted) { | |||
| 6519 | ||||
| 6520 | #ifndef NDEBUG | |||
| 6521 | if (!MapProvenSorted) { | |||
| 6522 | assert(llvm::is_sorted(IntrinsicMap))(static_cast <bool> (llvm::is_sorted(IntrinsicMap)) ? void (0) : __assert_fail ("llvm::is_sorted(IntrinsicMap)", "clang/lib/CodeGen/CGBuiltin.cpp" , 6522, __extension__ __PRETTY_FUNCTION__)); | |||
| 6523 | MapProvenSorted = true; | |||
| 6524 | } | |||
| 6525 | #endif | |||
| 6526 | ||||
| 6527 | const ARMVectorIntrinsicInfo *Builtin = | |||
| 6528 | llvm::lower_bound(IntrinsicMap, BuiltinID); | |||
| 6529 | ||||
| 6530 | if (Builtin != IntrinsicMap.end() && Builtin->BuiltinID == BuiltinID) | |||
| 6531 | return Builtin; | |||
| 6532 | ||||
| 6533 | return nullptr; | |||
| 6534 | } | |||
| 6535 | ||||
| 6536 | Function *CodeGenFunction::LookupNeonLLVMIntrinsic(unsigned IntrinsicID, | |||
| 6537 | unsigned Modifier, | |||
| 6538 | llvm::Type *ArgType, | |||
| 6539 | const CallExpr *E) { | |||
| 6540 | int VectorSize = 0; | |||
| 6541 | if (Modifier & Use64BitVectors) | |||
| 6542 | VectorSize = 64; | |||
| 6543 | else if (Modifier & Use128BitVectors) | |||
| 6544 | VectorSize = 128; | |||
| 6545 | ||||
| 6546 | // Return type. | |||
| 6547 | SmallVector<llvm::Type *, 3> Tys; | |||
| 6548 | if (Modifier & AddRetType) { | |||
| 6549 | llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext())); | |||
| 6550 | if (Modifier & VectorizeRetType) | |||
| 6551 | Ty = llvm::FixedVectorType::get( | |||
| 6552 | Ty, VectorSize ? VectorSize / Ty->getPrimitiveSizeInBits() : 1); | |||
| 6553 | ||||
| 6554 | Tys.push_back(Ty); | |||
| 6555 | } | |||
| 6556 | ||||
| 6557 | // Arguments. | |||
| 6558 | if (Modifier & VectorizeArgTypes) { | |||
| 6559 | int Elts = VectorSize ? VectorSize / ArgType->getPrimitiveSizeInBits() : 1; | |||
| 6560 | ArgType = llvm::FixedVectorType::get(ArgType, Elts); | |||
| 6561 | } | |||
| 6562 | ||||
| 6563 | if (Modifier & (Add1ArgType | Add2ArgTypes)) | |||
| 6564 | Tys.push_back(ArgType); | |||
| 6565 | ||||
| 6566 | if (Modifier & Add2ArgTypes) | |||
| 6567 | Tys.push_back(ArgType); | |||
| 6568 | ||||
| 6569 | if (Modifier & InventFloatType) | |||
| 6570 | Tys.push_back(FloatTy); | |||
| 6571 | ||||
| 6572 | return CGM.getIntrinsic(IntrinsicID, Tys); | |||
| 6573 | } | |||
| 6574 | ||||
| 6575 | static Value *EmitCommonNeonSISDBuiltinExpr( | |||
| 6576 | CodeGenFunction &CGF, const ARMVectorIntrinsicInfo &SISDInfo, | |||
| 6577 | SmallVectorImpl<Value *> &Ops, const CallExpr *E) { | |||
| 6578 | unsigned BuiltinID = SISDInfo.BuiltinID; | |||
| 6579 | unsigned int Int = SISDInfo.LLVMIntrinsic; | |||
| 6580 | unsigned Modifier = SISDInfo.TypeModifier; | |||
| 6581 | const char *s = SISDInfo.NameHint; | |||
| 6582 | ||||
| 6583 | switch (BuiltinID) { | |||
| 6584 | case NEON::BI__builtin_neon_vcled_s64: | |||
| 6585 | case NEON::BI__builtin_neon_vcled_u64: | |||
| 6586 | case NEON::BI__builtin_neon_vcles_f32: | |||
| 6587 | case NEON::BI__builtin_neon_vcled_f64: | |||
| 6588 | case NEON::BI__builtin_neon_vcltd_s64: | |||
| 6589 | case NEON::BI__builtin_neon_vcltd_u64: | |||
| 6590 | case NEON::BI__builtin_neon_vclts_f32: | |||
| 6591 | case NEON::BI__builtin_neon_vcltd_f64: | |||
| 6592 | case NEON::BI__builtin_neon_vcales_f32: | |||
| 6593 | case NEON::BI__builtin_neon_vcaled_f64: | |||
| 6594 | case NEON::BI__builtin_neon_vcalts_f32: | |||
| 6595 | case NEON::BI__builtin_neon_vcaltd_f64: | |||
| 6596 | // Only one direction of comparisons actually exist, cmle is actually a cmge | |||
| 6597 | // with swapped operands. The table gives us the right intrinsic but we | |||
| 6598 | // still need to do the swap. | |||
| 6599 | std::swap(Ops[0], Ops[1]); | |||
| 6600 | break; | |||
| 6601 | } | |||
| 6602 | ||||
| 6603 | assert(Int && "Generic code assumes a valid intrinsic")(static_cast <bool> (Int && "Generic code assumes a valid intrinsic" ) ? void (0) : __assert_fail ("Int && \"Generic code assumes a valid intrinsic\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 6603, __extension__ __PRETTY_FUNCTION__ )); | |||
| 6604 | ||||
| 6605 | // Determine the type(s) of this overloaded AArch64 intrinsic. | |||
| 6606 | const Expr *Arg = E->getArg(0); | |||
| 6607 | llvm::Type *ArgTy = CGF.ConvertType(Arg->getType()); | |||
| 6608 | Function *F = CGF.LookupNeonLLVMIntrinsic(Int, Modifier, ArgTy, E); | |||
| 6609 | ||||
| 6610 | int j = 0; | |||
| 6611 | ConstantInt *C0 = ConstantInt::get(CGF.SizeTy, 0); | |||
| 6612 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); | |||
| 6613 | ai != ae; ++ai, ++j) { | |||
| 6614 | llvm::Type *ArgTy = ai->getType(); | |||
| 6615 | if (Ops[j]->getType()->getPrimitiveSizeInBits() == | |||
| 6616 | ArgTy->getPrimitiveSizeInBits()) | |||
| 6617 | continue; | |||
| 6618 | ||||
| 6619 | assert(ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy())(static_cast <bool> (ArgTy->isVectorTy() && ! Ops[j]->getType()->isVectorTy()) ? void (0) : __assert_fail ("ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 6619, __extension__ __PRETTY_FUNCTION__ )); | |||
| 6620 | // The constant argument to an _n_ intrinsic always has Int32Ty, so truncate | |||
| 6621 | // it before inserting. | |||
| 6622 | Ops[j] = CGF.Builder.CreateTruncOrBitCast( | |||
| 6623 | Ops[j], cast<llvm::VectorType>(ArgTy)->getElementType()); | |||
| 6624 | Ops[j] = | |||
| 6625 | CGF.Builder.CreateInsertElement(UndefValue::get(ArgTy), Ops[j], C0); | |||
| 6626 | } | |||
| 6627 | ||||
| 6628 | Value *Result = CGF.EmitNeonCall(F, Ops, s); | |||
| 6629 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); | |||
| 6630 | if (ResultType->getPrimitiveSizeInBits().getFixedSize() < | |||
| 6631 | Result->getType()->getPrimitiveSizeInBits().getFixedSize()) | |||
| 6632 | return CGF.Builder.CreateExtractElement(Result, C0); | |||
| 6633 | ||||
| 6634 | return CGF.Builder.CreateBitCast(Result, ResultType, s); | |||
| 6635 | } | |||
| 6636 | ||||
| 6637 | Value *CodeGenFunction::EmitCommonNeonBuiltinExpr( | |||
| 6638 | unsigned BuiltinID, unsigned LLVMIntrinsic, unsigned AltLLVMIntrinsic, | |||
| 6639 | const char *NameHint, unsigned Modifier, const CallExpr *E, | |||
| 6640 | SmallVectorImpl<llvm::Value *> &Ops, Address PtrOp0, Address PtrOp1, | |||
| 6641 | llvm::Triple::ArchType Arch) { | |||
| 6642 | // Get the last argument, which specifies the vector type. | |||
| 6643 | const Expr *Arg = E->getArg(E->getNumArgs() - 1); | |||
| 6644 | Optional<llvm::APSInt> NeonTypeConst = | |||
| 6645 | Arg->getIntegerConstantExpr(getContext()); | |||
| 6646 | if (!NeonTypeConst) | |||
| 6647 | return nullptr; | |||
| 6648 | ||||
| 6649 | // Determine the type of this overloaded NEON intrinsic. | |||
| 6650 | NeonTypeFlags Type(NeonTypeConst->getZExtValue()); | |||
| 6651 | bool Usgn = Type.isUnsigned(); | |||
| 6652 | bool Quad = Type.isQuad(); | |||
| 6653 | const bool HasLegalHalfType = getTarget().hasLegalHalfType(); | |||
| 6654 | const bool AllowBFloatArgsAndRet = | |||
| 6655 | getTargetHooks().getABIInfo().allowBFloatArgsAndRet(); | |||
| 6656 | ||||
| 6657 | llvm::FixedVectorType *VTy = | |||
| 6658 | GetNeonType(this, Type, HasLegalHalfType, false, AllowBFloatArgsAndRet); | |||
| 6659 | llvm::Type *Ty = VTy; | |||
| 6660 | if (!Ty) | |||
| 6661 | return nullptr; | |||
| 6662 | ||||
| 6663 | auto getAlignmentValue32 = [&](Address addr) -> Value* { | |||
| 6664 | return Builder.getInt32(addr.getAlignment().getQuantity()); | |||
| 6665 | }; | |||
| 6666 | ||||
| 6667 | unsigned Int = LLVMIntrinsic; | |||
| 6668 | if ((Modifier & UnsignedAlts) && !Usgn) | |||
| 6669 | Int = AltLLVMIntrinsic; | |||
| 6670 | ||||
| 6671 | switch (BuiltinID) { | |||
| 6672 | default: break; | |||
| 6673 | case NEON::BI__builtin_neon_splat_lane_v: | |||
| 6674 | case NEON::BI__builtin_neon_splat_laneq_v: | |||
| 6675 | case NEON::BI__builtin_neon_splatq_lane_v: | |||
| 6676 | case NEON::BI__builtin_neon_splatq_laneq_v: { | |||
| 6677 | auto NumElements = VTy->getElementCount(); | |||
| 6678 | if (BuiltinID == NEON::BI__builtin_neon_splatq_lane_v) | |||
| 6679 | NumElements = NumElements * 2; | |||
| 6680 | if (BuiltinID == NEON::BI__builtin_neon_splat_laneq_v) | |||
| 6681 | NumElements = NumElements.divideCoefficientBy(2); | |||
| 6682 | ||||
| 6683 | Ops[0] = Builder.CreateBitCast(Ops[0], VTy); | |||
| 6684 | return EmitNeonSplat(Ops[0], cast<ConstantInt>(Ops[1]), NumElements); | |||
| 6685 | } | |||
| 6686 | case NEON::BI__builtin_neon_vpadd_v: | |||
| 6687 | case NEON::BI__builtin_neon_vpaddq_v: | |||
| 6688 | // We don't allow fp/int overloading of intrinsics. | |||
| 6689 | if (VTy->getElementType()->isFloatingPointTy() && | |||
| 6690 | Int == Intrinsic::aarch64_neon_addp) | |||
| 6691 | Int = Intrinsic::aarch64_neon_faddp; | |||
| 6692 | break; | |||
| 6693 | case NEON::BI__builtin_neon_vabs_v: | |||
| 6694 | case NEON::BI__builtin_neon_vabsq_v: | |||
| 6695 | if (VTy->getElementType()->isFloatingPointTy()) | |||
| 6696 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, Ty), Ops, "vabs"); | |||
| 6697 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), Ops, "vabs"); | |||
| 6698 | case NEON::BI__builtin_neon_vadd_v: | |||
| 6699 | case NEON::BI__builtin_neon_vaddq_v: { | |||
| 6700 | llvm::Type *VTy = llvm::FixedVectorType::get(Int8Ty, Quad ? 16 : 8); | |||
| 6701 | Ops[0] = Builder.CreateBitCast(Ops[0], VTy); | |||
| 6702 | Ops[1] = Builder.CreateBitCast(Ops[1], VTy); | |||
| 6703 | Ops[0] = Builder.CreateXor(Ops[0], Ops[1]); | |||
| 6704 | return Builder.CreateBitCast(Ops[0], Ty); | |||
| 6705 | } | |||
| 6706 | case NEON::BI__builtin_neon_vaddhn_v: { | |||
| 6707 | llvm::FixedVectorType *SrcTy = | |||
| 6708 | llvm::FixedVectorType::getExtendedElementVectorType(VTy); | |||
| 6709 | ||||
| 6710 | // %sum = add <4 x i32> %lhs, %rhs | |||
| 6711 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
| 6712 | Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy); | |||
| 6713 | Ops[0] = Builder.CreateAdd(Ops[0], Ops[1], "vaddhn"); | |||
| 6714 | ||||
| 6715 | // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16> | |||
| 6716 | Constant *ShiftAmt = | |||
| 6717 | ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2); | |||
| 6718 | Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vaddhn"); | |||
| 6719 | ||||
| 6720 | // %res = trunc <4 x i32> %high to <4 x i16> | |||
| 6721 | return Builder.CreateTrunc(Ops[0], VTy, "vaddhn"); | |||
| 6722 | } | |||
| 6723 | case NEON::BI__builtin_neon_vcale_v: | |||
| 6724 | case NEON::BI__builtin_neon_vcaleq_v: | |||
| 6725 | case NEON::BI__builtin_neon_vcalt_v: | |||
| 6726 | case NEON::BI__builtin_neon_vcaltq_v: | |||
| 6727 | std::swap(Ops[0], Ops[1]); | |||
| 6728 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 6729 | case NEON::BI__builtin_neon_vcage_v: | |||
| 6730 | case NEON::BI__builtin_neon_vcageq_v: | |||
| 6731 | case NEON::BI__builtin_neon_vcagt_v: | |||
| 6732 | case NEON::BI__builtin_neon_vcagtq_v: { | |||
| 6733 | llvm::Type *Ty; | |||
| 6734 | switch (VTy->getScalarSizeInBits()) { | |||
| 6735 | default: llvm_unreachable("unexpected type")::llvm::llvm_unreachable_internal("unexpected type", "clang/lib/CodeGen/CGBuiltin.cpp" , 6735); | |||
| 6736 | case 32: | |||
| 6737 | Ty = FloatTy; | |||
| 6738 | break; | |||
| 6739 | case 64: | |||
| 6740 | Ty = DoubleTy; | |||
| 6741 | break; | |||
| 6742 | case 16: | |||
| 6743 | Ty = HalfTy; | |||
| 6744 | break; | |||
| 6745 | } | |||
| 6746 | auto *VecFlt = llvm::FixedVectorType::get(Ty, VTy->getNumElements()); | |||
| 6747 | llvm::Type *Tys[] = { VTy, VecFlt }; | |||
| 6748 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
| 6749 | return EmitNeonCall(F, Ops, NameHint); | |||
| 6750 | } | |||
| 6751 | case NEON::BI__builtin_neon_vceqz_v: | |||
| 6752 | case NEON::BI__builtin_neon_vceqzq_v: | |||
| 6753 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OEQ, | |||
| 6754 | ICmpInst::ICMP_EQ, "vceqz"); | |||
| 6755 | case NEON::BI__builtin_neon_vcgez_v: | |||
| 6756 | case NEON::BI__builtin_neon_vcgezq_v: | |||
| 6757 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGE, | |||
| 6758 | ICmpInst::ICMP_SGE, "vcgez"); | |||
| 6759 | case NEON::BI__builtin_neon_vclez_v: | |||
| 6760 | case NEON::BI__builtin_neon_vclezq_v: | |||
| 6761 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLE, | |||
| 6762 | ICmpInst::ICMP_SLE, "vclez"); | |||
| 6763 | case NEON::BI__builtin_neon_vcgtz_v: | |||
| 6764 | case NEON::BI__builtin_neon_vcgtzq_v: | |||
| 6765 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGT, | |||
| 6766 | ICmpInst::ICMP_SGT, "vcgtz"); | |||
| 6767 | case NEON::BI__builtin_neon_vcltz_v: | |||
| 6768 | case NEON::BI__builtin_neon_vcltzq_v: | |||
| 6769 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLT, | |||
| 6770 | ICmpInst::ICMP_SLT, "vcltz"); | |||
| 6771 | case NEON::BI__builtin_neon_vclz_v: | |||
| 6772 | case NEON::BI__builtin_neon_vclzq_v: | |||
| 6773 | // We generate target-independent intrinsic, which needs a second argument | |||
| 6774 | // for whether or not clz of zero is undefined; on ARM it isn't. | |||
| 6775 | Ops.push_back(Builder.getInt1(getTarget().isCLZForZeroUndef())); | |||
| 6776 | break; | |||
| 6777 | case NEON::BI__builtin_neon_vcvt_f32_v: | |||
| 6778 | case NEON::BI__builtin_neon_vcvtq_f32_v: | |||
| 6779 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 6780 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float32, false, Quad), | |||
| 6781 | HasLegalHalfType); | |||
| 6782 | return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") | |||
| 6783 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); | |||
| 6784 | case NEON::BI__builtin_neon_vcvt_f16_v: | |||
| 6785 | case NEON::BI__builtin_neon_vcvtq_f16_v: | |||
| 6786 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 6787 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float16, false, Quad), | |||
| 6788 | HasLegalHalfType); | |||
| 6789 | return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") | |||
| 6790 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); | |||
| 6791 | case NEON::BI__builtin_neon_vcvt_n_f16_v: | |||
| 6792 | case NEON::BI__builtin_neon_vcvt_n_f32_v: | |||
| 6793 | case NEON::BI__builtin_neon_vcvt_n_f64_v: | |||
| 6794 | case NEON::BI__builtin_neon_vcvtq_n_f16_v: | |||
| 6795 | case NEON::BI__builtin_neon_vcvtq_n_f32_v: | |||
| 6796 | case NEON::BI__builtin_neon_vcvtq_n_f64_v: { | |||
| 6797 | llvm::Type *Tys[2] = { GetFloatNeonType(this, Type), Ty }; | |||
| 6798 | Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic; | |||
| 6799 | Function *F = CGM.getIntrinsic(Int, Tys); | |||
| 6800 | return EmitNeonCall(F, Ops, "vcvt_n"); | |||
| 6801 | } | |||
| 6802 | case NEON::BI__builtin_neon_vcvt_n_s16_v: | |||
| 6803 | case NEON::BI__builtin_neon_vcvt_n_s32_v: | |||
| 6804 | case NEON::BI__builtin_neon_vcvt_n_u16_v: | |||
| 6805 | case NEON::BI__builtin_neon_vcvt_n_u32_v: | |||
| 6806 | case NEON::BI__builtin_neon_vcvt_n_s64_v: | |||
| 6807 | case NEON::BI__builtin_neon_vcvt_n_u64_v: | |||
| 6808 | case NEON::BI__builtin_neon_vcvtq_n_s16_v: | |||
| 6809 | case NEON::BI__builtin_neon_vcvtq_n_s32_v: | |||
| 6810 | case NEON::BI__builtin_neon_vcvtq_n_u16_v: | |||
| 6811 | case NEON::BI__builtin_neon_vcvtq_n_u32_v: | |||
| 6812 | case NEON::BI__builtin_neon_vcvtq_n_s64_v: | |||
| 6813 | case NEON::BI__builtin_neon_vcvtq_n_u64_v: { | |||
| 6814 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
| 6815 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
| 6816 | return EmitNeonCall(F, Ops, "vcvt_n"); | |||
| 6817 | } | |||
| 6818 | case NEON::BI__builtin_neon_vcvt_s32_v: | |||
| 6819 | case NEON::BI__builtin_neon_vcvt_u32_v: | |||
| 6820 | case NEON::BI__builtin_neon_vcvt_s64_v: | |||
| 6821 | case NEON::BI__builtin_neon_vcvt_u64_v: | |||
| 6822 | case NEON::BI__builtin_neon_vcvt_s16_v: | |||
| 6823 | case NEON::BI__builtin_neon_vcvt_u16_v: | |||
| 6824 | case NEON::BI__builtin_neon_vcvtq_s32_v: | |||
| 6825 | case NEON::BI__builtin_neon_vcvtq_u32_v: | |||
| 6826 | case NEON::BI__builtin_neon_vcvtq_s64_v: | |||
| 6827 | case NEON::BI__builtin_neon_vcvtq_u64_v: | |||
| 6828 | case NEON::BI__builtin_neon_vcvtq_s16_v: | |||
| 6829 | case NEON::BI__builtin_neon_vcvtq_u16_v: { | |||
| 6830 | Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type)); | |||
| 6831 | return Usgn ? Builder.CreateFPToUI(Ops[0], Ty, "vcvt") | |||
| 6832 | : Builder.CreateFPToSI(Ops[0], Ty, "vcvt"); | |||
| 6833 | } | |||
| 6834 | case NEON::BI__builtin_neon_vcvta_s16_v: | |||
| 6835 | case NEON::BI__builtin_neon_vcvta_s32_v: | |||
| 6836 | case NEON::BI__builtin_neon_vcvta_s64_v: | |||
| 6837 | case NEON::BI__builtin_neon_vcvta_u16_v: | |||
| 6838 | case NEON::BI__builtin_neon_vcvta_u32_v: | |||
| 6839 | case NEON::BI__builtin_neon_vcvta_u64_v: | |||
| 6840 | case NEON::BI__builtin_neon_vcvtaq_s16_v: | |||
| 6841 | case NEON::BI__builtin_neon_vcvtaq_s32_v: | |||
| 6842 | case NEON::BI__builtin_neon_vcvtaq_s64_v: | |||
| 6843 | case NEON::BI__builtin_neon_vcvtaq_u16_v: | |||
| 6844 | case NEON::BI__builtin_neon_vcvtaq_u32_v: | |||
| 6845 | case NEON::BI__builtin_neon_vcvtaq_u64_v: | |||
| 6846 | case NEON::BI__builtin_neon_vcvtn_s16_v: | |||
| 6847 | case NEON::BI__builtin_neon_vcvtn_s32_v: | |||
| 6848 | case NEON::BI__builtin_neon_vcvtn_s64_v: | |||
| 6849 | case NEON::BI__builtin_neon_vcvtn_u16_v: | |||
| 6850 | case NEON::BI__builtin_neon_vcvtn_u32_v: | |||
| 6851 | case NEON::BI__builtin_neon_vcvtn_u64_v: | |||
| 6852 | case NEON::BI__builtin_neon_vcvtnq_s16_v: | |||
| 6853 | case NEON::BI__builtin_neon_vcvtnq_s32_v: | |||
| 6854 | case NEON::BI__builtin_neon_vcvtnq_s64_v: | |||
| 6855 | case NEON::BI__builtin_neon_vcvtnq_u16_v: | |||
| 6856 | case NEON::BI__builtin_neon_vcvtnq_u32_v: | |||
| 6857 | case NEON::BI__builtin_neon_vcvtnq_u64_v: | |||
| 6858 | case NEON::BI__builtin_neon_vcvtp_s16_v: | |||
| 6859 | case NEON::BI__builtin_neon_vcvtp_s32_v: | |||
| 6860 | case NEON::BI__builtin_neon_vcvtp_s64_v: | |||
| 6861 | case NEON::BI__builtin_neon_vcvtp_u16_v: | |||
| 6862 | case NEON::BI__builtin_neon_vcvtp_u32_v: | |||
| 6863 | case NEON::BI__builtin_neon_vcvtp_u64_v: | |||
| 6864 | case NEON::BI__builtin_neon_vcvtpq_s16_v: | |||
| 6865 | case NEON::BI__builtin_neon_vcvtpq_s32_v: | |||
| 6866 | case NEON::BI__builtin_neon_vcvtpq_s64_v: | |||
| 6867 | case NEON::BI__builtin_neon_vcvtpq_u16_v: | |||
| 6868 | case NEON::BI__builtin_neon_vcvtpq_u32_v: | |||
| 6869 | case NEON::BI__builtin_neon_vcvtpq_u64_v: | |||
| 6870 | case NEON::BI__builtin_neon_vcvtm_s16_v: | |||
| 6871 | case NEON::BI__builtin_neon_vcvtm_s32_v: | |||
| 6872 | case NEON::BI__builtin_neon_vcvtm_s64_v: | |||
| 6873 | case NEON::BI__builtin_neon_vcvtm_u16_v: | |||
| 6874 | case NEON::BI__builtin_neon_vcvtm_u32_v: | |||
| 6875 | case NEON::BI__builtin_neon_vcvtm_u64_v: | |||
| 6876 | case NEON::BI__builtin_neon_vcvtmq_s16_v: | |||
| 6877 | case NEON::BI__builtin_neon_vcvtmq_s32_v: | |||
| 6878 | case NEON::BI__builtin_neon_vcvtmq_s64_v: | |||
| 6879 | case NEON::BI__builtin_neon_vcvtmq_u16_v: | |||
| 6880 | case NEON::BI__builtin_neon_vcvtmq_u32_v: | |||
| 6881 | case NEON::BI__builtin_neon_vcvtmq_u64_v: { | |||
| 6882 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
| 6883 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint); | |||
| 6884 | } | |||
| 6885 | case NEON::BI__builtin_neon_vcvtx_f32_v: { | |||
| 6886 | llvm::Type *Tys[2] = { VTy->getTruncatedElementVectorType(VTy), Ty}; | |||
| 6887 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint); | |||
| 6888 | ||||
| 6889 | } | |||
| 6890 | case NEON::BI__builtin_neon_vext_v: | |||
| 6891 | case NEON::BI__builtin_neon_vextq_v: { | |||
| 6892 | int CV = cast<ConstantInt>(Ops[2])->getSExtValue(); | |||
| 6893 | SmallVector<int, 16> Indices; | |||
| 6894 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) | |||
| 6895 | Indices.push_back(i+CV); | |||
| 6896 | ||||
| 6897 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 6898 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 6899 | return Builder.CreateShuffleVector(Ops[0], Ops[1], Indices, "vext"); | |||
| 6900 | } | |||
| 6901 | case NEON::BI__builtin_neon_vfma_v: | |||
| 6902 | case NEON::BI__builtin_neon_vfmaq_v: { | |||
| 6903 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 6904 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 6905 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 6906 | ||||
| 6907 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. | |||
| 6908 | return emitCallMaybeConstrainedFPBuiltin( | |||
| 6909 | *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty, | |||
| 6910 | {Ops[1], Ops[2], Ops[0]}); | |||
| 6911 | } | |||
| 6912 | case NEON::BI__builtin_neon_vld1_v: | |||
| 6913 | case NEON::BI__builtin_neon_vld1q_v: { | |||
| 6914 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
| 6915 | Ops.push_back(getAlignmentValue32(PtrOp0)); | |||
| 6916 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "vld1"); | |||
| 6917 | } | |||
| 6918 | case NEON::BI__builtin_neon_vld1_x2_v: | |||
| 6919 | case NEON::BI__builtin_neon_vld1q_x2_v: | |||
| 6920 | case NEON::BI__builtin_neon_vld1_x3_v: | |||
| 6921 | case NEON::BI__builtin_neon_vld1q_x3_v: | |||
| 6922 | case NEON::BI__builtin_neon_vld1_x4_v: | |||
| 6923 | case NEON::BI__builtin_neon_vld1q_x4_v: { | |||
| 6924 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 6925 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
| 6926 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 6927 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
| 6928 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld1xN"); | |||
| 6929 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
| 6930 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 6931 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 6932 | } | |||
| 6933 | case NEON::BI__builtin_neon_vld2_v: | |||
| 6934 | case NEON::BI__builtin_neon_vld2q_v: | |||
| 6935 | case NEON::BI__builtin_neon_vld3_v: | |||
| 6936 | case NEON::BI__builtin_neon_vld3q_v: | |||
| 6937 | case NEON::BI__builtin_neon_vld4_v: | |||
| 6938 | case NEON::BI__builtin_neon_vld4q_v: | |||
| 6939 | case NEON::BI__builtin_neon_vld2_dup_v: | |||
| 6940 | case NEON::BI__builtin_neon_vld2q_dup_v: | |||
| 6941 | case NEON::BI__builtin_neon_vld3_dup_v: | |||
| 6942 | case NEON::BI__builtin_neon_vld3q_dup_v: | |||
| 6943 | case NEON::BI__builtin_neon_vld4_dup_v: | |||
| 6944 | case NEON::BI__builtin_neon_vld4q_dup_v: { | |||
| 6945 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
| 6946 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
| 6947 | Value *Align = getAlignmentValue32(PtrOp1); | |||
| 6948 | Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint); | |||
| 6949 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
| 6950 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 6951 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 6952 | } | |||
| 6953 | case NEON::BI__builtin_neon_vld1_dup_v: | |||
| 6954 | case NEON::BI__builtin_neon_vld1q_dup_v: { | |||
| 6955 | Value *V = UndefValue::get(Ty); | |||
| 6956 | PtrOp0 = Builder.CreateElementBitCast(PtrOp0, VTy->getElementType()); | |||
| 6957 | LoadInst *Ld = Builder.CreateLoad(PtrOp0); | |||
| 6958 | llvm::Constant *CI = ConstantInt::get(SizeTy, 0); | |||
| 6959 | Ops[0] = Builder.CreateInsertElement(V, Ld, CI); | |||
| 6960 | return EmitNeonSplat(Ops[0], CI); | |||
| 6961 | } | |||
| 6962 | case NEON::BI__builtin_neon_vld2_lane_v: | |||
| 6963 | case NEON::BI__builtin_neon_vld2q_lane_v: | |||
| 6964 | case NEON::BI__builtin_neon_vld3_lane_v: | |||
| 6965 | case NEON::BI__builtin_neon_vld3q_lane_v: | |||
| 6966 | case NEON::BI__builtin_neon_vld4_lane_v: | |||
| 6967 | case NEON::BI__builtin_neon_vld4q_lane_v: { | |||
| 6968 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
| 6969 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); | |||
| 6970 | for (unsigned I = 2; I < Ops.size() - 1; ++I) | |||
| 6971 | Ops[I] = Builder.CreateBitCast(Ops[I], Ty); | |||
| 6972 | Ops.push_back(getAlignmentValue32(PtrOp1)); | |||
| 6973 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), NameHint); | |||
| 6974 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
| 6975 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 6976 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 6977 | } | |||
| 6978 | case NEON::BI__builtin_neon_vmovl_v: { | |||
| 6979 | llvm::FixedVectorType *DTy = | |||
| 6980 | llvm::FixedVectorType::getTruncatedElementVectorType(VTy); | |||
| 6981 | Ops[0] = Builder.CreateBitCast(Ops[0], DTy); | |||
| 6982 | if (Usgn) | |||
| 6983 | return Builder.CreateZExt(Ops[0], Ty, "vmovl"); | |||
| 6984 | return Builder.CreateSExt(Ops[0], Ty, "vmovl"); | |||
| 6985 | } | |||
| 6986 | case NEON::BI__builtin_neon_vmovn_v: { | |||
| 6987 | llvm::FixedVectorType *QTy = | |||
| 6988 | llvm::FixedVectorType::getExtendedElementVectorType(VTy); | |||
| 6989 | Ops[0] = Builder.CreateBitCast(Ops[0], QTy); | |||
| 6990 | return Builder.CreateTrunc(Ops[0], Ty, "vmovn"); | |||
| 6991 | } | |||
| 6992 | case NEON::BI__builtin_neon_vmull_v: | |||
| 6993 | // FIXME: the integer vmull operations could be emitted in terms of pure | |||
| 6994 | // LLVM IR (2 exts followed by a mul). Unfortunately LLVM has a habit of | |||
| 6995 | // hoisting the exts outside loops. Until global ISel comes along that can | |||
| 6996 | // see through such movement this leads to bad CodeGen. So we need an | |||
| 6997 | // intrinsic for now. | |||
| 6998 | Int = Usgn ? Intrinsic::arm_neon_vmullu : Intrinsic::arm_neon_vmulls; | |||
| 6999 | Int = Type.isPoly() ? (unsigned)Intrinsic::arm_neon_vmullp : Int; | |||
| 7000 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull"); | |||
| 7001 | case NEON::BI__builtin_neon_vpadal_v: | |||
| 7002 | case NEON::BI__builtin_neon_vpadalq_v: { | |||
| 7003 | // The source operand type has twice as many elements of half the size. | |||
| 7004 | unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits(); | |||
| 7005 | llvm::Type *EltTy = | |||
| 7006 | llvm::IntegerType::get(getLLVMContext(), EltBits / 2); | |||
| 7007 | auto *NarrowTy = | |||
| 7008 | llvm::FixedVectorType::get(EltTy, VTy->getNumElements() * 2); | |||
| 7009 | llvm::Type *Tys[2] = { Ty, NarrowTy }; | |||
| 7010 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint); | |||
| 7011 | } | |||
| 7012 | case NEON::BI__builtin_neon_vpaddl_v: | |||
| 7013 | case NEON::BI__builtin_neon_vpaddlq_v: { | |||
| 7014 | // The source operand type has twice as many elements of half the size. | |||
| 7015 | unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits(); | |||
| 7016 | llvm::Type *EltTy = llvm::IntegerType::get(getLLVMContext(), EltBits / 2); | |||
| 7017 | auto *NarrowTy = | |||
| 7018 | llvm::FixedVectorType::get(EltTy, VTy->getNumElements() * 2); | |||
| 7019 | llvm::Type *Tys[2] = { Ty, NarrowTy }; | |||
| 7020 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vpaddl"); | |||
| 7021 | } | |||
| 7022 | case NEON::BI__builtin_neon_vqdmlal_v: | |||
| 7023 | case NEON::BI__builtin_neon_vqdmlsl_v: { | |||
| 7024 | SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end()); | |||
| 7025 | Ops[1] = | |||
| 7026 | EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), MulOps, "vqdmlal"); | |||
| 7027 | Ops.resize(2); | |||
| 7028 | return EmitNeonCall(CGM.getIntrinsic(AltLLVMIntrinsic, Ty), Ops, NameHint); | |||
| 7029 | } | |||
| 7030 | case NEON::BI__builtin_neon_vqdmulhq_lane_v: | |||
| 7031 | case NEON::BI__builtin_neon_vqdmulh_lane_v: | |||
| 7032 | case NEON::BI__builtin_neon_vqrdmulhq_lane_v: | |||
| 7033 | case NEON::BI__builtin_neon_vqrdmulh_lane_v: { | |||
| 7034 | auto *RTy = cast<llvm::FixedVectorType>(Ty); | |||
| 7035 | if (BuiltinID == NEON::BI__builtin_neon_vqdmulhq_lane_v || | |||
| 7036 | BuiltinID == NEON::BI__builtin_neon_vqrdmulhq_lane_v) | |||
| 7037 | RTy = llvm::FixedVectorType::get(RTy->getElementType(), | |||
| 7038 | RTy->getNumElements() * 2); | |||
| 7039 | llvm::Type *Tys[2] = { | |||
| 7040 | RTy, GetNeonType(this, NeonTypeFlags(Type.getEltType(), false, | |||
| 7041 | /*isQuad*/ false))}; | |||
| 7042 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint); | |||
| 7043 | } | |||
| 7044 | case NEON::BI__builtin_neon_vqdmulhq_laneq_v: | |||
| 7045 | case NEON::BI__builtin_neon_vqdmulh_laneq_v: | |||
| 7046 | case NEON::BI__builtin_neon_vqrdmulhq_laneq_v: | |||
| 7047 | case NEON::BI__builtin_neon_vqrdmulh_laneq_v: { | |||
| 7048 | llvm::Type *Tys[2] = { | |||
| 7049 | Ty, GetNeonType(this, NeonTypeFlags(Type.getEltType(), false, | |||
| 7050 | /*isQuad*/ true))}; | |||
| 7051 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint); | |||
| 7052 | } | |||
| 7053 | case NEON::BI__builtin_neon_vqshl_n_v: | |||
| 7054 | case NEON::BI__builtin_neon_vqshlq_n_v: | |||
| 7055 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshl_n", | |||
| 7056 | 1, false); | |||
| 7057 | case NEON::BI__builtin_neon_vqshlu_n_v: | |||
| 7058 | case NEON::BI__builtin_neon_vqshluq_n_v: | |||
| 7059 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshlu_n", | |||
| 7060 | 1, false); | |||
| 7061 | case NEON::BI__builtin_neon_vrecpe_v: | |||
| 7062 | case NEON::BI__builtin_neon_vrecpeq_v: | |||
| 7063 | case NEON::BI__builtin_neon_vrsqrte_v: | |||
| 7064 | case NEON::BI__builtin_neon_vrsqrteq_v: | |||
| 7065 | Int = Ty->isFPOrFPVectorTy() ? LLVMIntrinsic : AltLLVMIntrinsic; | |||
| 7066 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint); | |||
| 7067 | case NEON::BI__builtin_neon_vrndi_v: | |||
| 7068 | case NEON::BI__builtin_neon_vrndiq_v: | |||
| 7069 | Int = Builder.getIsFPConstrained() | |||
| 7070 | ? Intrinsic::experimental_constrained_nearbyint | |||
| 7071 | : Intrinsic::nearbyint; | |||
| 7072 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint); | |||
| 7073 | case NEON::BI__builtin_neon_vrshr_n_v: | |||
| 7074 | case NEON::BI__builtin_neon_vrshrq_n_v: | |||
| 7075 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshr_n", | |||
| 7076 | 1, true); | |||
| 7077 | case NEON::BI__builtin_neon_vsha512hq_v: | |||
| 7078 | case NEON::BI__builtin_neon_vsha512h2q_v: | |||
| 7079 | case NEON::BI__builtin_neon_vsha512su0q_v: | |||
| 7080 | case NEON::BI__builtin_neon_vsha512su1q_v: { | |||
| 7081 | Function *F = CGM.getIntrinsic(Int); | |||
| 7082 | return EmitNeonCall(F, Ops, ""); | |||
| 7083 | } | |||
| 7084 | case NEON::BI__builtin_neon_vshl_n_v: | |||
| 7085 | case NEON::BI__builtin_neon_vshlq_n_v: | |||
| 7086 | Ops[1] = EmitNeonShiftVector(Ops[1], Ty, false); | |||
| 7087 | return Builder.CreateShl(Builder.CreateBitCast(Ops[0],Ty), Ops[1], | |||
| 7088 | "vshl_n"); | |||
| 7089 | case NEON::BI__builtin_neon_vshll_n_v: { | |||
| 7090 | llvm::FixedVectorType *SrcTy = | |||
| 7091 | llvm::FixedVectorType::getTruncatedElementVectorType(VTy); | |||
| 7092 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
| 7093 | if (Usgn) | |||
| 7094 | Ops[0] = Builder.CreateZExt(Ops[0], VTy); | |||
| 7095 | else | |||
| 7096 | Ops[0] = Builder.CreateSExt(Ops[0], VTy); | |||
| 7097 | Ops[1] = EmitNeonShiftVector(Ops[1], VTy, false); | |||
| 7098 | return Builder.CreateShl(Ops[0], Ops[1], "vshll_n"); | |||
| 7099 | } | |||
| 7100 | case NEON::BI__builtin_neon_vshrn_n_v: { | |||
| 7101 | llvm::FixedVectorType *SrcTy = | |||
| 7102 | llvm::FixedVectorType::getExtendedElementVectorType(VTy); | |||
| 7103 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
| 7104 | Ops[1] = EmitNeonShiftVector(Ops[1], SrcTy, false); | |||
| 7105 | if (Usgn) | |||
| 7106 | Ops[0] = Builder.CreateLShr(Ops[0], Ops[1]); | |||
| 7107 | else | |||
| 7108 | Ops[0] = Builder.CreateAShr(Ops[0], Ops[1]); | |||
| 7109 | return Builder.CreateTrunc(Ops[0], Ty, "vshrn_n"); | |||
| 7110 | } | |||
| 7111 | case NEON::BI__builtin_neon_vshr_n_v: | |||
| 7112 | case NEON::BI__builtin_neon_vshrq_n_v: | |||
| 7113 | return EmitNeonRShiftImm(Ops[0], Ops[1], Ty, Usgn, "vshr_n"); | |||
| 7114 | case NEON::BI__builtin_neon_vst1_v: | |||
| 7115 | case NEON::BI__builtin_neon_vst1q_v: | |||
| 7116 | case NEON::BI__builtin_neon_vst2_v: | |||
| 7117 | case NEON::BI__builtin_neon_vst2q_v: | |||
| 7118 | case NEON::BI__builtin_neon_vst3_v: | |||
| 7119 | case NEON::BI__builtin_neon_vst3q_v: | |||
| 7120 | case NEON::BI__builtin_neon_vst4_v: | |||
| 7121 | case NEON::BI__builtin_neon_vst4q_v: | |||
| 7122 | case NEON::BI__builtin_neon_vst2_lane_v: | |||
| 7123 | case NEON::BI__builtin_neon_vst2q_lane_v: | |||
| 7124 | case NEON::BI__builtin_neon_vst3_lane_v: | |||
| 7125 | case NEON::BI__builtin_neon_vst3q_lane_v: | |||
| 7126 | case NEON::BI__builtin_neon_vst4_lane_v: | |||
| 7127 | case NEON::BI__builtin_neon_vst4q_lane_v: { | |||
| 7128 | llvm::Type *Tys[] = {Int8PtrTy, Ty}; | |||
| 7129 | Ops.push_back(getAlignmentValue32(PtrOp0)); | |||
| 7130 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, ""); | |||
| 7131 | } | |||
| 7132 | case NEON::BI__builtin_neon_vsm3partw1q_v: | |||
| 7133 | case NEON::BI__builtin_neon_vsm3partw2q_v: | |||
| 7134 | case NEON::BI__builtin_neon_vsm3ss1q_v: | |||
| 7135 | case NEON::BI__builtin_neon_vsm4ekeyq_v: | |||
| 7136 | case NEON::BI__builtin_neon_vsm4eq_v: { | |||
| 7137 | Function *F = CGM.getIntrinsic(Int); | |||
| 7138 | return EmitNeonCall(F, Ops, ""); | |||
| 7139 | } | |||
| 7140 | case NEON::BI__builtin_neon_vsm3tt1aq_v: | |||
| 7141 | case NEON::BI__builtin_neon_vsm3tt1bq_v: | |||
| 7142 | case NEON::BI__builtin_neon_vsm3tt2aq_v: | |||
| 7143 | case NEON::BI__builtin_neon_vsm3tt2bq_v: { | |||
| 7144 | Function *F = CGM.getIntrinsic(Int); | |||
| 7145 | Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty); | |||
| 7146 | return EmitNeonCall(F, Ops, ""); | |||
| 7147 | } | |||
| 7148 | case NEON::BI__builtin_neon_vst1_x2_v: | |||
| 7149 | case NEON::BI__builtin_neon_vst1q_x2_v: | |||
| 7150 | case NEON::BI__builtin_neon_vst1_x3_v: | |||
| 7151 | case NEON::BI__builtin_neon_vst1q_x3_v: | |||
| 7152 | case NEON::BI__builtin_neon_vst1_x4_v: | |||
| 7153 | case NEON::BI__builtin_neon_vst1q_x4_v: { | |||
| 7154 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 7155 | // TODO: Currently in AArch32 mode the pointer operand comes first, whereas | |||
| 7156 | // in AArch64 it comes last. We may want to stick to one or another. | |||
| 7157 | if (Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_be || | |||
| 7158 | Arch == llvm::Triple::aarch64_32) { | |||
| 7159 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 7160 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
| 7161 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, ""); | |||
| 7162 | } | |||
| 7163 | llvm::Type *Tys[2] = { PTy, VTy }; | |||
| 7164 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, ""); | |||
| 7165 | } | |||
| 7166 | case NEON::BI__builtin_neon_vsubhn_v: { | |||
| 7167 | llvm::FixedVectorType *SrcTy = | |||
| 7168 | llvm::FixedVectorType::getExtendedElementVectorType(VTy); | |||
| 7169 | ||||
| 7170 | // %sum = add <4 x i32> %lhs, %rhs | |||
| 7171 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); | |||
| 7172 | Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy); | |||
| 7173 | Ops[0] = Builder.CreateSub(Ops[0], Ops[1], "vsubhn"); | |||
| 7174 | ||||
| 7175 | // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16> | |||
| 7176 | Constant *ShiftAmt = | |||
| 7177 | ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2); | |||
| 7178 | Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vsubhn"); | |||
| 7179 | ||||
| 7180 | // %res = trunc <4 x i32> %high to <4 x i16> | |||
| 7181 | return Builder.CreateTrunc(Ops[0], VTy, "vsubhn"); | |||
| 7182 | } | |||
| 7183 | case NEON::BI__builtin_neon_vtrn_v: | |||
| 7184 | case NEON::BI__builtin_neon_vtrnq_v: { | |||
| 7185 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
| 7186 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 7187 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 7188 | Value *SV = nullptr; | |||
| 7189 | ||||
| 7190 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
| 7191 | SmallVector<int, 16> Indices; | |||
| 7192 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
| 7193 | Indices.push_back(i+vi); | |||
| 7194 | Indices.push_back(i+e+vi); | |||
| 7195 | } | |||
| 7196 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
| 7197 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn"); | |||
| 7198 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
| 7199 | } | |||
| 7200 | return SV; | |||
| 7201 | } | |||
| 7202 | case NEON::BI__builtin_neon_vtst_v: | |||
| 7203 | case NEON::BI__builtin_neon_vtstq_v: { | |||
| 7204 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 7205 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 7206 | Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]); | |||
| 7207 | Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0], | |||
| 7208 | ConstantAggregateZero::get(Ty)); | |||
| 7209 | return Builder.CreateSExt(Ops[0], Ty, "vtst"); | |||
| 7210 | } | |||
| 7211 | case NEON::BI__builtin_neon_vuzp_v: | |||
| 7212 | case NEON::BI__builtin_neon_vuzpq_v: { | |||
| 7213 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
| 7214 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 7215 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 7216 | Value *SV = nullptr; | |||
| 7217 | ||||
| 7218 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
| 7219 | SmallVector<int, 16> Indices; | |||
| 7220 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) | |||
| 7221 | Indices.push_back(2*i+vi); | |||
| 7222 | ||||
| 7223 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
| 7224 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp"); | |||
| 7225 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
| 7226 | } | |||
| 7227 | return SV; | |||
| 7228 | } | |||
| 7229 | case NEON::BI__builtin_neon_vxarq_v: { | |||
| 7230 | Function *F = CGM.getIntrinsic(Int); | |||
| 7231 | Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); | |||
| 7232 | return EmitNeonCall(F, Ops, ""); | |||
| 7233 | } | |||
| 7234 | case NEON::BI__builtin_neon_vzip_v: | |||
| 7235 | case NEON::BI__builtin_neon_vzipq_v: { | |||
| 7236 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
| 7237 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 7238 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 7239 | Value *SV = nullptr; | |||
| 7240 | ||||
| 7241 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
| 7242 | SmallVector<int, 16> Indices; | |||
| 7243 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
| 7244 | Indices.push_back((i + vi*e) >> 1); | |||
| 7245 | Indices.push_back(((i + vi*e) >> 1)+e); | |||
| 7246 | } | |||
| 7247 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
| 7248 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip"); | |||
| 7249 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
| 7250 | } | |||
| 7251 | return SV; | |||
| 7252 | } | |||
| 7253 | case NEON::BI__builtin_neon_vdot_v: | |||
| 7254 | case NEON::BI__builtin_neon_vdotq_v: { | |||
| 7255 | auto *InputTy = | |||
| 7256 | llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8); | |||
| 7257 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7258 | Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic; | |||
| 7259 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vdot"); | |||
| 7260 | } | |||
| 7261 | case NEON::BI__builtin_neon_vfmlal_low_v: | |||
| 7262 | case NEON::BI__builtin_neon_vfmlalq_low_v: { | |||
| 7263 | auto *InputTy = | |||
| 7264 | llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
| 7265 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7266 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_low"); | |||
| 7267 | } | |||
| 7268 | case NEON::BI__builtin_neon_vfmlsl_low_v: | |||
| 7269 | case NEON::BI__builtin_neon_vfmlslq_low_v: { | |||
| 7270 | auto *InputTy = | |||
| 7271 | llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
| 7272 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7273 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_low"); | |||
| 7274 | } | |||
| 7275 | case NEON::BI__builtin_neon_vfmlal_high_v: | |||
| 7276 | case NEON::BI__builtin_neon_vfmlalq_high_v: { | |||
| 7277 | auto *InputTy = | |||
| 7278 | llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
| 7279 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7280 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_high"); | |||
| 7281 | } | |||
| 7282 | case NEON::BI__builtin_neon_vfmlsl_high_v: | |||
| 7283 | case NEON::BI__builtin_neon_vfmlslq_high_v: { | |||
| 7284 | auto *InputTy = | |||
| 7285 | llvm::FixedVectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); | |||
| 7286 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7287 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_high"); | |||
| 7288 | } | |||
| 7289 | case NEON::BI__builtin_neon_vmmlaq_v: { | |||
| 7290 | auto *InputTy = | |||
| 7291 | llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8); | |||
| 7292 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7293 | Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic; | |||
| 7294 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmmla"); | |||
| 7295 | } | |||
| 7296 | case NEON::BI__builtin_neon_vusmmlaq_v: { | |||
| 7297 | auto *InputTy = | |||
| 7298 | llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8); | |||
| 7299 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7300 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vusmmla"); | |||
| 7301 | } | |||
| 7302 | case NEON::BI__builtin_neon_vusdot_v: | |||
| 7303 | case NEON::BI__builtin_neon_vusdotq_v: { | |||
| 7304 | auto *InputTy = | |||
| 7305 | llvm::FixedVectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8); | |||
| 7306 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7307 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vusdot"); | |||
| 7308 | } | |||
| 7309 | case NEON::BI__builtin_neon_vbfdot_v: | |||
| 7310 | case NEON::BI__builtin_neon_vbfdotq_v: { | |||
| 7311 | llvm::Type *InputTy = | |||
| 7312 | llvm::FixedVectorType::get(BFloatTy, Ty->getPrimitiveSizeInBits() / 16); | |||
| 7313 | llvm::Type *Tys[2] = { Ty, InputTy }; | |||
| 7314 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vbfdot"); | |||
| 7315 | } | |||
| 7316 | case NEON::BI__builtin_neon___a32_vcvt_bf16_v: { | |||
| 7317 | llvm::Type *Tys[1] = { Ty }; | |||
| 7318 | Function *F = CGM.getIntrinsic(Int, Tys); | |||
| 7319 | return EmitNeonCall(F, Ops, "vcvtfp2bf"); | |||
| 7320 | } | |||
| 7321 | ||||
| 7322 | } | |||
| 7323 | ||||
| 7324 | assert(Int && "Expected valid intrinsic number")(static_cast <bool> (Int && "Expected valid intrinsic number" ) ? void (0) : __assert_fail ("Int && \"Expected valid intrinsic number\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 7324, __extension__ __PRETTY_FUNCTION__ )); | |||
| 7325 | ||||
| 7326 | // Determine the type(s) of this overloaded AArch64 intrinsic. | |||
| 7327 | Function *F = LookupNeonLLVMIntrinsic(Int, Modifier, Ty, E); | |||
| 7328 | ||||
| 7329 | Value *Result = EmitNeonCall(F, Ops, NameHint); | |||
| 7330 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 7331 | // AArch64 intrinsic one-element vector type cast to | |||
| 7332 | // scalar type expected by the builtin | |||
| 7333 | return Builder.CreateBitCast(Result, ResultType, NameHint); | |||
| 7334 | } | |||
| 7335 | ||||
| 7336 | Value *CodeGenFunction::EmitAArch64CompareBuiltinExpr( | |||
| 7337 | Value *Op, llvm::Type *Ty, const CmpInst::Predicate Fp, | |||
| 7338 | const CmpInst::Predicate Ip, const Twine &Name) { | |||
| 7339 | llvm::Type *OTy = Op->getType(); | |||
| 7340 | ||||
| 7341 | // FIXME: this is utterly horrific. We should not be looking at previous | |||
| 7342 | // codegen context to find out what needs doing. Unfortunately TableGen | |||
| 7343 | // currently gives us exactly the same calls for vceqz_f32 and vceqz_s32 | |||
| 7344 | // (etc). | |||
| 7345 | if (BitCastInst *BI = dyn_cast<BitCastInst>(Op)) | |||
| 7346 | OTy = BI->getOperand(0)->getType(); | |||
| 7347 | ||||
| 7348 | Op = Builder.CreateBitCast(Op, OTy); | |||
| 7349 | if (OTy->getScalarType()->isFloatingPointTy()) { | |||
| 7350 | if (Fp == CmpInst::FCMP_OEQ) | |||
| 7351 | Op = Builder.CreateFCmp(Fp, Op, Constant::getNullValue(OTy)); | |||
| 7352 | else | |||
| 7353 | Op = Builder.CreateFCmpS(Fp, Op, Constant::getNullValue(OTy)); | |||
| 7354 | } else { | |||
| 7355 | Op = Builder.CreateICmp(Ip, Op, Constant::getNullValue(OTy)); | |||
| 7356 | } | |||
| 7357 | return Builder.CreateSExt(Op, Ty, Name); | |||
| 7358 | } | |||
| 7359 | ||||
| 7360 | static Value *packTBLDVectorList(CodeGenFunction &CGF, ArrayRef<Value *> Ops, | |||
| 7361 | Value *ExtOp, Value *IndexOp, | |||
| 7362 | llvm::Type *ResTy, unsigned IntID, | |||
| 7363 | const char *Name) { | |||
| 7364 | SmallVector<Value *, 2> TblOps; | |||
| 7365 | if (ExtOp) | |||
| 7366 | TblOps.push_back(ExtOp); | |||
| 7367 | ||||
| 7368 | // Build a vector containing sequential number like (0, 1, 2, ..., 15) | |||
| 7369 | SmallVector<int, 16> Indices; | |||
| 7370 | auto *TblTy = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 7371 | for (unsigned i = 0, e = TblTy->getNumElements(); i != e; ++i) { | |||
| 7372 | Indices.push_back(2*i); | |||
| 7373 | Indices.push_back(2*i+1); | |||
| 7374 | } | |||
| 7375 | ||||
| 7376 | int PairPos = 0, End = Ops.size() - 1; | |||
| 7377 | while (PairPos < End) { | |||
| 7378 | TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos], | |||
| 7379 | Ops[PairPos+1], Indices, | |||
| 7380 | Name)); | |||
| 7381 | PairPos += 2; | |||
| 7382 | } | |||
| 7383 | ||||
| 7384 | // If there's an odd number of 64-bit lookup table, fill the high 64-bit | |||
| 7385 | // of the 128-bit lookup table with zero. | |||
| 7386 | if (PairPos == End) { | |||
| 7387 | Value *ZeroTbl = ConstantAggregateZero::get(TblTy); | |||
| 7388 | TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos], | |||
| 7389 | ZeroTbl, Indices, Name)); | |||
| 7390 | } | |||
| 7391 | ||||
| 7392 | Function *TblF; | |||
| 7393 | TblOps.push_back(IndexOp); | |||
| 7394 | TblF = CGF.CGM.getIntrinsic(IntID, ResTy); | |||
| 7395 | ||||
| 7396 | return CGF.EmitNeonCall(TblF, TblOps, Name); | |||
| 7397 | } | |||
| 7398 | ||||
| 7399 | Value *CodeGenFunction::GetValueForARMHint(unsigned BuiltinID) { | |||
| 7400 | unsigned Value; | |||
| 7401 | switch (BuiltinID) { | |||
| 7402 | default: | |||
| 7403 | return nullptr; | |||
| 7404 | case ARM::BI__builtin_arm_nop: | |||
| 7405 | Value = 0; | |||
| 7406 | break; | |||
| 7407 | case ARM::BI__builtin_arm_yield: | |||
| 7408 | case ARM::BI__yield: | |||
| 7409 | Value = 1; | |||
| 7410 | break; | |||
| 7411 | case ARM::BI__builtin_arm_wfe: | |||
| 7412 | case ARM::BI__wfe: | |||
| 7413 | Value = 2; | |||
| 7414 | break; | |||
| 7415 | case ARM::BI__builtin_arm_wfi: | |||
| 7416 | case ARM::BI__wfi: | |||
| 7417 | Value = 3; | |||
| 7418 | break; | |||
| 7419 | case ARM::BI__builtin_arm_sev: | |||
| 7420 | case ARM::BI__sev: | |||
| 7421 | Value = 4; | |||
| 7422 | break; | |||
| 7423 | case ARM::BI__builtin_arm_sevl: | |||
| 7424 | case ARM::BI__sevl: | |||
| 7425 | Value = 5; | |||
| 7426 | break; | |||
| 7427 | } | |||
| 7428 | ||||
| 7429 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_hint), | |||
| 7430 | llvm::ConstantInt::get(Int32Ty, Value)); | |||
| 7431 | } | |||
| 7432 | ||||
| 7433 | enum SpecialRegisterAccessKind { | |||
| 7434 | NormalRead, | |||
| 7435 | VolatileRead, | |||
| 7436 | Write, | |||
| 7437 | }; | |||
| 7438 | ||||
| 7439 | // Generates the IR for the read/write special register builtin, | |||
| 7440 | // ValueType is the type of the value that is to be written or read, | |||
| 7441 | // RegisterType is the type of the register being written to or read from. | |||
| 7442 | static Value *EmitSpecialRegisterBuiltin(CodeGenFunction &CGF, | |||
| 7443 | const CallExpr *E, | |||
| 7444 | llvm::Type *RegisterType, | |||
| 7445 | llvm::Type *ValueType, | |||
| 7446 | SpecialRegisterAccessKind AccessKind, | |||
| 7447 | StringRef SysReg = "") { | |||
| 7448 | // write and register intrinsics only support 32 and 64 bit operations. | |||
| 7449 | assert((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64))(static_cast <bool> ((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && "Unsupported size for register." ) ? void (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 7450, __extension__ __PRETTY_FUNCTION__ )) | |||
| 7450 | && "Unsupported size for register.")(static_cast <bool> ((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && "Unsupported size for register." ) ? void (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 7450, __extension__ __PRETTY_FUNCTION__ )); | |||
| 7451 | ||||
| 7452 | CodeGen::CGBuilderTy &Builder = CGF.Builder; | |||
| 7453 | CodeGen::CodeGenModule &CGM = CGF.CGM; | |||
| 7454 | LLVMContext &Context = CGM.getLLVMContext(); | |||
| 7455 | ||||
| 7456 | if (SysReg.empty()) { | |||
| 7457 | const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts(); | |||
| 7458 | SysReg = cast<clang::StringLiteral>(SysRegStrExpr)->getString(); | |||
| 7459 | } | |||
| 7460 | ||||
| 7461 | llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysReg) }; | |||
| 7462 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); | |||
| 7463 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); | |||
| 7464 | ||||
| 7465 | llvm::Type *Types[] = { RegisterType }; | |||
| 7466 | ||||
| 7467 | bool MixedTypes = RegisterType->isIntegerTy(64) && ValueType->isIntegerTy(32); | |||
| 7468 | assert(!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64))(static_cast <bool> (!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register" ) ? void (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 7469, __extension__ __PRETTY_FUNCTION__ )) | |||
| 7469 | && "Can't fit 64-bit value in 32-bit register")(static_cast <bool> (!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register" ) ? void (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 7469, __extension__ __PRETTY_FUNCTION__ )); | |||
| 7470 | ||||
| 7471 | if (AccessKind != Write) { | |||
| 7472 | assert(AccessKind == NormalRead || AccessKind == VolatileRead)(static_cast <bool> (AccessKind == NormalRead || AccessKind == VolatileRead) ? void (0) : __assert_fail ("AccessKind == NormalRead || AccessKind == VolatileRead" , "clang/lib/CodeGen/CGBuiltin.cpp", 7472, __extension__ __PRETTY_FUNCTION__ )); | |||
| 7473 | llvm::Function *F = CGM.getIntrinsic( | |||
| 7474 | AccessKind == VolatileRead ? llvm::Intrinsic::read_volatile_register | |||
| 7475 | : llvm::Intrinsic::read_register, | |||
| 7476 | Types); | |||
| 7477 | llvm::Value *Call = Builder.CreateCall(F, Metadata); | |||
| 7478 | ||||
| 7479 | if (MixedTypes) | |||
| 7480 | // Read into 64 bit register and then truncate result to 32 bit. | |||
| 7481 | return Builder.CreateTrunc(Call, ValueType); | |||
| 7482 | ||||
| 7483 | if (ValueType->isPointerTy()) | |||
| 7484 | // Have i32/i64 result (Call) but want to return a VoidPtrTy (i8*). | |||
| 7485 | return Builder.CreateIntToPtr(Call, ValueType); | |||
| 7486 | ||||
| 7487 | return Call; | |||
| 7488 | } | |||
| 7489 | ||||
| 7490 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types); | |||
| 7491 | llvm::Value *ArgValue = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 7492 | if (MixedTypes) { | |||
| 7493 | // Extend 32 bit write value to 64 bit to pass to write. | |||
| 7494 | ArgValue = Builder.CreateZExt(ArgValue, RegisterType); | |||
| 7495 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
| 7496 | } | |||
| 7497 | ||||
| 7498 | if (ValueType->isPointerTy()) { | |||
| 7499 | // Have VoidPtrTy ArgValue but want to return an i32/i64. | |||
| 7500 | ArgValue = Builder.CreatePtrToInt(ArgValue, RegisterType); | |||
| 7501 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
| 7502 | } | |||
| 7503 | ||||
| 7504 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
| 7505 | } | |||
| 7506 | ||||
| 7507 | /// Return true if BuiltinID is an overloaded Neon intrinsic with an extra | |||
| 7508 | /// argument that specifies the vector type. | |||
| 7509 | static bool HasExtraNeonArgument(unsigned BuiltinID) { | |||
| 7510 | switch (BuiltinID) { | |||
| 7511 | default: break; | |||
| 7512 | case NEON::BI__builtin_neon_vget_lane_i8: | |||
| 7513 | case NEON::BI__builtin_neon_vget_lane_i16: | |||
| 7514 | case NEON::BI__builtin_neon_vget_lane_bf16: | |||
| 7515 | case NEON::BI__builtin_neon_vget_lane_i32: | |||
| 7516 | case NEON::BI__builtin_neon_vget_lane_i64: | |||
| 7517 | case NEON::BI__builtin_neon_vget_lane_f32: | |||
| 7518 | case NEON::BI__builtin_neon_vgetq_lane_i8: | |||
| 7519 | case NEON::BI__builtin_neon_vgetq_lane_i16: | |||
| 7520 | case NEON::BI__builtin_neon_vgetq_lane_bf16: | |||
| 7521 | case NEON::BI__builtin_neon_vgetq_lane_i32: | |||
| 7522 | case NEON::BI__builtin_neon_vgetq_lane_i64: | |||
| 7523 | case NEON::BI__builtin_neon_vgetq_lane_f32: | |||
| 7524 | case NEON::BI__builtin_neon_vduph_lane_bf16: | |||
| 7525 | case NEON::BI__builtin_neon_vduph_laneq_bf16: | |||
| 7526 | case NEON::BI__builtin_neon_vset_lane_i8: | |||
| 7527 | case NEON::BI__builtin_neon_vset_lane_i16: | |||
| 7528 | case NEON::BI__builtin_neon_vset_lane_bf16: | |||
| 7529 | case NEON::BI__builtin_neon_vset_lane_i32: | |||
| 7530 | case NEON::BI__builtin_neon_vset_lane_i64: | |||
| 7531 | case NEON::BI__builtin_neon_vset_lane_f32: | |||
| 7532 | case NEON::BI__builtin_neon_vsetq_lane_i8: | |||
| 7533 | case NEON::BI__builtin_neon_vsetq_lane_i16: | |||
| 7534 | case NEON::BI__builtin_neon_vsetq_lane_bf16: | |||
| 7535 | case NEON::BI__builtin_neon_vsetq_lane_i32: | |||
| 7536 | case NEON::BI__builtin_neon_vsetq_lane_i64: | |||
| 7537 | case NEON::BI__builtin_neon_vsetq_lane_f32: | |||
| 7538 | case NEON::BI__builtin_neon_vsha1h_u32: | |||
| 7539 | case NEON::BI__builtin_neon_vsha1cq_u32: | |||
| 7540 | case NEON::BI__builtin_neon_vsha1pq_u32: | |||
| 7541 | case NEON::BI__builtin_neon_vsha1mq_u32: | |||
| 7542 | case NEON::BI__builtin_neon_vcvth_bf16_f32: | |||
| 7543 | case clang::ARM::BI_MoveToCoprocessor: | |||
| 7544 | case clang::ARM::BI_MoveToCoprocessor2: | |||
| 7545 | return false; | |||
| 7546 | } | |||
| 7547 | return true; | |||
| 7548 | } | |||
| 7549 | ||||
| 7550 | Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID, | |||
| 7551 | const CallExpr *E, | |||
| 7552 | ReturnValueSlot ReturnValue, | |||
| 7553 | llvm::Triple::ArchType Arch) { | |||
| 7554 | if (auto Hint = GetValueForARMHint(BuiltinID)) | |||
| 7555 | return Hint; | |||
| 7556 | ||||
| 7557 | if (BuiltinID == ARM::BI__emit) { | |||
| 7558 | bool IsThumb = getTarget().getTriple().getArch() == llvm::Triple::thumb; | |||
| 7559 | llvm::FunctionType *FTy = | |||
| 7560 | llvm::FunctionType::get(VoidTy, /*Variadic=*/false); | |||
| 7561 | ||||
| 7562 | Expr::EvalResult Result; | |||
| 7563 | if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext())) | |||
| 7564 | llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant" , "clang/lib/CodeGen/CGBuiltin.cpp", 7564); | |||
| 7565 | ||||
| 7566 | llvm::APSInt Value = Result.Val.getInt(); | |||
| 7567 | uint64_t ZExtValue = Value.zextOrTrunc(IsThumb ? 16 : 32).getZExtValue(); | |||
| 7568 | ||||
| 7569 | llvm::InlineAsm *Emit = | |||
| 7570 | IsThumb ? InlineAsm::get(FTy, ".inst.n 0x" + utohexstr(ZExtValue), "", | |||
| 7571 | /*hasSideEffects=*/true) | |||
| 7572 | : InlineAsm::get(FTy, ".inst 0x" + utohexstr(ZExtValue), "", | |||
| 7573 | /*hasSideEffects=*/true); | |||
| 7574 | ||||
| 7575 | return Builder.CreateCall(Emit); | |||
| 7576 | } | |||
| 7577 | ||||
| 7578 | if (BuiltinID == ARM::BI__builtin_arm_dbg) { | |||
| 7579 | Value *Option = EmitScalarExpr(E->getArg(0)); | |||
| 7580 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_dbg), Option); | |||
| 7581 | } | |||
| 7582 | ||||
| 7583 | if (BuiltinID == ARM::BI__builtin_arm_prefetch) { | |||
| 7584 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
| 7585 | Value *RW = EmitScalarExpr(E->getArg(1)); | |||
| 7586 | Value *IsData = EmitScalarExpr(E->getArg(2)); | |||
| 7587 | ||||
| 7588 | // Locality is not supported on ARM target | |||
| 7589 | Value *Locality = llvm::ConstantInt::get(Int32Ty, 3); | |||
| 7590 | ||||
| 7591 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
| 7592 | return Builder.CreateCall(F, {Address, RW, Locality, IsData}); | |||
| 7593 | } | |||
| 7594 | ||||
| 7595 | if (BuiltinID == ARM::BI__builtin_arm_rbit) { | |||
| 7596 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 7597 | return Builder.CreateCall( | |||
| 7598 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); | |||
| 7599 | } | |||
| 7600 | ||||
| 7601 | if (BuiltinID == ARM::BI__builtin_arm_cls) { | |||
| 7602 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 7603 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_cls), Arg, "cls"); | |||
| 7604 | } | |||
| 7605 | if (BuiltinID == ARM::BI__builtin_arm_cls64) { | |||
| 7606 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 7607 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_cls64), Arg, | |||
| 7608 | "cls"); | |||
| 7609 | } | |||
| 7610 | ||||
| 7611 | if (BuiltinID == ARM::BI__clear_cache) { | |||
| 7612 | assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")(static_cast <bool> (E->getNumArgs() == 2 && "__clear_cache takes 2 arguments") ? void (0) : __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 7612, __extension__ __PRETTY_FUNCTION__ )); | |||
| 7613 | const FunctionDecl *FD = E->getDirectCallee(); | |||
| 7614 | Value *Ops[2]; | |||
| 7615 | for (unsigned i = 0; i < 2; i++) | |||
| 7616 | Ops[i] = EmitScalarExpr(E->getArg(i)); | |||
| 7617 | llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType()); | |||
| 7618 | llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty); | |||
| 7619 | StringRef Name = FD->getName(); | |||
| 7620 | return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops); | |||
| 7621 | } | |||
| 7622 | ||||
| 7623 | if (BuiltinID == ARM::BI__builtin_arm_mcrr || | |||
| 7624 | BuiltinID == ARM::BI__builtin_arm_mcrr2) { | |||
| 7625 | Function *F; | |||
| 7626 | ||||
| 7627 | switch (BuiltinID) { | |||
| 7628 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 7628); | |||
| 7629 | case ARM::BI__builtin_arm_mcrr: | |||
| 7630 | F = CGM.getIntrinsic(Intrinsic::arm_mcrr); | |||
| 7631 | break; | |||
| 7632 | case ARM::BI__builtin_arm_mcrr2: | |||
| 7633 | F = CGM.getIntrinsic(Intrinsic::arm_mcrr2); | |||
| 7634 | break; | |||
| 7635 | } | |||
| 7636 | ||||
| 7637 | // MCRR{2} instruction has 5 operands but | |||
| 7638 | // the intrinsic has 4 because Rt and Rt2 | |||
| 7639 | // are represented as a single unsigned 64 | |||
| 7640 | // bit integer in the intrinsic definition | |||
| 7641 | // but internally it's represented as 2 32 | |||
| 7642 | // bit integers. | |||
| 7643 | ||||
| 7644 | Value *Coproc = EmitScalarExpr(E->getArg(0)); | |||
| 7645 | Value *Opc1 = EmitScalarExpr(E->getArg(1)); | |||
| 7646 | Value *RtAndRt2 = EmitScalarExpr(E->getArg(2)); | |||
| 7647 | Value *CRm = EmitScalarExpr(E->getArg(3)); | |||
| 7648 | ||||
| 7649 | Value *C1 = llvm::ConstantInt::get(Int64Ty, 32); | |||
| 7650 | Value *Rt = Builder.CreateTruncOrBitCast(RtAndRt2, Int32Ty); | |||
| 7651 | Value *Rt2 = Builder.CreateLShr(RtAndRt2, C1); | |||
| 7652 | Rt2 = Builder.CreateTruncOrBitCast(Rt2, Int32Ty); | |||
| 7653 | ||||
| 7654 | return Builder.CreateCall(F, {Coproc, Opc1, Rt, Rt2, CRm}); | |||
| 7655 | } | |||
| 7656 | ||||
| 7657 | if (BuiltinID == ARM::BI__builtin_arm_mrrc || | |||
| 7658 | BuiltinID == ARM::BI__builtin_arm_mrrc2) { | |||
| 7659 | Function *F; | |||
| 7660 | ||||
| 7661 | switch (BuiltinID) { | |||
| 7662 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 7662); | |||
| 7663 | case ARM::BI__builtin_arm_mrrc: | |||
| 7664 | F = CGM.getIntrinsic(Intrinsic::arm_mrrc); | |||
| 7665 | break; | |||
| 7666 | case ARM::BI__builtin_arm_mrrc2: | |||
| 7667 | F = CGM.getIntrinsic(Intrinsic::arm_mrrc2); | |||
| 7668 | break; | |||
| 7669 | } | |||
| 7670 | ||||
| 7671 | Value *Coproc = EmitScalarExpr(E->getArg(0)); | |||
| 7672 | Value *Opc1 = EmitScalarExpr(E->getArg(1)); | |||
| 7673 | Value *CRm = EmitScalarExpr(E->getArg(2)); | |||
| 7674 | Value *RtAndRt2 = Builder.CreateCall(F, {Coproc, Opc1, CRm}); | |||
| 7675 | ||||
| 7676 | // Returns an unsigned 64 bit integer, represented | |||
| 7677 | // as two 32 bit integers. | |||
| 7678 | ||||
| 7679 | Value *Rt = Builder.CreateExtractValue(RtAndRt2, 1); | |||
| 7680 | Value *Rt1 = Builder.CreateExtractValue(RtAndRt2, 0); | |||
| 7681 | Rt = Builder.CreateZExt(Rt, Int64Ty); | |||
| 7682 | Rt1 = Builder.CreateZExt(Rt1, Int64Ty); | |||
| 7683 | ||||
| 7684 | Value *ShiftCast = llvm::ConstantInt::get(Int64Ty, 32); | |||
| 7685 | RtAndRt2 = Builder.CreateShl(Rt, ShiftCast, "shl", true); | |||
| 7686 | RtAndRt2 = Builder.CreateOr(RtAndRt2, Rt1); | |||
| 7687 | ||||
| 7688 | return Builder.CreateBitCast(RtAndRt2, ConvertType(E->getType())); | |||
| 7689 | } | |||
| 7690 | ||||
| 7691 | if (BuiltinID == ARM::BI__builtin_arm_ldrexd || | |||
| 7692 | ((BuiltinID == ARM::BI__builtin_arm_ldrex || | |||
| 7693 | BuiltinID == ARM::BI__builtin_arm_ldaex) && | |||
| 7694 | getContext().getTypeSize(E->getType()) == 64) || | |||
| 7695 | BuiltinID == ARM::BI__ldrexd) { | |||
| 7696 | Function *F; | |||
| 7697 | ||||
| 7698 | switch (BuiltinID) { | |||
| 7699 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 7699); | |||
| 7700 | case ARM::BI__builtin_arm_ldaex: | |||
| 7701 | F = CGM.getIntrinsic(Intrinsic::arm_ldaexd); | |||
| 7702 | break; | |||
| 7703 | case ARM::BI__builtin_arm_ldrexd: | |||
| 7704 | case ARM::BI__builtin_arm_ldrex: | |||
| 7705 | case ARM::BI__ldrexd: | |||
| 7706 | F = CGM.getIntrinsic(Intrinsic::arm_ldrexd); | |||
| 7707 | break; | |||
| 7708 | } | |||
| 7709 | ||||
| 7710 | Value *LdPtr = EmitScalarExpr(E->getArg(0)); | |||
| 7711 | Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy), | |||
| 7712 | "ldrexd"); | |||
| 7713 | ||||
| 7714 | Value *Val0 = Builder.CreateExtractValue(Val, 1); | |||
| 7715 | Value *Val1 = Builder.CreateExtractValue(Val, 0); | |||
| 7716 | Val0 = Builder.CreateZExt(Val0, Int64Ty); | |||
| 7717 | Val1 = Builder.CreateZExt(Val1, Int64Ty); | |||
| 7718 | ||||
| 7719 | Value *ShiftCst = llvm::ConstantInt::get(Int64Ty, 32); | |||
| 7720 | Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */); | |||
| 7721 | Val = Builder.CreateOr(Val, Val1); | |||
| 7722 | return Builder.CreateBitCast(Val, ConvertType(E->getType())); | |||
| 7723 | } | |||
| 7724 | ||||
| 7725 | if (BuiltinID == ARM::BI__builtin_arm_ldrex || | |||
| 7726 | BuiltinID == ARM::BI__builtin_arm_ldaex) { | |||
| 7727 | Value *LoadAddr = EmitScalarExpr(E->getArg(0)); | |||
| 7728 | ||||
| 7729 | QualType Ty = E->getType(); | |||
| 7730 | llvm::Type *RealResTy = ConvertType(Ty); | |||
| 7731 | llvm::Type *IntTy = | |||
| 7732 | llvm::IntegerType::get(getLLVMContext(), getContext().getTypeSize(Ty)); | |||
| 7733 | llvm::Type *PtrTy = IntTy->getPointerTo(); | |||
| 7734 | LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy); | |||
| 7735 | ||||
| 7736 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_ldaex | |||
| 7737 | ? Intrinsic::arm_ldaex | |||
| 7738 | : Intrinsic::arm_ldrex, | |||
| 7739 | PtrTy); | |||
| 7740 | CallInst *Val = Builder.CreateCall(F, LoadAddr, "ldrex"); | |||
| 7741 | Val->addParamAttr( | |||
| 7742 | 0, Attribute::get(getLLVMContext(), Attribute::ElementType, IntTy)); | |||
| 7743 | ||||
| 7744 | if (RealResTy->isPointerTy()) | |||
| 7745 | return Builder.CreateIntToPtr(Val, RealResTy); | |||
| 7746 | else { | |||
| 7747 | llvm::Type *IntResTy = llvm::IntegerType::get( | |||
| 7748 | getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy)); | |||
| 7749 | return Builder.CreateBitCast(Builder.CreateTruncOrBitCast(Val, IntResTy), | |||
| 7750 | RealResTy); | |||
| 7751 | } | |||
| 7752 | } | |||
| 7753 | ||||
| 7754 | if (BuiltinID == ARM::BI__builtin_arm_strexd || | |||
| 7755 | ((BuiltinID == ARM::BI__builtin_arm_stlex || | |||
| 7756 | BuiltinID == ARM::BI__builtin_arm_strex) && | |||
| 7757 | getContext().getTypeSize(E->getArg(0)->getType()) == 64)) { | |||
| 7758 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex | |||
| 7759 | ? Intrinsic::arm_stlexd | |||
| 7760 | : Intrinsic::arm_strexd); | |||
| 7761 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty); | |||
| 7762 | ||||
| 7763 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); | |||
| 7764 | Value *Val = EmitScalarExpr(E->getArg(0)); | |||
| 7765 | Builder.CreateStore(Val, Tmp); | |||
| 7766 | ||||
| 7767 | Address LdPtr = Builder.CreateElementBitCast(Tmp, STy); | |||
| 7768 | Val = Builder.CreateLoad(LdPtr); | |||
| 7769 | ||||
| 7770 | Value *Arg0 = Builder.CreateExtractValue(Val, 0); | |||
| 7771 | Value *Arg1 = Builder.CreateExtractValue(Val, 1); | |||
| 7772 | Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), Int8PtrTy); | |||
| 7773 | return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "strexd"); | |||
| 7774 | } | |||
| 7775 | ||||
| 7776 | if (BuiltinID == ARM::BI__builtin_arm_strex || | |||
| 7777 | BuiltinID == ARM::BI__builtin_arm_stlex) { | |||
| 7778 | Value *StoreVal = EmitScalarExpr(E->getArg(0)); | |||
| 7779 | Value *StoreAddr = EmitScalarExpr(E->getArg(1)); | |||
| 7780 | ||||
| 7781 | QualType Ty = E->getArg(0)->getType(); | |||
| 7782 | llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(), | |||
| 7783 | getContext().getTypeSize(Ty)); | |||
| 7784 | StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo()); | |||
| 7785 | ||||
| 7786 | if (StoreVal->getType()->isPointerTy()) | |||
| 7787 | StoreVal = Builder.CreatePtrToInt(StoreVal, Int32Ty); | |||
| 7788 | else { | |||
| 7789 | llvm::Type *IntTy = llvm::IntegerType::get( | |||
| 7790 | getLLVMContext(), | |||
| 7791 | CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType())); | |||
| 7792 | StoreVal = Builder.CreateBitCast(StoreVal, IntTy); | |||
| 7793 | StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int32Ty); | |||
| 7794 | } | |||
| 7795 | ||||
| 7796 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex | |||
| 7797 | ? Intrinsic::arm_stlex | |||
| 7798 | : Intrinsic::arm_strex, | |||
| 7799 | StoreAddr->getType()); | |||
| 7800 | ||||
| 7801 | CallInst *CI = Builder.CreateCall(F, {StoreVal, StoreAddr}, "strex"); | |||
| 7802 | CI->addParamAttr( | |||
| 7803 | 1, Attribute::get(getLLVMContext(), Attribute::ElementType, StoreTy)); | |||
| 7804 | return CI; | |||
| 7805 | } | |||
| 7806 | ||||
| 7807 | if (BuiltinID == ARM::BI__builtin_arm_clrex) { | |||
| 7808 | Function *F = CGM.getIntrinsic(Intrinsic::arm_clrex); | |||
| 7809 | return Builder.CreateCall(F); | |||
| 7810 | } | |||
| 7811 | ||||
| 7812 | // CRC32 | |||
| 7813 | Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic; | |||
| 7814 | switch (BuiltinID) { | |||
| 7815 | case ARM::BI__builtin_arm_crc32b: | |||
| 7816 | CRCIntrinsicID = Intrinsic::arm_crc32b; break; | |||
| 7817 | case ARM::BI__builtin_arm_crc32cb: | |||
| 7818 | CRCIntrinsicID = Intrinsic::arm_crc32cb; break; | |||
| 7819 | case ARM::BI__builtin_arm_crc32h: | |||
| 7820 | CRCIntrinsicID = Intrinsic::arm_crc32h; break; | |||
| 7821 | case ARM::BI__builtin_arm_crc32ch: | |||
| 7822 | CRCIntrinsicID = Intrinsic::arm_crc32ch; break; | |||
| 7823 | case ARM::BI__builtin_arm_crc32w: | |||
| 7824 | case ARM::BI__builtin_arm_crc32d: | |||
| 7825 | CRCIntrinsicID = Intrinsic::arm_crc32w; break; | |||
| 7826 | case ARM::BI__builtin_arm_crc32cw: | |||
| 7827 | case ARM::BI__builtin_arm_crc32cd: | |||
| 7828 | CRCIntrinsicID = Intrinsic::arm_crc32cw; break; | |||
| 7829 | } | |||
| 7830 | ||||
| 7831 | if (CRCIntrinsicID != Intrinsic::not_intrinsic) { | |||
| 7832 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
| 7833 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
| 7834 | ||||
| 7835 | // crc32{c,}d intrinsics are implemnted as two calls to crc32{c,}w | |||
| 7836 | // intrinsics, hence we need different codegen for these cases. | |||
| 7837 | if (BuiltinID == ARM::BI__builtin_arm_crc32d || | |||
| 7838 | BuiltinID == ARM::BI__builtin_arm_crc32cd) { | |||
| 7839 | Value *C1 = llvm::ConstantInt::get(Int64Ty, 32); | |||
| 7840 | Value *Arg1a = Builder.CreateTruncOrBitCast(Arg1, Int32Ty); | |||
| 7841 | Value *Arg1b = Builder.CreateLShr(Arg1, C1); | |||
| 7842 | Arg1b = Builder.CreateTruncOrBitCast(Arg1b, Int32Ty); | |||
| 7843 | ||||
| 7844 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); | |||
| 7845 | Value *Res = Builder.CreateCall(F, {Arg0, Arg1a}); | |||
| 7846 | return Builder.CreateCall(F, {Res, Arg1b}); | |||
| 7847 | } else { | |||
| 7848 | Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty); | |||
| 7849 | ||||
| 7850 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); | |||
| 7851 | return Builder.CreateCall(F, {Arg0, Arg1}); | |||
| 7852 | } | |||
| 7853 | } | |||
| 7854 | ||||
| 7855 | if (BuiltinID == ARM::BI__builtin_arm_rsr || | |||
| 7856 | BuiltinID == ARM::BI__builtin_arm_rsr64 || | |||
| 7857 | BuiltinID == ARM::BI__builtin_arm_rsrp || | |||
| 7858 | BuiltinID == ARM::BI__builtin_arm_wsr || | |||
| 7859 | BuiltinID == ARM::BI__builtin_arm_wsr64 || | |||
| 7860 | BuiltinID == ARM::BI__builtin_arm_wsrp) { | |||
| 7861 | ||||
| 7862 | SpecialRegisterAccessKind AccessKind = Write; | |||
| 7863 | if (BuiltinID == ARM::BI__builtin_arm_rsr || | |||
| 7864 | BuiltinID == ARM::BI__builtin_arm_rsr64 || | |||
| 7865 | BuiltinID == ARM::BI__builtin_arm_rsrp) | |||
| 7866 | AccessKind = VolatileRead; | |||
| 7867 | ||||
| 7868 | bool IsPointerBuiltin = BuiltinID == ARM::BI__builtin_arm_rsrp || | |||
| 7869 | BuiltinID == ARM::BI__builtin_arm_wsrp; | |||
| 7870 | ||||
| 7871 | bool Is64Bit = BuiltinID == ARM::BI__builtin_arm_rsr64 || | |||
| 7872 | BuiltinID == ARM::BI__builtin_arm_wsr64; | |||
| 7873 | ||||
| 7874 | llvm::Type *ValueType; | |||
| 7875 | llvm::Type *RegisterType; | |||
| 7876 | if (IsPointerBuiltin) { | |||
| 7877 | ValueType = VoidPtrTy; | |||
| 7878 | RegisterType = Int32Ty; | |||
| 7879 | } else if (Is64Bit) { | |||
| 7880 | ValueType = RegisterType = Int64Ty; | |||
| 7881 | } else { | |||
| 7882 | ValueType = RegisterType = Int32Ty; | |||
| 7883 | } | |||
| 7884 | ||||
| 7885 | return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, | |||
| 7886 | AccessKind); | |||
| 7887 | } | |||
| 7888 | ||||
| 7889 | // Handle MSVC intrinsics before argument evaluation to prevent double | |||
| 7890 | // evaluation. | |||
| 7891 | if (Optional<MSVCIntrin> MsvcIntId = translateArmToMsvcIntrin(BuiltinID)) | |||
| 7892 | return EmitMSVCBuiltinExpr(*MsvcIntId, E); | |||
| 7893 | ||||
| 7894 | // Deal with MVE builtins | |||
| 7895 | if (Value *Result = EmitARMMVEBuiltinExpr(BuiltinID, E, ReturnValue, Arch)) | |||
| 7896 | return Result; | |||
| 7897 | // Handle CDE builtins | |||
| 7898 | if (Value *Result = EmitARMCDEBuiltinExpr(BuiltinID, E, ReturnValue, Arch)) | |||
| 7899 | return Result; | |||
| 7900 | ||||
| 7901 | // Find out if any arguments are required to be integer constant | |||
| 7902 | // expressions. | |||
| 7903 | unsigned ICEArguments = 0; | |||
| 7904 | ASTContext::GetBuiltinTypeError Error; | |||
| 7905 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
| 7906 | assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None && "Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 7906, __extension__ __PRETTY_FUNCTION__ )); | |||
| 7907 | ||||
| 7908 | auto getAlignmentValue32 = [&](Address addr) -> Value* { | |||
| 7909 | return Builder.getInt32(addr.getAlignment().getQuantity()); | |||
| 7910 | }; | |||
| 7911 | ||||
| 7912 | Address PtrOp0 = Address::invalid(); | |||
| 7913 | Address PtrOp1 = Address::invalid(); | |||
| 7914 | SmallVector<Value*, 4> Ops; | |||
| 7915 | bool HasExtraArg = HasExtraNeonArgument(BuiltinID); | |||
| 7916 | unsigned NumArgs = E->getNumArgs() - (HasExtraArg ? 1 : 0); | |||
| 7917 | for (unsigned i = 0, e = NumArgs; i != e; i++) { | |||
| 7918 | if (i == 0) { | |||
| 7919 | switch (BuiltinID) { | |||
| 7920 | case NEON::BI__builtin_neon_vld1_v: | |||
| 7921 | case NEON::BI__builtin_neon_vld1q_v: | |||
| 7922 | case NEON::BI__builtin_neon_vld1q_lane_v: | |||
| 7923 | case NEON::BI__builtin_neon_vld1_lane_v: | |||
| 7924 | case NEON::BI__builtin_neon_vld1_dup_v: | |||
| 7925 | case NEON::BI__builtin_neon_vld1q_dup_v: | |||
| 7926 | case NEON::BI__builtin_neon_vst1_v: | |||
| 7927 | case NEON::BI__builtin_neon_vst1q_v: | |||
| 7928 | case NEON::BI__builtin_neon_vst1q_lane_v: | |||
| 7929 | case NEON::BI__builtin_neon_vst1_lane_v: | |||
| 7930 | case NEON::BI__builtin_neon_vst2_v: | |||
| 7931 | case NEON::BI__builtin_neon_vst2q_v: | |||
| 7932 | case NEON::BI__builtin_neon_vst2_lane_v: | |||
| 7933 | case NEON::BI__builtin_neon_vst2q_lane_v: | |||
| 7934 | case NEON::BI__builtin_neon_vst3_v: | |||
| 7935 | case NEON::BI__builtin_neon_vst3q_v: | |||
| 7936 | case NEON::BI__builtin_neon_vst3_lane_v: | |||
| 7937 | case NEON::BI__builtin_neon_vst3q_lane_v: | |||
| 7938 | case NEON::BI__builtin_neon_vst4_v: | |||
| 7939 | case NEON::BI__builtin_neon_vst4q_v: | |||
| 7940 | case NEON::BI__builtin_neon_vst4_lane_v: | |||
| 7941 | case NEON::BI__builtin_neon_vst4q_lane_v: | |||
| 7942 | // Get the alignment for the argument in addition to the value; | |||
| 7943 | // we'll use it later. | |||
| 7944 | PtrOp0 = EmitPointerWithAlignment(E->getArg(0)); | |||
| 7945 | Ops.push_back(PtrOp0.getPointer()); | |||
| 7946 | continue; | |||
| 7947 | } | |||
| 7948 | } | |||
| 7949 | if (i == 1) { | |||
| 7950 | switch (BuiltinID) { | |||
| 7951 | case NEON::BI__builtin_neon_vld2_v: | |||
| 7952 | case NEON::BI__builtin_neon_vld2q_v: | |||
| 7953 | case NEON::BI__builtin_neon_vld3_v: | |||
| 7954 | case NEON::BI__builtin_neon_vld3q_v: | |||
| 7955 | case NEON::BI__builtin_neon_vld4_v: | |||
| 7956 | case NEON::BI__builtin_neon_vld4q_v: | |||
| 7957 | case NEON::BI__builtin_neon_vld2_lane_v: | |||
| 7958 | case NEON::BI__builtin_neon_vld2q_lane_v: | |||
| 7959 | case NEON::BI__builtin_neon_vld3_lane_v: | |||
| 7960 | case NEON::BI__builtin_neon_vld3q_lane_v: | |||
| 7961 | case NEON::BI__builtin_neon_vld4_lane_v: | |||
| 7962 | case NEON::BI__builtin_neon_vld4q_lane_v: | |||
| 7963 | case NEON::BI__builtin_neon_vld2_dup_v: | |||
| 7964 | case NEON::BI__builtin_neon_vld2q_dup_v: | |||
| 7965 | case NEON::BI__builtin_neon_vld3_dup_v: | |||
| 7966 | case NEON::BI__builtin_neon_vld3q_dup_v: | |||
| 7967 | case NEON::BI__builtin_neon_vld4_dup_v: | |||
| 7968 | case NEON::BI__builtin_neon_vld4q_dup_v: | |||
| 7969 | // Get the alignment for the argument in addition to the value; | |||
| 7970 | // we'll use it later. | |||
| 7971 | PtrOp1 = EmitPointerWithAlignment(E->getArg(1)); | |||
| 7972 | Ops.push_back(PtrOp1.getPointer()); | |||
| 7973 | continue; | |||
| 7974 | } | |||
| 7975 | } | |||
| 7976 | ||||
| 7977 | if ((ICEArguments & (1 << i)) == 0) { | |||
| 7978 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 7979 | } else { | |||
| 7980 | // If this is required to be a constant, constant fold it so that we know | |||
| 7981 | // that the generated intrinsic gets a ConstantInt. | |||
| 7982 | Ops.push_back(llvm::ConstantInt::get( | |||
| 7983 | getLLVMContext(), | |||
| 7984 | *E->getArg(i)->getIntegerConstantExpr(getContext()))); | |||
| 7985 | } | |||
| 7986 | } | |||
| 7987 | ||||
| 7988 | switch (BuiltinID) { | |||
| 7989 | default: break; | |||
| 7990 | ||||
| 7991 | case NEON::BI__builtin_neon_vget_lane_i8: | |||
| 7992 | case NEON::BI__builtin_neon_vget_lane_i16: | |||
| 7993 | case NEON::BI__builtin_neon_vget_lane_i32: | |||
| 7994 | case NEON::BI__builtin_neon_vget_lane_i64: | |||
| 7995 | case NEON::BI__builtin_neon_vget_lane_bf16: | |||
| 7996 | case NEON::BI__builtin_neon_vget_lane_f32: | |||
| 7997 | case NEON::BI__builtin_neon_vgetq_lane_i8: | |||
| 7998 | case NEON::BI__builtin_neon_vgetq_lane_i16: | |||
| 7999 | case NEON::BI__builtin_neon_vgetq_lane_i32: | |||
| 8000 | case NEON::BI__builtin_neon_vgetq_lane_i64: | |||
| 8001 | case NEON::BI__builtin_neon_vgetq_lane_bf16: | |||
| 8002 | case NEON::BI__builtin_neon_vgetq_lane_f32: | |||
| 8003 | case NEON::BI__builtin_neon_vduph_lane_bf16: | |||
| 8004 | case NEON::BI__builtin_neon_vduph_laneq_bf16: | |||
| 8005 | return Builder.CreateExtractElement(Ops[0], Ops[1], "vget_lane"); | |||
| 8006 | ||||
| 8007 | case NEON::BI__builtin_neon_vrndns_f32: { | |||
| 8008 | Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 8009 | llvm::Type *Tys[] = {Arg->getType()}; | |||
| 8010 | Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vrintn, Tys); | |||
| 8011 | return Builder.CreateCall(F, {Arg}, "vrndn"); } | |||
| 8012 | ||||
| 8013 | case NEON::BI__builtin_neon_vset_lane_i8: | |||
| 8014 | case NEON::BI__builtin_neon_vset_lane_i16: | |||
| 8015 | case NEON::BI__builtin_neon_vset_lane_i32: | |||
| 8016 | case NEON::BI__builtin_neon_vset_lane_i64: | |||
| 8017 | case NEON::BI__builtin_neon_vset_lane_bf16: | |||
| 8018 | case NEON::BI__builtin_neon_vset_lane_f32: | |||
| 8019 | case NEON::BI__builtin_neon_vsetq_lane_i8: | |||
| 8020 | case NEON::BI__builtin_neon_vsetq_lane_i16: | |||
| 8021 | case NEON::BI__builtin_neon_vsetq_lane_i32: | |||
| 8022 | case NEON::BI__builtin_neon_vsetq_lane_i64: | |||
| 8023 | case NEON::BI__builtin_neon_vsetq_lane_bf16: | |||
| 8024 | case NEON::BI__builtin_neon_vsetq_lane_f32: | |||
| 8025 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
| 8026 | ||||
| 8027 | case NEON::BI__builtin_neon_vsha1h_u32: | |||
| 8028 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1h), Ops, | |||
| 8029 | "vsha1h"); | |||
| 8030 | case NEON::BI__builtin_neon_vsha1cq_u32: | |||
| 8031 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1c), Ops, | |||
| 8032 | "vsha1h"); | |||
| 8033 | case NEON::BI__builtin_neon_vsha1pq_u32: | |||
| 8034 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1p), Ops, | |||
| 8035 | "vsha1h"); | |||
| 8036 | case NEON::BI__builtin_neon_vsha1mq_u32: | |||
| 8037 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1m), Ops, | |||
| 8038 | "vsha1h"); | |||
| 8039 | ||||
| 8040 | case NEON::BI__builtin_neon_vcvth_bf16_f32: { | |||
| 8041 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vcvtbfp2bf), Ops, | |||
| 8042 | "vcvtbfp2bf"); | |||
| 8043 | } | |||
| 8044 | ||||
| 8045 | // The ARM _MoveToCoprocessor builtins put the input register value as | |||
| 8046 | // the first argument, but the LLVM intrinsic expects it as the third one. | |||
| 8047 | case ARM::BI_MoveToCoprocessor: | |||
| 8048 | case ARM::BI_MoveToCoprocessor2: { | |||
| 8049 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI_MoveToCoprocessor ? | |||
| 8050 | Intrinsic::arm_mcr : Intrinsic::arm_mcr2); | |||
| 8051 | return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0], | |||
| 8052 | Ops[3], Ops[4], Ops[5]}); | |||
| 8053 | } | |||
| 8054 | } | |||
| 8055 | ||||
| 8056 | // Get the last argument, which specifies the vector type. | |||
| 8057 | assert(HasExtraArg)(static_cast <bool> (HasExtraArg) ? void (0) : __assert_fail ("HasExtraArg", "clang/lib/CodeGen/CGBuiltin.cpp", 8057, __extension__ __PRETTY_FUNCTION__)); | |||
| 8058 | const Expr *Arg = E->getArg(E->getNumArgs()-1); | |||
| 8059 | Optional<llvm::APSInt> Result = Arg->getIntegerConstantExpr(getContext()); | |||
| 8060 | if (!Result) | |||
| 8061 | return nullptr; | |||
| 8062 | ||||
| 8063 | if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f || | |||
| 8064 | BuiltinID == ARM::BI__builtin_arm_vcvtr_d) { | |||
| 8065 | // Determine the overloaded type of this builtin. | |||
| 8066 | llvm::Type *Ty; | |||
| 8067 | if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f) | |||
| 8068 | Ty = FloatTy; | |||
| 8069 | else | |||
| 8070 | Ty = DoubleTy; | |||
| 8071 | ||||
| 8072 | // Determine whether this is an unsigned conversion or not. | |||
| 8073 | bool usgn = Result->getZExtValue() == 1; | |||
| 8074 | unsigned Int = usgn ? Intrinsic::arm_vcvtru : Intrinsic::arm_vcvtr; | |||
| 8075 | ||||
| 8076 | // Call the appropriate intrinsic. | |||
| 8077 | Function *F = CGM.getIntrinsic(Int, Ty); | |||
| 8078 | return Builder.CreateCall(F, Ops, "vcvtr"); | |||
| 8079 | } | |||
| 8080 | ||||
| 8081 | // Determine the type of this overloaded NEON intrinsic. | |||
| 8082 | NeonTypeFlags Type = Result->getZExtValue(); | |||
| 8083 | bool usgn = Type.isUnsigned(); | |||
| 8084 | bool rightShift = false; | |||
| 8085 | ||||
| 8086 | llvm::FixedVectorType *VTy = | |||
| 8087 | GetNeonType(this, Type, getTarget().hasLegalHalfType(), false, | |||
| 8088 | getTarget().hasBFloat16Type()); | |||
| 8089 | llvm::Type *Ty = VTy; | |||
| 8090 | if (!Ty) | |||
| 8091 | return nullptr; | |||
| 8092 | ||||
| 8093 | // Many NEON builtins have identical semantics and uses in ARM and | |||
| 8094 | // AArch64. Emit these in a single function. | |||
| 8095 | auto IntrinsicMap = makeArrayRef(ARMSIMDIntrinsicMap); | |||
| 8096 | const ARMVectorIntrinsicInfo *Builtin = findARMVectorIntrinsicInMap( | |||
| 8097 | IntrinsicMap, BuiltinID, NEONSIMDIntrinsicsProvenSorted); | |||
| 8098 | if (Builtin) | |||
| 8099 | return EmitCommonNeonBuiltinExpr( | |||
| 8100 | Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic, | |||
| 8101 | Builtin->NameHint, Builtin->TypeModifier, E, Ops, PtrOp0, PtrOp1, Arch); | |||
| 8102 | ||||
| 8103 | unsigned Int; | |||
| 8104 | switch (BuiltinID) { | |||
| 8105 | default: return nullptr; | |||
| 8106 | case NEON::BI__builtin_neon_vld1q_lane_v: | |||
| 8107 | // Handle 64-bit integer elements as a special case. Use shuffles of | |||
| 8108 | // one-element vectors to avoid poor code for i64 in the backend. | |||
| 8109 | if (VTy->getElementType()->isIntegerTy(64)) { | |||
| 8110 | // Extract the other lane. | |||
| 8111 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 8112 | int Lane = cast<ConstantInt>(Ops[2])->getZExtValue(); | |||
| 8113 | Value *SV = llvm::ConstantVector::get(ConstantInt::get(Int32Ty, 1-Lane)); | |||
| 8114 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV); | |||
| 8115 | // Load the value as a one-element vector. | |||
| 8116 | Ty = llvm::FixedVectorType::get(VTy->getElementType(), 1); | |||
| 8117 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; | |||
| 8118 | Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vld1, Tys); | |||
| 8119 | Value *Align = getAlignmentValue32(PtrOp0); | |||
| 8120 | Value *Ld = Builder.CreateCall(F, {Ops[0], Align}); | |||
| 8121 | // Combine them. | |||
| 8122 | int Indices[] = {1 - Lane, Lane}; | |||
| 8123 | return Builder.CreateShuffleVector(Ops[1], Ld, Indices, "vld1q_lane"); | |||
| 8124 | } | |||
| 8125 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 8126 | case NEON::BI__builtin_neon_vld1_lane_v: { | |||
| 8127 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 8128 | PtrOp0 = Builder.CreateElementBitCast(PtrOp0, VTy->getElementType()); | |||
| 8129 | Value *Ld = Builder.CreateLoad(PtrOp0); | |||
| 8130 | return Builder.CreateInsertElement(Ops[1], Ld, Ops[2], "vld1_lane"); | |||
| 8131 | } | |||
| 8132 | case NEON::BI__builtin_neon_vqrshrn_n_v: | |||
| 8133 | Int = | |||
| 8134 | usgn ? Intrinsic::arm_neon_vqrshiftnu : Intrinsic::arm_neon_vqrshiftns; | |||
| 8135 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n", | |||
| 8136 | 1, true); | |||
| 8137 | case NEON::BI__builtin_neon_vqrshrun_n_v: | |||
| 8138 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqrshiftnsu, Ty), | |||
| 8139 | Ops, "vqrshrun_n", 1, true); | |||
| 8140 | case NEON::BI__builtin_neon_vqshrn_n_v: | |||
| 8141 | Int = usgn ? Intrinsic::arm_neon_vqshiftnu : Intrinsic::arm_neon_vqshiftns; | |||
| 8142 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n", | |||
| 8143 | 1, true); | |||
| 8144 | case NEON::BI__builtin_neon_vqshrun_n_v: | |||
| 8145 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqshiftnsu, Ty), | |||
| 8146 | Ops, "vqshrun_n", 1, true); | |||
| 8147 | case NEON::BI__builtin_neon_vrecpe_v: | |||
| 8148 | case NEON::BI__builtin_neon_vrecpeq_v: | |||
| 8149 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrecpe, Ty), | |||
| 8150 | Ops, "vrecpe"); | |||
| 8151 | case NEON::BI__builtin_neon_vrshrn_n_v: | |||
| 8152 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrshiftn, Ty), | |||
| 8153 | Ops, "vrshrn_n", 1, true); | |||
| 8154 | case NEON::BI__builtin_neon_vrsra_n_v: | |||
| 8155 | case NEON::BI__builtin_neon_vrsraq_n_v: | |||
| 8156 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 8157 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 8158 | Ops[2] = EmitNeonShiftVector(Ops[2], Ty, true); | |||
| 8159 | Int = usgn ? Intrinsic::arm_neon_vrshiftu : Intrinsic::arm_neon_vrshifts; | |||
| 8160 | Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Ty), {Ops[1], Ops[2]}); | |||
| 8161 | return Builder.CreateAdd(Ops[0], Ops[1], "vrsra_n"); | |||
| 8162 | case NEON::BI__builtin_neon_vsri_n_v: | |||
| 8163 | case NEON::BI__builtin_neon_vsriq_n_v: | |||
| 8164 | rightShift = true; | |||
| 8165 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 8166 | case NEON::BI__builtin_neon_vsli_n_v: | |||
| 8167 | case NEON::BI__builtin_neon_vsliq_n_v: | |||
| 8168 | Ops[2] = EmitNeonShiftVector(Ops[2], Ty, rightShift); | |||
| 8169 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vshiftins, Ty), | |||
| 8170 | Ops, "vsli_n"); | |||
| 8171 | case NEON::BI__builtin_neon_vsra_n_v: | |||
| 8172 | case NEON::BI__builtin_neon_vsraq_n_v: | |||
| 8173 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 8174 | Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n"); | |||
| 8175 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
| 8176 | case NEON::BI__builtin_neon_vst1q_lane_v: | |||
| 8177 | // Handle 64-bit integer elements as a special case. Use a shuffle to get | |||
| 8178 | // a one-element vector and avoid poor code for i64 in the backend. | |||
| 8179 | if (VTy->getElementType()->isIntegerTy(64)) { | |||
| 8180 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 8181 | Value *SV = llvm::ConstantVector::get(cast<llvm::Constant>(Ops[2])); | |||
| 8182 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV); | |||
| 8183 | Ops[2] = getAlignmentValue32(PtrOp0); | |||
| 8184 | llvm::Type *Tys[] = {Int8PtrTy, Ops[1]->getType()}; | |||
| 8185 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_neon_vst1, | |||
| 8186 | Tys), Ops); | |||
| 8187 | } | |||
| 8188 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 8189 | case NEON::BI__builtin_neon_vst1_lane_v: { | |||
| 8190 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 8191 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]); | |||
| 8192 | auto St = Builder.CreateStore( | |||
| 8193 | Ops[1], Builder.CreateElementBitCast(PtrOp0, Ops[1]->getType())); | |||
| 8194 | return St; | |||
| 8195 | } | |||
| 8196 | case NEON::BI__builtin_neon_vtbl1_v: | |||
| 8197 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl1), | |||
| 8198 | Ops, "vtbl1"); | |||
| 8199 | case NEON::BI__builtin_neon_vtbl2_v: | |||
| 8200 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl2), | |||
| 8201 | Ops, "vtbl2"); | |||
| 8202 | case NEON::BI__builtin_neon_vtbl3_v: | |||
| 8203 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl3), | |||
| 8204 | Ops, "vtbl3"); | |||
| 8205 | case NEON::BI__builtin_neon_vtbl4_v: | |||
| 8206 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl4), | |||
| 8207 | Ops, "vtbl4"); | |||
| 8208 | case NEON::BI__builtin_neon_vtbx1_v: | |||
| 8209 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx1), | |||
| 8210 | Ops, "vtbx1"); | |||
| 8211 | case NEON::BI__builtin_neon_vtbx2_v: | |||
| 8212 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx2), | |||
| 8213 | Ops, "vtbx2"); | |||
| 8214 | case NEON::BI__builtin_neon_vtbx3_v: | |||
| 8215 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx3), | |||
| 8216 | Ops, "vtbx3"); | |||
| 8217 | case NEON::BI__builtin_neon_vtbx4_v: | |||
| 8218 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx4), | |||
| 8219 | Ops, "vtbx4"); | |||
| 8220 | } | |||
| 8221 | } | |||
| 8222 | ||||
| 8223 | template<typename Integer> | |||
| 8224 | static Integer GetIntegerConstantValue(const Expr *E, ASTContext &Context) { | |||
| 8225 | return E->getIntegerConstantExpr(Context)->getExtValue(); | |||
| 8226 | } | |||
| 8227 | ||||
| 8228 | static llvm::Value *SignOrZeroExtend(CGBuilderTy &Builder, llvm::Value *V, | |||
| 8229 | llvm::Type *T, bool Unsigned) { | |||
| 8230 | // Helper function called by Tablegen-constructed ARM MVE builtin codegen, | |||
| 8231 | // which finds it convenient to specify signed/unsigned as a boolean flag. | |||
| 8232 | return Unsigned ? Builder.CreateZExt(V, T) : Builder.CreateSExt(V, T); | |||
| 8233 | } | |||
| 8234 | ||||
| 8235 | static llvm::Value *MVEImmediateShr(CGBuilderTy &Builder, llvm::Value *V, | |||
| 8236 | uint32_t Shift, bool Unsigned) { | |||
| 8237 | // MVE helper function for integer shift right. This must handle signed vs | |||
| 8238 | // unsigned, and also deal specially with the case where the shift count is | |||
| 8239 | // equal to the lane size. In LLVM IR, an LShr with that parameter would be | |||
| 8240 | // undefined behavior, but in MVE it's legal, so we must convert it to code | |||
| 8241 | // that is not undefined in IR. | |||
| 8242 | unsigned LaneBits = cast<llvm::VectorType>(V->getType()) | |||
| 8243 | ->getElementType() | |||
| 8244 | ->getPrimitiveSizeInBits(); | |||
| 8245 | if (Shift == LaneBits) { | |||
| 8246 | // An unsigned shift of the full lane size always generates zero, so we can | |||
| 8247 | // simply emit a zero vector. A signed shift of the full lane size does the | |||
| 8248 | // same thing as shifting by one bit fewer. | |||
| 8249 | if (Unsigned) | |||
| 8250 | return llvm::Constant::getNullValue(V->getType()); | |||
| 8251 | else | |||
| 8252 | --Shift; | |||
| 8253 | } | |||
| 8254 | return Unsigned ? Builder.CreateLShr(V, Shift) : Builder.CreateAShr(V, Shift); | |||
| 8255 | } | |||
| 8256 | ||||
| 8257 | static llvm::Value *ARMMVEVectorSplat(CGBuilderTy &Builder, llvm::Value *V) { | |||
| 8258 | // MVE-specific helper function for a vector splat, which infers the element | |||
| 8259 | // count of the output vector by knowing that MVE vectors are all 128 bits | |||
| 8260 | // wide. | |||
| 8261 | unsigned Elements = 128 / V->getType()->getPrimitiveSizeInBits(); | |||
| 8262 | return Builder.CreateVectorSplat(Elements, V); | |||
| 8263 | } | |||
| 8264 | ||||
| 8265 | static llvm::Value *ARMMVEVectorReinterpret(CGBuilderTy &Builder, | |||
| 8266 | CodeGenFunction *CGF, | |||
| 8267 | llvm::Value *V, | |||
| 8268 | llvm::Type *DestType) { | |||
| 8269 | // Convert one MVE vector type into another by reinterpreting its in-register | |||
| 8270 | // format. | |||
| 8271 | // | |||
| 8272 | // Little-endian, this is identical to a bitcast (which reinterprets the | |||
| 8273 | // memory format). But big-endian, they're not necessarily the same, because | |||
| 8274 | // the register and memory formats map to each other differently depending on | |||
| 8275 | // the lane size. | |||
| 8276 | // | |||
| 8277 | // We generate a bitcast whenever we can (if we're little-endian, or if the | |||
| 8278 | // lane sizes are the same anyway). Otherwise we fall back to an IR intrinsic | |||
| 8279 | // that performs the different kind of reinterpretation. | |||
| 8280 | if (CGF->getTarget().isBigEndian() && | |||
| 8281 | V->getType()->getScalarSizeInBits() != DestType->getScalarSizeInBits()) { | |||
| 8282 | return Builder.CreateCall( | |||
| 8283 | CGF->CGM.getIntrinsic(Intrinsic::arm_mve_vreinterpretq, | |||
| 8284 | {DestType, V->getType()}), | |||
| 8285 | V); | |||
| 8286 | } else { | |||
| 8287 | return Builder.CreateBitCast(V, DestType); | |||
| 8288 | } | |||
| 8289 | } | |||
| 8290 | ||||
| 8291 | static llvm::Value *VectorUnzip(CGBuilderTy &Builder, llvm::Value *V, bool Odd) { | |||
| 8292 | // Make a shufflevector that extracts every other element of a vector (evens | |||
| 8293 | // or odds, as desired). | |||
| 8294 | SmallVector<int, 16> Indices; | |||
| 8295 | unsigned InputElements = | |||
| 8296 | cast<llvm::FixedVectorType>(V->getType())->getNumElements(); | |||
| 8297 | for (unsigned i = 0; i < InputElements; i += 2) | |||
| 8298 | Indices.push_back(i + Odd); | |||
| 8299 | return Builder.CreateShuffleVector(V, Indices); | |||
| 8300 | } | |||
| 8301 | ||||
| 8302 | static llvm::Value *VectorZip(CGBuilderTy &Builder, llvm::Value *V0, | |||
| 8303 | llvm::Value *V1) { | |||
| 8304 | // Make a shufflevector that interleaves two vectors element by element. | |||
| 8305 | assert(V0->getType() == V1->getType() && "Can't zip different vector types")(static_cast <bool> (V0->getType() == V1->getType () && "Can't zip different vector types") ? void (0) : __assert_fail ("V0->getType() == V1->getType() && \"Can't zip different vector types\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8305, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8306 | SmallVector<int, 16> Indices; | |||
| 8307 | unsigned InputElements = | |||
| 8308 | cast<llvm::FixedVectorType>(V0->getType())->getNumElements(); | |||
| 8309 | for (unsigned i = 0; i < InputElements; i++) { | |||
| 8310 | Indices.push_back(i); | |||
| 8311 | Indices.push_back(i + InputElements); | |||
| 8312 | } | |||
| 8313 | return Builder.CreateShuffleVector(V0, V1, Indices); | |||
| 8314 | } | |||
| 8315 | ||||
| 8316 | template<unsigned HighBit, unsigned OtherBits> | |||
| 8317 | static llvm::Value *ARMMVEConstantSplat(CGBuilderTy &Builder, llvm::Type *VT) { | |||
| 8318 | // MVE-specific helper function to make a vector splat of a constant such as | |||
| 8319 | // UINT_MAX or INT_MIN, in which all bits below the highest one are equal. | |||
| 8320 | llvm::Type *T = cast<llvm::VectorType>(VT)->getElementType(); | |||
| 8321 | unsigned LaneBits = T->getPrimitiveSizeInBits(); | |||
| 8322 | uint32_t Value = HighBit << (LaneBits - 1); | |||
| 8323 | if (OtherBits) | |||
| 8324 | Value |= (1UL << (LaneBits - 1)) - 1; | |||
| 8325 | llvm::Value *Lane = llvm::ConstantInt::get(T, Value); | |||
| 8326 | return ARMMVEVectorSplat(Builder, Lane); | |||
| 8327 | } | |||
| 8328 | ||||
| 8329 | static llvm::Value *ARMMVEVectorElementReverse(CGBuilderTy &Builder, | |||
| 8330 | llvm::Value *V, | |||
| 8331 | unsigned ReverseWidth) { | |||
| 8332 | // MVE-specific helper function which reverses the elements of a | |||
| 8333 | // vector within every (ReverseWidth)-bit collection of lanes. | |||
| 8334 | SmallVector<int, 16> Indices; | |||
| 8335 | unsigned LaneSize = V->getType()->getScalarSizeInBits(); | |||
| 8336 | unsigned Elements = 128 / LaneSize; | |||
| 8337 | unsigned Mask = ReverseWidth / LaneSize - 1; | |||
| 8338 | for (unsigned i = 0; i < Elements; i++) | |||
| 8339 | Indices.push_back(i ^ Mask); | |||
| 8340 | return Builder.CreateShuffleVector(V, Indices); | |||
| 8341 | } | |||
| 8342 | ||||
| 8343 | Value *CodeGenFunction::EmitARMMVEBuiltinExpr(unsigned BuiltinID, | |||
| 8344 | const CallExpr *E, | |||
| 8345 | ReturnValueSlot ReturnValue, | |||
| 8346 | llvm::Triple::ArchType Arch) { | |||
| 8347 | enum class CustomCodeGen { VLD24, VST24 } CustomCodeGenType; | |||
| 8348 | Intrinsic::ID IRIntr; | |||
| 8349 | unsigned NumVectors; | |||
| 8350 | ||||
| 8351 | // Code autogenerated by Tablegen will handle all the simple builtins. | |||
| 8352 | switch (BuiltinID) { | |||
| 8353 | #include "clang/Basic/arm_mve_builtin_cg.inc" | |||
| 8354 | ||||
| 8355 | // If we didn't match an MVE builtin id at all, go back to the | |||
| 8356 | // main EmitARMBuiltinExpr. | |||
| 8357 | default: | |||
| 8358 | return nullptr; | |||
| 8359 | } | |||
| 8360 | ||||
| 8361 | // Anything that breaks from that switch is an MVE builtin that | |||
| 8362 | // needs handwritten code to generate. | |||
| 8363 | ||||
| 8364 | switch (CustomCodeGenType) { | |||
| 8365 | ||||
| 8366 | case CustomCodeGen::VLD24: { | |||
| 8367 | llvm::SmallVector<Value *, 4> Ops; | |||
| 8368 | llvm::SmallVector<llvm::Type *, 4> Tys; | |||
| 8369 | ||||
| 8370 | auto MvecCType = E->getType(); | |||
| 8371 | auto MvecLType = ConvertType(MvecCType); | |||
| 8372 | assert(MvecLType->isStructTy() &&(static_cast <bool> (MvecLType->isStructTy() && "Return type for vld[24]q should be a struct") ? void (0) : __assert_fail ("MvecLType->isStructTy() && \"Return type for vld[24]q should be a struct\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8373, __extension__ __PRETTY_FUNCTION__ )) | |||
| 8373 | "Return type for vld[24]q should be a struct")(static_cast <bool> (MvecLType->isStructTy() && "Return type for vld[24]q should be a struct") ? void (0) : __assert_fail ("MvecLType->isStructTy() && \"Return type for vld[24]q should be a struct\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8373, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8374 | assert(MvecLType->getStructNumElements() == 1 &&(static_cast <bool> (MvecLType->getStructNumElements () == 1 && "Return-type struct for vld[24]q should have one element" ) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Return-type struct for vld[24]q should have one element\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8375, __extension__ __PRETTY_FUNCTION__ )) | |||
| 8375 | "Return-type struct for vld[24]q should have one element")(static_cast <bool> (MvecLType->getStructNumElements () == 1 && "Return-type struct for vld[24]q should have one element" ) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Return-type struct for vld[24]q should have one element\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8375, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8376 | auto MvecLTypeInner = MvecLType->getStructElementType(0); | |||
| 8377 | assert(MvecLTypeInner->isArrayTy() &&(static_cast <bool> (MvecLTypeInner->isArrayTy() && "Return-type struct for vld[24]q should contain an array") ? void (0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Return-type struct for vld[24]q should contain an array\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8378, __extension__ __PRETTY_FUNCTION__ )) | |||
| 8378 | "Return-type struct for vld[24]q should contain an array")(static_cast <bool> (MvecLTypeInner->isArrayTy() && "Return-type struct for vld[24]q should contain an array") ? void (0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Return-type struct for vld[24]q should contain an array\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8378, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8379 | assert(MvecLTypeInner->getArrayNumElements() == NumVectors &&(static_cast <bool> (MvecLTypeInner->getArrayNumElements () == NumVectors && "Array member of return-type struct vld[24]q has wrong length" ) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8380, __extension__ __PRETTY_FUNCTION__ )) | |||
| 8380 | "Array member of return-type struct vld[24]q has wrong length")(static_cast <bool> (MvecLTypeInner->getArrayNumElements () == NumVectors && "Array member of return-type struct vld[24]q has wrong length" ) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8380, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8381 | auto VecLType = MvecLTypeInner->getArrayElementType(); | |||
| 8382 | ||||
| 8383 | Tys.push_back(VecLType); | |||
| 8384 | ||||
| 8385 | auto Addr = E->getArg(0); | |||
| 8386 | Ops.push_back(EmitScalarExpr(Addr)); | |||
| 8387 | Tys.push_back(ConvertType(Addr->getType())); | |||
| 8388 | ||||
| 8389 | Function *F = CGM.getIntrinsic(IRIntr, makeArrayRef(Tys)); | |||
| 8390 | Value *LoadResult = Builder.CreateCall(F, Ops); | |||
| 8391 | Value *MvecOut = UndefValue::get(MvecLType); | |||
| 8392 | for (unsigned i = 0; i < NumVectors; ++i) { | |||
| 8393 | Value *Vec = Builder.CreateExtractValue(LoadResult, i); | |||
| 8394 | MvecOut = Builder.CreateInsertValue(MvecOut, Vec, {0, i}); | |||
| 8395 | } | |||
| 8396 | ||||
| 8397 | if (ReturnValue.isNull()) | |||
| 8398 | return MvecOut; | |||
| 8399 | else | |||
| 8400 | return Builder.CreateStore(MvecOut, ReturnValue.getValue()); | |||
| 8401 | } | |||
| 8402 | ||||
| 8403 | case CustomCodeGen::VST24: { | |||
| 8404 | llvm::SmallVector<Value *, 4> Ops; | |||
| 8405 | llvm::SmallVector<llvm::Type *, 4> Tys; | |||
| 8406 | ||||
| 8407 | auto Addr = E->getArg(0); | |||
| 8408 | Ops.push_back(EmitScalarExpr(Addr)); | |||
| 8409 | Tys.push_back(ConvertType(Addr->getType())); | |||
| 8410 | ||||
| 8411 | auto MvecCType = E->getArg(1)->getType(); | |||
| 8412 | auto MvecLType = ConvertType(MvecCType); | |||
| 8413 | assert(MvecLType->isStructTy() && "Data type for vst2q should be a struct")(static_cast <bool> (MvecLType->isStructTy() && "Data type for vst2q should be a struct") ? void (0) : __assert_fail ("MvecLType->isStructTy() && \"Data type for vst2q should be a struct\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8413, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8414 | assert(MvecLType->getStructNumElements() == 1 &&(static_cast <bool> (MvecLType->getStructNumElements () == 1 && "Data-type struct for vst2q should have one element" ) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Data-type struct for vst2q should have one element\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8415, __extension__ __PRETTY_FUNCTION__ )) | |||
| 8415 | "Data-type struct for vst2q should have one element")(static_cast <bool> (MvecLType->getStructNumElements () == 1 && "Data-type struct for vst2q should have one element" ) ? void (0) : __assert_fail ("MvecLType->getStructNumElements() == 1 && \"Data-type struct for vst2q should have one element\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8415, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8416 | auto MvecLTypeInner = MvecLType->getStructElementType(0); | |||
| 8417 | assert(MvecLTypeInner->isArrayTy() &&(static_cast <bool> (MvecLTypeInner->isArrayTy() && "Data-type struct for vst2q should contain an array") ? void (0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Data-type struct for vst2q should contain an array\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8418, __extension__ __PRETTY_FUNCTION__ )) | |||
| 8418 | "Data-type struct for vst2q should contain an array")(static_cast <bool> (MvecLTypeInner->isArrayTy() && "Data-type struct for vst2q should contain an array") ? void (0) : __assert_fail ("MvecLTypeInner->isArrayTy() && \"Data-type struct for vst2q should contain an array\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8418, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8419 | assert(MvecLTypeInner->getArrayNumElements() == NumVectors &&(static_cast <bool> (MvecLTypeInner->getArrayNumElements () == NumVectors && "Array member of return-type struct vld[24]q has wrong length" ) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8420, __extension__ __PRETTY_FUNCTION__ )) | |||
| 8420 | "Array member of return-type struct vld[24]q has wrong length")(static_cast <bool> (MvecLTypeInner->getArrayNumElements () == NumVectors && "Array member of return-type struct vld[24]q has wrong length" ) ? void (0) : __assert_fail ("MvecLTypeInner->getArrayNumElements() == NumVectors && \"Array member of return-type struct vld[24]q has wrong length\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8420, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8421 | auto VecLType = MvecLTypeInner->getArrayElementType(); | |||
| 8422 | ||||
| 8423 | Tys.push_back(VecLType); | |||
| 8424 | ||||
| 8425 | AggValueSlot MvecSlot = CreateAggTemp(MvecCType); | |||
| 8426 | EmitAggExpr(E->getArg(1), MvecSlot); | |||
| 8427 | auto Mvec = Builder.CreateLoad(MvecSlot.getAddress()); | |||
| 8428 | for (unsigned i = 0; i < NumVectors; i++) | |||
| 8429 | Ops.push_back(Builder.CreateExtractValue(Mvec, {0, i})); | |||
| 8430 | ||||
| 8431 | Function *F = CGM.getIntrinsic(IRIntr, makeArrayRef(Tys)); | |||
| 8432 | Value *ToReturn = nullptr; | |||
| 8433 | for (unsigned i = 0; i < NumVectors; i++) { | |||
| 8434 | Ops.push_back(llvm::ConstantInt::get(Int32Ty, i)); | |||
| 8435 | ToReturn = Builder.CreateCall(F, Ops); | |||
| 8436 | Ops.pop_back(); | |||
| 8437 | } | |||
| 8438 | return ToReturn; | |||
| 8439 | } | |||
| 8440 | } | |||
| 8441 | llvm_unreachable("unknown custom codegen type.")::llvm::llvm_unreachable_internal("unknown custom codegen type." , "clang/lib/CodeGen/CGBuiltin.cpp", 8441); | |||
| 8442 | } | |||
| 8443 | ||||
| 8444 | Value *CodeGenFunction::EmitARMCDEBuiltinExpr(unsigned BuiltinID, | |||
| 8445 | const CallExpr *E, | |||
| 8446 | ReturnValueSlot ReturnValue, | |||
| 8447 | llvm::Triple::ArchType Arch) { | |||
| 8448 | switch (BuiltinID) { | |||
| 8449 | default: | |||
| 8450 | return nullptr; | |||
| 8451 | #include "clang/Basic/arm_cde_builtin_cg.inc" | |||
| 8452 | } | |||
| 8453 | } | |||
| 8454 | ||||
| 8455 | static Value *EmitAArch64TblBuiltinExpr(CodeGenFunction &CGF, unsigned BuiltinID, | |||
| 8456 | const CallExpr *E, | |||
| 8457 | SmallVectorImpl<Value *> &Ops, | |||
| 8458 | llvm::Triple::ArchType Arch) { | |||
| 8459 | unsigned int Int = 0; | |||
| 8460 | const char *s = nullptr; | |||
| 8461 | ||||
| 8462 | switch (BuiltinID) { | |||
| 8463 | default: | |||
| 8464 | return nullptr; | |||
| 8465 | case NEON::BI__builtin_neon_vtbl1_v: | |||
| 8466 | case NEON::BI__builtin_neon_vqtbl1_v: | |||
| 8467 | case NEON::BI__builtin_neon_vqtbl1q_v: | |||
| 8468 | case NEON::BI__builtin_neon_vtbl2_v: | |||
| 8469 | case NEON::BI__builtin_neon_vqtbl2_v: | |||
| 8470 | case NEON::BI__builtin_neon_vqtbl2q_v: | |||
| 8471 | case NEON::BI__builtin_neon_vtbl3_v: | |||
| 8472 | case NEON::BI__builtin_neon_vqtbl3_v: | |||
| 8473 | case NEON::BI__builtin_neon_vqtbl3q_v: | |||
| 8474 | case NEON::BI__builtin_neon_vtbl4_v: | |||
| 8475 | case NEON::BI__builtin_neon_vqtbl4_v: | |||
| 8476 | case NEON::BI__builtin_neon_vqtbl4q_v: | |||
| 8477 | break; | |||
| 8478 | case NEON::BI__builtin_neon_vtbx1_v: | |||
| 8479 | case NEON::BI__builtin_neon_vqtbx1_v: | |||
| 8480 | case NEON::BI__builtin_neon_vqtbx1q_v: | |||
| 8481 | case NEON::BI__builtin_neon_vtbx2_v: | |||
| 8482 | case NEON::BI__builtin_neon_vqtbx2_v: | |||
| 8483 | case NEON::BI__builtin_neon_vqtbx2q_v: | |||
| 8484 | case NEON::BI__builtin_neon_vtbx3_v: | |||
| 8485 | case NEON::BI__builtin_neon_vqtbx3_v: | |||
| 8486 | case NEON::BI__builtin_neon_vqtbx3q_v: | |||
| 8487 | case NEON::BI__builtin_neon_vtbx4_v: | |||
| 8488 | case NEON::BI__builtin_neon_vqtbx4_v: | |||
| 8489 | case NEON::BI__builtin_neon_vqtbx4q_v: | |||
| 8490 | break; | |||
| 8491 | } | |||
| 8492 | ||||
| 8493 | assert(E->getNumArgs() >= 3)(static_cast <bool> (E->getNumArgs() >= 3) ? void (0) : __assert_fail ("E->getNumArgs() >= 3", "clang/lib/CodeGen/CGBuiltin.cpp" , 8493, __extension__ __PRETTY_FUNCTION__)); | |||
| 8494 | ||||
| 8495 | // Get the last argument, which specifies the vector type. | |||
| 8496 | const Expr *Arg = E->getArg(E->getNumArgs() - 1); | |||
| 8497 | Optional<llvm::APSInt> Result = Arg->getIntegerConstantExpr(CGF.getContext()); | |||
| 8498 | if (!Result) | |||
| 8499 | return nullptr; | |||
| 8500 | ||||
| 8501 | // Determine the type of this overloaded NEON intrinsic. | |||
| 8502 | NeonTypeFlags Type = Result->getZExtValue(); | |||
| 8503 | llvm::FixedVectorType *Ty = GetNeonType(&CGF, Type); | |||
| 8504 | if (!Ty) | |||
| 8505 | return nullptr; | |||
| 8506 | ||||
| 8507 | CodeGen::CGBuilderTy &Builder = CGF.Builder; | |||
| 8508 | ||||
| 8509 | // AArch64 scalar builtins are not overloaded, they do not have an extra | |||
| 8510 | // argument that specifies the vector type, need to handle each case. | |||
| 8511 | switch (BuiltinID) { | |||
| 8512 | case NEON::BI__builtin_neon_vtbl1_v: { | |||
| 8513 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 1), nullptr, | |||
| 8514 | Ops[1], Ty, Intrinsic::aarch64_neon_tbl1, | |||
| 8515 | "vtbl1"); | |||
| 8516 | } | |||
| 8517 | case NEON::BI__builtin_neon_vtbl2_v: { | |||
| 8518 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 2), nullptr, | |||
| 8519 | Ops[2], Ty, Intrinsic::aarch64_neon_tbl1, | |||
| 8520 | "vtbl1"); | |||
| 8521 | } | |||
| 8522 | case NEON::BI__builtin_neon_vtbl3_v: { | |||
| 8523 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 3), nullptr, | |||
| 8524 | Ops[3], Ty, Intrinsic::aarch64_neon_tbl2, | |||
| 8525 | "vtbl2"); | |||
| 8526 | } | |||
| 8527 | case NEON::BI__builtin_neon_vtbl4_v: { | |||
| 8528 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 4), nullptr, | |||
| 8529 | Ops[4], Ty, Intrinsic::aarch64_neon_tbl2, | |||
| 8530 | "vtbl2"); | |||
| 8531 | } | |||
| 8532 | case NEON::BI__builtin_neon_vtbx1_v: { | |||
| 8533 | Value *TblRes = | |||
| 8534 | packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 1), nullptr, Ops[2], | |||
| 8535 | Ty, Intrinsic::aarch64_neon_tbl1, "vtbl1"); | |||
| 8536 | ||||
| 8537 | llvm::Constant *EightV = ConstantInt::get(Ty, 8); | |||
| 8538 | Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[2], EightV); | |||
| 8539 | CmpRes = Builder.CreateSExt(CmpRes, Ty); | |||
| 8540 | ||||
| 8541 | Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]); | |||
| 8542 | Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes); | |||
| 8543 | return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx"); | |||
| 8544 | } | |||
| 8545 | case NEON::BI__builtin_neon_vtbx2_v: { | |||
| 8546 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 2), Ops[0], | |||
| 8547 | Ops[3], Ty, Intrinsic::aarch64_neon_tbx1, | |||
| 8548 | "vtbx1"); | |||
| 8549 | } | |||
| 8550 | case NEON::BI__builtin_neon_vtbx3_v: { | |||
| 8551 | Value *TblRes = | |||
| 8552 | packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 3), nullptr, Ops[4], | |||
| 8553 | Ty, Intrinsic::aarch64_neon_tbl2, "vtbl2"); | |||
| 8554 | ||||
| 8555 | llvm::Constant *TwentyFourV = ConstantInt::get(Ty, 24); | |||
| 8556 | Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[4], | |||
| 8557 | TwentyFourV); | |||
| 8558 | CmpRes = Builder.CreateSExt(CmpRes, Ty); | |||
| 8559 | ||||
| 8560 | Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]); | |||
| 8561 | Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes); | |||
| 8562 | return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx"); | |||
| 8563 | } | |||
| 8564 | case NEON::BI__builtin_neon_vtbx4_v: { | |||
| 8565 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 4), Ops[0], | |||
| 8566 | Ops[5], Ty, Intrinsic::aarch64_neon_tbx2, | |||
| 8567 | "vtbx2"); | |||
| 8568 | } | |||
| 8569 | case NEON::BI__builtin_neon_vqtbl1_v: | |||
| 8570 | case NEON::BI__builtin_neon_vqtbl1q_v: | |||
| 8571 | Int = Intrinsic::aarch64_neon_tbl1; s = "vtbl1"; break; | |||
| 8572 | case NEON::BI__builtin_neon_vqtbl2_v: | |||
| 8573 | case NEON::BI__builtin_neon_vqtbl2q_v: { | |||
| 8574 | Int = Intrinsic::aarch64_neon_tbl2; s = "vtbl2"; break; | |||
| 8575 | case NEON::BI__builtin_neon_vqtbl3_v: | |||
| 8576 | case NEON::BI__builtin_neon_vqtbl3q_v: | |||
| 8577 | Int = Intrinsic::aarch64_neon_tbl3; s = "vtbl3"; break; | |||
| 8578 | case NEON::BI__builtin_neon_vqtbl4_v: | |||
| 8579 | case NEON::BI__builtin_neon_vqtbl4q_v: | |||
| 8580 | Int = Intrinsic::aarch64_neon_tbl4; s = "vtbl4"; break; | |||
| 8581 | case NEON::BI__builtin_neon_vqtbx1_v: | |||
| 8582 | case NEON::BI__builtin_neon_vqtbx1q_v: | |||
| 8583 | Int = Intrinsic::aarch64_neon_tbx1; s = "vtbx1"; break; | |||
| 8584 | case NEON::BI__builtin_neon_vqtbx2_v: | |||
| 8585 | case NEON::BI__builtin_neon_vqtbx2q_v: | |||
| 8586 | Int = Intrinsic::aarch64_neon_tbx2; s = "vtbx2"; break; | |||
| 8587 | case NEON::BI__builtin_neon_vqtbx3_v: | |||
| 8588 | case NEON::BI__builtin_neon_vqtbx3q_v: | |||
| 8589 | Int = Intrinsic::aarch64_neon_tbx3; s = "vtbx3"; break; | |||
| 8590 | case NEON::BI__builtin_neon_vqtbx4_v: | |||
| 8591 | case NEON::BI__builtin_neon_vqtbx4q_v: | |||
| 8592 | Int = Intrinsic::aarch64_neon_tbx4; s = "vtbx4"; break; | |||
| 8593 | } | |||
| 8594 | } | |||
| 8595 | ||||
| 8596 | if (!Int) | |||
| 8597 | return nullptr; | |||
| 8598 | ||||
| 8599 | Function *F = CGF.CGM.getIntrinsic(Int, Ty); | |||
| 8600 | return CGF.EmitNeonCall(F, Ops, s); | |||
| 8601 | } | |||
| 8602 | ||||
| 8603 | Value *CodeGenFunction::vectorWrapScalar16(Value *Op) { | |||
| 8604 | auto *VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 8605 | Op = Builder.CreateBitCast(Op, Int16Ty); | |||
| 8606 | Value *V = UndefValue::get(VTy); | |||
| 8607 | llvm::Constant *CI = ConstantInt::get(SizeTy, 0); | |||
| 8608 | Op = Builder.CreateInsertElement(V, Op, CI); | |||
| 8609 | return Op; | |||
| 8610 | } | |||
| 8611 | ||||
| 8612 | /// SVEBuiltinMemEltTy - Returns the memory element type for this memory | |||
| 8613 | /// access builtin. Only required if it can't be inferred from the base pointer | |||
| 8614 | /// operand. | |||
| 8615 | llvm::Type *CodeGenFunction::SVEBuiltinMemEltTy(const SVETypeFlags &TypeFlags) { | |||
| 8616 | switch (TypeFlags.getMemEltType()) { | |||
| 8617 | case SVETypeFlags::MemEltTyDefault: | |||
| 8618 | return getEltType(TypeFlags); | |||
| 8619 | case SVETypeFlags::MemEltTyInt8: | |||
| 8620 | return Builder.getInt8Ty(); | |||
| 8621 | case SVETypeFlags::MemEltTyInt16: | |||
| 8622 | return Builder.getInt16Ty(); | |||
| 8623 | case SVETypeFlags::MemEltTyInt32: | |||
| 8624 | return Builder.getInt32Ty(); | |||
| 8625 | case SVETypeFlags::MemEltTyInt64: | |||
| 8626 | return Builder.getInt64Ty(); | |||
| 8627 | } | |||
| 8628 | llvm_unreachable("Unknown MemEltType")::llvm::llvm_unreachable_internal("Unknown MemEltType", "clang/lib/CodeGen/CGBuiltin.cpp" , 8628); | |||
| 8629 | } | |||
| 8630 | ||||
| 8631 | llvm::Type *CodeGenFunction::getEltType(const SVETypeFlags &TypeFlags) { | |||
| 8632 | switch (TypeFlags.getEltType()) { | |||
| 8633 | default: | |||
| 8634 | llvm_unreachable("Invalid SVETypeFlag!")::llvm::llvm_unreachable_internal("Invalid SVETypeFlag!", "clang/lib/CodeGen/CGBuiltin.cpp" , 8634); | |||
| 8635 | ||||
| 8636 | case SVETypeFlags::EltTyInt8: | |||
| 8637 | return Builder.getInt8Ty(); | |||
| 8638 | case SVETypeFlags::EltTyInt16: | |||
| 8639 | return Builder.getInt16Ty(); | |||
| 8640 | case SVETypeFlags::EltTyInt32: | |||
| 8641 | return Builder.getInt32Ty(); | |||
| 8642 | case SVETypeFlags::EltTyInt64: | |||
| 8643 | return Builder.getInt64Ty(); | |||
| 8644 | ||||
| 8645 | case SVETypeFlags::EltTyFloat16: | |||
| 8646 | return Builder.getHalfTy(); | |||
| 8647 | case SVETypeFlags::EltTyFloat32: | |||
| 8648 | return Builder.getFloatTy(); | |||
| 8649 | case SVETypeFlags::EltTyFloat64: | |||
| 8650 | return Builder.getDoubleTy(); | |||
| 8651 | ||||
| 8652 | case SVETypeFlags::EltTyBFloat16: | |||
| 8653 | return Builder.getBFloatTy(); | |||
| 8654 | ||||
| 8655 | case SVETypeFlags::EltTyBool8: | |||
| 8656 | case SVETypeFlags::EltTyBool16: | |||
| 8657 | case SVETypeFlags::EltTyBool32: | |||
| 8658 | case SVETypeFlags::EltTyBool64: | |||
| 8659 | return Builder.getInt1Ty(); | |||
| 8660 | } | |||
| 8661 | } | |||
| 8662 | ||||
| 8663 | // Return the llvm predicate vector type corresponding to the specified element | |||
| 8664 | // TypeFlags. | |||
| 8665 | llvm::ScalableVectorType * | |||
| 8666 | CodeGenFunction::getSVEPredType(const SVETypeFlags &TypeFlags) { | |||
| 8667 | switch (TypeFlags.getEltType()) { | |||
| 8668 | default: llvm_unreachable("Unhandled SVETypeFlag!")::llvm::llvm_unreachable_internal("Unhandled SVETypeFlag!", "clang/lib/CodeGen/CGBuiltin.cpp" , 8668); | |||
| 8669 | ||||
| 8670 | case SVETypeFlags::EltTyInt8: | |||
| 8671 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16); | |||
| 8672 | case SVETypeFlags::EltTyInt16: | |||
| 8673 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8); | |||
| 8674 | case SVETypeFlags::EltTyInt32: | |||
| 8675 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4); | |||
| 8676 | case SVETypeFlags::EltTyInt64: | |||
| 8677 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2); | |||
| 8678 | ||||
| 8679 | case SVETypeFlags::EltTyBFloat16: | |||
| 8680 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8); | |||
| 8681 | case SVETypeFlags::EltTyFloat16: | |||
| 8682 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8); | |||
| 8683 | case SVETypeFlags::EltTyFloat32: | |||
| 8684 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4); | |||
| 8685 | case SVETypeFlags::EltTyFloat64: | |||
| 8686 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2); | |||
| 8687 | ||||
| 8688 | case SVETypeFlags::EltTyBool8: | |||
| 8689 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16); | |||
| 8690 | case SVETypeFlags::EltTyBool16: | |||
| 8691 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8); | |||
| 8692 | case SVETypeFlags::EltTyBool32: | |||
| 8693 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4); | |||
| 8694 | case SVETypeFlags::EltTyBool64: | |||
| 8695 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2); | |||
| 8696 | } | |||
| 8697 | } | |||
| 8698 | ||||
| 8699 | // Return the llvm vector type corresponding to the specified element TypeFlags. | |||
| 8700 | llvm::ScalableVectorType * | |||
| 8701 | CodeGenFunction::getSVEType(const SVETypeFlags &TypeFlags) { | |||
| 8702 | switch (TypeFlags.getEltType()) { | |||
| 8703 | default: | |||
| 8704 | llvm_unreachable("Invalid SVETypeFlag!")::llvm::llvm_unreachable_internal("Invalid SVETypeFlag!", "clang/lib/CodeGen/CGBuiltin.cpp" , 8704); | |||
| 8705 | ||||
| 8706 | case SVETypeFlags::EltTyInt8: | |||
| 8707 | return llvm::ScalableVectorType::get(Builder.getInt8Ty(), 16); | |||
| 8708 | case SVETypeFlags::EltTyInt16: | |||
| 8709 | return llvm::ScalableVectorType::get(Builder.getInt16Ty(), 8); | |||
| 8710 | case SVETypeFlags::EltTyInt32: | |||
| 8711 | return llvm::ScalableVectorType::get(Builder.getInt32Ty(), 4); | |||
| 8712 | case SVETypeFlags::EltTyInt64: | |||
| 8713 | return llvm::ScalableVectorType::get(Builder.getInt64Ty(), 2); | |||
| 8714 | ||||
| 8715 | case SVETypeFlags::EltTyFloat16: | |||
| 8716 | return llvm::ScalableVectorType::get(Builder.getHalfTy(), 8); | |||
| 8717 | case SVETypeFlags::EltTyBFloat16: | |||
| 8718 | return llvm::ScalableVectorType::get(Builder.getBFloatTy(), 8); | |||
| 8719 | case SVETypeFlags::EltTyFloat32: | |||
| 8720 | return llvm::ScalableVectorType::get(Builder.getFloatTy(), 4); | |||
| 8721 | case SVETypeFlags::EltTyFloat64: | |||
| 8722 | return llvm::ScalableVectorType::get(Builder.getDoubleTy(), 2); | |||
| 8723 | ||||
| 8724 | case SVETypeFlags::EltTyBool8: | |||
| 8725 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16); | |||
| 8726 | case SVETypeFlags::EltTyBool16: | |||
| 8727 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8); | |||
| 8728 | case SVETypeFlags::EltTyBool32: | |||
| 8729 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4); | |||
| 8730 | case SVETypeFlags::EltTyBool64: | |||
| 8731 | return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2); | |||
| 8732 | } | |||
| 8733 | } | |||
| 8734 | ||||
| 8735 | llvm::Value * | |||
| 8736 | CodeGenFunction::EmitSVEAllTruePred(const SVETypeFlags &TypeFlags) { | |||
| 8737 | Function *Ptrue = | |||
| 8738 | CGM.getIntrinsic(Intrinsic::aarch64_sve_ptrue, getSVEPredType(TypeFlags)); | |||
| 8739 | return Builder.CreateCall(Ptrue, {Builder.getInt32(/*SV_ALL*/ 31)}); | |||
| 8740 | } | |||
| 8741 | ||||
| 8742 | constexpr unsigned SVEBitsPerBlock = 128; | |||
| 8743 | ||||
| 8744 | static llvm::ScalableVectorType *getSVEVectorForElementType(llvm::Type *EltTy) { | |||
| 8745 | unsigned NumElts = SVEBitsPerBlock / EltTy->getScalarSizeInBits(); | |||
| 8746 | return llvm::ScalableVectorType::get(EltTy, NumElts); | |||
| 8747 | } | |||
| 8748 | ||||
| 8749 | // Reinterpret the input predicate so that it can be used to correctly isolate | |||
| 8750 | // the elements of the specified datatype. | |||
| 8751 | Value *CodeGenFunction::EmitSVEPredicateCast(Value *Pred, | |||
| 8752 | llvm::ScalableVectorType *VTy) { | |||
| 8753 | auto *RTy = llvm::VectorType::get(IntegerType::get(getLLVMContext(), 1), VTy); | |||
| 8754 | if (Pred->getType() == RTy) | |||
| 8755 | return Pred; | |||
| 8756 | ||||
| 8757 | unsigned IntID; | |||
| 8758 | llvm::Type *IntrinsicTy; | |||
| 8759 | switch (VTy->getMinNumElements()) { | |||
| 8760 | default: | |||
| 8761 | llvm_unreachable("unsupported element count!")::llvm::llvm_unreachable_internal("unsupported element count!" , "clang/lib/CodeGen/CGBuiltin.cpp", 8761); | |||
| 8762 | case 2: | |||
| 8763 | case 4: | |||
| 8764 | case 8: | |||
| 8765 | IntID = Intrinsic::aarch64_sve_convert_from_svbool; | |||
| 8766 | IntrinsicTy = RTy; | |||
| 8767 | break; | |||
| 8768 | case 16: | |||
| 8769 | IntID = Intrinsic::aarch64_sve_convert_to_svbool; | |||
| 8770 | IntrinsicTy = Pred->getType(); | |||
| 8771 | break; | |||
| 8772 | } | |||
| 8773 | ||||
| 8774 | Function *F = CGM.getIntrinsic(IntID, IntrinsicTy); | |||
| 8775 | Value *C = Builder.CreateCall(F, Pred); | |||
| 8776 | assert(C->getType() == RTy && "Unexpected return type!")(static_cast <bool> (C->getType() == RTy && "Unexpected return type!" ) ? void (0) : __assert_fail ("C->getType() == RTy && \"Unexpected return type!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8776, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8777 | return C; | |||
| 8778 | } | |||
| 8779 | ||||
| 8780 | Value *CodeGenFunction::EmitSVEGatherLoad(const SVETypeFlags &TypeFlags, | |||
| 8781 | SmallVectorImpl<Value *> &Ops, | |||
| 8782 | unsigned IntID) { | |||
| 8783 | auto *ResultTy = getSVEType(TypeFlags); | |||
| 8784 | auto *OverloadedTy = | |||
| 8785 | llvm::ScalableVectorType::get(SVEBuiltinMemEltTy(TypeFlags), ResultTy); | |||
| 8786 | ||||
| 8787 | // At the ACLE level there's only one predicate type, svbool_t, which is | |||
| 8788 | // mapped to <n x 16 x i1>. However, this might be incompatible with the | |||
| 8789 | // actual type being loaded. For example, when loading doubles (i64) the | |||
| 8790 | // predicated should be <n x 2 x i1> instead. At the IR level the type of | |||
| 8791 | // the predicate and the data being loaded must match. Cast accordingly. | |||
| 8792 | Ops[0] = EmitSVEPredicateCast(Ops[0], OverloadedTy); | |||
| 8793 | ||||
| 8794 | Function *F = nullptr; | |||
| 8795 | if (Ops[1]->getType()->isVectorTy()) | |||
| 8796 | // This is the "vector base, scalar offset" case. In order to uniquely | |||
| 8797 | // map this built-in to an LLVM IR intrinsic, we need both the return type | |||
| 8798 | // and the type of the vector base. | |||
| 8799 | F = CGM.getIntrinsic(IntID, {OverloadedTy, Ops[1]->getType()}); | |||
| 8800 | else | |||
| 8801 | // This is the "scalar base, vector offset case". The type of the offset | |||
| 8802 | // is encoded in the name of the intrinsic. We only need to specify the | |||
| 8803 | // return type in order to uniquely map this built-in to an LLVM IR | |||
| 8804 | // intrinsic. | |||
| 8805 | F = CGM.getIntrinsic(IntID, OverloadedTy); | |||
| 8806 | ||||
| 8807 | // Pass 0 when the offset is missing. This can only be applied when using | |||
| 8808 | // the "vector base" addressing mode for which ACLE allows no offset. The | |||
| 8809 | // corresponding LLVM IR always requires an offset. | |||
| 8810 | if (Ops.size() == 2) { | |||
| 8811 | assert(Ops[1]->getType()->isVectorTy() && "Scalar base requires an offset")(static_cast <bool> (Ops[1]->getType()->isVectorTy () && "Scalar base requires an offset") ? void (0) : __assert_fail ("Ops[1]->getType()->isVectorTy() && \"Scalar base requires an offset\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8811, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8812 | Ops.push_back(ConstantInt::get(Int64Ty, 0)); | |||
| 8813 | } | |||
| 8814 | ||||
| 8815 | // For "vector base, scalar index" scale the index so that it becomes a | |||
| 8816 | // scalar offset. | |||
| 8817 | if (!TypeFlags.isByteIndexed() && Ops[1]->getType()->isVectorTy()) { | |||
| 8818 | unsigned BytesPerElt = | |||
| 8819 | OverloadedTy->getElementType()->getScalarSizeInBits() / 8; | |||
| 8820 | Value *Scale = ConstantInt::get(Int64Ty, BytesPerElt); | |||
| 8821 | Ops[2] = Builder.CreateMul(Ops[2], Scale); | |||
| 8822 | } | |||
| 8823 | ||||
| 8824 | Value *Call = Builder.CreateCall(F, Ops); | |||
| 8825 | ||||
| 8826 | // The following sext/zext is only needed when ResultTy != OverloadedTy. In | |||
| 8827 | // other cases it's folded into a nop. | |||
| 8828 | return TypeFlags.isZExtReturn() ? Builder.CreateZExt(Call, ResultTy) | |||
| 8829 | : Builder.CreateSExt(Call, ResultTy); | |||
| 8830 | } | |||
| 8831 | ||||
| 8832 | Value *CodeGenFunction::EmitSVEScatterStore(const SVETypeFlags &TypeFlags, | |||
| 8833 | SmallVectorImpl<Value *> &Ops, | |||
| 8834 | unsigned IntID) { | |||
| 8835 | auto *SrcDataTy = getSVEType(TypeFlags); | |||
| 8836 | auto *OverloadedTy = | |||
| 8837 | llvm::ScalableVectorType::get(SVEBuiltinMemEltTy(TypeFlags), SrcDataTy); | |||
| 8838 | ||||
| 8839 | // In ACLE the source data is passed in the last argument, whereas in LLVM IR | |||
| 8840 | // it's the first argument. Move it accordingly. | |||
| 8841 | Ops.insert(Ops.begin(), Ops.pop_back_val()); | |||
| 8842 | ||||
| 8843 | Function *F = nullptr; | |||
| 8844 | if (Ops[2]->getType()->isVectorTy()) | |||
| 8845 | // This is the "vector base, scalar offset" case. In order to uniquely | |||
| 8846 | // map this built-in to an LLVM IR intrinsic, we need both the return type | |||
| 8847 | // and the type of the vector base. | |||
| 8848 | F = CGM.getIntrinsic(IntID, {OverloadedTy, Ops[2]->getType()}); | |||
| 8849 | else | |||
| 8850 | // This is the "scalar base, vector offset case". The type of the offset | |||
| 8851 | // is encoded in the name of the intrinsic. We only need to specify the | |||
| 8852 | // return type in order to uniquely map this built-in to an LLVM IR | |||
| 8853 | // intrinsic. | |||
| 8854 | F = CGM.getIntrinsic(IntID, OverloadedTy); | |||
| 8855 | ||||
| 8856 | // Pass 0 when the offset is missing. This can only be applied when using | |||
| 8857 | // the "vector base" addressing mode for which ACLE allows no offset. The | |||
| 8858 | // corresponding LLVM IR always requires an offset. | |||
| 8859 | if (Ops.size() == 3) { | |||
| 8860 | assert(Ops[1]->getType()->isVectorTy() && "Scalar base requires an offset")(static_cast <bool> (Ops[1]->getType()->isVectorTy () && "Scalar base requires an offset") ? void (0) : __assert_fail ("Ops[1]->getType()->isVectorTy() && \"Scalar base requires an offset\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 8860, __extension__ __PRETTY_FUNCTION__ )); | |||
| 8861 | Ops.push_back(ConstantInt::get(Int64Ty, 0)); | |||
| 8862 | } | |||
| 8863 | ||||
| 8864 | // Truncation is needed when SrcDataTy != OverloadedTy. In other cases it's | |||
| 8865 | // folded into a nop. | |||
| 8866 | Ops[0] = Builder.CreateTrunc(Ops[0], OverloadedTy); | |||
| 8867 | ||||
| 8868 | // At the ACLE level there's only one predicate type, svbool_t, which is | |||
| 8869 | // mapped to <n x 16 x i1>. However, this might be incompatible with the | |||
| 8870 | // actual type being stored. For example, when storing doubles (i64) the | |||
| 8871 | // predicated should be <n x 2 x i1> instead. At the IR level the type of | |||
| 8872 | // the predicate and the data being stored must match. Cast accordingly. | |||
| 8873 | Ops[1] = EmitSVEPredicateCast(Ops[1], OverloadedTy); | |||
| 8874 | ||||
| 8875 | // For "vector base, scalar index" scale the index so that it becomes a | |||
| 8876 | // scalar offset. | |||
| 8877 | if (!TypeFlags.isByteIndexed() && Ops[2]->getType()->isVectorTy()) { | |||
| 8878 | unsigned BytesPerElt = | |||
| 8879 | OverloadedTy->getElementType()->getScalarSizeInBits() / 8; | |||
| 8880 | Value *Scale = ConstantInt::get(Int64Ty, BytesPerElt); | |||
| 8881 | Ops[3] = Builder.CreateMul(Ops[3], Scale); | |||
| 8882 | } | |||
| 8883 | ||||
| 8884 | return Builder.CreateCall(F, Ops); | |||
| 8885 | } | |||
| 8886 | ||||
| 8887 | Value *CodeGenFunction::EmitSVEGatherPrefetch(const SVETypeFlags &TypeFlags, | |||
| 8888 | SmallVectorImpl<Value *> &Ops, | |||
| 8889 | unsigned IntID) { | |||
| 8890 | // The gather prefetches are overloaded on the vector input - this can either | |||
| 8891 | // be the vector of base addresses or vector of offsets. | |||
| 8892 | auto *OverloadedTy = dyn_cast<llvm::ScalableVectorType>(Ops[1]->getType()); | |||
| 8893 | if (!OverloadedTy) | |||
| 8894 | OverloadedTy = cast<llvm::ScalableVectorType>(Ops[2]->getType()); | |||
| 8895 | ||||
| 8896 | // Cast the predicate from svbool_t to the right number of elements. | |||
| 8897 | Ops[0] = EmitSVEPredicateCast(Ops[0], OverloadedTy); | |||
| 8898 | ||||
| 8899 | // vector + imm addressing modes | |||
| 8900 | if (Ops[1]->getType()->isVectorTy()) { | |||
| 8901 | if (Ops.size() == 3) { | |||
| 8902 | // Pass 0 for 'vector+imm' when the index is omitted. | |||
| 8903 | Ops.push_back(ConstantInt::get(Int64Ty, 0)); | |||
| 8904 | ||||
| 8905 | // The sv_prfop is the last operand in the builtin and IR intrinsic. | |||
| 8906 | std::swap(Ops[2], Ops[3]); | |||
| 8907 | } else { | |||
| 8908 | // Index needs to be passed as scaled offset. | |||
| 8909 | llvm::Type *MemEltTy = SVEBuiltinMemEltTy(TypeFlags); | |||
| 8910 | unsigned BytesPerElt = MemEltTy->getPrimitiveSizeInBits() / 8; | |||
| 8911 | Value *Scale = ConstantInt::get(Int64Ty, BytesPerElt); | |||
| 8912 | Ops[2] = Builder.CreateMul(Ops[2], Scale); | |||
| 8913 | } | |||
| 8914 | } | |||
| 8915 | ||||
| 8916 | Function *F = CGM.getIntrinsic(IntID, OverloadedTy); | |||
| 8917 | return Builder.CreateCall(F, Ops); | |||
| 8918 | } | |||
| 8919 | ||||
| 8920 | Value *CodeGenFunction::EmitSVEStructLoad(const SVETypeFlags &TypeFlags, | |||
| 8921 | SmallVectorImpl<Value*> &Ops, | |||
| 8922 | unsigned IntID) { | |||
| 8923 | llvm::ScalableVectorType *VTy = getSVEType(TypeFlags); | |||
| 8924 | auto VecPtrTy = llvm::PointerType::getUnqual(VTy); | |||
| 8925 | auto EltPtrTy = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 8926 | ||||
| 8927 | unsigned N; | |||
| 8928 | switch (IntID) { | |||
| 8929 | case Intrinsic::aarch64_sve_ld2: | |||
| 8930 | N = 2; | |||
| 8931 | break; | |||
| 8932 | case Intrinsic::aarch64_sve_ld3: | |||
| 8933 | N = 3; | |||
| 8934 | break; | |||
| 8935 | case Intrinsic::aarch64_sve_ld4: | |||
| 8936 | N = 4; | |||
| 8937 | break; | |||
| 8938 | default: | |||
| 8939 | llvm_unreachable("unknown intrinsic!")::llvm::llvm_unreachable_internal("unknown intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 8939); | |||
| 8940 | } | |||
| 8941 | auto RetTy = llvm::VectorType::get(VTy->getElementType(), | |||
| 8942 | VTy->getElementCount() * N); | |||
| 8943 | ||||
| 8944 | Value *Predicate = EmitSVEPredicateCast(Ops[0], VTy); | |||
| 8945 | Value *BasePtr= Builder.CreateBitCast(Ops[1], VecPtrTy); | |||
| 8946 | Value *Offset = Ops.size() > 2 ? Ops[2] : Builder.getInt32(0); | |||
| 8947 | BasePtr = Builder.CreateGEP(VTy, BasePtr, Offset); | |||
| 8948 | BasePtr = Builder.CreateBitCast(BasePtr, EltPtrTy); | |||
| 8949 | ||||
| 8950 | Function *F = CGM.getIntrinsic(IntID, {RetTy, Predicate->getType()}); | |||
| 8951 | return Builder.CreateCall(F, { Predicate, BasePtr }); | |||
| 8952 | } | |||
| 8953 | ||||
| 8954 | Value *CodeGenFunction::EmitSVEStructStore(const SVETypeFlags &TypeFlags, | |||
| 8955 | SmallVectorImpl<Value*> &Ops, | |||
| 8956 | unsigned IntID) { | |||
| 8957 | llvm::ScalableVectorType *VTy = getSVEType(TypeFlags); | |||
| 8958 | auto VecPtrTy = llvm::PointerType::getUnqual(VTy); | |||
| 8959 | auto EltPtrTy = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 8960 | ||||
| 8961 | unsigned N; | |||
| 8962 | switch (IntID) { | |||
| 8963 | case Intrinsic::aarch64_sve_st2: | |||
| 8964 | N = 2; | |||
| 8965 | break; | |||
| 8966 | case Intrinsic::aarch64_sve_st3: | |||
| 8967 | N = 3; | |||
| 8968 | break; | |||
| 8969 | case Intrinsic::aarch64_sve_st4: | |||
| 8970 | N = 4; | |||
| 8971 | break; | |||
| 8972 | default: | |||
| 8973 | llvm_unreachable("unknown intrinsic!")::llvm::llvm_unreachable_internal("unknown intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 8973); | |||
| 8974 | } | |||
| 8975 | auto TupleTy = | |||
| 8976 | llvm::VectorType::get(VTy->getElementType(), VTy->getElementCount() * N); | |||
| 8977 | ||||
| 8978 | Value *Predicate = EmitSVEPredicateCast(Ops[0], VTy); | |||
| 8979 | Value *BasePtr = Builder.CreateBitCast(Ops[1], VecPtrTy); | |||
| 8980 | Value *Offset = Ops.size() > 3 ? Ops[2] : Builder.getInt32(0); | |||
| 8981 | Value *Val = Ops.back(); | |||
| 8982 | BasePtr = Builder.CreateGEP(VTy, BasePtr, Offset); | |||
| 8983 | BasePtr = Builder.CreateBitCast(BasePtr, EltPtrTy); | |||
| 8984 | ||||
| 8985 | // The llvm.aarch64.sve.st2/3/4 intrinsics take legal part vectors, so we | |||
| 8986 | // need to break up the tuple vector. | |||
| 8987 | SmallVector<llvm::Value*, 5> Operands; | |||
| 8988 | Function *FExtr = | |||
| 8989 | CGM.getIntrinsic(Intrinsic::aarch64_sve_tuple_get, {VTy, TupleTy}); | |||
| 8990 | for (unsigned I = 0; I < N; ++I) | |||
| 8991 | Operands.push_back(Builder.CreateCall(FExtr, {Val, Builder.getInt32(I)})); | |||
| 8992 | Operands.append({Predicate, BasePtr}); | |||
| 8993 | ||||
| 8994 | Function *F = CGM.getIntrinsic(IntID, { VTy }); | |||
| 8995 | return Builder.CreateCall(F, Operands); | |||
| 8996 | } | |||
| 8997 | ||||
| 8998 | // SVE2's svpmullb and svpmullt builtins are similar to the svpmullb_pair and | |||
| 8999 | // svpmullt_pair intrinsics, with the exception that their results are bitcast | |||
| 9000 | // to a wider type. | |||
| 9001 | Value *CodeGenFunction::EmitSVEPMull(const SVETypeFlags &TypeFlags, | |||
| 9002 | SmallVectorImpl<Value *> &Ops, | |||
| 9003 | unsigned BuiltinID) { | |||
| 9004 | // Splat scalar operand to vector (intrinsics with _n infix) | |||
| 9005 | if (TypeFlags.hasSplatOperand()) { | |||
| 9006 | unsigned OpNo = TypeFlags.getSplatOperand(); | |||
| 9007 | Ops[OpNo] = EmitSVEDupX(Ops[OpNo]); | |||
| 9008 | } | |||
| 9009 | ||||
| 9010 | // The pair-wise function has a narrower overloaded type. | |||
| 9011 | Function *F = CGM.getIntrinsic(BuiltinID, Ops[0]->getType()); | |||
| 9012 | Value *Call = Builder.CreateCall(F, {Ops[0], Ops[1]}); | |||
| 9013 | ||||
| 9014 | // Now bitcast to the wider result type. | |||
| 9015 | llvm::ScalableVectorType *Ty = getSVEType(TypeFlags); | |||
| 9016 | return EmitSVEReinterpret(Call, Ty); | |||
| 9017 | } | |||
| 9018 | ||||
| 9019 | Value *CodeGenFunction::EmitSVEMovl(const SVETypeFlags &TypeFlags, | |||
| 9020 | ArrayRef<Value *> Ops, unsigned BuiltinID) { | |||
| 9021 | llvm::Type *OverloadedTy = getSVEType(TypeFlags); | |||
| 9022 | Function *F = CGM.getIntrinsic(BuiltinID, OverloadedTy); | |||
| 9023 | return Builder.CreateCall(F, {Ops[0], Builder.getInt32(0)}); | |||
| 9024 | } | |||
| 9025 | ||||
| 9026 | Value *CodeGenFunction::EmitSVEPrefetchLoad(const SVETypeFlags &TypeFlags, | |||
| 9027 | SmallVectorImpl<Value *> &Ops, | |||
| 9028 | unsigned BuiltinID) { | |||
| 9029 | auto *MemEltTy = SVEBuiltinMemEltTy(TypeFlags); | |||
| 9030 | auto *VectorTy = getSVEVectorForElementType(MemEltTy); | |||
| 9031 | auto *MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy); | |||
| 9032 | ||||
| 9033 | Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy); | |||
| 9034 | Value *BasePtr = Ops[1]; | |||
| 9035 | ||||
| 9036 | // Implement the index operand if not omitted. | |||
| 9037 | if (Ops.size() > 3) { | |||
| 9038 | BasePtr = Builder.CreateBitCast(BasePtr, MemoryTy->getPointerTo()); | |||
| 9039 | BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Ops[2]); | |||
| 9040 | } | |||
| 9041 | ||||
| 9042 | // Prefetch intriniscs always expect an i8* | |||
| 9043 | BasePtr = Builder.CreateBitCast(BasePtr, llvm::PointerType::getUnqual(Int8Ty)); | |||
| 9044 | Value *PrfOp = Ops.back(); | |||
| 9045 | ||||
| 9046 | Function *F = CGM.getIntrinsic(BuiltinID, Predicate->getType()); | |||
| 9047 | return Builder.CreateCall(F, {Predicate, BasePtr, PrfOp}); | |||
| 9048 | } | |||
| 9049 | ||||
| 9050 | Value *CodeGenFunction::EmitSVEMaskedLoad(const CallExpr *E, | |||
| 9051 | llvm::Type *ReturnTy, | |||
| 9052 | SmallVectorImpl<Value *> &Ops, | |||
| 9053 | unsigned BuiltinID, | |||
| 9054 | bool IsZExtReturn) { | |||
| 9055 | QualType LangPTy = E->getArg(1)->getType(); | |||
| 9056 | llvm::Type *MemEltTy = CGM.getTypes().ConvertType( | |||
| 9057 | LangPTy->castAs<PointerType>()->getPointeeType()); | |||
| 9058 | ||||
| 9059 | // The vector type that is returned may be different from the | |||
| 9060 | // eventual type loaded from memory. | |||
| 9061 | auto VectorTy = cast<llvm::ScalableVectorType>(ReturnTy); | |||
| 9062 | auto MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy); | |||
| 9063 | ||||
| 9064 | Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy); | |||
| 9065 | Value *BasePtr = Builder.CreateBitCast(Ops[1], MemoryTy->getPointerTo()); | |||
| 9066 | Value *Offset = Ops.size() > 2 ? Ops[2] : Builder.getInt32(0); | |||
| 9067 | BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Offset); | |||
| 9068 | ||||
| 9069 | BasePtr = Builder.CreateBitCast(BasePtr, MemEltTy->getPointerTo()); | |||
| 9070 | Function *F = CGM.getIntrinsic(BuiltinID, MemoryTy); | |||
| 9071 | auto *Load = | |||
| 9072 | cast<llvm::Instruction>(Builder.CreateCall(F, {Predicate, BasePtr})); | |||
| 9073 | auto TBAAInfo = CGM.getTBAAAccessInfo(LangPTy->getPointeeType()); | |||
| 9074 | CGM.DecorateInstructionWithTBAA(Load, TBAAInfo); | |||
| 9075 | ||||
| 9076 | return IsZExtReturn ? Builder.CreateZExt(Load, VectorTy) | |||
| 9077 | : Builder.CreateSExt(Load, VectorTy); | |||
| 9078 | } | |||
| 9079 | ||||
| 9080 | Value *CodeGenFunction::EmitSVEMaskedStore(const CallExpr *E, | |||
| 9081 | SmallVectorImpl<Value *> &Ops, | |||
| 9082 | unsigned BuiltinID) { | |||
| 9083 | QualType LangPTy = E->getArg(1)->getType(); | |||
| 9084 | llvm::Type *MemEltTy = CGM.getTypes().ConvertType( | |||
| 9085 | LangPTy->castAs<PointerType>()->getPointeeType()); | |||
| 9086 | ||||
| 9087 | // The vector type that is stored may be different from the | |||
| 9088 | // eventual type stored to memory. | |||
| 9089 | auto VectorTy = cast<llvm::ScalableVectorType>(Ops.back()->getType()); | |||
| 9090 | auto MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy); | |||
| 9091 | ||||
| 9092 | Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy); | |||
| 9093 | Value *BasePtr = Builder.CreateBitCast(Ops[1], MemoryTy->getPointerTo()); | |||
| 9094 | Value *Offset = Ops.size() == 4 ? Ops[2] : Builder.getInt32(0); | |||
| 9095 | BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Offset); | |||
| 9096 | ||||
| 9097 | // Last value is always the data | |||
| 9098 | llvm::Value *Val = Builder.CreateTrunc(Ops.back(), MemoryTy); | |||
| 9099 | ||||
| 9100 | BasePtr = Builder.CreateBitCast(BasePtr, MemEltTy->getPointerTo()); | |||
| 9101 | Function *F = CGM.getIntrinsic(BuiltinID, MemoryTy); | |||
| 9102 | auto *Store = | |||
| 9103 | cast<llvm::Instruction>(Builder.CreateCall(F, {Val, Predicate, BasePtr})); | |||
| 9104 | auto TBAAInfo = CGM.getTBAAAccessInfo(LangPTy->getPointeeType()); | |||
| 9105 | CGM.DecorateInstructionWithTBAA(Store, TBAAInfo); | |||
| 9106 | return Store; | |||
| 9107 | } | |||
| 9108 | ||||
| 9109 | // Limit the usage of scalable llvm IR generated by the ACLE by using the | |||
| 9110 | // sve dup.x intrinsic instead of IRBuilder::CreateVectorSplat. | |||
| 9111 | Value *CodeGenFunction::EmitSVEDupX(Value *Scalar, llvm::Type *Ty) { | |||
| 9112 | auto F = CGM.getIntrinsic(Intrinsic::aarch64_sve_dup_x, Ty); | |||
| 9113 | return Builder.CreateCall(F, Scalar); | |||
| 9114 | } | |||
| 9115 | ||||
| 9116 | Value *CodeGenFunction::EmitSVEDupX(Value* Scalar) { | |||
| 9117 | return EmitSVEDupX(Scalar, getSVEVectorForElementType(Scalar->getType())); | |||
| 9118 | } | |||
| 9119 | ||||
| 9120 | Value *CodeGenFunction::EmitSVEReinterpret(Value *Val, llvm::Type *Ty) { | |||
| 9121 | // FIXME: For big endian this needs an additional REV, or needs a separate | |||
| 9122 | // intrinsic that is code-generated as a no-op, because the LLVM bitcast | |||
| 9123 | // instruction is defined as 'bitwise' equivalent from memory point of | |||
| 9124 | // view (when storing/reloading), whereas the svreinterpret builtin | |||
| 9125 | // implements bitwise equivalent cast from register point of view. | |||
| 9126 | // LLVM CodeGen for a bitcast must add an explicit REV for big-endian. | |||
| 9127 | return Builder.CreateBitCast(Val, Ty); | |||
| 9128 | } | |||
| 9129 | ||||
| 9130 | static void InsertExplicitZeroOperand(CGBuilderTy &Builder, llvm::Type *Ty, | |||
| 9131 | SmallVectorImpl<Value *> &Ops) { | |||
| 9132 | auto *SplatZero = Constant::getNullValue(Ty); | |||
| 9133 | Ops.insert(Ops.begin(), SplatZero); | |||
| 9134 | } | |||
| 9135 | ||||
| 9136 | static void InsertExplicitUndefOperand(CGBuilderTy &Builder, llvm::Type *Ty, | |||
| 9137 | SmallVectorImpl<Value *> &Ops) { | |||
| 9138 | auto *SplatUndef = UndefValue::get(Ty); | |||
| 9139 | Ops.insert(Ops.begin(), SplatUndef); | |||
| 9140 | } | |||
| 9141 | ||||
| 9142 | SmallVector<llvm::Type *, 2> | |||
| 9143 | CodeGenFunction::getSVEOverloadTypes(const SVETypeFlags &TypeFlags, | |||
| 9144 | llvm::Type *ResultType, | |||
| 9145 | ArrayRef<Value *> Ops) { | |||
| 9146 | if (TypeFlags.isOverloadNone()) | |||
| 9147 | return {}; | |||
| 9148 | ||||
| 9149 | llvm::Type *DefaultType = getSVEType(TypeFlags); | |||
| 9150 | ||||
| 9151 | if (TypeFlags.isOverloadWhile()) | |||
| 9152 | return {DefaultType, Ops[1]->getType()}; | |||
| 9153 | ||||
| 9154 | if (TypeFlags.isOverloadWhileRW()) | |||
| 9155 | return {getSVEPredType(TypeFlags), Ops[0]->getType()}; | |||
| 9156 | ||||
| 9157 | if (TypeFlags.isOverloadCvt() || TypeFlags.isTupleSet()) | |||
| 9158 | return {Ops[0]->getType(), Ops.back()->getType()}; | |||
| 9159 | ||||
| 9160 | if (TypeFlags.isTupleCreate() || TypeFlags.isTupleGet()) | |||
| 9161 | return {ResultType, Ops[0]->getType()}; | |||
| 9162 | ||||
| 9163 | assert(TypeFlags.isOverloadDefault() && "Unexpected value for overloads")(static_cast <bool> (TypeFlags.isOverloadDefault() && "Unexpected value for overloads") ? void (0) : __assert_fail ("TypeFlags.isOverloadDefault() && \"Unexpected value for overloads\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9163, __extension__ __PRETTY_FUNCTION__ )); | |||
| 9164 | return {DefaultType}; | |||
| 9165 | } | |||
| 9166 | ||||
| 9167 | Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, | |||
| 9168 | const CallExpr *E) { | |||
| 9169 | // Find out if any arguments are required to be integer constant expressions. | |||
| 9170 | unsigned ICEArguments = 0; | |||
| 9171 | ASTContext::GetBuiltinTypeError Error; | |||
| 9172 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
| 9173 | assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None && "Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9173, __extension__ __PRETTY_FUNCTION__ )); | |||
| 9174 | ||||
| 9175 | llvm::Type *Ty = ConvertType(E->getType()); | |||
| 9176 | if (BuiltinID >= SVE::BI__builtin_sve_reinterpret_s8_s8 && | |||
| 9177 | BuiltinID <= SVE::BI__builtin_sve_reinterpret_f64_f64) { | |||
| 9178 | Value *Val = EmitScalarExpr(E->getArg(0)); | |||
| 9179 | return EmitSVEReinterpret(Val, Ty); | |||
| 9180 | } | |||
| 9181 | ||||
| 9182 | llvm::SmallVector<Value *, 4> Ops; | |||
| 9183 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | |||
| 9184 | if ((ICEArguments & (1 << i)) == 0) | |||
| 9185 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 9186 | else { | |||
| 9187 | // If this is required to be a constant, constant fold it so that we know | |||
| 9188 | // that the generated intrinsic gets a ConstantInt. | |||
| 9189 | Optional<llvm::APSInt> Result = | |||
| 9190 | E->getArg(i)->getIntegerConstantExpr(getContext()); | |||
| 9191 | assert(Result && "Expected argument to be a constant")(static_cast <bool> (Result && "Expected argument to be a constant" ) ? void (0) : __assert_fail ("Result && \"Expected argument to be a constant\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9191, __extension__ __PRETTY_FUNCTION__ )); | |||
| 9192 | ||||
| 9193 | // Immediates for SVE llvm intrinsics are always 32bit. We can safely | |||
| 9194 | // truncate because the immediate has been range checked and no valid | |||
| 9195 | // immediate requires more than a handful of bits. | |||
| 9196 | *Result = Result->extOrTrunc(32); | |||
| 9197 | Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), *Result)); | |||
| 9198 | } | |||
| 9199 | } | |||
| 9200 | ||||
| 9201 | auto *Builtin = findARMVectorIntrinsicInMap(AArch64SVEIntrinsicMap, BuiltinID, | |||
| 9202 | AArch64SVEIntrinsicsProvenSorted); | |||
| 9203 | SVETypeFlags TypeFlags(Builtin->TypeModifier); | |||
| 9204 | if (TypeFlags.isLoad()) | |||
| 9205 | return EmitSVEMaskedLoad(E, Ty, Ops, Builtin->LLVMIntrinsic, | |||
| 9206 | TypeFlags.isZExtReturn()); | |||
| 9207 | else if (TypeFlags.isStore()) | |||
| 9208 | return EmitSVEMaskedStore(E, Ops, Builtin->LLVMIntrinsic); | |||
| 9209 | else if (TypeFlags.isGatherLoad()) | |||
| 9210 | return EmitSVEGatherLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic); | |||
| 9211 | else if (TypeFlags.isScatterStore()) | |||
| 9212 | return EmitSVEScatterStore(TypeFlags, Ops, Builtin->LLVMIntrinsic); | |||
| 9213 | else if (TypeFlags.isPrefetch()) | |||
| 9214 | return EmitSVEPrefetchLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic); | |||
| 9215 | else if (TypeFlags.isGatherPrefetch()) | |||
| 9216 | return EmitSVEGatherPrefetch(TypeFlags, Ops, Builtin->LLVMIntrinsic); | |||
| 9217 | else if (TypeFlags.isStructLoad()) | |||
| 9218 | return EmitSVEStructLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic); | |||
| 9219 | else if (TypeFlags.isStructStore()) | |||
| 9220 | return EmitSVEStructStore(TypeFlags, Ops, Builtin->LLVMIntrinsic); | |||
| 9221 | else if (TypeFlags.isUndef()) | |||
| 9222 | return UndefValue::get(Ty); | |||
| 9223 | else if (Builtin->LLVMIntrinsic != 0) { | |||
| 9224 | if (TypeFlags.getMergeType() == SVETypeFlags::MergeZeroExp) | |||
| 9225 | InsertExplicitZeroOperand(Builder, Ty, Ops); | |||
| 9226 | ||||
| 9227 | if (TypeFlags.getMergeType() == SVETypeFlags::MergeAnyExp) | |||
| 9228 | InsertExplicitUndefOperand(Builder, Ty, Ops); | |||
| 9229 | ||||
| 9230 | // Some ACLE builtins leave out the argument to specify the predicate | |||
| 9231 | // pattern, which is expected to be expanded to an SV_ALL pattern. | |||
| 9232 | if (TypeFlags.isAppendSVALL()) | |||
| 9233 | Ops.push_back(Builder.getInt32(/*SV_ALL*/ 31)); | |||
| 9234 | if (TypeFlags.isInsertOp1SVALL()) | |||
| 9235 | Ops.insert(&Ops[1], Builder.getInt32(/*SV_ALL*/ 31)); | |||
| 9236 | ||||
| 9237 | // Predicates must match the main datatype. | |||
| 9238 | for (unsigned i = 0, e = Ops.size(); i != e; ++i) | |||
| 9239 | if (auto PredTy = dyn_cast<llvm::VectorType>(Ops[i]->getType())) | |||
| 9240 | if (PredTy->getElementType()->isIntegerTy(1)) | |||
| 9241 | Ops[i] = EmitSVEPredicateCast(Ops[i], getSVEType(TypeFlags)); | |||
| 9242 | ||||
| 9243 | // Splat scalar operand to vector (intrinsics with _n infix) | |||
| 9244 | if (TypeFlags.hasSplatOperand()) { | |||
| 9245 | unsigned OpNo = TypeFlags.getSplatOperand(); | |||
| 9246 | Ops[OpNo] = EmitSVEDupX(Ops[OpNo]); | |||
| 9247 | } | |||
| 9248 | ||||
| 9249 | if (TypeFlags.isReverseCompare()) | |||
| 9250 | std::swap(Ops[1], Ops[2]); | |||
| 9251 | ||||
| 9252 | if (TypeFlags.isReverseUSDOT()) | |||
| 9253 | std::swap(Ops[1], Ops[2]); | |||
| 9254 | ||||
| 9255 | // Predicated intrinsics with _z suffix need a select w/ zeroinitializer. | |||
| 9256 | if (TypeFlags.getMergeType() == SVETypeFlags::MergeZero) { | |||
| 9257 | llvm::Type *OpndTy = Ops[1]->getType(); | |||
| 9258 | auto *SplatZero = Constant::getNullValue(OpndTy); | |||
| 9259 | Function *Sel = CGM.getIntrinsic(Intrinsic::aarch64_sve_sel, OpndTy); | |||
| 9260 | Ops[1] = Builder.CreateCall(Sel, {Ops[0], Ops[1], SplatZero}); | |||
| 9261 | } | |||
| 9262 | ||||
| 9263 | Function *F = CGM.getIntrinsic(Builtin->LLVMIntrinsic, | |||
| 9264 | getSVEOverloadTypes(TypeFlags, Ty, Ops)); | |||
| 9265 | Value *Call = Builder.CreateCall(F, Ops); | |||
| 9266 | ||||
| 9267 | // Predicate results must be converted to svbool_t. | |||
| 9268 | if (auto PredTy = dyn_cast<llvm::VectorType>(Call->getType())) | |||
| 9269 | if (PredTy->getScalarType()->isIntegerTy(1)) | |||
| 9270 | Call = EmitSVEPredicateCast(Call, cast<llvm::ScalableVectorType>(Ty)); | |||
| 9271 | ||||
| 9272 | return Call; | |||
| 9273 | } | |||
| 9274 | ||||
| 9275 | switch (BuiltinID) { | |||
| 9276 | default: | |||
| 9277 | return nullptr; | |||
| 9278 | ||||
| 9279 | case SVE::BI__builtin_sve_svmov_b_z: { | |||
| 9280 | // svmov_b_z(pg, op) <=> svand_b_z(pg, op, op) | |||
| 9281 | SVETypeFlags TypeFlags(Builtin->TypeModifier); | |||
| 9282 | llvm::Type* OverloadedTy = getSVEType(TypeFlags); | |||
| 9283 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_and_z, OverloadedTy); | |||
| 9284 | return Builder.CreateCall(F, {Ops[0], Ops[1], Ops[1]}); | |||
| 9285 | } | |||
| 9286 | ||||
| 9287 | case SVE::BI__builtin_sve_svnot_b_z: { | |||
| 9288 | // svnot_b_z(pg, op) <=> sveor_b_z(pg, op, pg) | |||
| 9289 | SVETypeFlags TypeFlags(Builtin->TypeModifier); | |||
| 9290 | llvm::Type* OverloadedTy = getSVEType(TypeFlags); | |||
| 9291 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_eor_z, OverloadedTy); | |||
| 9292 | return Builder.CreateCall(F, {Ops[0], Ops[1], Ops[0]}); | |||
| 9293 | } | |||
| 9294 | ||||
| 9295 | case SVE::BI__builtin_sve_svmovlb_u16: | |||
| 9296 | case SVE::BI__builtin_sve_svmovlb_u32: | |||
| 9297 | case SVE::BI__builtin_sve_svmovlb_u64: | |||
| 9298 | return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_ushllb); | |||
| 9299 | ||||
| 9300 | case SVE::BI__builtin_sve_svmovlb_s16: | |||
| 9301 | case SVE::BI__builtin_sve_svmovlb_s32: | |||
| 9302 | case SVE::BI__builtin_sve_svmovlb_s64: | |||
| 9303 | return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_sshllb); | |||
| 9304 | ||||
| 9305 | case SVE::BI__builtin_sve_svmovlt_u16: | |||
| 9306 | case SVE::BI__builtin_sve_svmovlt_u32: | |||
| 9307 | case SVE::BI__builtin_sve_svmovlt_u64: | |||
| 9308 | return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_ushllt); | |||
| 9309 | ||||
| 9310 | case SVE::BI__builtin_sve_svmovlt_s16: | |||
| 9311 | case SVE::BI__builtin_sve_svmovlt_s32: | |||
| 9312 | case SVE::BI__builtin_sve_svmovlt_s64: | |||
| 9313 | return EmitSVEMovl(TypeFlags, Ops, Intrinsic::aarch64_sve_sshllt); | |||
| 9314 | ||||
| 9315 | case SVE::BI__builtin_sve_svpmullt_u16: | |||
| 9316 | case SVE::BI__builtin_sve_svpmullt_u64: | |||
| 9317 | case SVE::BI__builtin_sve_svpmullt_n_u16: | |||
| 9318 | case SVE::BI__builtin_sve_svpmullt_n_u64: | |||
| 9319 | return EmitSVEPMull(TypeFlags, Ops, Intrinsic::aarch64_sve_pmullt_pair); | |||
| 9320 | ||||
| 9321 | case SVE::BI__builtin_sve_svpmullb_u16: | |||
| 9322 | case SVE::BI__builtin_sve_svpmullb_u64: | |||
| 9323 | case SVE::BI__builtin_sve_svpmullb_n_u16: | |||
| 9324 | case SVE::BI__builtin_sve_svpmullb_n_u64: | |||
| 9325 | return EmitSVEPMull(TypeFlags, Ops, Intrinsic::aarch64_sve_pmullb_pair); | |||
| 9326 | ||||
| 9327 | case SVE::BI__builtin_sve_svdup_n_b8: | |||
| 9328 | case SVE::BI__builtin_sve_svdup_n_b16: | |||
| 9329 | case SVE::BI__builtin_sve_svdup_n_b32: | |||
| 9330 | case SVE::BI__builtin_sve_svdup_n_b64: { | |||
| 9331 | Value *CmpNE = | |||
| 9332 | Builder.CreateICmpNE(Ops[0], Constant::getNullValue(Ops[0]->getType())); | |||
| 9333 | llvm::ScalableVectorType *OverloadedTy = getSVEType(TypeFlags); | |||
| 9334 | Value *Dup = EmitSVEDupX(CmpNE, OverloadedTy); | |||
| 9335 | return EmitSVEPredicateCast(Dup, cast<llvm::ScalableVectorType>(Ty)); | |||
| 9336 | } | |||
| 9337 | ||||
| 9338 | case SVE::BI__builtin_sve_svdupq_n_b8: | |||
| 9339 | case SVE::BI__builtin_sve_svdupq_n_b16: | |||
| 9340 | case SVE::BI__builtin_sve_svdupq_n_b32: | |||
| 9341 | case SVE::BI__builtin_sve_svdupq_n_b64: | |||
| 9342 | case SVE::BI__builtin_sve_svdupq_n_u8: | |||
| 9343 | case SVE::BI__builtin_sve_svdupq_n_s8: | |||
| 9344 | case SVE::BI__builtin_sve_svdupq_n_u64: | |||
| 9345 | case SVE::BI__builtin_sve_svdupq_n_f64: | |||
| 9346 | case SVE::BI__builtin_sve_svdupq_n_s64: | |||
| 9347 | case SVE::BI__builtin_sve_svdupq_n_u16: | |||
| 9348 | case SVE::BI__builtin_sve_svdupq_n_f16: | |||
| 9349 | case SVE::BI__builtin_sve_svdupq_n_bf16: | |||
| 9350 | case SVE::BI__builtin_sve_svdupq_n_s16: | |||
| 9351 | case SVE::BI__builtin_sve_svdupq_n_u32: | |||
| 9352 | case SVE::BI__builtin_sve_svdupq_n_f32: | |||
| 9353 | case SVE::BI__builtin_sve_svdupq_n_s32: { | |||
| 9354 | // These builtins are implemented by storing each element to an array and using | |||
| 9355 | // ld1rq to materialize a vector. | |||
| 9356 | unsigned NumOpnds = Ops.size(); | |||
| 9357 | ||||
| 9358 | bool IsBoolTy = | |||
| 9359 | cast<llvm::VectorType>(Ty)->getElementType()->isIntegerTy(1); | |||
| 9360 | ||||
| 9361 | // For svdupq_n_b* the element type of is an integer of type 128/numelts, | |||
| 9362 | // so that the compare can use the width that is natural for the expected | |||
| 9363 | // number of predicate lanes. | |||
| 9364 | llvm::Type *EltTy = Ops[0]->getType(); | |||
| 9365 | if (IsBoolTy) | |||
| 9366 | EltTy = IntegerType::get(getLLVMContext(), SVEBitsPerBlock / NumOpnds); | |||
| 9367 | ||||
| 9368 | SmallVector<llvm::Value *, 16> VecOps; | |||
| 9369 | for (unsigned I = 0; I < NumOpnds; ++I) | |||
| 9370 | VecOps.push_back(Builder.CreateZExt(Ops[I], EltTy)); | |||
| 9371 | Value *Vec = BuildVector(VecOps); | |||
| 9372 | ||||
| 9373 | SVETypeFlags TypeFlags(Builtin->TypeModifier); | |||
| 9374 | Value *Pred = EmitSVEAllTruePred(TypeFlags); | |||
| 9375 | ||||
| 9376 | llvm::Type *OverloadedTy = getSVEVectorForElementType(EltTy); | |||
| 9377 | Value *InsertSubVec = Builder.CreateInsertVector( | |||
| 9378 | OverloadedTy, UndefValue::get(OverloadedTy), Vec, Builder.getInt64(0)); | |||
| 9379 | ||||
| 9380 | Function *F = | |||
| 9381 | CGM.getIntrinsic(Intrinsic::aarch64_sve_dupq_lane, OverloadedTy); | |||
| 9382 | Value *DupQLane = | |||
| 9383 | Builder.CreateCall(F, {InsertSubVec, Builder.getInt64(0)}); | |||
| 9384 | ||||
| 9385 | if (!IsBoolTy) | |||
| 9386 | return DupQLane; | |||
| 9387 | ||||
| 9388 | // For svdupq_n_b* we need to add an additional 'cmpne' with '0'. | |||
| 9389 | F = CGM.getIntrinsic(NumOpnds == 2 ? Intrinsic::aarch64_sve_cmpne | |||
| 9390 | : Intrinsic::aarch64_sve_cmpne_wide, | |||
| 9391 | OverloadedTy); | |||
| 9392 | Value *Call = Builder.CreateCall( | |||
| 9393 | F, {Pred, DupQLane, EmitSVEDupX(Builder.getInt64(0))}); | |||
| 9394 | return EmitSVEPredicateCast(Call, cast<llvm::ScalableVectorType>(Ty)); | |||
| 9395 | } | |||
| 9396 | ||||
| 9397 | case SVE::BI__builtin_sve_svpfalse_b: | |||
| 9398 | return ConstantInt::getFalse(Ty); | |||
| 9399 | ||||
| 9400 | case SVE::BI__builtin_sve_svlen_bf16: | |||
| 9401 | case SVE::BI__builtin_sve_svlen_f16: | |||
| 9402 | case SVE::BI__builtin_sve_svlen_f32: | |||
| 9403 | case SVE::BI__builtin_sve_svlen_f64: | |||
| 9404 | case SVE::BI__builtin_sve_svlen_s8: | |||
| 9405 | case SVE::BI__builtin_sve_svlen_s16: | |||
| 9406 | case SVE::BI__builtin_sve_svlen_s32: | |||
| 9407 | case SVE::BI__builtin_sve_svlen_s64: | |||
| 9408 | case SVE::BI__builtin_sve_svlen_u8: | |||
| 9409 | case SVE::BI__builtin_sve_svlen_u16: | |||
| 9410 | case SVE::BI__builtin_sve_svlen_u32: | |||
| 9411 | case SVE::BI__builtin_sve_svlen_u64: { | |||
| 9412 | SVETypeFlags TF(Builtin->TypeModifier); | |||
| 9413 | auto VTy = cast<llvm::VectorType>(getSVEType(TF)); | |||
| 9414 | auto *NumEls = | |||
| 9415 | llvm::ConstantInt::get(Ty, VTy->getElementCount().getKnownMinValue()); | |||
| 9416 | ||||
| 9417 | Function *F = CGM.getIntrinsic(Intrinsic::vscale, Ty); | |||
| 9418 | return Builder.CreateMul(NumEls, Builder.CreateCall(F)); | |||
| 9419 | } | |||
| 9420 | ||||
| 9421 | case SVE::BI__builtin_sve_svtbl2_u8: | |||
| 9422 | case SVE::BI__builtin_sve_svtbl2_s8: | |||
| 9423 | case SVE::BI__builtin_sve_svtbl2_u16: | |||
| 9424 | case SVE::BI__builtin_sve_svtbl2_s16: | |||
| 9425 | case SVE::BI__builtin_sve_svtbl2_u32: | |||
| 9426 | case SVE::BI__builtin_sve_svtbl2_s32: | |||
| 9427 | case SVE::BI__builtin_sve_svtbl2_u64: | |||
| 9428 | case SVE::BI__builtin_sve_svtbl2_s64: | |||
| 9429 | case SVE::BI__builtin_sve_svtbl2_f16: | |||
| 9430 | case SVE::BI__builtin_sve_svtbl2_bf16: | |||
| 9431 | case SVE::BI__builtin_sve_svtbl2_f32: | |||
| 9432 | case SVE::BI__builtin_sve_svtbl2_f64: { | |||
| 9433 | SVETypeFlags TF(Builtin->TypeModifier); | |||
| 9434 | auto VTy = cast<llvm::VectorType>(getSVEType(TF)); | |||
| 9435 | auto TupleTy = llvm::VectorType::getDoubleElementsVectorType(VTy); | |||
| 9436 | Function *FExtr = | |||
| 9437 | CGM.getIntrinsic(Intrinsic::aarch64_sve_tuple_get, {VTy, TupleTy}); | |||
| 9438 | Value *V0 = Builder.CreateCall(FExtr, {Ops[0], Builder.getInt32(0)}); | |||
| 9439 | Value *V1 = Builder.CreateCall(FExtr, {Ops[0], Builder.getInt32(1)}); | |||
| 9440 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_tbl2, VTy); | |||
| 9441 | return Builder.CreateCall(F, {V0, V1, Ops[1]}); | |||
| 9442 | } | |||
| 9443 | ||||
| 9444 | case SVE::BI__builtin_sve_svset_neonq_s8: | |||
| 9445 | case SVE::BI__builtin_sve_svset_neonq_s16: | |||
| 9446 | case SVE::BI__builtin_sve_svset_neonq_s32: | |||
| 9447 | case SVE::BI__builtin_sve_svset_neonq_s64: | |||
| 9448 | case SVE::BI__builtin_sve_svset_neonq_u8: | |||
| 9449 | case SVE::BI__builtin_sve_svset_neonq_u16: | |||
| 9450 | case SVE::BI__builtin_sve_svset_neonq_u32: | |||
| 9451 | case SVE::BI__builtin_sve_svset_neonq_u64: | |||
| 9452 | case SVE::BI__builtin_sve_svset_neonq_f16: | |||
| 9453 | case SVE::BI__builtin_sve_svset_neonq_f32: | |||
| 9454 | case SVE::BI__builtin_sve_svset_neonq_f64: | |||
| 9455 | case SVE::BI__builtin_sve_svset_neonq_bf16: { | |||
| 9456 | return Builder.CreateInsertVector(Ty, Ops[0], Ops[1], Builder.getInt64(0)); | |||
| 9457 | } | |||
| 9458 | ||||
| 9459 | case SVE::BI__builtin_sve_svget_neonq_s8: | |||
| 9460 | case SVE::BI__builtin_sve_svget_neonq_s16: | |||
| 9461 | case SVE::BI__builtin_sve_svget_neonq_s32: | |||
| 9462 | case SVE::BI__builtin_sve_svget_neonq_s64: | |||
| 9463 | case SVE::BI__builtin_sve_svget_neonq_u8: | |||
| 9464 | case SVE::BI__builtin_sve_svget_neonq_u16: | |||
| 9465 | case SVE::BI__builtin_sve_svget_neonq_u32: | |||
| 9466 | case SVE::BI__builtin_sve_svget_neonq_u64: | |||
| 9467 | case SVE::BI__builtin_sve_svget_neonq_f16: | |||
| 9468 | case SVE::BI__builtin_sve_svget_neonq_f32: | |||
| 9469 | case SVE::BI__builtin_sve_svget_neonq_f64: | |||
| 9470 | case SVE::BI__builtin_sve_svget_neonq_bf16: { | |||
| 9471 | return Builder.CreateExtractVector(Ty, Ops[0], Builder.getInt64(0)); | |||
| 9472 | } | |||
| 9473 | ||||
| 9474 | case SVE::BI__builtin_sve_svdup_neonq_s8: | |||
| 9475 | case SVE::BI__builtin_sve_svdup_neonq_s16: | |||
| 9476 | case SVE::BI__builtin_sve_svdup_neonq_s32: | |||
| 9477 | case SVE::BI__builtin_sve_svdup_neonq_s64: | |||
| 9478 | case SVE::BI__builtin_sve_svdup_neonq_u8: | |||
| 9479 | case SVE::BI__builtin_sve_svdup_neonq_u16: | |||
| 9480 | case SVE::BI__builtin_sve_svdup_neonq_u32: | |||
| 9481 | case SVE::BI__builtin_sve_svdup_neonq_u64: | |||
| 9482 | case SVE::BI__builtin_sve_svdup_neonq_f16: | |||
| 9483 | case SVE::BI__builtin_sve_svdup_neonq_f32: | |||
| 9484 | case SVE::BI__builtin_sve_svdup_neonq_f64: | |||
| 9485 | case SVE::BI__builtin_sve_svdup_neonq_bf16: { | |||
| 9486 | Value *Insert = Builder.CreateInsertVector(Ty, UndefValue::get(Ty), Ops[0], | |||
| 9487 | Builder.getInt64(0)); | |||
| 9488 | return Builder.CreateIntrinsic(Intrinsic::aarch64_sve_dupq_lane, {Ty}, | |||
| 9489 | {Insert, Builder.getInt64(0)}); | |||
| 9490 | } | |||
| 9491 | } | |||
| 9492 | ||||
| 9493 | /// Should not happen | |||
| 9494 | return nullptr; | |||
| 9495 | } | |||
| 9496 | ||||
| 9497 | Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID, | |||
| 9498 | const CallExpr *E, | |||
| 9499 | llvm::Triple::ArchType Arch) { | |||
| 9500 | if (BuiltinID >= AArch64::FirstSVEBuiltin && | |||
| 9501 | BuiltinID <= AArch64::LastSVEBuiltin) | |||
| 9502 | return EmitAArch64SVEBuiltinExpr(BuiltinID, E); | |||
| 9503 | ||||
| 9504 | unsigned HintID = static_cast<unsigned>(-1); | |||
| 9505 | switch (BuiltinID) { | |||
| 9506 | default: break; | |||
| 9507 | case AArch64::BI__builtin_arm_nop: | |||
| 9508 | HintID = 0; | |||
| 9509 | break; | |||
| 9510 | case AArch64::BI__builtin_arm_yield: | |||
| 9511 | case AArch64::BI__yield: | |||
| 9512 | HintID = 1; | |||
| 9513 | break; | |||
| 9514 | case AArch64::BI__builtin_arm_wfe: | |||
| 9515 | case AArch64::BI__wfe: | |||
| 9516 | HintID = 2; | |||
| 9517 | break; | |||
| 9518 | case AArch64::BI__builtin_arm_wfi: | |||
| 9519 | case AArch64::BI__wfi: | |||
| 9520 | HintID = 3; | |||
| 9521 | break; | |||
| 9522 | case AArch64::BI__builtin_arm_sev: | |||
| 9523 | case AArch64::BI__sev: | |||
| 9524 | HintID = 4; | |||
| 9525 | break; | |||
| 9526 | case AArch64::BI__builtin_arm_sevl: | |||
| 9527 | case AArch64::BI__sevl: | |||
| 9528 | HintID = 5; | |||
| 9529 | break; | |||
| 9530 | } | |||
| 9531 | ||||
| 9532 | if (HintID != static_cast<unsigned>(-1)) { | |||
| 9533 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_hint); | |||
| 9534 | return Builder.CreateCall(F, llvm::ConstantInt::get(Int32Ty, HintID)); | |||
| 9535 | } | |||
| 9536 | ||||
| 9537 | if (BuiltinID == AArch64::BI__builtin_arm_prefetch) { | |||
| 9538 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
| 9539 | Value *RW = EmitScalarExpr(E->getArg(1)); | |||
| 9540 | Value *CacheLevel = EmitScalarExpr(E->getArg(2)); | |||
| 9541 | Value *RetentionPolicy = EmitScalarExpr(E->getArg(3)); | |||
| 9542 | Value *IsData = EmitScalarExpr(E->getArg(4)); | |||
| 9543 | ||||
| 9544 | Value *Locality = nullptr; | |||
| 9545 | if (cast<llvm::ConstantInt>(RetentionPolicy)->isZero()) { | |||
| 9546 | // Temporal fetch, needs to convert cache level to locality. | |||
| 9547 | Locality = llvm::ConstantInt::get(Int32Ty, | |||
| 9548 | -cast<llvm::ConstantInt>(CacheLevel)->getValue() + 3); | |||
| 9549 | } else { | |||
| 9550 | // Streaming fetch. | |||
| 9551 | Locality = llvm::ConstantInt::get(Int32Ty, 0); | |||
| 9552 | } | |||
| 9553 | ||||
| 9554 | // FIXME: We need AArch64 specific LLVM intrinsic if we want to specify | |||
| 9555 | // PLDL3STRM or PLDL2STRM. | |||
| 9556 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
| 9557 | return Builder.CreateCall(F, {Address, RW, Locality, IsData}); | |||
| 9558 | } | |||
| 9559 | ||||
| 9560 | if (BuiltinID == AArch64::BI__builtin_arm_rbit) { | |||
| 9561 | assert((getContext().getTypeSize(E->getType()) == 32) &&(static_cast <bool> ((getContext().getTypeSize(E->getType ()) == 32) && "rbit of unusual size!") ? void (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9562, __extension__ __PRETTY_FUNCTION__ )) | |||
| 9562 | "rbit of unusual size!")(static_cast <bool> ((getContext().getTypeSize(E->getType ()) == 32) && "rbit of unusual size!") ? void (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9562, __extension__ __PRETTY_FUNCTION__ )); | |||
| 9563 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9564 | return Builder.CreateCall( | |||
| 9565 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); | |||
| 9566 | } | |||
| 9567 | if (BuiltinID == AArch64::BI__builtin_arm_rbit64) { | |||
| 9568 | assert((getContext().getTypeSize(E->getType()) == 64) &&(static_cast <bool> ((getContext().getTypeSize(E->getType ()) == 64) && "rbit of unusual size!") ? void (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9569, __extension__ __PRETTY_FUNCTION__ )) | |||
| 9569 | "rbit of unusual size!")(static_cast <bool> ((getContext().getTypeSize(E->getType ()) == 64) && "rbit of unusual size!") ? void (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9569, __extension__ __PRETTY_FUNCTION__ )); | |||
| 9570 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9571 | return Builder.CreateCall( | |||
| 9572 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); | |||
| 9573 | } | |||
| 9574 | ||||
| 9575 | if (BuiltinID == AArch64::BI__builtin_arm_cls) { | |||
| 9576 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9577 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_cls), Arg, | |||
| 9578 | "cls"); | |||
| 9579 | } | |||
| 9580 | if (BuiltinID == AArch64::BI__builtin_arm_cls64) { | |||
| 9581 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9582 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_cls64), Arg, | |||
| 9583 | "cls"); | |||
| 9584 | } | |||
| 9585 | ||||
| 9586 | if (BuiltinID == AArch64::BI__builtin_arm_frint32zf || | |||
| 9587 | BuiltinID == AArch64::BI__builtin_arm_frint32z) { | |||
| 9588 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9589 | llvm::Type *Ty = Arg->getType(); | |||
| 9590 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint32z, Ty), | |||
| 9591 | Arg, "frint32z"); | |||
| 9592 | } | |||
| 9593 | ||||
| 9594 | if (BuiltinID == AArch64::BI__builtin_arm_frint64zf || | |||
| 9595 | BuiltinID == AArch64::BI__builtin_arm_frint64z) { | |||
| 9596 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9597 | llvm::Type *Ty = Arg->getType(); | |||
| 9598 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint64z, Ty), | |||
| 9599 | Arg, "frint64z"); | |||
| 9600 | } | |||
| 9601 | ||||
| 9602 | if (BuiltinID == AArch64::BI__builtin_arm_frint32xf || | |||
| 9603 | BuiltinID == AArch64::BI__builtin_arm_frint32x) { | |||
| 9604 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9605 | llvm::Type *Ty = Arg->getType(); | |||
| 9606 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint32x, Ty), | |||
| 9607 | Arg, "frint32x"); | |||
| 9608 | } | |||
| 9609 | ||||
| 9610 | if (BuiltinID == AArch64::BI__builtin_arm_frint64xf || | |||
| 9611 | BuiltinID == AArch64::BI__builtin_arm_frint64x) { | |||
| 9612 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9613 | llvm::Type *Ty = Arg->getType(); | |||
| 9614 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_frint64x, Ty), | |||
| 9615 | Arg, "frint64x"); | |||
| 9616 | } | |||
| 9617 | ||||
| 9618 | if (BuiltinID == AArch64::BI__builtin_arm_jcvt) { | |||
| 9619 | assert((getContext().getTypeSize(E->getType()) == 32) &&(static_cast <bool> ((getContext().getTypeSize(E->getType ()) == 32) && "__jcvt of unusual size!") ? void (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"__jcvt of unusual size!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9620, __extension__ __PRETTY_FUNCTION__ )) | |||
| 9620 | "__jcvt of unusual size!")(static_cast <bool> ((getContext().getTypeSize(E->getType ()) == 32) && "__jcvt of unusual size!") ? void (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"__jcvt of unusual size!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9620, __extension__ __PRETTY_FUNCTION__ )); | |||
| 9621 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); | |||
| 9622 | return Builder.CreateCall( | |||
| 9623 | CGM.getIntrinsic(Intrinsic::aarch64_fjcvtzs), Arg); | |||
| 9624 | } | |||
| 9625 | ||||
| 9626 | if (BuiltinID == AArch64::BI__builtin_arm_ld64b || | |||
| 9627 | BuiltinID == AArch64::BI__builtin_arm_st64b || | |||
| 9628 | BuiltinID == AArch64::BI__builtin_arm_st64bv || | |||
| 9629 | BuiltinID == AArch64::BI__builtin_arm_st64bv0) { | |||
| 9630 | llvm::Value *MemAddr = EmitScalarExpr(E->getArg(0)); | |||
| 9631 | llvm::Value *ValPtr = EmitScalarExpr(E->getArg(1)); | |||
| 9632 | ||||
| 9633 | if (BuiltinID == AArch64::BI__builtin_arm_ld64b) { | |||
| 9634 | // Load from the address via an LLVM intrinsic, receiving a | |||
| 9635 | // tuple of 8 i64 words, and store each one to ValPtr. | |||
| 9636 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_ld64b); | |||
| 9637 | llvm::Value *Val = Builder.CreateCall(F, MemAddr); | |||
| 9638 | llvm::Value *ToRet; | |||
| 9639 | for (size_t i = 0; i < 8; i++) { | |||
| 9640 | llvm::Value *ValOffsetPtr = | |||
| 9641 | Builder.CreateGEP(Int64Ty, ValPtr, Builder.getInt32(i)); | |||
| 9642 | Address Addr = | |||
| 9643 | Address(ValOffsetPtr, Int64Ty, CharUnits::fromQuantity(8)); | |||
| 9644 | ToRet = Builder.CreateStore(Builder.CreateExtractValue(Val, i), Addr); | |||
| 9645 | } | |||
| 9646 | return ToRet; | |||
| 9647 | } else { | |||
| 9648 | // Load 8 i64 words from ValPtr, and store them to the address | |||
| 9649 | // via an LLVM intrinsic. | |||
| 9650 | SmallVector<llvm::Value *, 9> Args; | |||
| 9651 | Args.push_back(MemAddr); | |||
| 9652 | for (size_t i = 0; i < 8; i++) { | |||
| 9653 | llvm::Value *ValOffsetPtr = | |||
| 9654 | Builder.CreateGEP(Int64Ty, ValPtr, Builder.getInt32(i)); | |||
| 9655 | Address Addr = | |||
| 9656 | Address(ValOffsetPtr, Int64Ty, CharUnits::fromQuantity(8)); | |||
| 9657 | Args.push_back(Builder.CreateLoad(Addr)); | |||
| 9658 | } | |||
| 9659 | ||||
| 9660 | auto Intr = (BuiltinID == AArch64::BI__builtin_arm_st64b | |||
| 9661 | ? Intrinsic::aarch64_st64b | |||
| 9662 | : BuiltinID == AArch64::BI__builtin_arm_st64bv | |||
| 9663 | ? Intrinsic::aarch64_st64bv | |||
| 9664 | : Intrinsic::aarch64_st64bv0); | |||
| 9665 | Function *F = CGM.getIntrinsic(Intr); | |||
| 9666 | return Builder.CreateCall(F, Args); | |||
| 9667 | } | |||
| 9668 | } | |||
| 9669 | ||||
| 9670 | if (BuiltinID == AArch64::BI__builtin_arm_rndr || | |||
| 9671 | BuiltinID == AArch64::BI__builtin_arm_rndrrs) { | |||
| 9672 | ||||
| 9673 | auto Intr = (BuiltinID == AArch64::BI__builtin_arm_rndr | |||
| 9674 | ? Intrinsic::aarch64_rndr | |||
| 9675 | : Intrinsic::aarch64_rndrrs); | |||
| 9676 | Function *F = CGM.getIntrinsic(Intr); | |||
| 9677 | llvm::Value *Val = Builder.CreateCall(F); | |||
| 9678 | Value *RandomValue = Builder.CreateExtractValue(Val, 0); | |||
| 9679 | Value *Status = Builder.CreateExtractValue(Val, 1); | |||
| 9680 | ||||
| 9681 | Address MemAddress = EmitPointerWithAlignment(E->getArg(0)); | |||
| 9682 | Builder.CreateStore(RandomValue, MemAddress); | |||
| 9683 | Status = Builder.CreateZExt(Status, Int32Ty); | |||
| 9684 | return Status; | |||
| 9685 | } | |||
| 9686 | ||||
| 9687 | if (BuiltinID == AArch64::BI__clear_cache) { | |||
| 9688 | assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")(static_cast <bool> (E->getNumArgs() == 2 && "__clear_cache takes 2 arguments") ? void (0) : __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 9688, __extension__ __PRETTY_FUNCTION__ )); | |||
| 9689 | const FunctionDecl *FD = E->getDirectCallee(); | |||
| 9690 | Value *Ops[2]; | |||
| 9691 | for (unsigned i = 0; i < 2; i++) | |||
| 9692 | Ops[i] = EmitScalarExpr(E->getArg(i)); | |||
| 9693 | llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType()); | |||
| 9694 | llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty); | |||
| 9695 | StringRef Name = FD->getName(); | |||
| 9696 | return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops); | |||
| 9697 | } | |||
| 9698 | ||||
| 9699 | if ((BuiltinID == AArch64::BI__builtin_arm_ldrex || | |||
| 9700 | BuiltinID == AArch64::BI__builtin_arm_ldaex) && | |||
| 9701 | getContext().getTypeSize(E->getType()) == 128) { | |||
| 9702 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex | |||
| 9703 | ? Intrinsic::aarch64_ldaxp | |||
| 9704 | : Intrinsic::aarch64_ldxp); | |||
| 9705 | ||||
| 9706 | Value *LdPtr = EmitScalarExpr(E->getArg(0)); | |||
| 9707 | Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy), | |||
| 9708 | "ldxp"); | |||
| 9709 | ||||
| 9710 | Value *Val0 = Builder.CreateExtractValue(Val, 1); | |||
| 9711 | Value *Val1 = Builder.CreateExtractValue(Val, 0); | |||
| 9712 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); | |||
| 9713 | Val0 = Builder.CreateZExt(Val0, Int128Ty); | |||
| 9714 | Val1 = Builder.CreateZExt(Val1, Int128Ty); | |||
| 9715 | ||||
| 9716 | Value *ShiftCst = llvm::ConstantInt::get(Int128Ty, 64); | |||
| 9717 | Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */); | |||
| 9718 | Val = Builder.CreateOr(Val, Val1); | |||
| 9719 | return Builder.CreateBitCast(Val, ConvertType(E->getType())); | |||
| 9720 | } else if (BuiltinID == AArch64::BI__builtin_arm_ldrex || | |||
| 9721 | BuiltinID == AArch64::BI__builtin_arm_ldaex) { | |||
| 9722 | Value *LoadAddr = EmitScalarExpr(E->getArg(0)); | |||
| 9723 | ||||
| 9724 | QualType Ty = E->getType(); | |||
| 9725 | llvm::Type *RealResTy = ConvertType(Ty); | |||
| 9726 | llvm::Type *IntTy = | |||
| 9727 | llvm::IntegerType::get(getLLVMContext(), getContext().getTypeSize(Ty)); | |||
| 9728 | llvm::Type *PtrTy = IntTy->getPointerTo(); | |||
| 9729 | LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy); | |||
| 9730 | ||||
| 9731 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex | |||
| 9732 | ? Intrinsic::aarch64_ldaxr | |||
| 9733 | : Intrinsic::aarch64_ldxr, | |||
| 9734 | PtrTy); | |||
| 9735 | CallInst *Val = Builder.CreateCall(F, LoadAddr, "ldxr"); | |||
| 9736 | Val->addParamAttr( | |||
| 9737 | 0, Attribute::get(getLLVMContext(), Attribute::ElementType, IntTy)); | |||
| 9738 | ||||
| 9739 | if (RealResTy->isPointerTy()) | |||
| 9740 | return Builder.CreateIntToPtr(Val, RealResTy); | |||
| 9741 | ||||
| 9742 | llvm::Type *IntResTy = llvm::IntegerType::get( | |||
| 9743 | getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy)); | |||
| 9744 | return Builder.CreateBitCast(Builder.CreateTruncOrBitCast(Val, IntResTy), | |||
| 9745 | RealResTy); | |||
| 9746 | } | |||
| 9747 | ||||
| 9748 | if ((BuiltinID == AArch64::BI__builtin_arm_strex || | |||
| 9749 | BuiltinID == AArch64::BI__builtin_arm_stlex) && | |||
| 9750 | getContext().getTypeSize(E->getArg(0)->getType()) == 128) { | |||
| 9751 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex | |||
| 9752 | ? Intrinsic::aarch64_stlxp | |||
| 9753 | : Intrinsic::aarch64_stxp); | |||
| 9754 | llvm::Type *STy = llvm::StructType::get(Int64Ty, Int64Ty); | |||
| 9755 | ||||
| 9756 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); | |||
| 9757 | EmitAnyExprToMem(E->getArg(0), Tmp, Qualifiers(), /*init*/ true); | |||
| 9758 | ||||
| 9759 | Tmp = Builder.CreateElementBitCast(Tmp, STy); | |||
| 9760 | llvm::Value *Val = Builder.CreateLoad(Tmp); | |||
| 9761 | ||||
| 9762 | Value *Arg0 = Builder.CreateExtractValue(Val, 0); | |||
| 9763 | Value *Arg1 = Builder.CreateExtractValue(Val, 1); | |||
| 9764 | Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), | |||
| 9765 | Int8PtrTy); | |||
| 9766 | return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "stxp"); | |||
| 9767 | } | |||
| 9768 | ||||
| 9769 | if (BuiltinID == AArch64::BI__builtin_arm_strex || | |||
| 9770 | BuiltinID == AArch64::BI__builtin_arm_stlex) { | |||
| 9771 | Value *StoreVal = EmitScalarExpr(E->getArg(0)); | |||
| 9772 | Value *StoreAddr = EmitScalarExpr(E->getArg(1)); | |||
| 9773 | ||||
| 9774 | QualType Ty = E->getArg(0)->getType(); | |||
| 9775 | llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(), | |||
| 9776 | getContext().getTypeSize(Ty)); | |||
| 9777 | StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo()); | |||
| 9778 | ||||
| 9779 | if (StoreVal->getType()->isPointerTy()) | |||
| 9780 | StoreVal = Builder.CreatePtrToInt(StoreVal, Int64Ty); | |||
| 9781 | else { | |||
| 9782 | llvm::Type *IntTy = llvm::IntegerType::get( | |||
| 9783 | getLLVMContext(), | |||
| 9784 | CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType())); | |||
| 9785 | StoreVal = Builder.CreateBitCast(StoreVal, IntTy); | |||
| 9786 | StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int64Ty); | |||
| 9787 | } | |||
| 9788 | ||||
| 9789 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex | |||
| 9790 | ? Intrinsic::aarch64_stlxr | |||
| 9791 | : Intrinsic::aarch64_stxr, | |||
| 9792 | StoreAddr->getType()); | |||
| 9793 | CallInst *CI = Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr"); | |||
| 9794 | CI->addParamAttr( | |||
| 9795 | 1, Attribute::get(getLLVMContext(), Attribute::ElementType, StoreTy)); | |||
| 9796 | return CI; | |||
| 9797 | } | |||
| 9798 | ||||
| 9799 | if (BuiltinID == AArch64::BI__getReg) { | |||
| 9800 | Expr::EvalResult Result; | |||
| 9801 | if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext())) | |||
| 9802 | llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant" , "clang/lib/CodeGen/CGBuiltin.cpp", 9802); | |||
| 9803 | ||||
| 9804 | llvm::APSInt Value = Result.Val.getInt(); | |||
| 9805 | LLVMContext &Context = CGM.getLLVMContext(); | |||
| 9806 | std::string Reg = Value == 31 ? "sp" : "x" + toString(Value, 10); | |||
| 9807 | ||||
| 9808 | llvm::Metadata *Ops[] = {llvm::MDString::get(Context, Reg)}; | |||
| 9809 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); | |||
| 9810 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); | |||
| 9811 | ||||
| 9812 | llvm::Function *F = | |||
| 9813 | CGM.getIntrinsic(llvm::Intrinsic::read_register, {Int64Ty}); | |||
| 9814 | return Builder.CreateCall(F, Metadata); | |||
| 9815 | } | |||
| 9816 | ||||
| 9817 | if (BuiltinID == AArch64::BI__builtin_arm_clrex) { | |||
| 9818 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_clrex); | |||
| 9819 | return Builder.CreateCall(F); | |||
| 9820 | } | |||
| 9821 | ||||
| 9822 | if (BuiltinID == AArch64::BI_ReadWriteBarrier) | |||
| 9823 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, | |||
| 9824 | llvm::SyncScope::SingleThread); | |||
| 9825 | ||||
| 9826 | // CRC32 | |||
| 9827 | Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic; | |||
| 9828 | switch (BuiltinID) { | |||
| 9829 | case AArch64::BI__builtin_arm_crc32b: | |||
| 9830 | CRCIntrinsicID = Intrinsic::aarch64_crc32b; break; | |||
| 9831 | case AArch64::BI__builtin_arm_crc32cb: | |||
| 9832 | CRCIntrinsicID = Intrinsic::aarch64_crc32cb; break; | |||
| 9833 | case AArch64::BI__builtin_arm_crc32h: | |||
| 9834 | CRCIntrinsicID = Intrinsic::aarch64_crc32h; break; | |||
| 9835 | case AArch64::BI__builtin_arm_crc32ch: | |||
| 9836 | CRCIntrinsicID = Intrinsic::aarch64_crc32ch; break; | |||
| 9837 | case AArch64::BI__builtin_arm_crc32w: | |||
| 9838 | CRCIntrinsicID = Intrinsic::aarch64_crc32w; break; | |||
| 9839 | case AArch64::BI__builtin_arm_crc32cw: | |||
| 9840 | CRCIntrinsicID = Intrinsic::aarch64_crc32cw; break; | |||
| 9841 | case AArch64::BI__builtin_arm_crc32d: | |||
| 9842 | CRCIntrinsicID = Intrinsic::aarch64_crc32x; break; | |||
| 9843 | case AArch64::BI__builtin_arm_crc32cd: | |||
| 9844 | CRCIntrinsicID = Intrinsic::aarch64_crc32cx; break; | |||
| 9845 | } | |||
| 9846 | ||||
| 9847 | if (CRCIntrinsicID != Intrinsic::not_intrinsic) { | |||
| 9848 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
| 9849 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
| 9850 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); | |||
| 9851 | ||||
| 9852 | llvm::Type *DataTy = F->getFunctionType()->getParamType(1); | |||
| 9853 | Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy); | |||
| 9854 | ||||
| 9855 | return Builder.CreateCall(F, {Arg0, Arg1}); | |||
| 9856 | } | |||
| 9857 | ||||
| 9858 | // Memory Operations (MOPS) | |||
| 9859 | if (BuiltinID == AArch64::BI__builtin_arm_mops_memset_tag) { | |||
| 9860 | Value *Dst = EmitScalarExpr(E->getArg(0)); | |||
| 9861 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 9862 | Value *Size = EmitScalarExpr(E->getArg(2)); | |||
| 9863 | Dst = Builder.CreatePointerCast(Dst, Int8PtrTy); | |||
| 9864 | Val = Builder.CreateTrunc(Val, Int8Ty); | |||
| 9865 | Size = Builder.CreateIntCast(Size, Int64Ty, false); | |||
| 9866 | return Builder.CreateCall( | |||
| 9867 | CGM.getIntrinsic(Intrinsic::aarch64_mops_memset_tag), {Dst, Val, Size}); | |||
| 9868 | } | |||
| 9869 | ||||
| 9870 | // Memory Tagging Extensions (MTE) Intrinsics | |||
| 9871 | Intrinsic::ID MTEIntrinsicID = Intrinsic::not_intrinsic; | |||
| 9872 | switch (BuiltinID) { | |||
| 9873 | case AArch64::BI__builtin_arm_irg: | |||
| 9874 | MTEIntrinsicID = Intrinsic::aarch64_irg; break; | |||
| 9875 | case AArch64::BI__builtin_arm_addg: | |||
| 9876 | MTEIntrinsicID = Intrinsic::aarch64_addg; break; | |||
| 9877 | case AArch64::BI__builtin_arm_gmi: | |||
| 9878 | MTEIntrinsicID = Intrinsic::aarch64_gmi; break; | |||
| 9879 | case AArch64::BI__builtin_arm_ldg: | |||
| 9880 | MTEIntrinsicID = Intrinsic::aarch64_ldg; break; | |||
| 9881 | case AArch64::BI__builtin_arm_stg: | |||
| 9882 | MTEIntrinsicID = Intrinsic::aarch64_stg; break; | |||
| 9883 | case AArch64::BI__builtin_arm_subp: | |||
| 9884 | MTEIntrinsicID = Intrinsic::aarch64_subp; break; | |||
| 9885 | } | |||
| 9886 | ||||
| 9887 | if (MTEIntrinsicID != Intrinsic::not_intrinsic) { | |||
| 9888 | llvm::Type *T = ConvertType(E->getType()); | |||
| 9889 | ||||
| 9890 | if (MTEIntrinsicID == Intrinsic::aarch64_irg) { | |||
| 9891 | Value *Pointer = EmitScalarExpr(E->getArg(0)); | |||
| 9892 | Value *Mask = EmitScalarExpr(E->getArg(1)); | |||
| 9893 | ||||
| 9894 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); | |||
| 9895 | Mask = Builder.CreateZExt(Mask, Int64Ty); | |||
| 9896 | Value *RV = Builder.CreateCall( | |||
| 9897 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, Mask}); | |||
| 9898 | return Builder.CreatePointerCast(RV, T); | |||
| 9899 | } | |||
| 9900 | if (MTEIntrinsicID == Intrinsic::aarch64_addg) { | |||
| 9901 | Value *Pointer = EmitScalarExpr(E->getArg(0)); | |||
| 9902 | Value *TagOffset = EmitScalarExpr(E->getArg(1)); | |||
| 9903 | ||||
| 9904 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); | |||
| 9905 | TagOffset = Builder.CreateZExt(TagOffset, Int64Ty); | |||
| 9906 | Value *RV = Builder.CreateCall( | |||
| 9907 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, TagOffset}); | |||
| 9908 | return Builder.CreatePointerCast(RV, T); | |||
| 9909 | } | |||
| 9910 | if (MTEIntrinsicID == Intrinsic::aarch64_gmi) { | |||
| 9911 | Value *Pointer = EmitScalarExpr(E->getArg(0)); | |||
| 9912 | Value *ExcludedMask = EmitScalarExpr(E->getArg(1)); | |||
| 9913 | ||||
| 9914 | ExcludedMask = Builder.CreateZExt(ExcludedMask, Int64Ty); | |||
| 9915 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); | |||
| 9916 | return Builder.CreateCall( | |||
| 9917 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, ExcludedMask}); | |||
| 9918 | } | |||
| 9919 | // Although it is possible to supply a different return | |||
| 9920 | // address (first arg) to this intrinsic, for now we set | |||
| 9921 | // return address same as input address. | |||
| 9922 | if (MTEIntrinsicID == Intrinsic::aarch64_ldg) { | |||
| 9923 | Value *TagAddress = EmitScalarExpr(E->getArg(0)); | |||
| 9924 | TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy); | |||
| 9925 | Value *RV = Builder.CreateCall( | |||
| 9926 | CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress}); | |||
| 9927 | return Builder.CreatePointerCast(RV, T); | |||
| 9928 | } | |||
| 9929 | // Although it is possible to supply a different tag (to set) | |||
| 9930 | // to this intrinsic (as first arg), for now we supply | |||
| 9931 | // the tag that is in input address arg (common use case). | |||
| 9932 | if (MTEIntrinsicID == Intrinsic::aarch64_stg) { | |||
| 9933 | Value *TagAddress = EmitScalarExpr(E->getArg(0)); | |||
| 9934 | TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy); | |||
| 9935 | return Builder.CreateCall( | |||
| 9936 | CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress}); | |||
| 9937 | } | |||
| 9938 | if (MTEIntrinsicID == Intrinsic::aarch64_subp) { | |||
| 9939 | Value *PointerA = EmitScalarExpr(E->getArg(0)); | |||
| 9940 | Value *PointerB = EmitScalarExpr(E->getArg(1)); | |||
| 9941 | PointerA = Builder.CreatePointerCast(PointerA, Int8PtrTy); | |||
| 9942 | PointerB = Builder.CreatePointerCast(PointerB, Int8PtrTy); | |||
| 9943 | return Builder.CreateCall( | |||
| 9944 | CGM.getIntrinsic(MTEIntrinsicID), {PointerA, PointerB}); | |||
| 9945 | } | |||
| 9946 | } | |||
| 9947 | ||||
| 9948 | if (BuiltinID == AArch64::BI__builtin_arm_rsr || | |||
| 9949 | BuiltinID == AArch64::BI__builtin_arm_rsr64 || | |||
| 9950 | BuiltinID == AArch64::BI__builtin_arm_rsrp || | |||
| 9951 | BuiltinID == AArch64::BI__builtin_arm_wsr || | |||
| 9952 | BuiltinID == AArch64::BI__builtin_arm_wsr64 || | |||
| 9953 | BuiltinID == AArch64::BI__builtin_arm_wsrp) { | |||
| 9954 | ||||
| 9955 | SpecialRegisterAccessKind AccessKind = Write; | |||
| 9956 | if (BuiltinID == AArch64::BI__builtin_arm_rsr || | |||
| 9957 | BuiltinID == AArch64::BI__builtin_arm_rsr64 || | |||
| 9958 | BuiltinID == AArch64::BI__builtin_arm_rsrp) | |||
| 9959 | AccessKind = VolatileRead; | |||
| 9960 | ||||
| 9961 | bool IsPointerBuiltin = BuiltinID == AArch64::BI__builtin_arm_rsrp || | |||
| 9962 | BuiltinID == AArch64::BI__builtin_arm_wsrp; | |||
| 9963 | ||||
| 9964 | bool Is64Bit = BuiltinID != AArch64::BI__builtin_arm_rsr && | |||
| 9965 | BuiltinID != AArch64::BI__builtin_arm_wsr; | |||
| 9966 | ||||
| 9967 | llvm::Type *ValueType; | |||
| 9968 | llvm::Type *RegisterType = Int64Ty; | |||
| 9969 | if (IsPointerBuiltin) { | |||
| 9970 | ValueType = VoidPtrTy; | |||
| 9971 | } else if (Is64Bit) { | |||
| 9972 | ValueType = Int64Ty; | |||
| 9973 | } else { | |||
| 9974 | ValueType = Int32Ty; | |||
| 9975 | } | |||
| 9976 | ||||
| 9977 | return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, | |||
| 9978 | AccessKind); | |||
| 9979 | } | |||
| 9980 | ||||
| 9981 | if (BuiltinID == AArch64::BI_ReadStatusReg || | |||
| 9982 | BuiltinID == AArch64::BI_WriteStatusReg) { | |||
| 9983 | LLVMContext &Context = CGM.getLLVMContext(); | |||
| 9984 | ||||
| 9985 | unsigned SysReg = | |||
| 9986 | E->getArg(0)->EvaluateKnownConstInt(getContext()).getZExtValue(); | |||
| 9987 | ||||
| 9988 | std::string SysRegStr; | |||
| 9989 | llvm::raw_string_ostream(SysRegStr) << | |||
| 9990 | ((1 << 1) | ((SysReg >> 14) & 1)) << ":" << | |||
| 9991 | ((SysReg >> 11) & 7) << ":" << | |||
| 9992 | ((SysReg >> 7) & 15) << ":" << | |||
| 9993 | ((SysReg >> 3) & 15) << ":" << | |||
| 9994 | ( SysReg & 7); | |||
| 9995 | ||||
| 9996 | llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysRegStr) }; | |||
| 9997 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); | |||
| 9998 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); | |||
| 9999 | ||||
| 10000 | llvm::Type *RegisterType = Int64Ty; | |||
| 10001 | llvm::Type *Types[] = { RegisterType }; | |||
| 10002 | ||||
| 10003 | if (BuiltinID == AArch64::BI_ReadStatusReg) { | |||
| 10004 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types); | |||
| 10005 | ||||
| 10006 | return Builder.CreateCall(F, Metadata); | |||
| 10007 | } | |||
| 10008 | ||||
| 10009 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types); | |||
| 10010 | llvm::Value *ArgValue = EmitScalarExpr(E->getArg(1)); | |||
| 10011 | ||||
| 10012 | return Builder.CreateCall(F, { Metadata, ArgValue }); | |||
| 10013 | } | |||
| 10014 | ||||
| 10015 | if (BuiltinID == AArch64::BI_AddressOfReturnAddress) { | |||
| 10016 | llvm::Function *F = | |||
| 10017 | CGM.getIntrinsic(Intrinsic::addressofreturnaddress, AllocaInt8PtrTy); | |||
| 10018 | return Builder.CreateCall(F); | |||
| 10019 | } | |||
| 10020 | ||||
| 10021 | if (BuiltinID == AArch64::BI__builtin_sponentry) { | |||
| 10022 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::sponentry, AllocaInt8PtrTy); | |||
| 10023 | return Builder.CreateCall(F); | |||
| 10024 | } | |||
| 10025 | ||||
| 10026 | if (BuiltinID == AArch64::BI__mulh || BuiltinID == AArch64::BI__umulh) { | |||
| 10027 | llvm::Type *ResType = ConvertType(E->getType()); | |||
| 10028 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); | |||
| 10029 | ||||
| 10030 | bool IsSigned = BuiltinID == AArch64::BI__mulh; | |||
| 10031 | Value *LHS = | |||
| 10032 | Builder.CreateIntCast(EmitScalarExpr(E->getArg(0)), Int128Ty, IsSigned); | |||
| 10033 | Value *RHS = | |||
| 10034 | Builder.CreateIntCast(EmitScalarExpr(E->getArg(1)), Int128Ty, IsSigned); | |||
| 10035 | ||||
| 10036 | Value *MulResult, *HigherBits; | |||
| 10037 | if (IsSigned) { | |||
| 10038 | MulResult = Builder.CreateNSWMul(LHS, RHS); | |||
| 10039 | HigherBits = Builder.CreateAShr(MulResult, 64); | |||
| 10040 | } else { | |||
| 10041 | MulResult = Builder.CreateNUWMul(LHS, RHS); | |||
| 10042 | HigherBits = Builder.CreateLShr(MulResult, 64); | |||
| 10043 | } | |||
| 10044 | HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned); | |||
| 10045 | ||||
| 10046 | return HigherBits; | |||
| 10047 | } | |||
| 10048 | ||||
| 10049 | // Handle MSVC intrinsics before argument evaluation to prevent double | |||
| 10050 | // evaluation. | |||
| 10051 | if (Optional<MSVCIntrin> MsvcIntId = translateAarch64ToMsvcIntrin(BuiltinID)) | |||
| 10052 | return EmitMSVCBuiltinExpr(*MsvcIntId, E); | |||
| 10053 | ||||
| 10054 | // Find out if any arguments are required to be integer constant | |||
| 10055 | // expressions. | |||
| 10056 | unsigned ICEArguments = 0; | |||
| 10057 | ASTContext::GetBuiltinTypeError Error; | |||
| 10058 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
| 10059 | assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None && "Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 10059, __extension__ __PRETTY_FUNCTION__ )); | |||
| 10060 | ||||
| 10061 | llvm::SmallVector<Value*, 4> Ops; | |||
| 10062 | Address PtrOp0 = Address::invalid(); | |||
| 10063 | for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++) { | |||
| 10064 | if (i == 0) { | |||
| 10065 | switch (BuiltinID) { | |||
| 10066 | case NEON::BI__builtin_neon_vld1_v: | |||
| 10067 | case NEON::BI__builtin_neon_vld1q_v: | |||
| 10068 | case NEON::BI__builtin_neon_vld1_dup_v: | |||
| 10069 | case NEON::BI__builtin_neon_vld1q_dup_v: | |||
| 10070 | case NEON::BI__builtin_neon_vld1_lane_v: | |||
| 10071 | case NEON::BI__builtin_neon_vld1q_lane_v: | |||
| 10072 | case NEON::BI__builtin_neon_vst1_v: | |||
| 10073 | case NEON::BI__builtin_neon_vst1q_v: | |||
| 10074 | case NEON::BI__builtin_neon_vst1_lane_v: | |||
| 10075 | case NEON::BI__builtin_neon_vst1q_lane_v: | |||
| 10076 | // Get the alignment for the argument in addition to the value; | |||
| 10077 | // we'll use it later. | |||
| 10078 | PtrOp0 = EmitPointerWithAlignment(E->getArg(0)); | |||
| 10079 | Ops.push_back(PtrOp0.getPointer()); | |||
| 10080 | continue; | |||
| 10081 | } | |||
| 10082 | } | |||
| 10083 | if ((ICEArguments & (1 << i)) == 0) { | |||
| 10084 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 10085 | } else { | |||
| 10086 | // If this is required to be a constant, constant fold it so that we know | |||
| 10087 | // that the generated intrinsic gets a ConstantInt. | |||
| 10088 | Ops.push_back(llvm::ConstantInt::get( | |||
| 10089 | getLLVMContext(), | |||
| 10090 | *E->getArg(i)->getIntegerConstantExpr(getContext()))); | |||
| 10091 | } | |||
| 10092 | } | |||
| 10093 | ||||
| 10094 | auto SISDMap = makeArrayRef(AArch64SISDIntrinsicMap); | |||
| 10095 | const ARMVectorIntrinsicInfo *Builtin = findARMVectorIntrinsicInMap( | |||
| 10096 | SISDMap, BuiltinID, AArch64SISDIntrinsicsProvenSorted); | |||
| 10097 | ||||
| 10098 | if (Builtin) { | |||
| 10099 | Ops.push_back(EmitScalarExpr(E->getArg(E->getNumArgs() - 1))); | |||
| 10100 | Value *Result = EmitCommonNeonSISDBuiltinExpr(*this, *Builtin, Ops, E); | |||
| 10101 | assert(Result && "SISD intrinsic should have been handled")(static_cast <bool> (Result && "SISD intrinsic should have been handled" ) ? void (0) : __assert_fail ("Result && \"SISD intrinsic should have been handled\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 10101, __extension__ __PRETTY_FUNCTION__ )); | |||
| 10102 | return Result; | |||
| 10103 | } | |||
| 10104 | ||||
| 10105 | const Expr *Arg = E->getArg(E->getNumArgs()-1); | |||
| 10106 | NeonTypeFlags Type(0); | |||
| 10107 | if (Optional<llvm::APSInt> Result = Arg->getIntegerConstantExpr(getContext())) | |||
| 10108 | // Determine the type of this overloaded NEON intrinsic. | |||
| 10109 | Type = NeonTypeFlags(Result->getZExtValue()); | |||
| 10110 | ||||
| 10111 | bool usgn = Type.isUnsigned(); | |||
| 10112 | bool quad = Type.isQuad(); | |||
| 10113 | ||||
| 10114 | // Handle non-overloaded intrinsics first. | |||
| 10115 | switch (BuiltinID) { | |||
| 10116 | default: break; | |||
| 10117 | case NEON::BI__builtin_neon_vabsh_f16: | |||
| 10118 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10119 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, HalfTy), Ops, "vabs"); | |||
| 10120 | case NEON::BI__builtin_neon_vaddq_p128: { | |||
| 10121 | llvm::Type *Ty = GetNeonType(this, NeonTypeFlags::Poly128); | |||
| 10122 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10123 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 10124 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 10125 | Ops[0] = Builder.CreateXor(Ops[0], Ops[1]); | |||
| 10126 | llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128); | |||
| 10127 | return Builder.CreateBitCast(Ops[0], Int128Ty); | |||
| 10128 | } | |||
| 10129 | case NEON::BI__builtin_neon_vldrq_p128: { | |||
| 10130 | llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128); | |||
| 10131 | llvm::Type *Int128PTy = llvm::PointerType::get(Int128Ty, 0); | |||
| 10132 | Value *Ptr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int128PTy); | |||
| 10133 | return Builder.CreateAlignedLoad(Int128Ty, Ptr, | |||
| 10134 | CharUnits::fromQuantity(16)); | |||
| 10135 | } | |||
| 10136 | case NEON::BI__builtin_neon_vstrq_p128: { | |||
| 10137 | llvm::Type *Int128PTy = llvm::Type::getIntNPtrTy(getLLVMContext(), 128); | |||
| 10138 | Value *Ptr = Builder.CreateBitCast(Ops[0], Int128PTy); | |||
| 10139 | return Builder.CreateDefaultAlignedStore(EmitScalarExpr(E->getArg(1)), Ptr); | |||
| 10140 | } | |||
| 10141 | case NEON::BI__builtin_neon_vcvts_f32_u32: | |||
| 10142 | case NEON::BI__builtin_neon_vcvtd_f64_u64: | |||
| 10143 | usgn = true; | |||
| 10144 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 10145 | case NEON::BI__builtin_neon_vcvts_f32_s32: | |||
| 10146 | case NEON::BI__builtin_neon_vcvtd_f64_s64: { | |||
| 10147 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10148 | bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64; | |||
| 10149 | llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty; | |||
| 10150 | llvm::Type *FTy = Is64 ? DoubleTy : FloatTy; | |||
| 10151 | Ops[0] = Builder.CreateBitCast(Ops[0], InTy); | |||
| 10152 | if (usgn) | |||
| 10153 | return Builder.CreateUIToFP(Ops[0], FTy); | |||
| 10154 | return Builder.CreateSIToFP(Ops[0], FTy); | |||
| 10155 | } | |||
| 10156 | case NEON::BI__builtin_neon_vcvth_f16_u16: | |||
| 10157 | case NEON::BI__builtin_neon_vcvth_f16_u32: | |||
| 10158 | case NEON::BI__builtin_neon_vcvth_f16_u64: | |||
| 10159 | usgn = true; | |||
| 10160 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 10161 | case NEON::BI__builtin_neon_vcvth_f16_s16: | |||
| 10162 | case NEON::BI__builtin_neon_vcvth_f16_s32: | |||
| 10163 | case NEON::BI__builtin_neon_vcvth_f16_s64: { | |||
| 10164 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10165 | llvm::Type *FTy = HalfTy; | |||
| 10166 | llvm::Type *InTy; | |||
| 10167 | if (Ops[0]->getType()->getPrimitiveSizeInBits() == 64) | |||
| 10168 | InTy = Int64Ty; | |||
| 10169 | else if (Ops[0]->getType()->getPrimitiveSizeInBits() == 32) | |||
| 10170 | InTy = Int32Ty; | |||
| 10171 | else | |||
| 10172 | InTy = Int16Ty; | |||
| 10173 | Ops[0] = Builder.CreateBitCast(Ops[0], InTy); | |||
| 10174 | if (usgn) | |||
| 10175 | return Builder.CreateUIToFP(Ops[0], FTy); | |||
| 10176 | return Builder.CreateSIToFP(Ops[0], FTy); | |||
| 10177 | } | |||
| 10178 | case NEON::BI__builtin_neon_vcvtah_u16_f16: | |||
| 10179 | case NEON::BI__builtin_neon_vcvtmh_u16_f16: | |||
| 10180 | case NEON::BI__builtin_neon_vcvtnh_u16_f16: | |||
| 10181 | case NEON::BI__builtin_neon_vcvtph_u16_f16: | |||
| 10182 | case NEON::BI__builtin_neon_vcvth_u16_f16: | |||
| 10183 | case NEON::BI__builtin_neon_vcvtah_s16_f16: | |||
| 10184 | case NEON::BI__builtin_neon_vcvtmh_s16_f16: | |||
| 10185 | case NEON::BI__builtin_neon_vcvtnh_s16_f16: | |||
| 10186 | case NEON::BI__builtin_neon_vcvtph_s16_f16: | |||
| 10187 | case NEON::BI__builtin_neon_vcvth_s16_f16: { | |||
| 10188 | unsigned Int; | |||
| 10189 | llvm::Type* InTy = Int32Ty; | |||
| 10190 | llvm::Type* FTy = HalfTy; | |||
| 10191 | llvm::Type *Tys[2] = {InTy, FTy}; | |||
| 10192 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10193 | switch (BuiltinID) { | |||
| 10194 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10194); | |||
| 10195 | case NEON::BI__builtin_neon_vcvtah_u16_f16: | |||
| 10196 | Int = Intrinsic::aarch64_neon_fcvtau; break; | |||
| 10197 | case NEON::BI__builtin_neon_vcvtmh_u16_f16: | |||
| 10198 | Int = Intrinsic::aarch64_neon_fcvtmu; break; | |||
| 10199 | case NEON::BI__builtin_neon_vcvtnh_u16_f16: | |||
| 10200 | Int = Intrinsic::aarch64_neon_fcvtnu; break; | |||
| 10201 | case NEON::BI__builtin_neon_vcvtph_u16_f16: | |||
| 10202 | Int = Intrinsic::aarch64_neon_fcvtpu; break; | |||
| 10203 | case NEON::BI__builtin_neon_vcvth_u16_f16: | |||
| 10204 | Int = Intrinsic::aarch64_neon_fcvtzu; break; | |||
| 10205 | case NEON::BI__builtin_neon_vcvtah_s16_f16: | |||
| 10206 | Int = Intrinsic::aarch64_neon_fcvtas; break; | |||
| 10207 | case NEON::BI__builtin_neon_vcvtmh_s16_f16: | |||
| 10208 | Int = Intrinsic::aarch64_neon_fcvtms; break; | |||
| 10209 | case NEON::BI__builtin_neon_vcvtnh_s16_f16: | |||
| 10210 | Int = Intrinsic::aarch64_neon_fcvtns; break; | |||
| 10211 | case NEON::BI__builtin_neon_vcvtph_s16_f16: | |||
| 10212 | Int = Intrinsic::aarch64_neon_fcvtps; break; | |||
| 10213 | case NEON::BI__builtin_neon_vcvth_s16_f16: | |||
| 10214 | Int = Intrinsic::aarch64_neon_fcvtzs; break; | |||
| 10215 | } | |||
| 10216 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvt"); | |||
| 10217 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 10218 | } | |||
| 10219 | case NEON::BI__builtin_neon_vcaleh_f16: | |||
| 10220 | case NEON::BI__builtin_neon_vcalth_f16: | |||
| 10221 | case NEON::BI__builtin_neon_vcageh_f16: | |||
| 10222 | case NEON::BI__builtin_neon_vcagth_f16: { | |||
| 10223 | unsigned Int; | |||
| 10224 | llvm::Type* InTy = Int32Ty; | |||
| 10225 | llvm::Type* FTy = HalfTy; | |||
| 10226 | llvm::Type *Tys[2] = {InTy, FTy}; | |||
| 10227 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10228 | switch (BuiltinID) { | |||
| 10229 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10229); | |||
| 10230 | case NEON::BI__builtin_neon_vcageh_f16: | |||
| 10231 | Int = Intrinsic::aarch64_neon_facge; break; | |||
| 10232 | case NEON::BI__builtin_neon_vcagth_f16: | |||
| 10233 | Int = Intrinsic::aarch64_neon_facgt; break; | |||
| 10234 | case NEON::BI__builtin_neon_vcaleh_f16: | |||
| 10235 | Int = Intrinsic::aarch64_neon_facge; std::swap(Ops[0], Ops[1]); break; | |||
| 10236 | case NEON::BI__builtin_neon_vcalth_f16: | |||
| 10237 | Int = Intrinsic::aarch64_neon_facgt; std::swap(Ops[0], Ops[1]); break; | |||
| 10238 | } | |||
| 10239 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "facg"); | |||
| 10240 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 10241 | } | |||
| 10242 | case NEON::BI__builtin_neon_vcvth_n_s16_f16: | |||
| 10243 | case NEON::BI__builtin_neon_vcvth_n_u16_f16: { | |||
| 10244 | unsigned Int; | |||
| 10245 | llvm::Type* InTy = Int32Ty; | |||
| 10246 | llvm::Type* FTy = HalfTy; | |||
| 10247 | llvm::Type *Tys[2] = {InTy, FTy}; | |||
| 10248 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10249 | switch (BuiltinID) { | |||
| 10250 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10250); | |||
| 10251 | case NEON::BI__builtin_neon_vcvth_n_s16_f16: | |||
| 10252 | Int = Intrinsic::aarch64_neon_vcvtfp2fxs; break; | |||
| 10253 | case NEON::BI__builtin_neon_vcvth_n_u16_f16: | |||
| 10254 | Int = Intrinsic::aarch64_neon_vcvtfp2fxu; break; | |||
| 10255 | } | |||
| 10256 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n"); | |||
| 10257 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 10258 | } | |||
| 10259 | case NEON::BI__builtin_neon_vcvth_n_f16_s16: | |||
| 10260 | case NEON::BI__builtin_neon_vcvth_n_f16_u16: { | |||
| 10261 | unsigned Int; | |||
| 10262 | llvm::Type* FTy = HalfTy; | |||
| 10263 | llvm::Type* InTy = Int32Ty; | |||
| 10264 | llvm::Type *Tys[2] = {FTy, InTy}; | |||
| 10265 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10266 | switch (BuiltinID) { | |||
| 10267 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10267); | |||
| 10268 | case NEON::BI__builtin_neon_vcvth_n_f16_s16: | |||
| 10269 | Int = Intrinsic::aarch64_neon_vcvtfxs2fp; | |||
| 10270 | Ops[0] = Builder.CreateSExt(Ops[0], InTy, "sext"); | |||
| 10271 | break; | |||
| 10272 | case NEON::BI__builtin_neon_vcvth_n_f16_u16: | |||
| 10273 | Int = Intrinsic::aarch64_neon_vcvtfxu2fp; | |||
| 10274 | Ops[0] = Builder.CreateZExt(Ops[0], InTy); | |||
| 10275 | break; | |||
| 10276 | } | |||
| 10277 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n"); | |||
| 10278 | } | |||
| 10279 | case NEON::BI__builtin_neon_vpaddd_s64: { | |||
| 10280 | auto *Ty = llvm::FixedVectorType::get(Int64Ty, 2); | |||
| 10281 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 10282 | // The vector is v2f64, so make sure it's bitcast to that. | |||
| 10283 | Vec = Builder.CreateBitCast(Vec, Ty, "v2i64"); | |||
| 10284 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); | |||
| 10285 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); | |||
| 10286 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); | |||
| 10287 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); | |||
| 10288 | // Pairwise addition of a v2f64 into a scalar f64. | |||
| 10289 | return Builder.CreateAdd(Op0, Op1, "vpaddd"); | |||
| 10290 | } | |||
| 10291 | case NEON::BI__builtin_neon_vpaddd_f64: { | |||
| 10292 | auto *Ty = llvm::FixedVectorType::get(DoubleTy, 2); | |||
| 10293 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 10294 | // The vector is v2f64, so make sure it's bitcast to that. | |||
| 10295 | Vec = Builder.CreateBitCast(Vec, Ty, "v2f64"); | |||
| 10296 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); | |||
| 10297 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); | |||
| 10298 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); | |||
| 10299 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); | |||
| 10300 | // Pairwise addition of a v2f64 into a scalar f64. | |||
| 10301 | return Builder.CreateFAdd(Op0, Op1, "vpaddd"); | |||
| 10302 | } | |||
| 10303 | case NEON::BI__builtin_neon_vpadds_f32: { | |||
| 10304 | auto *Ty = llvm::FixedVectorType::get(FloatTy, 2); | |||
| 10305 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 10306 | // The vector is v2f32, so make sure it's bitcast to that. | |||
| 10307 | Vec = Builder.CreateBitCast(Vec, Ty, "v2f32"); | |||
| 10308 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); | |||
| 10309 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); | |||
| 10310 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); | |||
| 10311 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); | |||
| 10312 | // Pairwise addition of a v2f32 into a scalar f32. | |||
| 10313 | return Builder.CreateFAdd(Op0, Op1, "vpaddd"); | |||
| 10314 | } | |||
| 10315 | case NEON::BI__builtin_neon_vceqzd_s64: | |||
| 10316 | case NEON::BI__builtin_neon_vceqzd_f64: | |||
| 10317 | case NEON::BI__builtin_neon_vceqzs_f32: | |||
| 10318 | case NEON::BI__builtin_neon_vceqzh_f16: | |||
| 10319 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10320 | return EmitAArch64CompareBuiltinExpr( | |||
| 10321 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
| 10322 | ICmpInst::FCMP_OEQ, ICmpInst::ICMP_EQ, "vceqz"); | |||
| 10323 | case NEON::BI__builtin_neon_vcgezd_s64: | |||
| 10324 | case NEON::BI__builtin_neon_vcgezd_f64: | |||
| 10325 | case NEON::BI__builtin_neon_vcgezs_f32: | |||
| 10326 | case NEON::BI__builtin_neon_vcgezh_f16: | |||
| 10327 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10328 | return EmitAArch64CompareBuiltinExpr( | |||
| 10329 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
| 10330 | ICmpInst::FCMP_OGE, ICmpInst::ICMP_SGE, "vcgez"); | |||
| 10331 | case NEON::BI__builtin_neon_vclezd_s64: | |||
| 10332 | case NEON::BI__builtin_neon_vclezd_f64: | |||
| 10333 | case NEON::BI__builtin_neon_vclezs_f32: | |||
| 10334 | case NEON::BI__builtin_neon_vclezh_f16: | |||
| 10335 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10336 | return EmitAArch64CompareBuiltinExpr( | |||
| 10337 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
| 10338 | ICmpInst::FCMP_OLE, ICmpInst::ICMP_SLE, "vclez"); | |||
| 10339 | case NEON::BI__builtin_neon_vcgtzd_s64: | |||
| 10340 | case NEON::BI__builtin_neon_vcgtzd_f64: | |||
| 10341 | case NEON::BI__builtin_neon_vcgtzs_f32: | |||
| 10342 | case NEON::BI__builtin_neon_vcgtzh_f16: | |||
| 10343 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10344 | return EmitAArch64CompareBuiltinExpr( | |||
| 10345 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
| 10346 | ICmpInst::FCMP_OGT, ICmpInst::ICMP_SGT, "vcgtz"); | |||
| 10347 | case NEON::BI__builtin_neon_vcltzd_s64: | |||
| 10348 | case NEON::BI__builtin_neon_vcltzd_f64: | |||
| 10349 | case NEON::BI__builtin_neon_vcltzs_f32: | |||
| 10350 | case NEON::BI__builtin_neon_vcltzh_f16: | |||
| 10351 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10352 | return EmitAArch64CompareBuiltinExpr( | |||
| 10353 | Ops[0], ConvertType(E->getCallReturnType(getContext())), | |||
| 10354 | ICmpInst::FCMP_OLT, ICmpInst::ICMP_SLT, "vcltz"); | |||
| 10355 | ||||
| 10356 | case NEON::BI__builtin_neon_vceqzd_u64: { | |||
| 10357 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 10358 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); | |||
| 10359 | Ops[0] = | |||
| 10360 | Builder.CreateICmpEQ(Ops[0], llvm::Constant::getNullValue(Int64Ty)); | |||
| 10361 | return Builder.CreateSExt(Ops[0], Int64Ty, "vceqzd"); | |||
| 10362 | } | |||
| 10363 | case NEON::BI__builtin_neon_vceqd_f64: | |||
| 10364 | case NEON::BI__builtin_neon_vcled_f64: | |||
| 10365 | case NEON::BI__builtin_neon_vcltd_f64: | |||
| 10366 | case NEON::BI__builtin_neon_vcged_f64: | |||
| 10367 | case NEON::BI__builtin_neon_vcgtd_f64: { | |||
| 10368 | llvm::CmpInst::Predicate P; | |||
| 10369 | switch (BuiltinID) { | |||
| 10370 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10370); | |||
| 10371 | case NEON::BI__builtin_neon_vceqd_f64: P = llvm::FCmpInst::FCMP_OEQ; break; | |||
| 10372 | case NEON::BI__builtin_neon_vcled_f64: P = llvm::FCmpInst::FCMP_OLE; break; | |||
| 10373 | case NEON::BI__builtin_neon_vcltd_f64: P = llvm::FCmpInst::FCMP_OLT; break; | |||
| 10374 | case NEON::BI__builtin_neon_vcged_f64: P = llvm::FCmpInst::FCMP_OGE; break; | |||
| 10375 | case NEON::BI__builtin_neon_vcgtd_f64: P = llvm::FCmpInst::FCMP_OGT; break; | |||
| 10376 | } | |||
| 10377 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10378 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
| 10379 | Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy); | |||
| 10380 | if (P == llvm::FCmpInst::FCMP_OEQ) | |||
| 10381 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); | |||
| 10382 | else | |||
| 10383 | Ops[0] = Builder.CreateFCmpS(P, Ops[0], Ops[1]); | |||
| 10384 | return Builder.CreateSExt(Ops[0], Int64Ty, "vcmpd"); | |||
| 10385 | } | |||
| 10386 | case NEON::BI__builtin_neon_vceqs_f32: | |||
| 10387 | case NEON::BI__builtin_neon_vcles_f32: | |||
| 10388 | case NEON::BI__builtin_neon_vclts_f32: | |||
| 10389 | case NEON::BI__builtin_neon_vcges_f32: | |||
| 10390 | case NEON::BI__builtin_neon_vcgts_f32: { | |||
| 10391 | llvm::CmpInst::Predicate P; | |||
| 10392 | switch (BuiltinID) { | |||
| 10393 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10393); | |||
| 10394 | case NEON::BI__builtin_neon_vceqs_f32: P = llvm::FCmpInst::FCMP_OEQ; break; | |||
| 10395 | case NEON::BI__builtin_neon_vcles_f32: P = llvm::FCmpInst::FCMP_OLE; break; | |||
| 10396 | case NEON::BI__builtin_neon_vclts_f32: P = llvm::FCmpInst::FCMP_OLT; break; | |||
| 10397 | case NEON::BI__builtin_neon_vcges_f32: P = llvm::FCmpInst::FCMP_OGE; break; | |||
| 10398 | case NEON::BI__builtin_neon_vcgts_f32: P = llvm::FCmpInst::FCMP_OGT; break; | |||
| 10399 | } | |||
| 10400 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10401 | Ops[0] = Builder.CreateBitCast(Ops[0], FloatTy); | |||
| 10402 | Ops[1] = Builder.CreateBitCast(Ops[1], FloatTy); | |||
| 10403 | if (P == llvm::FCmpInst::FCMP_OEQ) | |||
| 10404 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); | |||
| 10405 | else | |||
| 10406 | Ops[0] = Builder.CreateFCmpS(P, Ops[0], Ops[1]); | |||
| 10407 | return Builder.CreateSExt(Ops[0], Int32Ty, "vcmpd"); | |||
| 10408 | } | |||
| 10409 | case NEON::BI__builtin_neon_vceqh_f16: | |||
| 10410 | case NEON::BI__builtin_neon_vcleh_f16: | |||
| 10411 | case NEON::BI__builtin_neon_vclth_f16: | |||
| 10412 | case NEON::BI__builtin_neon_vcgeh_f16: | |||
| 10413 | case NEON::BI__builtin_neon_vcgth_f16: { | |||
| 10414 | llvm::CmpInst::Predicate P; | |||
| 10415 | switch (BuiltinID) { | |||
| 10416 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10416); | |||
| 10417 | case NEON::BI__builtin_neon_vceqh_f16: P = llvm::FCmpInst::FCMP_OEQ; break; | |||
| 10418 | case NEON::BI__builtin_neon_vcleh_f16: P = llvm::FCmpInst::FCMP_OLE; break; | |||
| 10419 | case NEON::BI__builtin_neon_vclth_f16: P = llvm::FCmpInst::FCMP_OLT; break; | |||
| 10420 | case NEON::BI__builtin_neon_vcgeh_f16: P = llvm::FCmpInst::FCMP_OGE; break; | |||
| 10421 | case NEON::BI__builtin_neon_vcgth_f16: P = llvm::FCmpInst::FCMP_OGT; break; | |||
| 10422 | } | |||
| 10423 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10424 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); | |||
| 10425 | Ops[1] = Builder.CreateBitCast(Ops[1], HalfTy); | |||
| 10426 | if (P == llvm::FCmpInst::FCMP_OEQ) | |||
| 10427 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); | |||
| 10428 | else | |||
| 10429 | Ops[0] = Builder.CreateFCmpS(P, Ops[0], Ops[1]); | |||
| 10430 | return Builder.CreateSExt(Ops[0], Int16Ty, "vcmpd"); | |||
| 10431 | } | |||
| 10432 | case NEON::BI__builtin_neon_vceqd_s64: | |||
| 10433 | case NEON::BI__builtin_neon_vceqd_u64: | |||
| 10434 | case NEON::BI__builtin_neon_vcgtd_s64: | |||
| 10435 | case NEON::BI__builtin_neon_vcgtd_u64: | |||
| 10436 | case NEON::BI__builtin_neon_vcltd_s64: | |||
| 10437 | case NEON::BI__builtin_neon_vcltd_u64: | |||
| 10438 | case NEON::BI__builtin_neon_vcged_u64: | |||
| 10439 | case NEON::BI__builtin_neon_vcged_s64: | |||
| 10440 | case NEON::BI__builtin_neon_vcled_u64: | |||
| 10441 | case NEON::BI__builtin_neon_vcled_s64: { | |||
| 10442 | llvm::CmpInst::Predicate P; | |||
| 10443 | switch (BuiltinID) { | |||
| 10444 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "clang/lib/CodeGen/CGBuiltin.cpp", 10444); | |||
| 10445 | case NEON::BI__builtin_neon_vceqd_s64: | |||
| 10446 | case NEON::BI__builtin_neon_vceqd_u64:P = llvm::ICmpInst::ICMP_EQ;break; | |||
| 10447 | case NEON::BI__builtin_neon_vcgtd_s64:P = llvm::ICmpInst::ICMP_SGT;break; | |||
| 10448 | case NEON::BI__builtin_neon_vcgtd_u64:P = llvm::ICmpInst::ICMP_UGT;break; | |||
| 10449 | case NEON::BI__builtin_neon_vcltd_s64:P = llvm::ICmpInst::ICMP_SLT;break; | |||
| 10450 | case NEON::BI__builtin_neon_vcltd_u64:P = llvm::ICmpInst::ICMP_ULT;break; | |||
| 10451 | case NEON::BI__builtin_neon_vcged_u64:P = llvm::ICmpInst::ICMP_UGE;break; | |||
| 10452 | case NEON::BI__builtin_neon_vcged_s64:P = llvm::ICmpInst::ICMP_SGE;break; | |||
| 10453 | case NEON::BI__builtin_neon_vcled_u64:P = llvm::ICmpInst::ICMP_ULE;break; | |||
| 10454 | case NEON::BI__builtin_neon_vcled_s64:P = llvm::ICmpInst::ICMP_SLE;break; | |||
| 10455 | } | |||
| 10456 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10457 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); | |||
| 10458 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); | |||
| 10459 | Ops[0] = Builder.CreateICmp(P, Ops[0], Ops[1]); | |||
| 10460 | return Builder.CreateSExt(Ops[0], Int64Ty, "vceqd"); | |||
| 10461 | } | |||
| 10462 | case NEON::BI__builtin_neon_vtstd_s64: | |||
| 10463 | case NEON::BI__builtin_neon_vtstd_u64: { | |||
| 10464 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10465 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); | |||
| 10466 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); | |||
| 10467 | Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]); | |||
| 10468 | Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0], | |||
| 10469 | llvm::Constant::getNullValue(Int64Ty)); | |||
| 10470 | return Builder.CreateSExt(Ops[0], Int64Ty, "vtstd"); | |||
| 10471 | } | |||
| 10472 | case NEON::BI__builtin_neon_vset_lane_i8: | |||
| 10473 | case NEON::BI__builtin_neon_vset_lane_i16: | |||
| 10474 | case NEON::BI__builtin_neon_vset_lane_i32: | |||
| 10475 | case NEON::BI__builtin_neon_vset_lane_i64: | |||
| 10476 | case NEON::BI__builtin_neon_vset_lane_bf16: | |||
| 10477 | case NEON::BI__builtin_neon_vset_lane_f32: | |||
| 10478 | case NEON::BI__builtin_neon_vsetq_lane_i8: | |||
| 10479 | case NEON::BI__builtin_neon_vsetq_lane_i16: | |||
| 10480 | case NEON::BI__builtin_neon_vsetq_lane_i32: | |||
| 10481 | case NEON::BI__builtin_neon_vsetq_lane_i64: | |||
| 10482 | case NEON::BI__builtin_neon_vsetq_lane_bf16: | |||
| 10483 | case NEON::BI__builtin_neon_vsetq_lane_f32: | |||
| 10484 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
| 10485 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
| 10486 | case NEON::BI__builtin_neon_vset_lane_f64: | |||
| 10487 | // The vector type needs a cast for the v1f64 variant. | |||
| 10488 | Ops[1] = | |||
| 10489 | Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(DoubleTy, 1)); | |||
| 10490 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
| 10491 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
| 10492 | case NEON::BI__builtin_neon_vsetq_lane_f64: | |||
| 10493 | // The vector type needs a cast for the v2f64 variant. | |||
| 10494 | Ops[1] = | |||
| 10495 | Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(DoubleTy, 2)); | |||
| 10496 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
| 10497 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); | |||
| 10498 | ||||
| 10499 | case NEON::BI__builtin_neon_vget_lane_i8: | |||
| 10500 | case NEON::BI__builtin_neon_vdupb_lane_i8: | |||
| 10501 | Ops[0] = | |||
| 10502 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int8Ty, 8)); | |||
| 10503 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10504 | "vget_lane"); | |||
| 10505 | case NEON::BI__builtin_neon_vgetq_lane_i8: | |||
| 10506 | case NEON::BI__builtin_neon_vdupb_laneq_i8: | |||
| 10507 | Ops[0] = | |||
| 10508 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int8Ty, 16)); | |||
| 10509 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10510 | "vgetq_lane"); | |||
| 10511 | case NEON::BI__builtin_neon_vget_lane_i16: | |||
| 10512 | case NEON::BI__builtin_neon_vduph_lane_i16: | |||
| 10513 | Ops[0] = | |||
| 10514 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int16Ty, 4)); | |||
| 10515 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10516 | "vget_lane"); | |||
| 10517 | case NEON::BI__builtin_neon_vgetq_lane_i16: | |||
| 10518 | case NEON::BI__builtin_neon_vduph_laneq_i16: | |||
| 10519 | Ops[0] = | |||
| 10520 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int16Ty, 8)); | |||
| 10521 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10522 | "vgetq_lane"); | |||
| 10523 | case NEON::BI__builtin_neon_vget_lane_i32: | |||
| 10524 | case NEON::BI__builtin_neon_vdups_lane_i32: | |||
| 10525 | Ops[0] = | |||
| 10526 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 2)); | |||
| 10527 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10528 | "vget_lane"); | |||
| 10529 | case NEON::BI__builtin_neon_vdups_lane_f32: | |||
| 10530 | Ops[0] = | |||
| 10531 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(FloatTy, 2)); | |||
| 10532 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10533 | "vdups_lane"); | |||
| 10534 | case NEON::BI__builtin_neon_vgetq_lane_i32: | |||
| 10535 | case NEON::BI__builtin_neon_vdups_laneq_i32: | |||
| 10536 | Ops[0] = | |||
| 10537 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 4)); | |||
| 10538 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10539 | "vgetq_lane"); | |||
| 10540 | case NEON::BI__builtin_neon_vget_lane_i64: | |||
| 10541 | case NEON::BI__builtin_neon_vdupd_lane_i64: | |||
| 10542 | Ops[0] = | |||
| 10543 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 1)); | |||
| 10544 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10545 | "vget_lane"); | |||
| 10546 | case NEON::BI__builtin_neon_vdupd_lane_f64: | |||
| 10547 | Ops[0] = | |||
| 10548 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(DoubleTy, 1)); | |||
| 10549 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10550 | "vdupd_lane"); | |||
| 10551 | case NEON::BI__builtin_neon_vgetq_lane_i64: | |||
| 10552 | case NEON::BI__builtin_neon_vdupd_laneq_i64: | |||
| 10553 | Ops[0] = | |||
| 10554 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2)); | |||
| 10555 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10556 | "vgetq_lane"); | |||
| 10557 | case NEON::BI__builtin_neon_vget_lane_f32: | |||
| 10558 | Ops[0] = | |||
| 10559 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(FloatTy, 2)); | |||
| 10560 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10561 | "vget_lane"); | |||
| 10562 | case NEON::BI__builtin_neon_vget_lane_f64: | |||
| 10563 | Ops[0] = | |||
| 10564 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(DoubleTy, 1)); | |||
| 10565 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10566 | "vget_lane"); | |||
| 10567 | case NEON::BI__builtin_neon_vgetq_lane_f32: | |||
| 10568 | case NEON::BI__builtin_neon_vdups_laneq_f32: | |||
| 10569 | Ops[0] = | |||
| 10570 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(FloatTy, 4)); | |||
| 10571 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10572 | "vgetq_lane"); | |||
| 10573 | case NEON::BI__builtin_neon_vgetq_lane_f64: | |||
| 10574 | case NEON::BI__builtin_neon_vdupd_laneq_f64: | |||
| 10575 | Ops[0] = | |||
| 10576 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(DoubleTy, 2)); | |||
| 10577 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10578 | "vgetq_lane"); | |||
| 10579 | case NEON::BI__builtin_neon_vaddh_f16: | |||
| 10580 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10581 | return Builder.CreateFAdd(Ops[0], Ops[1], "vaddh"); | |||
| 10582 | case NEON::BI__builtin_neon_vsubh_f16: | |||
| 10583 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10584 | return Builder.CreateFSub(Ops[0], Ops[1], "vsubh"); | |||
| 10585 | case NEON::BI__builtin_neon_vmulh_f16: | |||
| 10586 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10587 | return Builder.CreateFMul(Ops[0], Ops[1], "vmulh"); | |||
| 10588 | case NEON::BI__builtin_neon_vdivh_f16: | |||
| 10589 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10590 | return Builder.CreateFDiv(Ops[0], Ops[1], "vdivh"); | |||
| 10591 | case NEON::BI__builtin_neon_vfmah_f16: | |||
| 10592 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. | |||
| 10593 | return emitCallMaybeConstrainedFPBuiltin( | |||
| 10594 | *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, HalfTy, | |||
| 10595 | {EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), Ops[0]}); | |||
| 10596 | case NEON::BI__builtin_neon_vfmsh_f16: { | |||
| 10597 | // FIXME: This should be an fneg instruction: | |||
| 10598 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(HalfTy); | |||
| 10599 | Value* Sub = Builder.CreateFSub(Zero, EmitScalarExpr(E->getArg(1)), "vsubh"); | |||
| 10600 | ||||
| 10601 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. | |||
| 10602 | return emitCallMaybeConstrainedFPBuiltin( | |||
| 10603 | *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, HalfTy, | |||
| 10604 | {Sub, EmitScalarExpr(E->getArg(2)), Ops[0]}); | |||
| 10605 | } | |||
| 10606 | case NEON::BI__builtin_neon_vaddd_s64: | |||
| 10607 | case NEON::BI__builtin_neon_vaddd_u64: | |||
| 10608 | return Builder.CreateAdd(Ops[0], EmitScalarExpr(E->getArg(1)), "vaddd"); | |||
| 10609 | case NEON::BI__builtin_neon_vsubd_s64: | |||
| 10610 | case NEON::BI__builtin_neon_vsubd_u64: | |||
| 10611 | return Builder.CreateSub(Ops[0], EmitScalarExpr(E->getArg(1)), "vsubd"); | |||
| 10612 | case NEON::BI__builtin_neon_vqdmlalh_s16: | |||
| 10613 | case NEON::BI__builtin_neon_vqdmlslh_s16: { | |||
| 10614 | SmallVector<Value *, 2> ProductOps; | |||
| 10615 | ProductOps.push_back(vectorWrapScalar16(Ops[1])); | |||
| 10616 | ProductOps.push_back(vectorWrapScalar16(EmitScalarExpr(E->getArg(2)))); | |||
| 10617 | auto *VTy = llvm::FixedVectorType::get(Int32Ty, 4); | |||
| 10618 | Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy), | |||
| 10619 | ProductOps, "vqdmlXl"); | |||
| 10620 | Constant *CI = ConstantInt::get(SizeTy, 0); | |||
| 10621 | Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0"); | |||
| 10622 | ||||
| 10623 | unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlalh_s16 | |||
| 10624 | ? Intrinsic::aarch64_neon_sqadd | |||
| 10625 | : Intrinsic::aarch64_neon_sqsub; | |||
| 10626 | return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int32Ty), Ops, "vqdmlXl"); | |||
| 10627 | } | |||
| 10628 | case NEON::BI__builtin_neon_vqshlud_n_s64: { | |||
| 10629 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10630 | Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty); | |||
| 10631 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqshlu, Int64Ty), | |||
| 10632 | Ops, "vqshlu_n"); | |||
| 10633 | } | |||
| 10634 | case NEON::BI__builtin_neon_vqshld_n_u64: | |||
| 10635 | case NEON::BI__builtin_neon_vqshld_n_s64: { | |||
| 10636 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vqshld_n_u64 | |||
| 10637 | ? Intrinsic::aarch64_neon_uqshl | |||
| 10638 | : Intrinsic::aarch64_neon_sqshl; | |||
| 10639 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10640 | Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty); | |||
| 10641 | return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vqshl_n"); | |||
| 10642 | } | |||
| 10643 | case NEON::BI__builtin_neon_vrshrd_n_u64: | |||
| 10644 | case NEON::BI__builtin_neon_vrshrd_n_s64: { | |||
| 10645 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrshrd_n_u64 | |||
| 10646 | ? Intrinsic::aarch64_neon_urshl | |||
| 10647 | : Intrinsic::aarch64_neon_srshl; | |||
| 10648 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10649 | int SV = cast<ConstantInt>(Ops[1])->getSExtValue(); | |||
| 10650 | Ops[1] = ConstantInt::get(Int64Ty, -SV); | |||
| 10651 | return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vrshr_n"); | |||
| 10652 | } | |||
| 10653 | case NEON::BI__builtin_neon_vrsrad_n_u64: | |||
| 10654 | case NEON::BI__builtin_neon_vrsrad_n_s64: { | |||
| 10655 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrsrad_n_u64 | |||
| 10656 | ? Intrinsic::aarch64_neon_urshl | |||
| 10657 | : Intrinsic::aarch64_neon_srshl; | |||
| 10658 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); | |||
| 10659 | Ops.push_back(Builder.CreateNeg(EmitScalarExpr(E->getArg(2)))); | |||
| 10660 | Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Int64Ty), | |||
| 10661 | {Ops[1], Builder.CreateSExt(Ops[2], Int64Ty)}); | |||
| 10662 | return Builder.CreateAdd(Ops[0], Builder.CreateBitCast(Ops[1], Int64Ty)); | |||
| 10663 | } | |||
| 10664 | case NEON::BI__builtin_neon_vshld_n_s64: | |||
| 10665 | case NEON::BI__builtin_neon_vshld_n_u64: { | |||
| 10666 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
| 10667 | return Builder.CreateShl( | |||
| 10668 | Ops[0], ConstantInt::get(Int64Ty, Amt->getZExtValue()), "shld_n"); | |||
| 10669 | } | |||
| 10670 | case NEON::BI__builtin_neon_vshrd_n_s64: { | |||
| 10671 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
| 10672 | return Builder.CreateAShr( | |||
| 10673 | Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63), | |||
| 10674 | Amt->getZExtValue())), | |||
| 10675 | "shrd_n"); | |||
| 10676 | } | |||
| 10677 | case NEON::BI__builtin_neon_vshrd_n_u64: { | |||
| 10678 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
| 10679 | uint64_t ShiftAmt = Amt->getZExtValue(); | |||
| 10680 | // Right-shifting an unsigned value by its size yields 0. | |||
| 10681 | if (ShiftAmt == 64) | |||
| 10682 | return ConstantInt::get(Int64Ty, 0); | |||
| 10683 | return Builder.CreateLShr(Ops[0], ConstantInt::get(Int64Ty, ShiftAmt), | |||
| 10684 | "shrd_n"); | |||
| 10685 | } | |||
| 10686 | case NEON::BI__builtin_neon_vsrad_n_s64: { | |||
| 10687 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2))); | |||
| 10688 | Ops[1] = Builder.CreateAShr( | |||
| 10689 | Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63), | |||
| 10690 | Amt->getZExtValue())), | |||
| 10691 | "shrd_n"); | |||
| 10692 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
| 10693 | } | |||
| 10694 | case NEON::BI__builtin_neon_vsrad_n_u64: { | |||
| 10695 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2))); | |||
| 10696 | uint64_t ShiftAmt = Amt->getZExtValue(); | |||
| 10697 | // Right-shifting an unsigned value by its size yields 0. | |||
| 10698 | // As Op + 0 = Op, return Ops[0] directly. | |||
| 10699 | if (ShiftAmt == 64) | |||
| 10700 | return Ops[0]; | |||
| 10701 | Ops[1] = Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, ShiftAmt), | |||
| 10702 | "shrd_n"); | |||
| 10703 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
| 10704 | } | |||
| 10705 | case NEON::BI__builtin_neon_vqdmlalh_lane_s16: | |||
| 10706 | case NEON::BI__builtin_neon_vqdmlalh_laneq_s16: | |||
| 10707 | case NEON::BI__builtin_neon_vqdmlslh_lane_s16: | |||
| 10708 | case NEON::BI__builtin_neon_vqdmlslh_laneq_s16: { | |||
| 10709 | Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)), | |||
| 10710 | "lane"); | |||
| 10711 | SmallVector<Value *, 2> ProductOps; | |||
| 10712 | ProductOps.push_back(vectorWrapScalar16(Ops[1])); | |||
| 10713 | ProductOps.push_back(vectorWrapScalar16(Ops[2])); | |||
| 10714 | auto *VTy = llvm::FixedVectorType::get(Int32Ty, 4); | |||
| 10715 | Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy), | |||
| 10716 | ProductOps, "vqdmlXl"); | |||
| 10717 | Constant *CI = ConstantInt::get(SizeTy, 0); | |||
| 10718 | Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0"); | |||
| 10719 | Ops.pop_back(); | |||
| 10720 | ||||
| 10721 | unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlalh_lane_s16 || | |||
| 10722 | BuiltinID == NEON::BI__builtin_neon_vqdmlalh_laneq_s16) | |||
| 10723 | ? Intrinsic::aarch64_neon_sqadd | |||
| 10724 | : Intrinsic::aarch64_neon_sqsub; | |||
| 10725 | return EmitNeonCall(CGM.getIntrinsic(AccInt, Int32Ty), Ops, "vqdmlXl"); | |||
| 10726 | } | |||
| 10727 | case NEON::BI__builtin_neon_vqdmlals_s32: | |||
| 10728 | case NEON::BI__builtin_neon_vqdmlsls_s32: { | |||
| 10729 | SmallVector<Value *, 2> ProductOps; | |||
| 10730 | ProductOps.push_back(Ops[1]); | |||
| 10731 | ProductOps.push_back(EmitScalarExpr(E->getArg(2))); | |||
| 10732 | Ops[1] = | |||
| 10733 | EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar), | |||
| 10734 | ProductOps, "vqdmlXl"); | |||
| 10735 | ||||
| 10736 | unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlals_s32 | |||
| 10737 | ? Intrinsic::aarch64_neon_sqadd | |||
| 10738 | : Intrinsic::aarch64_neon_sqsub; | |||
| 10739 | return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int64Ty), Ops, "vqdmlXl"); | |||
| 10740 | } | |||
| 10741 | case NEON::BI__builtin_neon_vqdmlals_lane_s32: | |||
| 10742 | case NEON::BI__builtin_neon_vqdmlals_laneq_s32: | |||
| 10743 | case NEON::BI__builtin_neon_vqdmlsls_lane_s32: | |||
| 10744 | case NEON::BI__builtin_neon_vqdmlsls_laneq_s32: { | |||
| 10745 | Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)), | |||
| 10746 | "lane"); | |||
| 10747 | SmallVector<Value *, 2> ProductOps; | |||
| 10748 | ProductOps.push_back(Ops[1]); | |||
| 10749 | ProductOps.push_back(Ops[2]); | |||
| 10750 | Ops[1] = | |||
| 10751 | EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar), | |||
| 10752 | ProductOps, "vqdmlXl"); | |||
| 10753 | Ops.pop_back(); | |||
| 10754 | ||||
| 10755 | unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlals_lane_s32 || | |||
| 10756 | BuiltinID == NEON::BI__builtin_neon_vqdmlals_laneq_s32) | |||
| 10757 | ? Intrinsic::aarch64_neon_sqadd | |||
| 10758 | : Intrinsic::aarch64_neon_sqsub; | |||
| 10759 | return EmitNeonCall(CGM.getIntrinsic(AccInt, Int64Ty), Ops, "vqdmlXl"); | |||
| 10760 | } | |||
| 10761 | case NEON::BI__builtin_neon_vget_lane_bf16: | |||
| 10762 | case NEON::BI__builtin_neon_vduph_lane_bf16: | |||
| 10763 | case NEON::BI__builtin_neon_vduph_lane_f16: { | |||
| 10764 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10765 | "vget_lane"); | |||
| 10766 | } | |||
| 10767 | case NEON::BI__builtin_neon_vgetq_lane_bf16: | |||
| 10768 | case NEON::BI__builtin_neon_vduph_laneq_bf16: | |||
| 10769 | case NEON::BI__builtin_neon_vduph_laneq_f16: { | |||
| 10770 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), | |||
| 10771 | "vgetq_lane"); | |||
| 10772 | } | |||
| 10773 | ||||
| 10774 | case AArch64::BI_InterlockedAdd: { | |||
| 10775 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); | |||
| 10776 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); | |||
| 10777 | AtomicRMWInst *RMWI = Builder.CreateAtomicRMW( | |||
| 10778 | AtomicRMWInst::Add, Arg0, Arg1, | |||
| 10779 | llvm::AtomicOrdering::SequentiallyConsistent); | |||
| 10780 | return Builder.CreateAdd(RMWI, Arg1); | |||
| 10781 | } | |||
| 10782 | } | |||
| 10783 | ||||
| 10784 | llvm::FixedVectorType *VTy = GetNeonType(this, Type); | |||
| 10785 | llvm::Type *Ty = VTy; | |||
| 10786 | if (!Ty) | |||
| 10787 | return nullptr; | |||
| 10788 | ||||
| 10789 | // Not all intrinsics handled by the common case work for AArch64 yet, so only | |||
| 10790 | // defer to common code if it's been added to our special map. | |||
| 10791 | Builtin = findARMVectorIntrinsicInMap(AArch64SIMDIntrinsicMap, BuiltinID, | |||
| 10792 | AArch64SIMDIntrinsicsProvenSorted); | |||
| 10793 | ||||
| 10794 | if (Builtin) | |||
| 10795 | return EmitCommonNeonBuiltinExpr( | |||
| 10796 | Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic, | |||
| 10797 | Builtin->NameHint, Builtin->TypeModifier, E, Ops, | |||
| 10798 | /*never use addresses*/ Address::invalid(), Address::invalid(), Arch); | |||
| 10799 | ||||
| 10800 | if (Value *V = EmitAArch64TblBuiltinExpr(*this, BuiltinID, E, Ops, Arch)) | |||
| 10801 | return V; | |||
| 10802 | ||||
| 10803 | unsigned Int; | |||
| 10804 | switch (BuiltinID) { | |||
| 10805 | default: return nullptr; | |||
| 10806 | case NEON::BI__builtin_neon_vbsl_v: | |||
| 10807 | case NEON::BI__builtin_neon_vbslq_v: { | |||
| 10808 | llvm::Type *BitTy = llvm::VectorType::getInteger(VTy); | |||
| 10809 | Ops[0] = Builder.CreateBitCast(Ops[0], BitTy, "vbsl"); | |||
| 10810 | Ops[1] = Builder.CreateBitCast(Ops[1], BitTy, "vbsl"); | |||
| 10811 | Ops[2] = Builder.CreateBitCast(Ops[2], BitTy, "vbsl"); | |||
| 10812 | ||||
| 10813 | Ops[1] = Builder.CreateAnd(Ops[0], Ops[1], "vbsl"); | |||
| 10814 | Ops[2] = Builder.CreateAnd(Builder.CreateNot(Ops[0]), Ops[2], "vbsl"); | |||
| 10815 | Ops[0] = Builder.CreateOr(Ops[1], Ops[2], "vbsl"); | |||
| 10816 | return Builder.CreateBitCast(Ops[0], Ty); | |||
| 10817 | } | |||
| 10818 | case NEON::BI__builtin_neon_vfma_lane_v: | |||
| 10819 | case NEON::BI__builtin_neon_vfmaq_lane_v: { // Only used for FP types | |||
| 10820 | // The ARM builtins (and instructions) have the addend as the first | |||
| 10821 | // operand, but the 'fma' intrinsics have it last. Swap it around here. | |||
| 10822 | Value *Addend = Ops[0]; | |||
| 10823 | Value *Multiplicand = Ops[1]; | |||
| 10824 | Value *LaneSource = Ops[2]; | |||
| 10825 | Ops[0] = Multiplicand; | |||
| 10826 | Ops[1] = LaneSource; | |||
| 10827 | Ops[2] = Addend; | |||
| 10828 | ||||
| 10829 | // Now adjust things to handle the lane access. | |||
| 10830 | auto *SourceTy = BuiltinID == NEON::BI__builtin_neon_vfmaq_lane_v | |||
| 10831 | ? llvm::FixedVectorType::get(VTy->getElementType(), | |||
| 10832 | VTy->getNumElements() / 2) | |||
| 10833 | : VTy; | |||
| 10834 | llvm::Constant *cst = cast<Constant>(Ops[3]); | |||
| 10835 | Value *SV = llvm::ConstantVector::getSplat(VTy->getElementCount(), cst); | |||
| 10836 | Ops[1] = Builder.CreateBitCast(Ops[1], SourceTy); | |||
| 10837 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV, "lane"); | |||
| 10838 | ||||
| 10839 | Ops.pop_back(); | |||
| 10840 | Int = Builder.getIsFPConstrained() ? Intrinsic::experimental_constrained_fma | |||
| 10841 | : Intrinsic::fma; | |||
| 10842 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "fmla"); | |||
| 10843 | } | |||
| 10844 | case NEON::BI__builtin_neon_vfma_laneq_v: { | |||
| 10845 | auto *VTy = cast<llvm::FixedVectorType>(Ty); | |||
| 10846 | // v1f64 fma should be mapped to Neon scalar f64 fma | |||
| 10847 | if (VTy && VTy->getElementType() == DoubleTy) { | |||
| 10848 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
| 10849 | Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy); | |||
| 10850 | llvm::FixedVectorType *VTy = | |||
| 10851 | GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, true)); | |||
| 10852 | Ops[2] = Builder.CreateBitCast(Ops[2], VTy); | |||
| 10853 | Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract"); | |||
| 10854 | Value *Result; | |||
| 10855 | Result = emitCallMaybeConstrainedFPBuiltin( | |||
| 10856 | *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, | |||
| 10857 | DoubleTy, {Ops[1], Ops[2], Ops[0]}); | |||
| 10858 | return Builder.CreateBitCast(Result, Ty); | |||
| 10859 | } | |||
| 10860 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 10861 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 10862 | ||||
| 10863 | auto *STy = llvm::FixedVectorType::get(VTy->getElementType(), | |||
| 10864 | VTy->getNumElements() * 2); | |||
| 10865 | Ops[2] = Builder.CreateBitCast(Ops[2], STy); | |||
| 10866 | Value *SV = llvm::ConstantVector::getSplat(VTy->getElementCount(), | |||
| 10867 | cast<ConstantInt>(Ops[3])); | |||
| 10868 | Ops[2] = Builder.CreateShuffleVector(Ops[2], Ops[2], SV, "lane"); | |||
| 10869 | ||||
| 10870 | return emitCallMaybeConstrainedFPBuiltin( | |||
| 10871 | *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty, | |||
| 10872 | {Ops[2], Ops[1], Ops[0]}); | |||
| 10873 | } | |||
| 10874 | case NEON::BI__builtin_neon_vfmaq_laneq_v: { | |||
| 10875 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 10876 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 10877 | ||||
| 10878 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 10879 | Ops[2] = EmitNeonSplat(Ops[2], cast<ConstantInt>(Ops[3])); | |||
| 10880 | return emitCallMaybeConstrainedFPBuiltin( | |||
| 10881 | *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty, | |||
| 10882 | {Ops[2], Ops[1], Ops[0]}); | |||
| 10883 | } | |||
| 10884 | case NEON::BI__builtin_neon_vfmah_lane_f16: | |||
| 10885 | case NEON::BI__builtin_neon_vfmas_lane_f32: | |||
| 10886 | case NEON::BI__builtin_neon_vfmah_laneq_f16: | |||
| 10887 | case NEON::BI__builtin_neon_vfmas_laneq_f32: | |||
| 10888 | case NEON::BI__builtin_neon_vfmad_lane_f64: | |||
| 10889 | case NEON::BI__builtin_neon_vfmad_laneq_f64: { | |||
| 10890 | Ops.push_back(EmitScalarExpr(E->getArg(3))); | |||
| 10891 | llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext())); | |||
| 10892 | Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract"); | |||
| 10893 | return emitCallMaybeConstrainedFPBuiltin( | |||
| 10894 | *this, Intrinsic::fma, Intrinsic::experimental_constrained_fma, Ty, | |||
| 10895 | {Ops[1], Ops[2], Ops[0]}); | |||
| 10896 | } | |||
| 10897 | case NEON::BI__builtin_neon_vmull_v: | |||
| 10898 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
| 10899 | Int = usgn ? Intrinsic::aarch64_neon_umull : Intrinsic::aarch64_neon_smull; | |||
| 10900 | if (Type.isPoly()) Int = Intrinsic::aarch64_neon_pmull; | |||
| 10901 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull"); | |||
| 10902 | case NEON::BI__builtin_neon_vmax_v: | |||
| 10903 | case NEON::BI__builtin_neon_vmaxq_v: | |||
| 10904 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
| 10905 | Int = usgn ? Intrinsic::aarch64_neon_umax : Intrinsic::aarch64_neon_smax; | |||
| 10906 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmax; | |||
| 10907 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmax"); | |||
| 10908 | case NEON::BI__builtin_neon_vmaxh_f16: { | |||
| 10909 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10910 | Int = Intrinsic::aarch64_neon_fmax; | |||
| 10911 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmax"); | |||
| 10912 | } | |||
| 10913 | case NEON::BI__builtin_neon_vmin_v: | |||
| 10914 | case NEON::BI__builtin_neon_vminq_v: | |||
| 10915 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
| 10916 | Int = usgn ? Intrinsic::aarch64_neon_umin : Intrinsic::aarch64_neon_smin; | |||
| 10917 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmin; | |||
| 10918 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmin"); | |||
| 10919 | case NEON::BI__builtin_neon_vminh_f16: { | |||
| 10920 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10921 | Int = Intrinsic::aarch64_neon_fmin; | |||
| 10922 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmin"); | |||
| 10923 | } | |||
| 10924 | case NEON::BI__builtin_neon_vabd_v: | |||
| 10925 | case NEON::BI__builtin_neon_vabdq_v: | |||
| 10926 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
| 10927 | Int = usgn ? Intrinsic::aarch64_neon_uabd : Intrinsic::aarch64_neon_sabd; | |||
| 10928 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fabd; | |||
| 10929 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vabd"); | |||
| 10930 | case NEON::BI__builtin_neon_vpadal_v: | |||
| 10931 | case NEON::BI__builtin_neon_vpadalq_v: { | |||
| 10932 | unsigned ArgElts = VTy->getNumElements(); | |||
| 10933 | llvm::IntegerType *EltTy = cast<IntegerType>(VTy->getElementType()); | |||
| 10934 | unsigned BitWidth = EltTy->getBitWidth(); | |||
| 10935 | auto *ArgTy = llvm::FixedVectorType::get( | |||
| 10936 | llvm::IntegerType::get(getLLVMContext(), BitWidth / 2), 2 * ArgElts); | |||
| 10937 | llvm::Type* Tys[2] = { VTy, ArgTy }; | |||
| 10938 | Int = usgn ? Intrinsic::aarch64_neon_uaddlp : Intrinsic::aarch64_neon_saddlp; | |||
| 10939 | SmallVector<llvm::Value*, 1> TmpOps; | |||
| 10940 | TmpOps.push_back(Ops[1]); | |||
| 10941 | Function *F = CGM.getIntrinsic(Int, Tys); | |||
| 10942 | llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vpadal"); | |||
| 10943 | llvm::Value *addend = Builder.CreateBitCast(Ops[0], tmp->getType()); | |||
| 10944 | return Builder.CreateAdd(tmp, addend); | |||
| 10945 | } | |||
| 10946 | case NEON::BI__builtin_neon_vpmin_v: | |||
| 10947 | case NEON::BI__builtin_neon_vpminq_v: | |||
| 10948 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
| 10949 | Int = usgn ? Intrinsic::aarch64_neon_uminp : Intrinsic::aarch64_neon_sminp; | |||
| 10950 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fminp; | |||
| 10951 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmin"); | |||
| 10952 | case NEON::BI__builtin_neon_vpmax_v: | |||
| 10953 | case NEON::BI__builtin_neon_vpmaxq_v: | |||
| 10954 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. | |||
| 10955 | Int = usgn ? Intrinsic::aarch64_neon_umaxp : Intrinsic::aarch64_neon_smaxp; | |||
| 10956 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmaxp; | |||
| 10957 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmax"); | |||
| 10958 | case NEON::BI__builtin_neon_vminnm_v: | |||
| 10959 | case NEON::BI__builtin_neon_vminnmq_v: | |||
| 10960 | Int = Intrinsic::aarch64_neon_fminnm; | |||
| 10961 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vminnm"); | |||
| 10962 | case NEON::BI__builtin_neon_vminnmh_f16: | |||
| 10963 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10964 | Int = Intrinsic::aarch64_neon_fminnm; | |||
| 10965 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vminnm"); | |||
| 10966 | case NEON::BI__builtin_neon_vmaxnm_v: | |||
| 10967 | case NEON::BI__builtin_neon_vmaxnmq_v: | |||
| 10968 | Int = Intrinsic::aarch64_neon_fmaxnm; | |||
| 10969 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmaxnm"); | |||
| 10970 | case NEON::BI__builtin_neon_vmaxnmh_f16: | |||
| 10971 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10972 | Int = Intrinsic::aarch64_neon_fmaxnm; | |||
| 10973 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmaxnm"); | |||
| 10974 | case NEON::BI__builtin_neon_vrecpss_f32: { | |||
| 10975 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10976 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, FloatTy), | |||
| 10977 | Ops, "vrecps"); | |||
| 10978 | } | |||
| 10979 | case NEON::BI__builtin_neon_vrecpsd_f64: | |||
| 10980 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10981 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, DoubleTy), | |||
| 10982 | Ops, "vrecps"); | |||
| 10983 | case NEON::BI__builtin_neon_vrecpsh_f16: | |||
| 10984 | Ops.push_back(EmitScalarExpr(E->getArg(1))); | |||
| 10985 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, HalfTy), | |||
| 10986 | Ops, "vrecps"); | |||
| 10987 | case NEON::BI__builtin_neon_vqshrun_n_v: | |||
| 10988 | Int = Intrinsic::aarch64_neon_sqshrun; | |||
| 10989 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrun_n"); | |||
| 10990 | case NEON::BI__builtin_neon_vqrshrun_n_v: | |||
| 10991 | Int = Intrinsic::aarch64_neon_sqrshrun; | |||
| 10992 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrun_n"); | |||
| 10993 | case NEON::BI__builtin_neon_vqshrn_n_v: | |||
| 10994 | Int = usgn ? Intrinsic::aarch64_neon_uqshrn : Intrinsic::aarch64_neon_sqshrn; | |||
| 10995 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n"); | |||
| 10996 | case NEON::BI__builtin_neon_vrshrn_n_v: | |||
| 10997 | Int = Intrinsic::aarch64_neon_rshrn; | |||
| 10998 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshrn_n"); | |||
| 10999 | case NEON::BI__builtin_neon_vqrshrn_n_v: | |||
| 11000 | Int = usgn ? Intrinsic::aarch64_neon_uqrshrn : Intrinsic::aarch64_neon_sqrshrn; | |||
| 11001 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n"); | |||
| 11002 | case NEON::BI__builtin_neon_vrndah_f16: { | |||
| 11003 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11004 | Int = Builder.getIsFPConstrained() | |||
| 11005 | ? Intrinsic::experimental_constrained_round | |||
| 11006 | : Intrinsic::round; | |||
| 11007 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrnda"); | |||
| 11008 | } | |||
| 11009 | case NEON::BI__builtin_neon_vrnda_v: | |||
| 11010 | case NEON::BI__builtin_neon_vrndaq_v: { | |||
| 11011 | Int = Builder.getIsFPConstrained() | |||
| 11012 | ? Intrinsic::experimental_constrained_round | |||
| 11013 | : Intrinsic::round; | |||
| 11014 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnda"); | |||
| 11015 | } | |||
| 11016 | case NEON::BI__builtin_neon_vrndih_f16: { | |||
| 11017 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11018 | Int = Builder.getIsFPConstrained() | |||
| 11019 | ? Intrinsic::experimental_constrained_nearbyint | |||
| 11020 | : Intrinsic::nearbyint; | |||
| 11021 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndi"); | |||
| 11022 | } | |||
| 11023 | case NEON::BI__builtin_neon_vrndmh_f16: { | |||
| 11024 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11025 | Int = Builder.getIsFPConstrained() | |||
| 11026 | ? Intrinsic::experimental_constrained_floor | |||
| 11027 | : Intrinsic::floor; | |||
| 11028 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndm"); | |||
| 11029 | } | |||
| 11030 | case NEON::BI__builtin_neon_vrndm_v: | |||
| 11031 | case NEON::BI__builtin_neon_vrndmq_v: { | |||
| 11032 | Int = Builder.getIsFPConstrained() | |||
| 11033 | ? Intrinsic::experimental_constrained_floor | |||
| 11034 | : Intrinsic::floor; | |||
| 11035 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndm"); | |||
| 11036 | } | |||
| 11037 | case NEON::BI__builtin_neon_vrndnh_f16: { | |||
| 11038 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11039 | Int = Builder.getIsFPConstrained() | |||
| 11040 | ? Intrinsic::experimental_constrained_roundeven | |||
| 11041 | : Intrinsic::roundeven; | |||
| 11042 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndn"); | |||
| 11043 | } | |||
| 11044 | case NEON::BI__builtin_neon_vrndn_v: | |||
| 11045 | case NEON::BI__builtin_neon_vrndnq_v: { | |||
| 11046 | Int = Builder.getIsFPConstrained() | |||
| 11047 | ? Intrinsic::experimental_constrained_roundeven | |||
| 11048 | : Intrinsic::roundeven; | |||
| 11049 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndn"); | |||
| 11050 | } | |||
| 11051 | case NEON::BI__builtin_neon_vrndns_f32: { | |||
| 11052 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11053 | Int = Builder.getIsFPConstrained() | |||
| 11054 | ? Intrinsic::experimental_constrained_roundeven | |||
| 11055 | : Intrinsic::roundeven; | |||
| 11056 | return EmitNeonCall(CGM.getIntrinsic(Int, FloatTy), Ops, "vrndn"); | |||
| 11057 | } | |||
| 11058 | case NEON::BI__builtin_neon_vrndph_f16: { | |||
| 11059 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11060 | Int = Builder.getIsFPConstrained() | |||
| 11061 | ? Intrinsic::experimental_constrained_ceil | |||
| 11062 | : Intrinsic::ceil; | |||
| 11063 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndp"); | |||
| 11064 | } | |||
| 11065 | case NEON::BI__builtin_neon_vrndp_v: | |||
| 11066 | case NEON::BI__builtin_neon_vrndpq_v: { | |||
| 11067 | Int = Builder.getIsFPConstrained() | |||
| 11068 | ? Intrinsic::experimental_constrained_ceil | |||
| 11069 | : Intrinsic::ceil; | |||
| 11070 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndp"); | |||
| 11071 | } | |||
| 11072 | case NEON::BI__builtin_neon_vrndxh_f16: { | |||
| 11073 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11074 | Int = Builder.getIsFPConstrained() | |||
| 11075 | ? Intrinsic::experimental_constrained_rint | |||
| 11076 | : Intrinsic::rint; | |||
| 11077 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndx"); | |||
| 11078 | } | |||
| 11079 | case NEON::BI__builtin_neon_vrndx_v: | |||
| 11080 | case NEON::BI__builtin_neon_vrndxq_v: { | |||
| 11081 | Int = Builder.getIsFPConstrained() | |||
| 11082 | ? Intrinsic::experimental_constrained_rint | |||
| 11083 | : Intrinsic::rint; | |||
| 11084 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndx"); | |||
| 11085 | } | |||
| 11086 | case NEON::BI__builtin_neon_vrndh_f16: { | |||
| 11087 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11088 | Int = Builder.getIsFPConstrained() | |||
| 11089 | ? Intrinsic::experimental_constrained_trunc | |||
| 11090 | : Intrinsic::trunc; | |||
| 11091 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndz"); | |||
| 11092 | } | |||
| 11093 | case NEON::BI__builtin_neon_vrnd32x_v: | |||
| 11094 | case NEON::BI__builtin_neon_vrnd32xq_v: { | |||
| 11095 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11096 | Int = Intrinsic::aarch64_neon_frint32x; | |||
| 11097 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd32x"); | |||
| 11098 | } | |||
| 11099 | case NEON::BI__builtin_neon_vrnd32z_v: | |||
| 11100 | case NEON::BI__builtin_neon_vrnd32zq_v: { | |||
| 11101 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11102 | Int = Intrinsic::aarch64_neon_frint32z; | |||
| 11103 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd32z"); | |||
| 11104 | } | |||
| 11105 | case NEON::BI__builtin_neon_vrnd64x_v: | |||
| 11106 | case NEON::BI__builtin_neon_vrnd64xq_v: { | |||
| 11107 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11108 | Int = Intrinsic::aarch64_neon_frint64x; | |||
| 11109 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd64x"); | |||
| 11110 | } | |||
| 11111 | case NEON::BI__builtin_neon_vrnd64z_v: | |||
| 11112 | case NEON::BI__builtin_neon_vrnd64zq_v: { | |||
| 11113 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11114 | Int = Intrinsic::aarch64_neon_frint64z; | |||
| 11115 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnd64z"); | |||
| 11116 | } | |||
| 11117 | case NEON::BI__builtin_neon_vrnd_v: | |||
| 11118 | case NEON::BI__builtin_neon_vrndq_v: { | |||
| 11119 | Int = Builder.getIsFPConstrained() | |||
| 11120 | ? Intrinsic::experimental_constrained_trunc | |||
| 11121 | : Intrinsic::trunc; | |||
| 11122 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndz"); | |||
| 11123 | } | |||
| 11124 | case NEON::BI__builtin_neon_vcvt_f64_v: | |||
| 11125 | case NEON::BI__builtin_neon_vcvtq_f64_v: | |||
| 11126 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11127 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, quad)); | |||
| 11128 | return usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") | |||
| 11129 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); | |||
| 11130 | case NEON::BI__builtin_neon_vcvt_f64_f32: { | |||
| 11131 | assert(Type.getEltType() == NeonTypeFlags::Float64 && quad &&(static_cast <bool> (Type.getEltType() == NeonTypeFlags ::Float64 && quad && "unexpected vcvt_f64_f32 builtin" ) ? void (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11132, __extension__ __PRETTY_FUNCTION__ )) | |||
| 11132 | "unexpected vcvt_f64_f32 builtin")(static_cast <bool> (Type.getEltType() == NeonTypeFlags ::Float64 && quad && "unexpected vcvt_f64_f32 builtin" ) ? void (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11132, __extension__ __PRETTY_FUNCTION__ )); | |||
| 11133 | NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float32, false, false); | |||
| 11134 | Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag)); | |||
| 11135 | ||||
| 11136 | return Builder.CreateFPExt(Ops[0], Ty, "vcvt"); | |||
| 11137 | } | |||
| 11138 | case NEON::BI__builtin_neon_vcvt_f32_f64: { | |||
| 11139 | assert(Type.getEltType() == NeonTypeFlags::Float32 &&(static_cast <bool> (Type.getEltType() == NeonTypeFlags ::Float32 && "unexpected vcvt_f32_f64 builtin") ? void (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11140, __extension__ __PRETTY_FUNCTION__ )) | |||
| 11140 | "unexpected vcvt_f32_f64 builtin")(static_cast <bool> (Type.getEltType() == NeonTypeFlags ::Float32 && "unexpected vcvt_f32_f64 builtin") ? void (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11140, __extension__ __PRETTY_FUNCTION__ )); | |||
| 11141 | NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float64, false, true); | |||
| 11142 | Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag)); | |||
| 11143 | ||||
| 11144 | return Builder.CreateFPTrunc(Ops[0], Ty, "vcvt"); | |||
| 11145 | } | |||
| 11146 | case NEON::BI__builtin_neon_vcvt_s32_v: | |||
| 11147 | case NEON::BI__builtin_neon_vcvt_u32_v: | |||
| 11148 | case NEON::BI__builtin_neon_vcvt_s64_v: | |||
| 11149 | case NEON::BI__builtin_neon_vcvt_u64_v: | |||
| 11150 | case NEON::BI__builtin_neon_vcvt_s16_v: | |||
| 11151 | case NEON::BI__builtin_neon_vcvt_u16_v: | |||
| 11152 | case NEON::BI__builtin_neon_vcvtq_s32_v: | |||
| 11153 | case NEON::BI__builtin_neon_vcvtq_u32_v: | |||
| 11154 | case NEON::BI__builtin_neon_vcvtq_s64_v: | |||
| 11155 | case NEON::BI__builtin_neon_vcvtq_u64_v: | |||
| 11156 | case NEON::BI__builtin_neon_vcvtq_s16_v: | |||
| 11157 | case NEON::BI__builtin_neon_vcvtq_u16_v: { | |||
| 11158 | Int = | |||
| 11159 | usgn ? Intrinsic::aarch64_neon_fcvtzu : Intrinsic::aarch64_neon_fcvtzs; | |||
| 11160 | llvm::Type *Tys[2] = {Ty, GetFloatNeonType(this, Type)}; | |||
| 11161 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtz"); | |||
| 11162 | } | |||
| 11163 | case NEON::BI__builtin_neon_vcvta_s16_v: | |||
| 11164 | case NEON::BI__builtin_neon_vcvta_u16_v: | |||
| 11165 | case NEON::BI__builtin_neon_vcvta_s32_v: | |||
| 11166 | case NEON::BI__builtin_neon_vcvtaq_s16_v: | |||
| 11167 | case NEON::BI__builtin_neon_vcvtaq_s32_v: | |||
| 11168 | case NEON::BI__builtin_neon_vcvta_u32_v: | |||
| 11169 | case NEON::BI__builtin_neon_vcvtaq_u16_v: | |||
| 11170 | case NEON::BI__builtin_neon_vcvtaq_u32_v: | |||
| 11171 | case NEON::BI__builtin_neon_vcvta_s64_v: | |||
| 11172 | case NEON::BI__builtin_neon_vcvtaq_s64_v: | |||
| 11173 | case NEON::BI__builtin_neon_vcvta_u64_v: | |||
| 11174 | case NEON::BI__builtin_neon_vcvtaq_u64_v: { | |||
| 11175 | Int = usgn ? Intrinsic::aarch64_neon_fcvtau : Intrinsic::aarch64_neon_fcvtas; | |||
| 11176 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
| 11177 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvta"); | |||
| 11178 | } | |||
| 11179 | case NEON::BI__builtin_neon_vcvtm_s16_v: | |||
| 11180 | case NEON::BI__builtin_neon_vcvtm_s32_v: | |||
| 11181 | case NEON::BI__builtin_neon_vcvtmq_s16_v: | |||
| 11182 | case NEON::BI__builtin_neon_vcvtmq_s32_v: | |||
| 11183 | case NEON::BI__builtin_neon_vcvtm_u16_v: | |||
| 11184 | case NEON::BI__builtin_neon_vcvtm_u32_v: | |||
| 11185 | case NEON::BI__builtin_neon_vcvtmq_u16_v: | |||
| 11186 | case NEON::BI__builtin_neon_vcvtmq_u32_v: | |||
| 11187 | case NEON::BI__builtin_neon_vcvtm_s64_v: | |||
| 11188 | case NEON::BI__builtin_neon_vcvtmq_s64_v: | |||
| 11189 | case NEON::BI__builtin_neon_vcvtm_u64_v: | |||
| 11190 | case NEON::BI__builtin_neon_vcvtmq_u64_v: { | |||
| 11191 | Int = usgn ? Intrinsic::aarch64_neon_fcvtmu : Intrinsic::aarch64_neon_fcvtms; | |||
| 11192 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
| 11193 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtm"); | |||
| 11194 | } | |||
| 11195 | case NEON::BI__builtin_neon_vcvtn_s16_v: | |||
| 11196 | case NEON::BI__builtin_neon_vcvtn_s32_v: | |||
| 11197 | case NEON::BI__builtin_neon_vcvtnq_s16_v: | |||
| 11198 | case NEON::BI__builtin_neon_vcvtnq_s32_v: | |||
| 11199 | case NEON::BI__builtin_neon_vcvtn_u16_v: | |||
| 11200 | case NEON::BI__builtin_neon_vcvtn_u32_v: | |||
| 11201 | case NEON::BI__builtin_neon_vcvtnq_u16_v: | |||
| 11202 | case NEON::BI__builtin_neon_vcvtnq_u32_v: | |||
| 11203 | case NEON::BI__builtin_neon_vcvtn_s64_v: | |||
| 11204 | case NEON::BI__builtin_neon_vcvtnq_s64_v: | |||
| 11205 | case NEON::BI__builtin_neon_vcvtn_u64_v: | |||
| 11206 | case NEON::BI__builtin_neon_vcvtnq_u64_v: { | |||
| 11207 | Int = usgn ? Intrinsic::aarch64_neon_fcvtnu : Intrinsic::aarch64_neon_fcvtns; | |||
| 11208 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
| 11209 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtn"); | |||
| 11210 | } | |||
| 11211 | case NEON::BI__builtin_neon_vcvtp_s16_v: | |||
| 11212 | case NEON::BI__builtin_neon_vcvtp_s32_v: | |||
| 11213 | case NEON::BI__builtin_neon_vcvtpq_s16_v: | |||
| 11214 | case NEON::BI__builtin_neon_vcvtpq_s32_v: | |||
| 11215 | case NEON::BI__builtin_neon_vcvtp_u16_v: | |||
| 11216 | case NEON::BI__builtin_neon_vcvtp_u32_v: | |||
| 11217 | case NEON::BI__builtin_neon_vcvtpq_u16_v: | |||
| 11218 | case NEON::BI__builtin_neon_vcvtpq_u32_v: | |||
| 11219 | case NEON::BI__builtin_neon_vcvtp_s64_v: | |||
| 11220 | case NEON::BI__builtin_neon_vcvtpq_s64_v: | |||
| 11221 | case NEON::BI__builtin_neon_vcvtp_u64_v: | |||
| 11222 | case NEON::BI__builtin_neon_vcvtpq_u64_v: { | |||
| 11223 | Int = usgn ? Intrinsic::aarch64_neon_fcvtpu : Intrinsic::aarch64_neon_fcvtps; | |||
| 11224 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; | |||
| 11225 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtp"); | |||
| 11226 | } | |||
| 11227 | case NEON::BI__builtin_neon_vmulx_v: | |||
| 11228 | case NEON::BI__builtin_neon_vmulxq_v: { | |||
| 11229 | Int = Intrinsic::aarch64_neon_fmulx; | |||
| 11230 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmulx"); | |||
| 11231 | } | |||
| 11232 | case NEON::BI__builtin_neon_vmulxh_lane_f16: | |||
| 11233 | case NEON::BI__builtin_neon_vmulxh_laneq_f16: { | |||
| 11234 | // vmulx_lane should be mapped to Neon scalar mulx after | |||
| 11235 | // extracting the scalar element | |||
| 11236 | Ops.push_back(EmitScalarExpr(E->getArg(2))); | |||
| 11237 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract"); | |||
| 11238 | Ops.pop_back(); | |||
| 11239 | Int = Intrinsic::aarch64_neon_fmulx; | |||
| 11240 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmulx"); | |||
| 11241 | } | |||
| 11242 | case NEON::BI__builtin_neon_vmul_lane_v: | |||
| 11243 | case NEON::BI__builtin_neon_vmul_laneq_v: { | |||
| 11244 | // v1f64 vmul_lane should be mapped to Neon scalar mul lane | |||
| 11245 | bool Quad = false; | |||
| 11246 | if (BuiltinID == NEON::BI__builtin_neon_vmul_laneq_v) | |||
| 11247 | Quad = true; | |||
| 11248 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
| 11249 | llvm::FixedVectorType *VTy = | |||
| 11250 | GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, Quad)); | |||
| 11251 | Ops[1] = Builder.CreateBitCast(Ops[1], VTy); | |||
| 11252 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract"); | |||
| 11253 | Value *Result = Builder.CreateFMul(Ops[0], Ops[1]); | |||
| 11254 | return Builder.CreateBitCast(Result, Ty); | |||
| 11255 | } | |||
| 11256 | case NEON::BI__builtin_neon_vnegd_s64: | |||
| 11257 | return Builder.CreateNeg(EmitScalarExpr(E->getArg(0)), "vnegd"); | |||
| 11258 | case NEON::BI__builtin_neon_vnegh_f16: | |||
| 11259 | return Builder.CreateFNeg(EmitScalarExpr(E->getArg(0)), "vnegh"); | |||
| 11260 | case NEON::BI__builtin_neon_vpmaxnm_v: | |||
| 11261 | case NEON::BI__builtin_neon_vpmaxnmq_v: { | |||
| 11262 | Int = Intrinsic::aarch64_neon_fmaxnmp; | |||
| 11263 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmaxnm"); | |||
| 11264 | } | |||
| 11265 | case NEON::BI__builtin_neon_vpminnm_v: | |||
| 11266 | case NEON::BI__builtin_neon_vpminnmq_v: { | |||
| 11267 | Int = Intrinsic::aarch64_neon_fminnmp; | |||
| 11268 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpminnm"); | |||
| 11269 | } | |||
| 11270 | case NEON::BI__builtin_neon_vsqrth_f16: { | |||
| 11271 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11272 | Int = Builder.getIsFPConstrained() | |||
| 11273 | ? Intrinsic::experimental_constrained_sqrt | |||
| 11274 | : Intrinsic::sqrt; | |||
| 11275 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vsqrt"); | |||
| 11276 | } | |||
| 11277 | case NEON::BI__builtin_neon_vsqrt_v: | |||
| 11278 | case NEON::BI__builtin_neon_vsqrtq_v: { | |||
| 11279 | Int = Builder.getIsFPConstrained() | |||
| 11280 | ? Intrinsic::experimental_constrained_sqrt | |||
| 11281 | : Intrinsic::sqrt; | |||
| 11282 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11283 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqrt"); | |||
| 11284 | } | |||
| 11285 | case NEON::BI__builtin_neon_vrbit_v: | |||
| 11286 | case NEON::BI__builtin_neon_vrbitq_v: { | |||
| 11287 | Int = Intrinsic::bitreverse; | |||
| 11288 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrbit"); | |||
| 11289 | } | |||
| 11290 | case NEON::BI__builtin_neon_vaddv_u8: | |||
| 11291 | // FIXME: These are handled by the AArch64 scalar code. | |||
| 11292 | usgn = true; | |||
| 11293 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 11294 | case NEON::BI__builtin_neon_vaddv_s8: { | |||
| 11295 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
| 11296 | Ty = Int32Ty; | |||
| 11297 | VTy = llvm::FixedVectorType::get(Int8Ty, 8); | |||
| 11298 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11299 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11300 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
| 11301 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11302 | } | |||
| 11303 | case NEON::BI__builtin_neon_vaddv_u16: | |||
| 11304 | usgn = true; | |||
| 11305 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 11306 | case NEON::BI__builtin_neon_vaddv_s16: { | |||
| 11307 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
| 11308 | Ty = Int32Ty; | |||
| 11309 | VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 11310 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11311 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11312 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
| 11313 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11314 | } | |||
| 11315 | case NEON::BI__builtin_neon_vaddvq_u8: | |||
| 11316 | usgn = true; | |||
| 11317 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 11318 | case NEON::BI__builtin_neon_vaddvq_s8: { | |||
| 11319 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
| 11320 | Ty = Int32Ty; | |||
| 11321 | VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 11322 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11323 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11324 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
| 11325 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11326 | } | |||
| 11327 | case NEON::BI__builtin_neon_vaddvq_u16: | |||
| 11328 | usgn = true; | |||
| 11329 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 11330 | case NEON::BI__builtin_neon_vaddvq_s16: { | |||
| 11331 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; | |||
| 11332 | Ty = Int32Ty; | |||
| 11333 | VTy = llvm::FixedVectorType::get(Int16Ty, 8); | |||
| 11334 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11335 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11336 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); | |||
| 11337 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11338 | } | |||
| 11339 | case NEON::BI__builtin_neon_vmaxv_u8: { | |||
| 11340 | Int = Intrinsic::aarch64_neon_umaxv; | |||
| 11341 | Ty = Int32Ty; | |||
| 11342 | VTy = llvm::FixedVectorType::get(Int8Ty, 8); | |||
| 11343 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11344 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11345 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11346 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11347 | } | |||
| 11348 | case NEON::BI__builtin_neon_vmaxv_u16: { | |||
| 11349 | Int = Intrinsic::aarch64_neon_umaxv; | |||
| 11350 | Ty = Int32Ty; | |||
| 11351 | VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 11352 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11353 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11354 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11355 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11356 | } | |||
| 11357 | case NEON::BI__builtin_neon_vmaxvq_u8: { | |||
| 11358 | Int = Intrinsic::aarch64_neon_umaxv; | |||
| 11359 | Ty = Int32Ty; | |||
| 11360 | VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 11361 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11362 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11363 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11364 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11365 | } | |||
| 11366 | case NEON::BI__builtin_neon_vmaxvq_u16: { | |||
| 11367 | Int = Intrinsic::aarch64_neon_umaxv; | |||
| 11368 | Ty = Int32Ty; | |||
| 11369 | VTy = llvm::FixedVectorType::get(Int16Ty, 8); | |||
| 11370 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11371 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11372 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11373 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11374 | } | |||
| 11375 | case NEON::BI__builtin_neon_vmaxv_s8: { | |||
| 11376 | Int = Intrinsic::aarch64_neon_smaxv; | |||
| 11377 | Ty = Int32Ty; | |||
| 11378 | VTy = llvm::FixedVectorType::get(Int8Ty, 8); | |||
| 11379 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11380 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11381 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11382 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11383 | } | |||
| 11384 | case NEON::BI__builtin_neon_vmaxv_s16: { | |||
| 11385 | Int = Intrinsic::aarch64_neon_smaxv; | |||
| 11386 | Ty = Int32Ty; | |||
| 11387 | VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 11388 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11389 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11390 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11391 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11392 | } | |||
| 11393 | case NEON::BI__builtin_neon_vmaxvq_s8: { | |||
| 11394 | Int = Intrinsic::aarch64_neon_smaxv; | |||
| 11395 | Ty = Int32Ty; | |||
| 11396 | VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 11397 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11398 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11399 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11400 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11401 | } | |||
| 11402 | case NEON::BI__builtin_neon_vmaxvq_s16: { | |||
| 11403 | Int = Intrinsic::aarch64_neon_smaxv; | |||
| 11404 | Ty = Int32Ty; | |||
| 11405 | VTy = llvm::FixedVectorType::get(Int16Ty, 8); | |||
| 11406 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11407 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11408 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11409 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11410 | } | |||
| 11411 | case NEON::BI__builtin_neon_vmaxv_f16: { | |||
| 11412 | Int = Intrinsic::aarch64_neon_fmaxv; | |||
| 11413 | Ty = HalfTy; | |||
| 11414 | VTy = llvm::FixedVectorType::get(HalfTy, 4); | |||
| 11415 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11416 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11417 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11418 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11419 | } | |||
| 11420 | case NEON::BI__builtin_neon_vmaxvq_f16: { | |||
| 11421 | Int = Intrinsic::aarch64_neon_fmaxv; | |||
| 11422 | Ty = HalfTy; | |||
| 11423 | VTy = llvm::FixedVectorType::get(HalfTy, 8); | |||
| 11424 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11425 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11426 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); | |||
| 11427 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11428 | } | |||
| 11429 | case NEON::BI__builtin_neon_vminv_u8: { | |||
| 11430 | Int = Intrinsic::aarch64_neon_uminv; | |||
| 11431 | Ty = Int32Ty; | |||
| 11432 | VTy = llvm::FixedVectorType::get(Int8Ty, 8); | |||
| 11433 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11434 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11435 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11436 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11437 | } | |||
| 11438 | case NEON::BI__builtin_neon_vminv_u16: { | |||
| 11439 | Int = Intrinsic::aarch64_neon_uminv; | |||
| 11440 | Ty = Int32Ty; | |||
| 11441 | VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 11442 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11443 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11444 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11445 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11446 | } | |||
| 11447 | case NEON::BI__builtin_neon_vminvq_u8: { | |||
| 11448 | Int = Intrinsic::aarch64_neon_uminv; | |||
| 11449 | Ty = Int32Ty; | |||
| 11450 | VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 11451 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11452 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11453 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11454 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11455 | } | |||
| 11456 | case NEON::BI__builtin_neon_vminvq_u16: { | |||
| 11457 | Int = Intrinsic::aarch64_neon_uminv; | |||
| 11458 | Ty = Int32Ty; | |||
| 11459 | VTy = llvm::FixedVectorType::get(Int16Ty, 8); | |||
| 11460 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11461 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11462 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11463 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11464 | } | |||
| 11465 | case NEON::BI__builtin_neon_vminv_s8: { | |||
| 11466 | Int = Intrinsic::aarch64_neon_sminv; | |||
| 11467 | Ty = Int32Ty; | |||
| 11468 | VTy = llvm::FixedVectorType::get(Int8Ty, 8); | |||
| 11469 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11470 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11471 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11472 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11473 | } | |||
| 11474 | case NEON::BI__builtin_neon_vminv_s16: { | |||
| 11475 | Int = Intrinsic::aarch64_neon_sminv; | |||
| 11476 | Ty = Int32Ty; | |||
| 11477 | VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 11478 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11479 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11480 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11481 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11482 | } | |||
| 11483 | case NEON::BI__builtin_neon_vminvq_s8: { | |||
| 11484 | Int = Intrinsic::aarch64_neon_sminv; | |||
| 11485 | Ty = Int32Ty; | |||
| 11486 | VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 11487 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11488 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11489 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11490 | return Builder.CreateTrunc(Ops[0], Int8Ty); | |||
| 11491 | } | |||
| 11492 | case NEON::BI__builtin_neon_vminvq_s16: { | |||
| 11493 | Int = Intrinsic::aarch64_neon_sminv; | |||
| 11494 | Ty = Int32Ty; | |||
| 11495 | VTy = llvm::FixedVectorType::get(Int16Ty, 8); | |||
| 11496 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11497 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11498 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11499 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11500 | } | |||
| 11501 | case NEON::BI__builtin_neon_vminv_f16: { | |||
| 11502 | Int = Intrinsic::aarch64_neon_fminv; | |||
| 11503 | Ty = HalfTy; | |||
| 11504 | VTy = llvm::FixedVectorType::get(HalfTy, 4); | |||
| 11505 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11506 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11507 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11508 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11509 | } | |||
| 11510 | case NEON::BI__builtin_neon_vminvq_f16: { | |||
| 11511 | Int = Intrinsic::aarch64_neon_fminv; | |||
| 11512 | Ty = HalfTy; | |||
| 11513 | VTy = llvm::FixedVectorType::get(HalfTy, 8); | |||
| 11514 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11515 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11516 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); | |||
| 11517 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11518 | } | |||
| 11519 | case NEON::BI__builtin_neon_vmaxnmv_f16: { | |||
| 11520 | Int = Intrinsic::aarch64_neon_fmaxnmv; | |||
| 11521 | Ty = HalfTy; | |||
| 11522 | VTy = llvm::FixedVectorType::get(HalfTy, 4); | |||
| 11523 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11524 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11525 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv"); | |||
| 11526 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11527 | } | |||
| 11528 | case NEON::BI__builtin_neon_vmaxnmvq_f16: { | |||
| 11529 | Int = Intrinsic::aarch64_neon_fmaxnmv; | |||
| 11530 | Ty = HalfTy; | |||
| 11531 | VTy = llvm::FixedVectorType::get(HalfTy, 8); | |||
| 11532 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11533 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11534 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv"); | |||
| 11535 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11536 | } | |||
| 11537 | case NEON::BI__builtin_neon_vminnmv_f16: { | |||
| 11538 | Int = Intrinsic::aarch64_neon_fminnmv; | |||
| 11539 | Ty = HalfTy; | |||
| 11540 | VTy = llvm::FixedVectorType::get(HalfTy, 4); | |||
| 11541 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11542 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11543 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv"); | |||
| 11544 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11545 | } | |||
| 11546 | case NEON::BI__builtin_neon_vminnmvq_f16: { | |||
| 11547 | Int = Intrinsic::aarch64_neon_fminnmv; | |||
| 11548 | Ty = HalfTy; | |||
| 11549 | VTy = llvm::FixedVectorType::get(HalfTy, 8); | |||
| 11550 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11551 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11552 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv"); | |||
| 11553 | return Builder.CreateTrunc(Ops[0], HalfTy); | |||
| 11554 | } | |||
| 11555 | case NEON::BI__builtin_neon_vmul_n_f64: { | |||
| 11556 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); | |||
| 11557 | Value *RHS = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), DoubleTy); | |||
| 11558 | return Builder.CreateFMul(Ops[0], RHS); | |||
| 11559 | } | |||
| 11560 | case NEON::BI__builtin_neon_vaddlv_u8: { | |||
| 11561 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
| 11562 | Ty = Int32Ty; | |||
| 11563 | VTy = llvm::FixedVectorType::get(Int8Ty, 8); | |||
| 11564 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11565 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11566 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11567 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11568 | } | |||
| 11569 | case NEON::BI__builtin_neon_vaddlv_u16: { | |||
| 11570 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
| 11571 | Ty = Int32Ty; | |||
| 11572 | VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 11573 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11574 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11575 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11576 | } | |||
| 11577 | case NEON::BI__builtin_neon_vaddlvq_u8: { | |||
| 11578 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
| 11579 | Ty = Int32Ty; | |||
| 11580 | VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 11581 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11582 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11583 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11584 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11585 | } | |||
| 11586 | case NEON::BI__builtin_neon_vaddlvq_u16: { | |||
| 11587 | Int = Intrinsic::aarch64_neon_uaddlv; | |||
| 11588 | Ty = Int32Ty; | |||
| 11589 | VTy = llvm::FixedVectorType::get(Int16Ty, 8); | |||
| 11590 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11591 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11592 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11593 | } | |||
| 11594 | case NEON::BI__builtin_neon_vaddlv_s8: { | |||
| 11595 | Int = Intrinsic::aarch64_neon_saddlv; | |||
| 11596 | Ty = Int32Ty; | |||
| 11597 | VTy = llvm::FixedVectorType::get(Int8Ty, 8); | |||
| 11598 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11599 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11600 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11601 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11602 | } | |||
| 11603 | case NEON::BI__builtin_neon_vaddlv_s16: { | |||
| 11604 | Int = Intrinsic::aarch64_neon_saddlv; | |||
| 11605 | Ty = Int32Ty; | |||
| 11606 | VTy = llvm::FixedVectorType::get(Int16Ty, 4); | |||
| 11607 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11608 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11609 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11610 | } | |||
| 11611 | case NEON::BI__builtin_neon_vaddlvq_s8: { | |||
| 11612 | Int = Intrinsic::aarch64_neon_saddlv; | |||
| 11613 | Ty = Int32Ty; | |||
| 11614 | VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 11615 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11616 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11617 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11618 | return Builder.CreateTrunc(Ops[0], Int16Ty); | |||
| 11619 | } | |||
| 11620 | case NEON::BI__builtin_neon_vaddlvq_s16: { | |||
| 11621 | Int = Intrinsic::aarch64_neon_saddlv; | |||
| 11622 | Ty = Int32Ty; | |||
| 11623 | VTy = llvm::FixedVectorType::get(Int16Ty, 8); | |||
| 11624 | llvm::Type *Tys[2] = { Ty, VTy }; | |||
| 11625 | Ops.push_back(EmitScalarExpr(E->getArg(0))); | |||
| 11626 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); | |||
| 11627 | } | |||
| 11628 | case NEON::BI__builtin_neon_vsri_n_v: | |||
| 11629 | case NEON::BI__builtin_neon_vsriq_n_v: { | |||
| 11630 | Int = Intrinsic::aarch64_neon_vsri; | |||
| 11631 | llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty); | |||
| 11632 | return EmitNeonCall(Intrin, Ops, "vsri_n"); | |||
| 11633 | } | |||
| 11634 | case NEON::BI__builtin_neon_vsli_n_v: | |||
| 11635 | case NEON::BI__builtin_neon_vsliq_n_v: { | |||
| 11636 | Int = Intrinsic::aarch64_neon_vsli; | |||
| 11637 | llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty); | |||
| 11638 | return EmitNeonCall(Intrin, Ops, "vsli_n"); | |||
| 11639 | } | |||
| 11640 | case NEON::BI__builtin_neon_vsra_n_v: | |||
| 11641 | case NEON::BI__builtin_neon_vsraq_n_v: | |||
| 11642 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11643 | Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n"); | |||
| 11644 | return Builder.CreateAdd(Ops[0], Ops[1]); | |||
| 11645 | case NEON::BI__builtin_neon_vrsra_n_v: | |||
| 11646 | case NEON::BI__builtin_neon_vrsraq_n_v: { | |||
| 11647 | Int = usgn ? Intrinsic::aarch64_neon_urshl : Intrinsic::aarch64_neon_srshl; | |||
| 11648 | SmallVector<llvm::Value*,2> TmpOps; | |||
| 11649 | TmpOps.push_back(Ops[1]); | |||
| 11650 | TmpOps.push_back(Ops[2]); | |||
| 11651 | Function* F = CGM.getIntrinsic(Int, Ty); | |||
| 11652 | llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vrshr_n", 1, true); | |||
| 11653 | Ops[0] = Builder.CreateBitCast(Ops[0], VTy); | |||
| 11654 | return Builder.CreateAdd(Ops[0], tmp); | |||
| 11655 | } | |||
| 11656 | case NEON::BI__builtin_neon_vld1_v: | |||
| 11657 | case NEON::BI__builtin_neon_vld1q_v: { | |||
| 11658 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy)); | |||
| 11659 | return Builder.CreateAlignedLoad(VTy, Ops[0], PtrOp0.getAlignment()); | |||
| 11660 | } | |||
| 11661 | case NEON::BI__builtin_neon_vst1_v: | |||
| 11662 | case NEON::BI__builtin_neon_vst1q_v: | |||
| 11663 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy)); | |||
| 11664 | Ops[1] = Builder.CreateBitCast(Ops[1], VTy); | |||
| 11665 | return Builder.CreateAlignedStore(Ops[1], Ops[0], PtrOp0.getAlignment()); | |||
| 11666 | case NEON::BI__builtin_neon_vld1_lane_v: | |||
| 11667 | case NEON::BI__builtin_neon_vld1q_lane_v: { | |||
| 11668 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11669 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 11670 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11671 | Ops[0] = Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], | |||
| 11672 | PtrOp0.getAlignment()); | |||
| 11673 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vld1_lane"); | |||
| 11674 | } | |||
| 11675 | case NEON::BI__builtin_neon_vld1_dup_v: | |||
| 11676 | case NEON::BI__builtin_neon_vld1q_dup_v: { | |||
| 11677 | Value *V = UndefValue::get(Ty); | |||
| 11678 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 11679 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11680 | Ops[0] = Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], | |||
| 11681 | PtrOp0.getAlignment()); | |||
| 11682 | llvm::Constant *CI = ConstantInt::get(Int32Ty, 0); | |||
| 11683 | Ops[0] = Builder.CreateInsertElement(V, Ops[0], CI); | |||
| 11684 | return EmitNeonSplat(Ops[0], CI); | |||
| 11685 | } | |||
| 11686 | case NEON::BI__builtin_neon_vst1_lane_v: | |||
| 11687 | case NEON::BI__builtin_neon_vst1q_lane_v: | |||
| 11688 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11689 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]); | |||
| 11690 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
| 11691 | return Builder.CreateAlignedStore(Ops[1], Builder.CreateBitCast(Ops[0], Ty), | |||
| 11692 | PtrOp0.getAlignment()); | |||
| 11693 | case NEON::BI__builtin_neon_vld2_v: | |||
| 11694 | case NEON::BI__builtin_neon_vld2q_v: { | |||
| 11695 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); | |||
| 11696 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
| 11697 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 11698 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2, Tys); | |||
| 11699 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld2"); | |||
| 11700 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
| 11701 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
| 11702 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11703 | } | |||
| 11704 | case NEON::BI__builtin_neon_vld3_v: | |||
| 11705 | case NEON::BI__builtin_neon_vld3q_v: { | |||
| 11706 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); | |||
| 11707 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
| 11708 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 11709 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3, Tys); | |||
| 11710 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld3"); | |||
| 11711 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
| 11712 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
| 11713 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11714 | } | |||
| 11715 | case NEON::BI__builtin_neon_vld4_v: | |||
| 11716 | case NEON::BI__builtin_neon_vld4q_v: { | |||
| 11717 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); | |||
| 11718 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
| 11719 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 11720 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4, Tys); | |||
| 11721 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld4"); | |||
| 11722 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
| 11723 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
| 11724 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11725 | } | |||
| 11726 | case NEON::BI__builtin_neon_vld2_dup_v: | |||
| 11727 | case NEON::BI__builtin_neon_vld2q_dup_v: { | |||
| 11728 | llvm::Type *PTy = | |||
| 11729 | llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 11730 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
| 11731 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 11732 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2r, Tys); | |||
| 11733 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld2"); | |||
| 11734 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
| 11735 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
| 11736 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11737 | } | |||
| 11738 | case NEON::BI__builtin_neon_vld3_dup_v: | |||
| 11739 | case NEON::BI__builtin_neon_vld3q_dup_v: { | |||
| 11740 | llvm::Type *PTy = | |||
| 11741 | llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 11742 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
| 11743 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 11744 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3r, Tys); | |||
| 11745 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld3"); | |||
| 11746 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
| 11747 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
| 11748 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11749 | } | |||
| 11750 | case NEON::BI__builtin_neon_vld4_dup_v: | |||
| 11751 | case NEON::BI__builtin_neon_vld4q_dup_v: { | |||
| 11752 | llvm::Type *PTy = | |||
| 11753 | llvm::PointerType::getUnqual(VTy->getElementType()); | |||
| 11754 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); | |||
| 11755 | llvm::Type *Tys[2] = { VTy, PTy }; | |||
| 11756 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4r, Tys); | |||
| 11757 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld4"); | |||
| 11758 | Ops[0] = Builder.CreateBitCast(Ops[0], | |||
| 11759 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
| 11760 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11761 | } | |||
| 11762 | case NEON::BI__builtin_neon_vld2_lane_v: | |||
| 11763 | case NEON::BI__builtin_neon_vld2q_lane_v: { | |||
| 11764 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; | |||
| 11765 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2lane, Tys); | |||
| 11766 | std::rotate(Ops.begin() + 1, Ops.begin() + 2, Ops.end()); | |||
| 11767 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11768 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 11769 | Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty); | |||
| 11770 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld2_lane"); | |||
| 11771 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
| 11772 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11773 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11774 | } | |||
| 11775 | case NEON::BI__builtin_neon_vld3_lane_v: | |||
| 11776 | case NEON::BI__builtin_neon_vld3q_lane_v: { | |||
| 11777 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; | |||
| 11778 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3lane, Tys); | |||
| 11779 | std::rotate(Ops.begin() + 1, Ops.begin() + 2, Ops.end()); | |||
| 11780 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11781 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 11782 | Ops[3] = Builder.CreateBitCast(Ops[3], Ty); | |||
| 11783 | Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty); | |||
| 11784 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld3_lane"); | |||
| 11785 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
| 11786 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11787 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11788 | } | |||
| 11789 | case NEON::BI__builtin_neon_vld4_lane_v: | |||
| 11790 | case NEON::BI__builtin_neon_vld4q_lane_v: { | |||
| 11791 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; | |||
| 11792 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4lane, Tys); | |||
| 11793 | std::rotate(Ops.begin() + 1, Ops.begin() + 2, Ops.end()); | |||
| 11794 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11795 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 11796 | Ops[3] = Builder.CreateBitCast(Ops[3], Ty); | |||
| 11797 | Ops[4] = Builder.CreateBitCast(Ops[4], Ty); | |||
| 11798 | Ops[5] = Builder.CreateZExt(Ops[5], Int64Ty); | |||
| 11799 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld4_lane"); | |||
| 11800 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); | |||
| 11801 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); | |||
| 11802 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); | |||
| 11803 | } | |||
| 11804 | case NEON::BI__builtin_neon_vst2_v: | |||
| 11805 | case NEON::BI__builtin_neon_vst2q_v: { | |||
| 11806 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
| 11807 | llvm::Type *Tys[2] = { VTy, Ops[2]->getType() }; | |||
| 11808 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2, Tys), | |||
| 11809 | Ops, ""); | |||
| 11810 | } | |||
| 11811 | case NEON::BI__builtin_neon_vst2_lane_v: | |||
| 11812 | case NEON::BI__builtin_neon_vst2q_lane_v: { | |||
| 11813 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
| 11814 | Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); | |||
| 11815 | llvm::Type *Tys[2] = { VTy, Ops[3]->getType() }; | |||
| 11816 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2lane, Tys), | |||
| 11817 | Ops, ""); | |||
| 11818 | } | |||
| 11819 | case NEON::BI__builtin_neon_vst3_v: | |||
| 11820 | case NEON::BI__builtin_neon_vst3q_v: { | |||
| 11821 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
| 11822 | llvm::Type *Tys[2] = { VTy, Ops[3]->getType() }; | |||
| 11823 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3, Tys), | |||
| 11824 | Ops, ""); | |||
| 11825 | } | |||
| 11826 | case NEON::BI__builtin_neon_vst3_lane_v: | |||
| 11827 | case NEON::BI__builtin_neon_vst3q_lane_v: { | |||
| 11828 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
| 11829 | Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty); | |||
| 11830 | llvm::Type *Tys[2] = { VTy, Ops[4]->getType() }; | |||
| 11831 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3lane, Tys), | |||
| 11832 | Ops, ""); | |||
| 11833 | } | |||
| 11834 | case NEON::BI__builtin_neon_vst4_v: | |||
| 11835 | case NEON::BI__builtin_neon_vst4q_v: { | |||
| 11836 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
| 11837 | llvm::Type *Tys[2] = { VTy, Ops[4]->getType() }; | |||
| 11838 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4, Tys), | |||
| 11839 | Ops, ""); | |||
| 11840 | } | |||
| 11841 | case NEON::BI__builtin_neon_vst4_lane_v: | |||
| 11842 | case NEON::BI__builtin_neon_vst4q_lane_v: { | |||
| 11843 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); | |||
| 11844 | Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty); | |||
| 11845 | llvm::Type *Tys[2] = { VTy, Ops[5]->getType() }; | |||
| 11846 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4lane, Tys), | |||
| 11847 | Ops, ""); | |||
| 11848 | } | |||
| 11849 | case NEON::BI__builtin_neon_vtrn_v: | |||
| 11850 | case NEON::BI__builtin_neon_vtrnq_v: { | |||
| 11851 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
| 11852 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11853 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 11854 | Value *SV = nullptr; | |||
| 11855 | ||||
| 11856 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
| 11857 | SmallVector<int, 16> Indices; | |||
| 11858 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
| 11859 | Indices.push_back(i+vi); | |||
| 11860 | Indices.push_back(i+e+vi); | |||
| 11861 | } | |||
| 11862 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
| 11863 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn"); | |||
| 11864 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
| 11865 | } | |||
| 11866 | return SV; | |||
| 11867 | } | |||
| 11868 | case NEON::BI__builtin_neon_vuzp_v: | |||
| 11869 | case NEON::BI__builtin_neon_vuzpq_v: { | |||
| 11870 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
| 11871 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11872 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 11873 | Value *SV = nullptr; | |||
| 11874 | ||||
| 11875 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
| 11876 | SmallVector<int, 16> Indices; | |||
| 11877 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) | |||
| 11878 | Indices.push_back(2*i+vi); | |||
| 11879 | ||||
| 11880 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
| 11881 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp"); | |||
| 11882 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
| 11883 | } | |||
| 11884 | return SV; | |||
| 11885 | } | |||
| 11886 | case NEON::BI__builtin_neon_vzip_v: | |||
| 11887 | case NEON::BI__builtin_neon_vzipq_v: { | |||
| 11888 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
| 11889 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); | |||
| 11890 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); | |||
| 11891 | Value *SV = nullptr; | |||
| 11892 | ||||
| 11893 | for (unsigned vi = 0; vi != 2; ++vi) { | |||
| 11894 | SmallVector<int, 16> Indices; | |||
| 11895 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { | |||
| 11896 | Indices.push_back((i + vi*e) >> 1); | |||
| 11897 | Indices.push_back(((i + vi*e) >> 1)+e); | |||
| 11898 | } | |||
| 11899 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); | |||
| 11900 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip"); | |||
| 11901 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); | |||
| 11902 | } | |||
| 11903 | return SV; | |||
| 11904 | } | |||
| 11905 | case NEON::BI__builtin_neon_vqtbl1q_v: { | |||
| 11906 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl1, Ty), | |||
| 11907 | Ops, "vtbl1"); | |||
| 11908 | } | |||
| 11909 | case NEON::BI__builtin_neon_vqtbl2q_v: { | |||
| 11910 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl2, Ty), | |||
| 11911 | Ops, "vtbl2"); | |||
| 11912 | } | |||
| 11913 | case NEON::BI__builtin_neon_vqtbl3q_v: { | |||
| 11914 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl3, Ty), | |||
| 11915 | Ops, "vtbl3"); | |||
| 11916 | } | |||
| 11917 | case NEON::BI__builtin_neon_vqtbl4q_v: { | |||
| 11918 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl4, Ty), | |||
| 11919 | Ops, "vtbl4"); | |||
| 11920 | } | |||
| 11921 | case NEON::BI__builtin_neon_vqtbx1q_v: { | |||
| 11922 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx1, Ty), | |||
| 11923 | Ops, "vtbx1"); | |||
| 11924 | } | |||
| 11925 | case NEON::BI__builtin_neon_vqtbx2q_v: { | |||
| 11926 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx2, Ty), | |||
| 11927 | Ops, "vtbx2"); | |||
| 11928 | } | |||
| 11929 | case NEON::BI__builtin_neon_vqtbx3q_v: { | |||
| 11930 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx3, Ty), | |||
| 11931 | Ops, "vtbx3"); | |||
| 11932 | } | |||
| 11933 | case NEON::BI__builtin_neon_vqtbx4q_v: { | |||
| 11934 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx4, Ty), | |||
| 11935 | Ops, "vtbx4"); | |||
| 11936 | } | |||
| 11937 | case NEON::BI__builtin_neon_vsqadd_v: | |||
| 11938 | case NEON::BI__builtin_neon_vsqaddq_v: { | |||
| 11939 | Int = Intrinsic::aarch64_neon_usqadd; | |||
| 11940 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqadd"); | |||
| 11941 | } | |||
| 11942 | case NEON::BI__builtin_neon_vuqadd_v: | |||
| 11943 | case NEON::BI__builtin_neon_vuqaddq_v: { | |||
| 11944 | Int = Intrinsic::aarch64_neon_suqadd; | |||
| 11945 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vuqadd"); | |||
| 11946 | } | |||
| 11947 | } | |||
| 11948 | } | |||
| 11949 | ||||
| 11950 | Value *CodeGenFunction::EmitBPFBuiltinExpr(unsigned BuiltinID, | |||
| 11951 | const CallExpr *E) { | |||
| 11952 | assert((BuiltinID == BPF::BI__builtin_preserve_field_info ||(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value ) && "unexpected BPF builtin") ? void (0) : __assert_fail ("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11956, __extension__ __PRETTY_FUNCTION__ )) | |||
| 11953 | BuiltinID == BPF::BI__builtin_btf_type_id ||(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value ) && "unexpected BPF builtin") ? void (0) : __assert_fail ("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11956, __extension__ __PRETTY_FUNCTION__ )) | |||
| 11954 | BuiltinID == BPF::BI__builtin_preserve_type_info ||(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value ) && "unexpected BPF builtin") ? void (0) : __assert_fail ("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11956, __extension__ __PRETTY_FUNCTION__ )) | |||
| 11955 | BuiltinID == BPF::BI__builtin_preserve_enum_value) &&(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value ) && "unexpected BPF builtin") ? void (0) : __assert_fail ("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11956, __extension__ __PRETTY_FUNCTION__ )) | |||
| 11956 | "unexpected BPF builtin")(static_cast <bool> ((BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value ) && "unexpected BPF builtin") ? void (0) : __assert_fail ("(BuiltinID == BPF::BI__builtin_preserve_field_info || BuiltinID == BPF::BI__builtin_btf_type_id || BuiltinID == BPF::BI__builtin_preserve_type_info || BuiltinID == BPF::BI__builtin_preserve_enum_value) && \"unexpected BPF builtin\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 11956, __extension__ __PRETTY_FUNCTION__ )); | |||
| 11957 | ||||
| 11958 | // A sequence number, injected into IR builtin functions, to | |||
| 11959 | // prevent CSE given the only difference of the funciton | |||
| 11960 | // may just be the debuginfo metadata. | |||
| 11961 | static uint32_t BuiltinSeqNum; | |||
| 11962 | ||||
| 11963 | switch (BuiltinID) { | |||
| 11964 | default: | |||
| 11965 | llvm_unreachable("Unexpected BPF builtin")::llvm::llvm_unreachable_internal("Unexpected BPF builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 11965); | |||
| 11966 | case BPF::BI__builtin_preserve_field_info: { | |||
| 11967 | const Expr *Arg = E->getArg(0); | |||
| 11968 | bool IsBitField = Arg->IgnoreParens()->getObjectKind() == OK_BitField; | |||
| 11969 | ||||
| 11970 | if (!getDebugInfo()) { | |||
| 11971 | CGM.Error(E->getExprLoc(), | |||
| 11972 | "using __builtin_preserve_field_info() without -g"); | |||
| 11973 | return IsBitField ? EmitLValue(Arg).getBitFieldPointer() | |||
| 11974 | : EmitLValue(Arg).getPointer(*this); | |||
| 11975 | } | |||
| 11976 | ||||
| 11977 | // Enable underlying preserve_*_access_index() generation. | |||
| 11978 | bool OldIsInPreservedAIRegion = IsInPreservedAIRegion; | |||
| 11979 | IsInPreservedAIRegion = true; | |||
| 11980 | Value *FieldAddr = IsBitField ? EmitLValue(Arg).getBitFieldPointer() | |||
| 11981 | : EmitLValue(Arg).getPointer(*this); | |||
| 11982 | IsInPreservedAIRegion = OldIsInPreservedAIRegion; | |||
| 11983 | ||||
| 11984 | ConstantInt *C = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
| 11985 | Value *InfoKind = ConstantInt::get(Int64Ty, C->getSExtValue()); | |||
| 11986 | ||||
| 11987 | // Built the IR for the preserve_field_info intrinsic. | |||
| 11988 | llvm::Function *FnGetFieldInfo = llvm::Intrinsic::getDeclaration( | |||
| 11989 | &CGM.getModule(), llvm::Intrinsic::bpf_preserve_field_info, | |||
| 11990 | {FieldAddr->getType()}); | |||
| 11991 | return Builder.CreateCall(FnGetFieldInfo, {FieldAddr, InfoKind}); | |||
| 11992 | } | |||
| 11993 | case BPF::BI__builtin_btf_type_id: | |||
| 11994 | case BPF::BI__builtin_preserve_type_info: { | |||
| 11995 | if (!getDebugInfo()) { | |||
| 11996 | CGM.Error(E->getExprLoc(), "using builtin function without -g"); | |||
| 11997 | return nullptr; | |||
| 11998 | } | |||
| 11999 | ||||
| 12000 | const Expr *Arg0 = E->getArg(0); | |||
| 12001 | llvm::DIType *DbgInfo = getDebugInfo()->getOrCreateStandaloneType( | |||
| 12002 | Arg0->getType(), Arg0->getExprLoc()); | |||
| 12003 | ||||
| 12004 | ConstantInt *Flag = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
| 12005 | Value *FlagValue = ConstantInt::get(Int64Ty, Flag->getSExtValue()); | |||
| 12006 | Value *SeqNumVal = ConstantInt::get(Int32Ty, BuiltinSeqNum++); | |||
| 12007 | ||||
| 12008 | llvm::Function *FnDecl; | |||
| 12009 | if (BuiltinID == BPF::BI__builtin_btf_type_id) | |||
| 12010 | FnDecl = llvm::Intrinsic::getDeclaration( | |||
| 12011 | &CGM.getModule(), llvm::Intrinsic::bpf_btf_type_id, {}); | |||
| 12012 | else | |||
| 12013 | FnDecl = llvm::Intrinsic::getDeclaration( | |||
| 12014 | &CGM.getModule(), llvm::Intrinsic::bpf_preserve_type_info, {}); | |||
| 12015 | CallInst *Fn = Builder.CreateCall(FnDecl, {SeqNumVal, FlagValue}); | |||
| 12016 | Fn->setMetadata(LLVMContext::MD_preserve_access_index, DbgInfo); | |||
| 12017 | return Fn; | |||
| 12018 | } | |||
| 12019 | case BPF::BI__builtin_preserve_enum_value: { | |||
| 12020 | if (!getDebugInfo()) { | |||
| 12021 | CGM.Error(E->getExprLoc(), "using builtin function without -g"); | |||
| 12022 | return nullptr; | |||
| 12023 | } | |||
| 12024 | ||||
| 12025 | const Expr *Arg0 = E->getArg(0); | |||
| 12026 | llvm::DIType *DbgInfo = getDebugInfo()->getOrCreateStandaloneType( | |||
| 12027 | Arg0->getType(), Arg0->getExprLoc()); | |||
| 12028 | ||||
| 12029 | // Find enumerator | |||
| 12030 | const auto *UO = cast<UnaryOperator>(Arg0->IgnoreParens()); | |||
| 12031 | const auto *CE = cast<CStyleCastExpr>(UO->getSubExpr()); | |||
| 12032 | const auto *DR = cast<DeclRefExpr>(CE->getSubExpr()); | |||
| 12033 | const auto *Enumerator = cast<EnumConstantDecl>(DR->getDecl()); | |||
| 12034 | ||||
| 12035 | auto &InitVal = Enumerator->getInitVal(); | |||
| 12036 | std::string InitValStr; | |||
| 12037 | if (InitVal.isNegative() || InitVal > uint64_t(INT64_MAX(9223372036854775807L))) | |||
| 12038 | InitValStr = std::to_string(InitVal.getSExtValue()); | |||
| 12039 | else | |||
| 12040 | InitValStr = std::to_string(InitVal.getZExtValue()); | |||
| 12041 | std::string EnumStr = Enumerator->getNameAsString() + ":" + InitValStr; | |||
| 12042 | Value *EnumStrVal = Builder.CreateGlobalStringPtr(EnumStr); | |||
| 12043 | ||||
| 12044 | ConstantInt *Flag = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); | |||
| 12045 | Value *FlagValue = ConstantInt::get(Int64Ty, Flag->getSExtValue()); | |||
| 12046 | Value *SeqNumVal = ConstantInt::get(Int32Ty, BuiltinSeqNum++); | |||
| 12047 | ||||
| 12048 | llvm::Function *IntrinsicFn = llvm::Intrinsic::getDeclaration( | |||
| 12049 | &CGM.getModule(), llvm::Intrinsic::bpf_preserve_enum_value, {}); | |||
| 12050 | CallInst *Fn = | |||
| 12051 | Builder.CreateCall(IntrinsicFn, {SeqNumVal, EnumStrVal, FlagValue}); | |||
| 12052 | Fn->setMetadata(LLVMContext::MD_preserve_access_index, DbgInfo); | |||
| 12053 | return Fn; | |||
| 12054 | } | |||
| 12055 | } | |||
| 12056 | } | |||
| 12057 | ||||
| 12058 | llvm::Value *CodeGenFunction:: | |||
| 12059 | BuildVector(ArrayRef<llvm::Value*> Ops) { | |||
| 12060 | assert((Ops.size() & (Ops.size() - 1)) == 0 &&(static_cast <bool> ((Ops.size() & (Ops.size() - 1) ) == 0 && "Not a power-of-two sized vector!") ? void ( 0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12061, __extension__ __PRETTY_FUNCTION__ )) | |||
| 12061 | "Not a power-of-two sized vector!")(static_cast <bool> ((Ops.size() & (Ops.size() - 1) ) == 0 && "Not a power-of-two sized vector!") ? void ( 0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12061, __extension__ __PRETTY_FUNCTION__ )); | |||
| 12062 | bool AllConstants = true; | |||
| 12063 | for (unsigned i = 0, e = Ops.size(); i != e && AllConstants; ++i) | |||
| 12064 | AllConstants &= isa<Constant>(Ops[i]); | |||
| 12065 | ||||
| 12066 | // If this is a constant vector, create a ConstantVector. | |||
| 12067 | if (AllConstants) { | |||
| 12068 | SmallVector<llvm::Constant*, 16> CstOps; | |||
| 12069 | for (unsigned i = 0, e = Ops.size(); i != e; ++i) | |||
| 12070 | CstOps.push_back(cast<Constant>(Ops[i])); | |||
| 12071 | return llvm::ConstantVector::get(CstOps); | |||
| 12072 | } | |||
| 12073 | ||||
| 12074 | // Otherwise, insertelement the values to build the vector. | |||
| 12075 | Value *Result = llvm::UndefValue::get( | |||
| 12076 | llvm::FixedVectorType::get(Ops[0]->getType(), Ops.size())); | |||
| 12077 | ||||
| 12078 | for (unsigned i = 0, e = Ops.size(); i != e; ++i) | |||
| 12079 | Result = Builder.CreateInsertElement(Result, Ops[i], Builder.getInt32(i)); | |||
| 12080 | ||||
| 12081 | return Result; | |||
| 12082 | } | |||
| 12083 | ||||
| 12084 | // Convert the mask from an integer type to a vector of i1. | |||
| 12085 | static Value *getMaskVecValue(CodeGenFunction &CGF, Value *Mask, | |||
| 12086 | unsigned NumElts) { | |||
| 12087 | ||||
| 12088 | auto *MaskTy = llvm::FixedVectorType::get( | |||
| 12089 | CGF.Builder.getInt1Ty(), | |||
| 12090 | cast<IntegerType>(Mask->getType())->getBitWidth()); | |||
| 12091 | Value *MaskVec = CGF.Builder.CreateBitCast(Mask, MaskTy); | |||
| 12092 | ||||
| 12093 | // If we have less than 8 elements, then the starting mask was an i8 and | |||
| 12094 | // we need to extract down to the right number of elements. | |||
| 12095 | if (NumElts < 8) { | |||
| 12096 | int Indices[4]; | |||
| 12097 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 12098 | Indices[i] = i; | |||
| 12099 | MaskVec = CGF.Builder.CreateShuffleVector(MaskVec, MaskVec, | |||
| 12100 | makeArrayRef(Indices, NumElts), | |||
| 12101 | "extract"); | |||
| 12102 | } | |||
| 12103 | return MaskVec; | |||
| 12104 | } | |||
| 12105 | ||||
| 12106 | static Value *EmitX86MaskedStore(CodeGenFunction &CGF, ArrayRef<Value *> Ops, | |||
| 12107 | Align Alignment) { | |||
| 12108 | // Cast the pointer to right type. | |||
| 12109 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], | |||
| 12110 | llvm::PointerType::getUnqual(Ops[1]->getType())); | |||
| 12111 | ||||
| 12112 | Value *MaskVec = getMaskVecValue( | |||
| 12113 | CGF, Ops[2], | |||
| 12114 | cast<llvm::FixedVectorType>(Ops[1]->getType())->getNumElements()); | |||
| 12115 | ||||
| 12116 | return CGF.Builder.CreateMaskedStore(Ops[1], Ptr, Alignment, MaskVec); | |||
| 12117 | } | |||
| 12118 | ||||
| 12119 | static Value *EmitX86MaskedLoad(CodeGenFunction &CGF, ArrayRef<Value *> Ops, | |||
| 12120 | Align Alignment) { | |||
| 12121 | // Cast the pointer to right type. | |||
| 12122 | llvm::Type *Ty = Ops[1]->getType(); | |||
| 12123 | Value *Ptr = | |||
| 12124 | CGF.Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); | |||
| 12125 | ||||
| 12126 | Value *MaskVec = getMaskVecValue( | |||
| 12127 | CGF, Ops[2], cast<llvm::FixedVectorType>(Ty)->getNumElements()); | |||
| 12128 | ||||
| 12129 | return CGF.Builder.CreateMaskedLoad(Ty, Ptr, Alignment, MaskVec, Ops[1]); | |||
| 12130 | } | |||
| 12131 | ||||
| 12132 | static Value *EmitX86ExpandLoad(CodeGenFunction &CGF, | |||
| 12133 | ArrayRef<Value *> Ops) { | |||
| 12134 | auto *ResultTy = cast<llvm::VectorType>(Ops[1]->getType()); | |||
| 12135 | llvm::Type *PtrTy = ResultTy->getElementType(); | |||
| 12136 | ||||
| 12137 | // Cast the pointer to element type. | |||
| 12138 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], | |||
| 12139 | llvm::PointerType::getUnqual(PtrTy)); | |||
| 12140 | ||||
| 12141 | Value *MaskVec = getMaskVecValue( | |||
| 12142 | CGF, Ops[2], cast<FixedVectorType>(ResultTy)->getNumElements()); | |||
| 12143 | ||||
| 12144 | llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_expandload, | |||
| 12145 | ResultTy); | |||
| 12146 | return CGF.Builder.CreateCall(F, { Ptr, MaskVec, Ops[1] }); | |||
| 12147 | } | |||
| 12148 | ||||
| 12149 | static Value *EmitX86CompressExpand(CodeGenFunction &CGF, | |||
| 12150 | ArrayRef<Value *> Ops, | |||
| 12151 | bool IsCompress) { | |||
| 12152 | auto *ResultTy = cast<llvm::FixedVectorType>(Ops[1]->getType()); | |||
| 12153 | ||||
| 12154 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], ResultTy->getNumElements()); | |||
| 12155 | ||||
| 12156 | Intrinsic::ID IID = IsCompress ? Intrinsic::x86_avx512_mask_compress | |||
| 12157 | : Intrinsic::x86_avx512_mask_expand; | |||
| 12158 | llvm::Function *F = CGF.CGM.getIntrinsic(IID, ResultTy); | |||
| 12159 | return CGF.Builder.CreateCall(F, { Ops[0], Ops[1], MaskVec }); | |||
| 12160 | } | |||
| 12161 | ||||
| 12162 | static Value *EmitX86CompressStore(CodeGenFunction &CGF, | |||
| 12163 | ArrayRef<Value *> Ops) { | |||
| 12164 | auto *ResultTy = cast<llvm::FixedVectorType>(Ops[1]->getType()); | |||
| 12165 | llvm::Type *PtrTy = ResultTy->getElementType(); | |||
| 12166 | ||||
| 12167 | // Cast the pointer to element type. | |||
| 12168 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], | |||
| 12169 | llvm::PointerType::getUnqual(PtrTy)); | |||
| 12170 | ||||
| 12171 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], ResultTy->getNumElements()); | |||
| 12172 | ||||
| 12173 | llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_compressstore, | |||
| 12174 | ResultTy); | |||
| 12175 | return CGF.Builder.CreateCall(F, { Ops[1], Ptr, MaskVec }); | |||
| 12176 | } | |||
| 12177 | ||||
| 12178 | static Value *EmitX86MaskLogic(CodeGenFunction &CGF, Instruction::BinaryOps Opc, | |||
| 12179 | ArrayRef<Value *> Ops, | |||
| 12180 | bool InvertLHS = false) { | |||
| 12181 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 12182 | Value *LHS = getMaskVecValue(CGF, Ops[0], NumElts); | |||
| 12183 | Value *RHS = getMaskVecValue(CGF, Ops[1], NumElts); | |||
| 12184 | ||||
| 12185 | if (InvertLHS) | |||
| 12186 | LHS = CGF.Builder.CreateNot(LHS); | |||
| 12187 | ||||
| 12188 | return CGF.Builder.CreateBitCast(CGF.Builder.CreateBinOp(Opc, LHS, RHS), | |||
| 12189 | Ops[0]->getType()); | |||
| 12190 | } | |||
| 12191 | ||||
| 12192 | static Value *EmitX86FunnelShift(CodeGenFunction &CGF, Value *Op0, Value *Op1, | |||
| 12193 | Value *Amt, bool IsRight) { | |||
| 12194 | llvm::Type *Ty = Op0->getType(); | |||
| 12195 | ||||
| 12196 | // Amount may be scalar immediate, in which case create a splat vector. | |||
| 12197 | // Funnel shifts amounts are treated as modulo and types are all power-of-2 so | |||
| 12198 | // we only care about the lowest log2 bits anyway. | |||
| 12199 | if (Amt->getType() != Ty) { | |||
| 12200 | unsigned NumElts = cast<llvm::FixedVectorType>(Ty)->getNumElements(); | |||
| 12201 | Amt = CGF.Builder.CreateIntCast(Amt, Ty->getScalarType(), false); | |||
| 12202 | Amt = CGF.Builder.CreateVectorSplat(NumElts, Amt); | |||
| 12203 | } | |||
| 12204 | ||||
| 12205 | unsigned IID = IsRight ? Intrinsic::fshr : Intrinsic::fshl; | |||
| 12206 | Function *F = CGF.CGM.getIntrinsic(IID, Ty); | |||
| 12207 | return CGF.Builder.CreateCall(F, {Op0, Op1, Amt}); | |||
| 12208 | } | |||
| 12209 | ||||
| 12210 | static Value *EmitX86vpcom(CodeGenFunction &CGF, ArrayRef<Value *> Ops, | |||
| 12211 | bool IsSigned) { | |||
| 12212 | Value *Op0 = Ops[0]; | |||
| 12213 | Value *Op1 = Ops[1]; | |||
| 12214 | llvm::Type *Ty = Op0->getType(); | |||
| 12215 | uint64_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; | |||
| 12216 | ||||
| 12217 | CmpInst::Predicate Pred; | |||
| 12218 | switch (Imm) { | |||
| 12219 | case 0x0: | |||
| 12220 | Pred = IsSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; | |||
| 12221 | break; | |||
| 12222 | case 0x1: | |||
| 12223 | Pred = IsSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; | |||
| 12224 | break; | |||
| 12225 | case 0x2: | |||
| 12226 | Pred = IsSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; | |||
| 12227 | break; | |||
| 12228 | case 0x3: | |||
| 12229 | Pred = IsSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; | |||
| 12230 | break; | |||
| 12231 | case 0x4: | |||
| 12232 | Pred = ICmpInst::ICMP_EQ; | |||
| 12233 | break; | |||
| 12234 | case 0x5: | |||
| 12235 | Pred = ICmpInst::ICMP_NE; | |||
| 12236 | break; | |||
| 12237 | case 0x6: | |||
| 12238 | return llvm::Constant::getNullValue(Ty); // FALSE | |||
| 12239 | case 0x7: | |||
| 12240 | return llvm::Constant::getAllOnesValue(Ty); // TRUE | |||
| 12241 | default: | |||
| 12242 | llvm_unreachable("Unexpected XOP vpcom/vpcomu predicate")::llvm::llvm_unreachable_internal("Unexpected XOP vpcom/vpcomu predicate" , "clang/lib/CodeGen/CGBuiltin.cpp", 12242); | |||
| 12243 | } | |||
| 12244 | ||||
| 12245 | Value *Cmp = CGF.Builder.CreateICmp(Pred, Op0, Op1); | |||
| 12246 | Value *Res = CGF.Builder.CreateSExt(Cmp, Ty); | |||
| 12247 | return Res; | |||
| 12248 | } | |||
| 12249 | ||||
| 12250 | static Value *EmitX86Select(CodeGenFunction &CGF, | |||
| 12251 | Value *Mask, Value *Op0, Value *Op1) { | |||
| 12252 | ||||
| 12253 | // If the mask is all ones just return first argument. | |||
| 12254 | if (const auto *C = dyn_cast<Constant>(Mask)) | |||
| 12255 | if (C->isAllOnesValue()) | |||
| 12256 | return Op0; | |||
| 12257 | ||||
| 12258 | Mask = getMaskVecValue( | |||
| 12259 | CGF, Mask, cast<llvm::FixedVectorType>(Op0->getType())->getNumElements()); | |||
| 12260 | ||||
| 12261 | return CGF.Builder.CreateSelect(Mask, Op0, Op1); | |||
| 12262 | } | |||
| 12263 | ||||
| 12264 | static Value *EmitX86ScalarSelect(CodeGenFunction &CGF, | |||
| 12265 | Value *Mask, Value *Op0, Value *Op1) { | |||
| 12266 | // If the mask is all ones just return first argument. | |||
| 12267 | if (const auto *C = dyn_cast<Constant>(Mask)) | |||
| 12268 | if (C->isAllOnesValue()) | |||
| 12269 | return Op0; | |||
| 12270 | ||||
| 12271 | auto *MaskTy = llvm::FixedVectorType::get( | |||
| 12272 | CGF.Builder.getInt1Ty(), Mask->getType()->getIntegerBitWidth()); | |||
| 12273 | Mask = CGF.Builder.CreateBitCast(Mask, MaskTy); | |||
| 12274 | Mask = CGF.Builder.CreateExtractElement(Mask, (uint64_t)0); | |||
| 12275 | return CGF.Builder.CreateSelect(Mask, Op0, Op1); | |||
| 12276 | } | |||
| 12277 | ||||
| 12278 | static Value *EmitX86MaskedCompareResult(CodeGenFunction &CGF, Value *Cmp, | |||
| 12279 | unsigned NumElts, Value *MaskIn) { | |||
| 12280 | if (MaskIn) { | |||
| 12281 | const auto *C = dyn_cast<Constant>(MaskIn); | |||
| 12282 | if (!C || !C->isAllOnesValue()) | |||
| 12283 | Cmp = CGF.Builder.CreateAnd(Cmp, getMaskVecValue(CGF, MaskIn, NumElts)); | |||
| 12284 | } | |||
| 12285 | ||||
| 12286 | if (NumElts < 8) { | |||
| 12287 | int Indices[8]; | |||
| 12288 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 12289 | Indices[i] = i; | |||
| 12290 | for (unsigned i = NumElts; i != 8; ++i) | |||
| 12291 | Indices[i] = i % NumElts + NumElts; | |||
| 12292 | Cmp = CGF.Builder.CreateShuffleVector( | |||
| 12293 | Cmp, llvm::Constant::getNullValue(Cmp->getType()), Indices); | |||
| 12294 | } | |||
| 12295 | ||||
| 12296 | return CGF.Builder.CreateBitCast(Cmp, | |||
| 12297 | IntegerType::get(CGF.getLLVMContext(), | |||
| 12298 | std::max(NumElts, 8U))); | |||
| 12299 | } | |||
| 12300 | ||||
| 12301 | static Value *EmitX86MaskedCompare(CodeGenFunction &CGF, unsigned CC, | |||
| 12302 | bool Signed, ArrayRef<Value *> Ops) { | |||
| 12303 | assert((Ops.size() == 2 || Ops.size() == 4) &&(static_cast <bool> ((Ops.size() == 2 || Ops.size() == 4 ) && "Unexpected number of arguments") ? void (0) : __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12304, __extension__ __PRETTY_FUNCTION__ )) | |||
| 12304 | "Unexpected number of arguments")(static_cast <bool> ((Ops.size() == 2 || Ops.size() == 4 ) && "Unexpected number of arguments") ? void (0) : __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12304, __extension__ __PRETTY_FUNCTION__ )); | |||
| 12305 | unsigned NumElts = | |||
| 12306 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 12307 | Value *Cmp; | |||
| 12308 | ||||
| 12309 | if (CC == 3) { | |||
| 12310 | Cmp = Constant::getNullValue( | |||
| 12311 | llvm::FixedVectorType::get(CGF.Builder.getInt1Ty(), NumElts)); | |||
| 12312 | } else if (CC == 7) { | |||
| 12313 | Cmp = Constant::getAllOnesValue( | |||
| 12314 | llvm::FixedVectorType::get(CGF.Builder.getInt1Ty(), NumElts)); | |||
| 12315 | } else { | |||
| 12316 | ICmpInst::Predicate Pred; | |||
| 12317 | switch (CC) { | |||
| 12318 | default: llvm_unreachable("Unknown condition code")::llvm::llvm_unreachable_internal("Unknown condition code", "clang/lib/CodeGen/CGBuiltin.cpp" , 12318); | |||
| 12319 | case 0: Pred = ICmpInst::ICMP_EQ; break; | |||
| 12320 | case 1: Pred = Signed ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; break; | |||
| 12321 | case 2: Pred = Signed ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; break; | |||
| 12322 | case 4: Pred = ICmpInst::ICMP_NE; break; | |||
| 12323 | case 5: Pred = Signed ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; break; | |||
| 12324 | case 6: Pred = Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; break; | |||
| 12325 | } | |||
| 12326 | Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]); | |||
| 12327 | } | |||
| 12328 | ||||
| 12329 | Value *MaskIn = nullptr; | |||
| 12330 | if (Ops.size() == 4) | |||
| 12331 | MaskIn = Ops[3]; | |||
| 12332 | ||||
| 12333 | return EmitX86MaskedCompareResult(CGF, Cmp, NumElts, MaskIn); | |||
| 12334 | } | |||
| 12335 | ||||
| 12336 | static Value *EmitX86ConvertToMask(CodeGenFunction &CGF, Value *In) { | |||
| 12337 | Value *Zero = Constant::getNullValue(In->getType()); | |||
| 12338 | return EmitX86MaskedCompare(CGF, 1, true, { In, Zero }); | |||
| 12339 | } | |||
| 12340 | ||||
| 12341 | static Value *EmitX86ConvertIntToFp(CodeGenFunction &CGF, const CallExpr *E, | |||
| 12342 | ArrayRef<Value *> Ops, bool IsSigned) { | |||
| 12343 | unsigned Rnd = cast<llvm::ConstantInt>(Ops[3])->getZExtValue(); | |||
| 12344 | llvm::Type *Ty = Ops[1]->getType(); | |||
| 12345 | ||||
| 12346 | Value *Res; | |||
| 12347 | if (Rnd != 4) { | |||
| 12348 | Intrinsic::ID IID = IsSigned ? Intrinsic::x86_avx512_sitofp_round | |||
| 12349 | : Intrinsic::x86_avx512_uitofp_round; | |||
| 12350 | Function *F = CGF.CGM.getIntrinsic(IID, { Ty, Ops[0]->getType() }); | |||
| 12351 | Res = CGF.Builder.CreateCall(F, { Ops[0], Ops[3] }); | |||
| 12352 | } else { | |||
| 12353 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); | |||
| 12354 | Res = IsSigned ? CGF.Builder.CreateSIToFP(Ops[0], Ty) | |||
| 12355 | : CGF.Builder.CreateUIToFP(Ops[0], Ty); | |||
| 12356 | } | |||
| 12357 | ||||
| 12358 | return EmitX86Select(CGF, Ops[2], Res, Ops[1]); | |||
| 12359 | } | |||
| 12360 | ||||
| 12361 | // Lowers X86 FMA intrinsics to IR. | |||
| 12362 | static Value *EmitX86FMAExpr(CodeGenFunction &CGF, const CallExpr *E, | |||
| 12363 | ArrayRef<Value *> Ops, unsigned BuiltinID, | |||
| 12364 | bool IsAddSub) { | |||
| 12365 | ||||
| 12366 | bool Subtract = false; | |||
| 12367 | Intrinsic::ID IID = Intrinsic::not_intrinsic; | |||
| 12368 | switch (BuiltinID) { | |||
| 12369 | default: break; | |||
| 12370 | case clang::X86::BI__builtin_ia32_vfmsubph512_mask3: | |||
| 12371 | Subtract = true; | |||
| 12372 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 12373 | case clang::X86::BI__builtin_ia32_vfmaddph512_mask: | |||
| 12374 | case clang::X86::BI__builtin_ia32_vfmaddph512_maskz: | |||
| 12375 | case clang::X86::BI__builtin_ia32_vfmaddph512_mask3: | |||
| 12376 | IID = llvm::Intrinsic::x86_avx512fp16_vfmadd_ph_512; | |||
| 12377 | break; | |||
| 12378 | case clang::X86::BI__builtin_ia32_vfmsubaddph512_mask3: | |||
| 12379 | Subtract = true; | |||
| 12380 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 12381 | case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask: | |||
| 12382 | case clang::X86::BI__builtin_ia32_vfmaddsubph512_maskz: | |||
| 12383 | case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask3: | |||
| 12384 | IID = llvm::Intrinsic::x86_avx512fp16_vfmaddsub_ph_512; | |||
| 12385 | break; | |||
| 12386 | case clang::X86::BI__builtin_ia32_vfmsubps512_mask3: | |||
| 12387 | Subtract = true; | |||
| 12388 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 12389 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask: | |||
| 12390 | case clang::X86::BI__builtin_ia32_vfmaddps512_maskz: | |||
| 12391 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask3: | |||
| 12392 | IID = llvm::Intrinsic::x86_avx512_vfmadd_ps_512; break; | |||
| 12393 | case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3: | |||
| 12394 | Subtract = true; | |||
| 12395 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 12396 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask: | |||
| 12397 | case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz: | |||
| 12398 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3: | |||
| 12399 | IID = llvm::Intrinsic::x86_avx512_vfmadd_pd_512; break; | |||
| 12400 | case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3: | |||
| 12401 | Subtract = true; | |||
| 12402 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 12403 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask: | |||
| 12404 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz: | |||
| 12405 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3: | |||
| 12406 | IID = llvm::Intrinsic::x86_avx512_vfmaddsub_ps_512; | |||
| 12407 | break; | |||
| 12408 | case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3: | |||
| 12409 | Subtract = true; | |||
| 12410 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 12411 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask: | |||
| 12412 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz: | |||
| 12413 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3: | |||
| 12414 | IID = llvm::Intrinsic::x86_avx512_vfmaddsub_pd_512; | |||
| 12415 | break; | |||
| 12416 | } | |||
| 12417 | ||||
| 12418 | Value *A = Ops[0]; | |||
| 12419 | Value *B = Ops[1]; | |||
| 12420 | Value *C = Ops[2]; | |||
| 12421 | ||||
| 12422 | if (Subtract) | |||
| 12423 | C = CGF.Builder.CreateFNeg(C); | |||
| 12424 | ||||
| 12425 | Value *Res; | |||
| 12426 | ||||
| 12427 | // Only handle in case of _MM_FROUND_CUR_DIRECTION/4 (no rounding). | |||
| 12428 | if (IID != Intrinsic::not_intrinsic && | |||
| 12429 | (cast<llvm::ConstantInt>(Ops.back())->getZExtValue() != (uint64_t)4 || | |||
| 12430 | IsAddSub)) { | |||
| 12431 | Function *Intr = CGF.CGM.getIntrinsic(IID); | |||
| 12432 | Res = CGF.Builder.CreateCall(Intr, {A, B, C, Ops.back() }); | |||
| 12433 | } else { | |||
| 12434 | llvm::Type *Ty = A->getType(); | |||
| 12435 | Function *FMA; | |||
| 12436 | if (CGF.Builder.getIsFPConstrained()) { | |||
| 12437 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); | |||
| 12438 | FMA = CGF.CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, Ty); | |||
| 12439 | Res = CGF.Builder.CreateConstrainedFPCall(FMA, {A, B, C}); | |||
| 12440 | } else { | |||
| 12441 | FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ty); | |||
| 12442 | Res = CGF.Builder.CreateCall(FMA, {A, B, C}); | |||
| 12443 | } | |||
| 12444 | } | |||
| 12445 | ||||
| 12446 | // Handle any required masking. | |||
| 12447 | Value *MaskFalseVal = nullptr; | |||
| 12448 | switch (BuiltinID) { | |||
| 12449 | case clang::X86::BI__builtin_ia32_vfmaddph512_mask: | |||
| 12450 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask: | |||
| 12451 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask: | |||
| 12452 | case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask: | |||
| 12453 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask: | |||
| 12454 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask: | |||
| 12455 | MaskFalseVal = Ops[0]; | |||
| 12456 | break; | |||
| 12457 | case clang::X86::BI__builtin_ia32_vfmaddph512_maskz: | |||
| 12458 | case clang::X86::BI__builtin_ia32_vfmaddps512_maskz: | |||
| 12459 | case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz: | |||
| 12460 | case clang::X86::BI__builtin_ia32_vfmaddsubph512_maskz: | |||
| 12461 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz: | |||
| 12462 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz: | |||
| 12463 | MaskFalseVal = Constant::getNullValue(Ops[0]->getType()); | |||
| 12464 | break; | |||
| 12465 | case clang::X86::BI__builtin_ia32_vfmsubph512_mask3: | |||
| 12466 | case clang::X86::BI__builtin_ia32_vfmaddph512_mask3: | |||
| 12467 | case clang::X86::BI__builtin_ia32_vfmsubps512_mask3: | |||
| 12468 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask3: | |||
| 12469 | case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3: | |||
| 12470 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3: | |||
| 12471 | case clang::X86::BI__builtin_ia32_vfmsubaddph512_mask3: | |||
| 12472 | case clang::X86::BI__builtin_ia32_vfmaddsubph512_mask3: | |||
| 12473 | case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3: | |||
| 12474 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3: | |||
| 12475 | case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3: | |||
| 12476 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3: | |||
| 12477 | MaskFalseVal = Ops[2]; | |||
| 12478 | break; | |||
| 12479 | } | |||
| 12480 | ||||
| 12481 | if (MaskFalseVal) | |||
| 12482 | return EmitX86Select(CGF, Ops[3], Res, MaskFalseVal); | |||
| 12483 | ||||
| 12484 | return Res; | |||
| 12485 | } | |||
| 12486 | ||||
| 12487 | static Value *EmitScalarFMAExpr(CodeGenFunction &CGF, const CallExpr *E, | |||
| 12488 | MutableArrayRef<Value *> Ops, Value *Upper, | |||
| 12489 | bool ZeroMask = false, unsigned PTIdx = 0, | |||
| 12490 | bool NegAcc = false) { | |||
| 12491 | unsigned Rnd = 4; | |||
| 12492 | if (Ops.size() > 4) | |||
| 12493 | Rnd = cast<llvm::ConstantInt>(Ops[4])->getZExtValue(); | |||
| 12494 | ||||
| 12495 | if (NegAcc) | |||
| 12496 | Ops[2] = CGF.Builder.CreateFNeg(Ops[2]); | |||
| 12497 | ||||
| 12498 | Ops[0] = CGF.Builder.CreateExtractElement(Ops[0], (uint64_t)0); | |||
| 12499 | Ops[1] = CGF.Builder.CreateExtractElement(Ops[1], (uint64_t)0); | |||
| 12500 | Ops[2] = CGF.Builder.CreateExtractElement(Ops[2], (uint64_t)0); | |||
| 12501 | Value *Res; | |||
| 12502 | if (Rnd != 4) { | |||
| 12503 | Intrinsic::ID IID; | |||
| 12504 | ||||
| 12505 | switch (Ops[0]->getType()->getPrimitiveSizeInBits()) { | |||
| 12506 | case 16: | |||
| 12507 | IID = Intrinsic::x86_avx512fp16_vfmadd_f16; | |||
| 12508 | break; | |||
| 12509 | case 32: | |||
| 12510 | IID = Intrinsic::x86_avx512_vfmadd_f32; | |||
| 12511 | break; | |||
| 12512 | case 64: | |||
| 12513 | IID = Intrinsic::x86_avx512_vfmadd_f64; | |||
| 12514 | break; | |||
| 12515 | default: | |||
| 12516 | llvm_unreachable("Unexpected size")::llvm::llvm_unreachable_internal("Unexpected size", "clang/lib/CodeGen/CGBuiltin.cpp" , 12516); | |||
| 12517 | } | |||
| 12518 | Res = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID), | |||
| 12519 | {Ops[0], Ops[1], Ops[2], Ops[4]}); | |||
| 12520 | } else if (CGF.Builder.getIsFPConstrained()) { | |||
| 12521 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); | |||
| 12522 | Function *FMA = CGF.CGM.getIntrinsic( | |||
| 12523 | Intrinsic::experimental_constrained_fma, Ops[0]->getType()); | |||
| 12524 | Res = CGF.Builder.CreateConstrainedFPCall(FMA, Ops.slice(0, 3)); | |||
| 12525 | } else { | |||
| 12526 | Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ops[0]->getType()); | |||
| 12527 | Res = CGF.Builder.CreateCall(FMA, Ops.slice(0, 3)); | |||
| 12528 | } | |||
| 12529 | // If we have more than 3 arguments, we need to do masking. | |||
| 12530 | if (Ops.size() > 3) { | |||
| 12531 | Value *PassThru = ZeroMask ? Constant::getNullValue(Res->getType()) | |||
| 12532 | : Ops[PTIdx]; | |||
| 12533 | ||||
| 12534 | // If we negated the accumulator and the its the PassThru value we need to | |||
| 12535 | // bypass the negate. Conveniently Upper should be the same thing in this | |||
| 12536 | // case. | |||
| 12537 | if (NegAcc && PTIdx == 2) | |||
| 12538 | PassThru = CGF.Builder.CreateExtractElement(Upper, (uint64_t)0); | |||
| 12539 | ||||
| 12540 | Res = EmitX86ScalarSelect(CGF, Ops[3], Res, PassThru); | |||
| 12541 | } | |||
| 12542 | return CGF.Builder.CreateInsertElement(Upper, Res, (uint64_t)0); | |||
| 12543 | } | |||
| 12544 | ||||
| 12545 | static Value *EmitX86Muldq(CodeGenFunction &CGF, bool IsSigned, | |||
| 12546 | ArrayRef<Value *> Ops) { | |||
| 12547 | llvm::Type *Ty = Ops[0]->getType(); | |||
| 12548 | // Arguments have a vXi32 type so cast to vXi64. | |||
| 12549 | Ty = llvm::FixedVectorType::get(CGF.Int64Ty, | |||
| 12550 | Ty->getPrimitiveSizeInBits() / 64); | |||
| 12551 | Value *LHS = CGF.Builder.CreateBitCast(Ops[0], Ty); | |||
| 12552 | Value *RHS = CGF.Builder.CreateBitCast(Ops[1], Ty); | |||
| 12553 | ||||
| 12554 | if (IsSigned) { | |||
| 12555 | // Shift left then arithmetic shift right. | |||
| 12556 | Constant *ShiftAmt = ConstantInt::get(Ty, 32); | |||
| 12557 | LHS = CGF.Builder.CreateShl(LHS, ShiftAmt); | |||
| 12558 | LHS = CGF.Builder.CreateAShr(LHS, ShiftAmt); | |||
| 12559 | RHS = CGF.Builder.CreateShl(RHS, ShiftAmt); | |||
| 12560 | RHS = CGF.Builder.CreateAShr(RHS, ShiftAmt); | |||
| 12561 | } else { | |||
| 12562 | // Clear the upper bits. | |||
| 12563 | Constant *Mask = ConstantInt::get(Ty, 0xffffffff); | |||
| 12564 | LHS = CGF.Builder.CreateAnd(LHS, Mask); | |||
| 12565 | RHS = CGF.Builder.CreateAnd(RHS, Mask); | |||
| 12566 | } | |||
| 12567 | ||||
| 12568 | return CGF.Builder.CreateMul(LHS, RHS); | |||
| 12569 | } | |||
| 12570 | ||||
| 12571 | // Emit a masked pternlog intrinsic. This only exists because the header has to | |||
| 12572 | // use a macro and we aren't able to pass the input argument to a pternlog | |||
| 12573 | // builtin and a select builtin without evaluating it twice. | |||
| 12574 | static Value *EmitX86Ternlog(CodeGenFunction &CGF, bool ZeroMask, | |||
| 12575 | ArrayRef<Value *> Ops) { | |||
| 12576 | llvm::Type *Ty = Ops[0]->getType(); | |||
| 12577 | ||||
| 12578 | unsigned VecWidth = Ty->getPrimitiveSizeInBits(); | |||
| 12579 | unsigned EltWidth = Ty->getScalarSizeInBits(); | |||
| 12580 | Intrinsic::ID IID; | |||
| 12581 | if (VecWidth == 128 && EltWidth == 32) | |||
| 12582 | IID = Intrinsic::x86_avx512_pternlog_d_128; | |||
| 12583 | else if (VecWidth == 256 && EltWidth == 32) | |||
| 12584 | IID = Intrinsic::x86_avx512_pternlog_d_256; | |||
| 12585 | else if (VecWidth == 512 && EltWidth == 32) | |||
| 12586 | IID = Intrinsic::x86_avx512_pternlog_d_512; | |||
| 12587 | else if (VecWidth == 128 && EltWidth == 64) | |||
| 12588 | IID = Intrinsic::x86_avx512_pternlog_q_128; | |||
| 12589 | else if (VecWidth == 256 && EltWidth == 64) | |||
| 12590 | IID = Intrinsic::x86_avx512_pternlog_q_256; | |||
| 12591 | else if (VecWidth == 512 && EltWidth == 64) | |||
| 12592 | IID = Intrinsic::x86_avx512_pternlog_q_512; | |||
| 12593 | else | |||
| 12594 | llvm_unreachable("Unexpected intrinsic")::llvm::llvm_unreachable_internal("Unexpected intrinsic", "clang/lib/CodeGen/CGBuiltin.cpp" , 12594); | |||
| 12595 | ||||
| 12596 | Value *Ternlog = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID), | |||
| 12597 | Ops.drop_back()); | |||
| 12598 | Value *PassThru = ZeroMask ? ConstantAggregateZero::get(Ty) : Ops[0]; | |||
| 12599 | return EmitX86Select(CGF, Ops[4], Ternlog, PassThru); | |||
| 12600 | } | |||
| 12601 | ||||
| 12602 | static Value *EmitX86SExtMask(CodeGenFunction &CGF, Value *Op, | |||
| 12603 | llvm::Type *DstTy) { | |||
| 12604 | unsigned NumberOfElements = | |||
| 12605 | cast<llvm::FixedVectorType>(DstTy)->getNumElements(); | |||
| 12606 | Value *Mask = getMaskVecValue(CGF, Op, NumberOfElements); | |||
| 12607 | return CGF.Builder.CreateSExt(Mask, DstTy, "vpmovm2"); | |||
| 12608 | } | |||
| 12609 | ||||
| 12610 | Value *CodeGenFunction::EmitX86CpuIs(const CallExpr *E) { | |||
| 12611 | const Expr *CPUExpr = E->getArg(0)->IgnoreParenCasts(); | |||
| 12612 | StringRef CPUStr = cast<clang::StringLiteral>(CPUExpr)->getString(); | |||
| 12613 | return EmitX86CpuIs(CPUStr); | |||
| 12614 | } | |||
| 12615 | ||||
| 12616 | // Convert F16 halfs to floats. | |||
| 12617 | static Value *EmitX86CvtF16ToFloatExpr(CodeGenFunction &CGF, | |||
| 12618 | ArrayRef<Value *> Ops, | |||
| 12619 | llvm::Type *DstTy) { | |||
| 12620 | assert((Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) &&(static_cast <bool> ((Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) && "Unknown cvtph2ps intrinsic") ? void (0) : __assert_fail ("(Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) && \"Unknown cvtph2ps intrinsic\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12621, __extension__ __PRETTY_FUNCTION__ )) | |||
| 12621 | "Unknown cvtph2ps intrinsic")(static_cast <bool> ((Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) && "Unknown cvtph2ps intrinsic") ? void (0) : __assert_fail ("(Ops.size() == 1 || Ops.size() == 3 || Ops.size() == 4) && \"Unknown cvtph2ps intrinsic\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12621, __extension__ __PRETTY_FUNCTION__ )); | |||
| 12622 | ||||
| 12623 | // If the SAE intrinsic doesn't use default rounding then we can't upgrade. | |||
| 12624 | if (Ops.size() == 4 && cast<llvm::ConstantInt>(Ops[3])->getZExtValue() != 4) { | |||
| 12625 | Function *F = | |||
| 12626 | CGF.CGM.getIntrinsic(Intrinsic::x86_avx512_mask_vcvtph2ps_512); | |||
| 12627 | return CGF.Builder.CreateCall(F, {Ops[0], Ops[1], Ops[2], Ops[3]}); | |||
| 12628 | } | |||
| 12629 | ||||
| 12630 | unsigned NumDstElts = cast<llvm::FixedVectorType>(DstTy)->getNumElements(); | |||
| 12631 | Value *Src = Ops[0]; | |||
| 12632 | ||||
| 12633 | // Extract the subvector. | |||
| 12634 | if (NumDstElts != | |||
| 12635 | cast<llvm::FixedVectorType>(Src->getType())->getNumElements()) { | |||
| 12636 | assert(NumDstElts == 4 && "Unexpected vector size")(static_cast <bool> (NumDstElts == 4 && "Unexpected vector size" ) ? void (0) : __assert_fail ("NumDstElts == 4 && \"Unexpected vector size\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12636, __extension__ __PRETTY_FUNCTION__ )); | |||
| 12637 | Src = CGF.Builder.CreateShuffleVector(Src, ArrayRef<int>{0, 1, 2, 3}); | |||
| 12638 | } | |||
| 12639 | ||||
| 12640 | // Bitcast from vXi16 to vXf16. | |||
| 12641 | auto *HalfTy = llvm::FixedVectorType::get( | |||
| 12642 | llvm::Type::getHalfTy(CGF.getLLVMContext()), NumDstElts); | |||
| 12643 | Src = CGF.Builder.CreateBitCast(Src, HalfTy); | |||
| 12644 | ||||
| 12645 | // Perform the fp-extension. | |||
| 12646 | Value *Res = CGF.Builder.CreateFPExt(Src, DstTy, "cvtph2ps"); | |||
| 12647 | ||||
| 12648 | if (Ops.size() >= 3) | |||
| 12649 | Res = EmitX86Select(CGF, Ops[2], Res, Ops[1]); | |||
| 12650 | return Res; | |||
| 12651 | } | |||
| 12652 | ||||
| 12653 | // Convert a BF16 to a float. | |||
| 12654 | static Value *EmitX86CvtBF16ToFloatExpr(CodeGenFunction &CGF, | |||
| 12655 | const CallExpr *E, | |||
| 12656 | ArrayRef<Value *> Ops) { | |||
| 12657 | llvm::Type *Int32Ty = CGF.Builder.getInt32Ty(); | |||
| 12658 | Value *ZeroExt = CGF.Builder.CreateZExt(Ops[0], Int32Ty); | |||
| 12659 | Value *Shl = CGF.Builder.CreateShl(ZeroExt, 16); | |||
| 12660 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); | |||
| 12661 | Value *BitCast = CGF.Builder.CreateBitCast(Shl, ResultType); | |||
| 12662 | return BitCast; | |||
| 12663 | } | |||
| 12664 | ||||
| 12665 | Value *CodeGenFunction::EmitX86CpuIs(StringRef CPUStr) { | |||
| 12666 | ||||
| 12667 | llvm::Type *Int32Ty = Builder.getInt32Ty(); | |||
| 12668 | ||||
| 12669 | // Matching the struct layout from the compiler-rt/libgcc structure that is | |||
| 12670 | // filled in: | |||
| 12671 | // unsigned int __cpu_vendor; | |||
| 12672 | // unsigned int __cpu_type; | |||
| 12673 | // unsigned int __cpu_subtype; | |||
| 12674 | // unsigned int __cpu_features[1]; | |||
| 12675 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty, | |||
| 12676 | llvm::ArrayType::get(Int32Ty, 1)); | |||
| 12677 | ||||
| 12678 | // Grab the global __cpu_model. | |||
| 12679 | llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model"); | |||
| 12680 | cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true); | |||
| 12681 | ||||
| 12682 | // Calculate the index needed to access the correct field based on the | |||
| 12683 | // range. Also adjust the expected value. | |||
| 12684 | unsigned Index; | |||
| 12685 | unsigned Value; | |||
| 12686 | std::tie(Index, Value) = StringSwitch<std::pair<unsigned, unsigned>>(CPUStr) | |||
| 12687 | #define X86_VENDOR(ENUM, STRING) \ | |||
| 12688 | .Case(STRING, {0u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
| 12689 | #define X86_CPU_TYPE_ALIAS(ENUM, ALIAS) \ | |||
| 12690 | .Case(ALIAS, {1u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
| 12691 | #define X86_CPU_TYPE(ENUM, STR) \ | |||
| 12692 | .Case(STR, {1u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
| 12693 | #define X86_CPU_SUBTYPE(ENUM, STR) \ | |||
| 12694 | .Case(STR, {2u, static_cast<unsigned>(llvm::X86::ENUM)}) | |||
| 12695 | #include "llvm/Support/X86TargetParser.def" | |||
| 12696 | .Default({0, 0}); | |||
| 12697 | assert(Value != 0 && "Invalid CPUStr passed to CpuIs")(static_cast <bool> (Value != 0 && "Invalid CPUStr passed to CpuIs" ) ? void (0) : __assert_fail ("Value != 0 && \"Invalid CPUStr passed to CpuIs\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12697, __extension__ __PRETTY_FUNCTION__ )); | |||
| 12698 | ||||
| 12699 | // Grab the appropriate field from __cpu_model. | |||
| 12700 | llvm::Value *Idxs[] = {ConstantInt::get(Int32Ty, 0), | |||
| 12701 | ConstantInt::get(Int32Ty, Index)}; | |||
| 12702 | llvm::Value *CpuValue = Builder.CreateGEP(STy, CpuModel, Idxs); | |||
| 12703 | CpuValue = Builder.CreateAlignedLoad(Int32Ty, CpuValue, | |||
| 12704 | CharUnits::fromQuantity(4)); | |||
| 12705 | ||||
| 12706 | // Check the value of the field against the requested value. | |||
| 12707 | return Builder.CreateICmpEQ(CpuValue, | |||
| 12708 | llvm::ConstantInt::get(Int32Ty, Value)); | |||
| 12709 | } | |||
| 12710 | ||||
| 12711 | Value *CodeGenFunction::EmitX86CpuSupports(const CallExpr *E) { | |||
| 12712 | const Expr *FeatureExpr = E->getArg(0)->IgnoreParenCasts(); | |||
| 12713 | StringRef FeatureStr = cast<StringLiteral>(FeatureExpr)->getString(); | |||
| 12714 | return EmitX86CpuSupports(FeatureStr); | |||
| 12715 | } | |||
| 12716 | ||||
| 12717 | Value *CodeGenFunction::EmitX86CpuSupports(ArrayRef<StringRef> FeatureStrs) { | |||
| 12718 | return EmitX86CpuSupports(llvm::X86::getCpuSupportsMask(FeatureStrs)); | |||
| 12719 | } | |||
| 12720 | ||||
| 12721 | llvm::Value *CodeGenFunction::EmitX86CpuSupports(uint64_t FeaturesMask) { | |||
| 12722 | uint32_t Features1 = Lo_32(FeaturesMask); | |||
| 12723 | uint32_t Features2 = Hi_32(FeaturesMask); | |||
| 12724 | ||||
| 12725 | Value *Result = Builder.getTrue(); | |||
| 12726 | ||||
| 12727 | if (Features1 != 0) { | |||
| 12728 | // Matching the struct layout from the compiler-rt/libgcc structure that is | |||
| 12729 | // filled in: | |||
| 12730 | // unsigned int __cpu_vendor; | |||
| 12731 | // unsigned int __cpu_type; | |||
| 12732 | // unsigned int __cpu_subtype; | |||
| 12733 | // unsigned int __cpu_features[1]; | |||
| 12734 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty, | |||
| 12735 | llvm::ArrayType::get(Int32Ty, 1)); | |||
| 12736 | ||||
| 12737 | // Grab the global __cpu_model. | |||
| 12738 | llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model"); | |||
| 12739 | cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true); | |||
| 12740 | ||||
| 12741 | // Grab the first (0th) element from the field __cpu_features off of the | |||
| 12742 | // global in the struct STy. | |||
| 12743 | Value *Idxs[] = {Builder.getInt32(0), Builder.getInt32(3), | |||
| 12744 | Builder.getInt32(0)}; | |||
| 12745 | Value *CpuFeatures = Builder.CreateGEP(STy, CpuModel, Idxs); | |||
| 12746 | Value *Features = Builder.CreateAlignedLoad(Int32Ty, CpuFeatures, | |||
| 12747 | CharUnits::fromQuantity(4)); | |||
| 12748 | ||||
| 12749 | // Check the value of the bit corresponding to the feature requested. | |||
| 12750 | Value *Mask = Builder.getInt32(Features1); | |||
| 12751 | Value *Bitset = Builder.CreateAnd(Features, Mask); | |||
| 12752 | Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask); | |||
| 12753 | Result = Builder.CreateAnd(Result, Cmp); | |||
| 12754 | } | |||
| 12755 | ||||
| 12756 | if (Features2 != 0) { | |||
| 12757 | llvm::Constant *CpuFeatures2 = CGM.CreateRuntimeVariable(Int32Ty, | |||
| 12758 | "__cpu_features2"); | |||
| 12759 | cast<llvm::GlobalValue>(CpuFeatures2)->setDSOLocal(true); | |||
| 12760 | ||||
| 12761 | Value *Features = Builder.CreateAlignedLoad(Int32Ty, CpuFeatures2, | |||
| 12762 | CharUnits::fromQuantity(4)); | |||
| 12763 | ||||
| 12764 | // Check the value of the bit corresponding to the feature requested. | |||
| 12765 | Value *Mask = Builder.getInt32(Features2); | |||
| 12766 | Value *Bitset = Builder.CreateAnd(Features, Mask); | |||
| 12767 | Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask); | |||
| 12768 | Result = Builder.CreateAnd(Result, Cmp); | |||
| 12769 | } | |||
| 12770 | ||||
| 12771 | return Result; | |||
| 12772 | } | |||
| 12773 | ||||
| 12774 | Value *CodeGenFunction::EmitX86CpuInit() { | |||
| 12775 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, | |||
| 12776 | /*Variadic*/ false); | |||
| 12777 | llvm::FunctionCallee Func = | |||
| 12778 | CGM.CreateRuntimeFunction(FTy, "__cpu_indicator_init"); | |||
| 12779 | cast<llvm::GlobalValue>(Func.getCallee())->setDSOLocal(true); | |||
| 12780 | cast<llvm::GlobalValue>(Func.getCallee()) | |||
| 12781 | ->setDLLStorageClass(llvm::GlobalValue::DefaultStorageClass); | |||
| 12782 | return Builder.CreateCall(Func); | |||
| 12783 | } | |||
| 12784 | ||||
| 12785 | Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID, | |||
| 12786 | const CallExpr *E) { | |||
| 12787 | if (BuiltinID == X86::BI__builtin_cpu_is) | |||
| 12788 | return EmitX86CpuIs(E); | |||
| 12789 | if (BuiltinID == X86::BI__builtin_cpu_supports) | |||
| 12790 | return EmitX86CpuSupports(E); | |||
| 12791 | if (BuiltinID == X86::BI__builtin_cpu_init) | |||
| 12792 | return EmitX86CpuInit(); | |||
| 12793 | ||||
| 12794 | // Handle MSVC intrinsics before argument evaluation to prevent double | |||
| 12795 | // evaluation. | |||
| 12796 | if (Optional<MSVCIntrin> MsvcIntId = translateX86ToMsvcIntrin(BuiltinID)) | |||
| 12797 | return EmitMSVCBuiltinExpr(*MsvcIntId, E); | |||
| 12798 | ||||
| 12799 | SmallVector<Value*, 4> Ops; | |||
| 12800 | bool IsMaskFCmp = false; | |||
| 12801 | bool IsConjFMA = false; | |||
| 12802 | ||||
| 12803 | // Find out if any arguments are required to be integer constant expressions. | |||
| 12804 | unsigned ICEArguments = 0; | |||
| 12805 | ASTContext::GetBuiltinTypeError Error; | |||
| 12806 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
| 12807 | assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None && "Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 12807, __extension__ __PRETTY_FUNCTION__ )); | |||
| 12808 | ||||
| 12809 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | |||
| 12810 | // If this is a normal argument, just emit it as a scalar. | |||
| 12811 | if ((ICEArguments & (1 << i)) == 0) { | |||
| 12812 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 12813 | continue; | |||
| 12814 | } | |||
| 12815 | ||||
| 12816 | // If this is required to be a constant, constant fold it so that we know | |||
| 12817 | // that the generated intrinsic gets a ConstantInt. | |||
| 12818 | Ops.push_back(llvm::ConstantInt::get( | |||
| 12819 | getLLVMContext(), *E->getArg(i)->getIntegerConstantExpr(getContext()))); | |||
| 12820 | } | |||
| 12821 | ||||
| 12822 | // These exist so that the builtin that takes an immediate can be bounds | |||
| 12823 | // checked by clang to avoid passing bad immediates to the backend. Since | |||
| 12824 | // AVX has a larger immediate than SSE we would need separate builtins to | |||
| 12825 | // do the different bounds checking. Rather than create a clang specific | |||
| 12826 | // SSE only builtin, this implements eight separate builtins to match gcc | |||
| 12827 | // implementation. | |||
| 12828 | auto getCmpIntrinsicCall = [this, &Ops](Intrinsic::ID ID, unsigned Imm) { | |||
| 12829 | Ops.push_back(llvm::ConstantInt::get(Int8Ty, Imm)); | |||
| 12830 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
| 12831 | return Builder.CreateCall(F, Ops); | |||
| 12832 | }; | |||
| 12833 | ||||
| 12834 | // For the vector forms of FP comparisons, translate the builtins directly to | |||
| 12835 | // IR. | |||
| 12836 | // TODO: The builtins could be removed if the SSE header files used vector | |||
| 12837 | // extension comparisons directly (vector ordered/unordered may need | |||
| 12838 | // additional support via __builtin_isnan()). | |||
| 12839 | auto getVectorFCmpIR = [this, &Ops, E](CmpInst::Predicate Pred, | |||
| 12840 | bool IsSignaling) { | |||
| 12841 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 12842 | Value *Cmp; | |||
| 12843 | if (IsSignaling) | |||
| 12844 | Cmp = Builder.CreateFCmpS(Pred, Ops[0], Ops[1]); | |||
| 12845 | else | |||
| 12846 | Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]); | |||
| 12847 | llvm::VectorType *FPVecTy = cast<llvm::VectorType>(Ops[0]->getType()); | |||
| 12848 | llvm::VectorType *IntVecTy = llvm::VectorType::getInteger(FPVecTy); | |||
| 12849 | Value *Sext = Builder.CreateSExt(Cmp, IntVecTy); | |||
| 12850 | return Builder.CreateBitCast(Sext, FPVecTy); | |||
| 12851 | }; | |||
| 12852 | ||||
| 12853 | switch (BuiltinID) { | |||
| 12854 | default: return nullptr; | |||
| 12855 | case X86::BI_mm_prefetch: { | |||
| 12856 | Value *Address = Ops[0]; | |||
| 12857 | ConstantInt *C = cast<ConstantInt>(Ops[1]); | |||
| 12858 | Value *RW = ConstantInt::get(Int32Ty, (C->getZExtValue() >> 2) & 0x1); | |||
| 12859 | Value *Locality = ConstantInt::get(Int32Ty, C->getZExtValue() & 0x3); | |||
| 12860 | Value *Data = ConstantInt::get(Int32Ty, 1); | |||
| 12861 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); | |||
| 12862 | return Builder.CreateCall(F, {Address, RW, Locality, Data}); | |||
| 12863 | } | |||
| 12864 | case X86::BI_mm_clflush: { | |||
| 12865 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_clflush), | |||
| 12866 | Ops[0]); | |||
| 12867 | } | |||
| 12868 | case X86::BI_mm_lfence: { | |||
| 12869 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_lfence)); | |||
| 12870 | } | |||
| 12871 | case X86::BI_mm_mfence: { | |||
| 12872 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_mfence)); | |||
| 12873 | } | |||
| 12874 | case X86::BI_mm_sfence: { | |||
| 12875 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_sfence)); | |||
| 12876 | } | |||
| 12877 | case X86::BI_mm_pause: { | |||
| 12878 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_pause)); | |||
| 12879 | } | |||
| 12880 | case X86::BI__rdtsc: { | |||
| 12881 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtsc)); | |||
| 12882 | } | |||
| 12883 | case X86::BI__builtin_ia32_rdtscp: { | |||
| 12884 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtscp)); | |||
| 12885 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1), | |||
| 12886 | Ops[0]); | |||
| 12887 | return Builder.CreateExtractValue(Call, 0); | |||
| 12888 | } | |||
| 12889 | case X86::BI__builtin_ia32_lzcnt_u16: | |||
| 12890 | case X86::BI__builtin_ia32_lzcnt_u32: | |||
| 12891 | case X86::BI__builtin_ia32_lzcnt_u64: { | |||
| 12892 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType()); | |||
| 12893 | return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)}); | |||
| 12894 | } | |||
| 12895 | case X86::BI__builtin_ia32_tzcnt_u16: | |||
| 12896 | case X86::BI__builtin_ia32_tzcnt_u32: | |||
| 12897 | case X86::BI__builtin_ia32_tzcnt_u64: { | |||
| 12898 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, Ops[0]->getType()); | |||
| 12899 | return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)}); | |||
| 12900 | } | |||
| 12901 | case X86::BI__builtin_ia32_undef128: | |||
| 12902 | case X86::BI__builtin_ia32_undef256: | |||
| 12903 | case X86::BI__builtin_ia32_undef512: | |||
| 12904 | // The x86 definition of "undef" is not the same as the LLVM definition | |||
| 12905 | // (PR32176). We leave optimizing away an unnecessary zero constant to the | |||
| 12906 | // IR optimizer and backend. | |||
| 12907 | // TODO: If we had a "freeze" IR instruction to generate a fixed undef | |||
| 12908 | // value, we should use that here instead of a zero. | |||
| 12909 | return llvm::Constant::getNullValue(ConvertType(E->getType())); | |||
| 12910 | case X86::BI__builtin_ia32_vec_init_v8qi: | |||
| 12911 | case X86::BI__builtin_ia32_vec_init_v4hi: | |||
| 12912 | case X86::BI__builtin_ia32_vec_init_v2si: | |||
| 12913 | return Builder.CreateBitCast(BuildVector(Ops), | |||
| 12914 | llvm::Type::getX86_MMXTy(getLLVMContext())); | |||
| 12915 | case X86::BI__builtin_ia32_vec_ext_v2si: | |||
| 12916 | case X86::BI__builtin_ia32_vec_ext_v16qi: | |||
| 12917 | case X86::BI__builtin_ia32_vec_ext_v8hi: | |||
| 12918 | case X86::BI__builtin_ia32_vec_ext_v4si: | |||
| 12919 | case X86::BI__builtin_ia32_vec_ext_v4sf: | |||
| 12920 | case X86::BI__builtin_ia32_vec_ext_v2di: | |||
| 12921 | case X86::BI__builtin_ia32_vec_ext_v32qi: | |||
| 12922 | case X86::BI__builtin_ia32_vec_ext_v16hi: | |||
| 12923 | case X86::BI__builtin_ia32_vec_ext_v8si: | |||
| 12924 | case X86::BI__builtin_ia32_vec_ext_v4di: { | |||
| 12925 | unsigned NumElts = | |||
| 12926 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 12927 | uint64_t Index = cast<ConstantInt>(Ops[1])->getZExtValue(); | |||
| 12928 | Index &= NumElts - 1; | |||
| 12929 | // These builtins exist so we can ensure the index is an ICE and in range. | |||
| 12930 | // Otherwise we could just do this in the header file. | |||
| 12931 | return Builder.CreateExtractElement(Ops[0], Index); | |||
| 12932 | } | |||
| 12933 | case X86::BI__builtin_ia32_vec_set_v16qi: | |||
| 12934 | case X86::BI__builtin_ia32_vec_set_v8hi: | |||
| 12935 | case X86::BI__builtin_ia32_vec_set_v4si: | |||
| 12936 | case X86::BI__builtin_ia32_vec_set_v2di: | |||
| 12937 | case X86::BI__builtin_ia32_vec_set_v32qi: | |||
| 12938 | case X86::BI__builtin_ia32_vec_set_v16hi: | |||
| 12939 | case X86::BI__builtin_ia32_vec_set_v8si: | |||
| 12940 | case X86::BI__builtin_ia32_vec_set_v4di: { | |||
| 12941 | unsigned NumElts = | |||
| 12942 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 12943 | unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue(); | |||
| 12944 | Index &= NumElts - 1; | |||
| 12945 | // These builtins exist so we can ensure the index is an ICE and in range. | |||
| 12946 | // Otherwise we could just do this in the header file. | |||
| 12947 | return Builder.CreateInsertElement(Ops[0], Ops[1], Index); | |||
| 12948 | } | |||
| 12949 | case X86::BI_mm_setcsr: | |||
| 12950 | case X86::BI__builtin_ia32_ldmxcsr: { | |||
| 12951 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); | |||
| 12952 | Builder.CreateStore(Ops[0], Tmp); | |||
| 12953 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_ldmxcsr), | |||
| 12954 | Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy)); | |||
| 12955 | } | |||
| 12956 | case X86::BI_mm_getcsr: | |||
| 12957 | case X86::BI__builtin_ia32_stmxcsr: { | |||
| 12958 | Address Tmp = CreateMemTemp(E->getType()); | |||
| 12959 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_stmxcsr), | |||
| 12960 | Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy)); | |||
| 12961 | return Builder.CreateLoad(Tmp, "stmxcsr"); | |||
| 12962 | } | |||
| 12963 | case X86::BI__builtin_ia32_xsave: | |||
| 12964 | case X86::BI__builtin_ia32_xsave64: | |||
| 12965 | case X86::BI__builtin_ia32_xrstor: | |||
| 12966 | case X86::BI__builtin_ia32_xrstor64: | |||
| 12967 | case X86::BI__builtin_ia32_xsaveopt: | |||
| 12968 | case X86::BI__builtin_ia32_xsaveopt64: | |||
| 12969 | case X86::BI__builtin_ia32_xrstors: | |||
| 12970 | case X86::BI__builtin_ia32_xrstors64: | |||
| 12971 | case X86::BI__builtin_ia32_xsavec: | |||
| 12972 | case X86::BI__builtin_ia32_xsavec64: | |||
| 12973 | case X86::BI__builtin_ia32_xsaves: | |||
| 12974 | case X86::BI__builtin_ia32_xsaves64: | |||
| 12975 | case X86::BI__builtin_ia32_xsetbv: | |||
| 12976 | case X86::BI_xsetbv: { | |||
| 12977 | Intrinsic::ID ID; | |||
| 12978 | #define INTRINSIC_X86_XSAVE_ID(NAME) \ | |||
| 12979 | case X86::BI__builtin_ia32_##NAME: \ | |||
| 12980 | ID = Intrinsic::x86_##NAME; \ | |||
| 12981 | break | |||
| 12982 | switch (BuiltinID) { | |||
| 12983 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 12983); | |||
| 12984 | INTRINSIC_X86_XSAVE_ID(xsave); | |||
| 12985 | INTRINSIC_X86_XSAVE_ID(xsave64); | |||
| 12986 | INTRINSIC_X86_XSAVE_ID(xrstor); | |||
| 12987 | INTRINSIC_X86_XSAVE_ID(xrstor64); | |||
| 12988 | INTRINSIC_X86_XSAVE_ID(xsaveopt); | |||
| 12989 | INTRINSIC_X86_XSAVE_ID(xsaveopt64); | |||
| 12990 | INTRINSIC_X86_XSAVE_ID(xrstors); | |||
| 12991 | INTRINSIC_X86_XSAVE_ID(xrstors64); | |||
| 12992 | INTRINSIC_X86_XSAVE_ID(xsavec); | |||
| 12993 | INTRINSIC_X86_XSAVE_ID(xsavec64); | |||
| 12994 | INTRINSIC_X86_XSAVE_ID(xsaves); | |||
| 12995 | INTRINSIC_X86_XSAVE_ID(xsaves64); | |||
| 12996 | INTRINSIC_X86_XSAVE_ID(xsetbv); | |||
| 12997 | case X86::BI_xsetbv: | |||
| 12998 | ID = Intrinsic::x86_xsetbv; | |||
| 12999 | break; | |||
| 13000 | } | |||
| 13001 | #undef INTRINSIC_X86_XSAVE_ID | |||
| 13002 | Value *Mhi = Builder.CreateTrunc( | |||
| 13003 | Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, 32)), Int32Ty); | |||
| 13004 | Value *Mlo = Builder.CreateTrunc(Ops[1], Int32Ty); | |||
| 13005 | Ops[1] = Mhi; | |||
| 13006 | Ops.push_back(Mlo); | |||
| 13007 | return Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
| 13008 | } | |||
| 13009 | case X86::BI__builtin_ia32_xgetbv: | |||
| 13010 | case X86::BI_xgetbv: | |||
| 13011 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_xgetbv), Ops); | |||
| 13012 | case X86::BI__builtin_ia32_storedqudi128_mask: | |||
| 13013 | case X86::BI__builtin_ia32_storedqusi128_mask: | |||
| 13014 | case X86::BI__builtin_ia32_storedquhi128_mask: | |||
| 13015 | case X86::BI__builtin_ia32_storedquqi128_mask: | |||
| 13016 | case X86::BI__builtin_ia32_storeupd128_mask: | |||
| 13017 | case X86::BI__builtin_ia32_storeups128_mask: | |||
| 13018 | case X86::BI__builtin_ia32_storedqudi256_mask: | |||
| 13019 | case X86::BI__builtin_ia32_storedqusi256_mask: | |||
| 13020 | case X86::BI__builtin_ia32_storedquhi256_mask: | |||
| 13021 | case X86::BI__builtin_ia32_storedquqi256_mask: | |||
| 13022 | case X86::BI__builtin_ia32_storeupd256_mask: | |||
| 13023 | case X86::BI__builtin_ia32_storeups256_mask: | |||
| 13024 | case X86::BI__builtin_ia32_storedqudi512_mask: | |||
| 13025 | case X86::BI__builtin_ia32_storedqusi512_mask: | |||
| 13026 | case X86::BI__builtin_ia32_storedquhi512_mask: | |||
| 13027 | case X86::BI__builtin_ia32_storedquqi512_mask: | |||
| 13028 | case X86::BI__builtin_ia32_storeupd512_mask: | |||
| 13029 | case X86::BI__builtin_ia32_storeups512_mask: | |||
| 13030 | return EmitX86MaskedStore(*this, Ops, Align(1)); | |||
| 13031 | ||||
| 13032 | case X86::BI__builtin_ia32_storesh128_mask: | |||
| 13033 | case X86::BI__builtin_ia32_storess128_mask: | |||
| 13034 | case X86::BI__builtin_ia32_storesd128_mask: | |||
| 13035 | return EmitX86MaskedStore(*this, Ops, Align(1)); | |||
| 13036 | ||||
| 13037 | case X86::BI__builtin_ia32_vpopcntb_128: | |||
| 13038 | case X86::BI__builtin_ia32_vpopcntd_128: | |||
| 13039 | case X86::BI__builtin_ia32_vpopcntq_128: | |||
| 13040 | case X86::BI__builtin_ia32_vpopcntw_128: | |||
| 13041 | case X86::BI__builtin_ia32_vpopcntb_256: | |||
| 13042 | case X86::BI__builtin_ia32_vpopcntd_256: | |||
| 13043 | case X86::BI__builtin_ia32_vpopcntq_256: | |||
| 13044 | case X86::BI__builtin_ia32_vpopcntw_256: | |||
| 13045 | case X86::BI__builtin_ia32_vpopcntb_512: | |||
| 13046 | case X86::BI__builtin_ia32_vpopcntd_512: | |||
| 13047 | case X86::BI__builtin_ia32_vpopcntq_512: | |||
| 13048 | case X86::BI__builtin_ia32_vpopcntw_512: { | |||
| 13049 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 13050 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); | |||
| 13051 | return Builder.CreateCall(F, Ops); | |||
| 13052 | } | |||
| 13053 | case X86::BI__builtin_ia32_cvtmask2b128: | |||
| 13054 | case X86::BI__builtin_ia32_cvtmask2b256: | |||
| 13055 | case X86::BI__builtin_ia32_cvtmask2b512: | |||
| 13056 | case X86::BI__builtin_ia32_cvtmask2w128: | |||
| 13057 | case X86::BI__builtin_ia32_cvtmask2w256: | |||
| 13058 | case X86::BI__builtin_ia32_cvtmask2w512: | |||
| 13059 | case X86::BI__builtin_ia32_cvtmask2d128: | |||
| 13060 | case X86::BI__builtin_ia32_cvtmask2d256: | |||
| 13061 | case X86::BI__builtin_ia32_cvtmask2d512: | |||
| 13062 | case X86::BI__builtin_ia32_cvtmask2q128: | |||
| 13063 | case X86::BI__builtin_ia32_cvtmask2q256: | |||
| 13064 | case X86::BI__builtin_ia32_cvtmask2q512: | |||
| 13065 | return EmitX86SExtMask(*this, Ops[0], ConvertType(E->getType())); | |||
| 13066 | ||||
| 13067 | case X86::BI__builtin_ia32_cvtb2mask128: | |||
| 13068 | case X86::BI__builtin_ia32_cvtb2mask256: | |||
| 13069 | case X86::BI__builtin_ia32_cvtb2mask512: | |||
| 13070 | case X86::BI__builtin_ia32_cvtw2mask128: | |||
| 13071 | case X86::BI__builtin_ia32_cvtw2mask256: | |||
| 13072 | case X86::BI__builtin_ia32_cvtw2mask512: | |||
| 13073 | case X86::BI__builtin_ia32_cvtd2mask128: | |||
| 13074 | case X86::BI__builtin_ia32_cvtd2mask256: | |||
| 13075 | case X86::BI__builtin_ia32_cvtd2mask512: | |||
| 13076 | case X86::BI__builtin_ia32_cvtq2mask128: | |||
| 13077 | case X86::BI__builtin_ia32_cvtq2mask256: | |||
| 13078 | case X86::BI__builtin_ia32_cvtq2mask512: | |||
| 13079 | return EmitX86ConvertToMask(*this, Ops[0]); | |||
| 13080 | ||||
| 13081 | case X86::BI__builtin_ia32_cvtdq2ps512_mask: | |||
| 13082 | case X86::BI__builtin_ia32_cvtqq2ps512_mask: | |||
| 13083 | case X86::BI__builtin_ia32_cvtqq2pd512_mask: | |||
| 13084 | case X86::BI__builtin_ia32_vcvtw2ph512_mask: | |||
| 13085 | case X86::BI__builtin_ia32_vcvtdq2ph512_mask: | |||
| 13086 | case X86::BI__builtin_ia32_vcvtqq2ph512_mask: | |||
| 13087 | return EmitX86ConvertIntToFp(*this, E, Ops, /*IsSigned*/ true); | |||
| 13088 | case X86::BI__builtin_ia32_cvtudq2ps512_mask: | |||
| 13089 | case X86::BI__builtin_ia32_cvtuqq2ps512_mask: | |||
| 13090 | case X86::BI__builtin_ia32_cvtuqq2pd512_mask: | |||
| 13091 | case X86::BI__builtin_ia32_vcvtuw2ph512_mask: | |||
| 13092 | case X86::BI__builtin_ia32_vcvtudq2ph512_mask: | |||
| 13093 | case X86::BI__builtin_ia32_vcvtuqq2ph512_mask: | |||
| 13094 | return EmitX86ConvertIntToFp(*this, E, Ops, /*IsSigned*/ false); | |||
| 13095 | ||||
| 13096 | case X86::BI__builtin_ia32_vfmaddss3: | |||
| 13097 | case X86::BI__builtin_ia32_vfmaddsd3: | |||
| 13098 | case X86::BI__builtin_ia32_vfmaddsh3_mask: | |||
| 13099 | case X86::BI__builtin_ia32_vfmaddss3_mask: | |||
| 13100 | case X86::BI__builtin_ia32_vfmaddsd3_mask: | |||
| 13101 | return EmitScalarFMAExpr(*this, E, Ops, Ops[0]); | |||
| 13102 | case X86::BI__builtin_ia32_vfmaddss: | |||
| 13103 | case X86::BI__builtin_ia32_vfmaddsd: | |||
| 13104 | return EmitScalarFMAExpr(*this, E, Ops, | |||
| 13105 | Constant::getNullValue(Ops[0]->getType())); | |||
| 13106 | case X86::BI__builtin_ia32_vfmaddsh3_maskz: | |||
| 13107 | case X86::BI__builtin_ia32_vfmaddss3_maskz: | |||
| 13108 | case X86::BI__builtin_ia32_vfmaddsd3_maskz: | |||
| 13109 | return EmitScalarFMAExpr(*this, E, Ops, Ops[0], /*ZeroMask*/ true); | |||
| 13110 | case X86::BI__builtin_ia32_vfmaddsh3_mask3: | |||
| 13111 | case X86::BI__builtin_ia32_vfmaddss3_mask3: | |||
| 13112 | case X86::BI__builtin_ia32_vfmaddsd3_mask3: | |||
| 13113 | return EmitScalarFMAExpr(*this, E, Ops, Ops[2], /*ZeroMask*/ false, 2); | |||
| 13114 | case X86::BI__builtin_ia32_vfmsubsh3_mask3: | |||
| 13115 | case X86::BI__builtin_ia32_vfmsubss3_mask3: | |||
| 13116 | case X86::BI__builtin_ia32_vfmsubsd3_mask3: | |||
| 13117 | return EmitScalarFMAExpr(*this, E, Ops, Ops[2], /*ZeroMask*/ false, 2, | |||
| 13118 | /*NegAcc*/ true); | |||
| 13119 | case X86::BI__builtin_ia32_vfmaddph: | |||
| 13120 | case X86::BI__builtin_ia32_vfmaddps: | |||
| 13121 | case X86::BI__builtin_ia32_vfmaddpd: | |||
| 13122 | case X86::BI__builtin_ia32_vfmaddph256: | |||
| 13123 | case X86::BI__builtin_ia32_vfmaddps256: | |||
| 13124 | case X86::BI__builtin_ia32_vfmaddpd256: | |||
| 13125 | case X86::BI__builtin_ia32_vfmaddph512_mask: | |||
| 13126 | case X86::BI__builtin_ia32_vfmaddph512_maskz: | |||
| 13127 | case X86::BI__builtin_ia32_vfmaddph512_mask3: | |||
| 13128 | case X86::BI__builtin_ia32_vfmaddps512_mask: | |||
| 13129 | case X86::BI__builtin_ia32_vfmaddps512_maskz: | |||
| 13130 | case X86::BI__builtin_ia32_vfmaddps512_mask3: | |||
| 13131 | case X86::BI__builtin_ia32_vfmsubps512_mask3: | |||
| 13132 | case X86::BI__builtin_ia32_vfmaddpd512_mask: | |||
| 13133 | case X86::BI__builtin_ia32_vfmaddpd512_maskz: | |||
| 13134 | case X86::BI__builtin_ia32_vfmaddpd512_mask3: | |||
| 13135 | case X86::BI__builtin_ia32_vfmsubpd512_mask3: | |||
| 13136 | case X86::BI__builtin_ia32_vfmsubph512_mask3: | |||
| 13137 | return EmitX86FMAExpr(*this, E, Ops, BuiltinID, /*IsAddSub*/ false); | |||
| 13138 | case X86::BI__builtin_ia32_vfmaddsubph512_mask: | |||
| 13139 | case X86::BI__builtin_ia32_vfmaddsubph512_maskz: | |||
| 13140 | case X86::BI__builtin_ia32_vfmaddsubph512_mask3: | |||
| 13141 | case X86::BI__builtin_ia32_vfmsubaddph512_mask3: | |||
| 13142 | case X86::BI__builtin_ia32_vfmaddsubps512_mask: | |||
| 13143 | case X86::BI__builtin_ia32_vfmaddsubps512_maskz: | |||
| 13144 | case X86::BI__builtin_ia32_vfmaddsubps512_mask3: | |||
| 13145 | case X86::BI__builtin_ia32_vfmsubaddps512_mask3: | |||
| 13146 | case X86::BI__builtin_ia32_vfmaddsubpd512_mask: | |||
| 13147 | case X86::BI__builtin_ia32_vfmaddsubpd512_maskz: | |||
| 13148 | case X86::BI__builtin_ia32_vfmaddsubpd512_mask3: | |||
| 13149 | case X86::BI__builtin_ia32_vfmsubaddpd512_mask3: | |||
| 13150 | return EmitX86FMAExpr(*this, E, Ops, BuiltinID, /*IsAddSub*/ true); | |||
| 13151 | ||||
| 13152 | case X86::BI__builtin_ia32_movdqa32store128_mask: | |||
| 13153 | case X86::BI__builtin_ia32_movdqa64store128_mask: | |||
| 13154 | case X86::BI__builtin_ia32_storeaps128_mask: | |||
| 13155 | case X86::BI__builtin_ia32_storeapd128_mask: | |||
| 13156 | case X86::BI__builtin_ia32_movdqa32store256_mask: | |||
| 13157 | case X86::BI__builtin_ia32_movdqa64store256_mask: | |||
| 13158 | case X86::BI__builtin_ia32_storeaps256_mask: | |||
| 13159 | case X86::BI__builtin_ia32_storeapd256_mask: | |||
| 13160 | case X86::BI__builtin_ia32_movdqa32store512_mask: | |||
| 13161 | case X86::BI__builtin_ia32_movdqa64store512_mask: | |||
| 13162 | case X86::BI__builtin_ia32_storeaps512_mask: | |||
| 13163 | case X86::BI__builtin_ia32_storeapd512_mask: | |||
| 13164 | return EmitX86MaskedStore( | |||
| 13165 | *this, Ops, | |||
| 13166 | getContext().getTypeAlignInChars(E->getArg(1)->getType()).getAsAlign()); | |||
| 13167 | ||||
| 13168 | case X86::BI__builtin_ia32_loadups128_mask: | |||
| 13169 | case X86::BI__builtin_ia32_loadups256_mask: | |||
| 13170 | case X86::BI__builtin_ia32_loadups512_mask: | |||
| 13171 | case X86::BI__builtin_ia32_loadupd128_mask: | |||
| 13172 | case X86::BI__builtin_ia32_loadupd256_mask: | |||
| 13173 | case X86::BI__builtin_ia32_loadupd512_mask: | |||
| 13174 | case X86::BI__builtin_ia32_loaddquqi128_mask: | |||
| 13175 | case X86::BI__builtin_ia32_loaddquqi256_mask: | |||
| 13176 | case X86::BI__builtin_ia32_loaddquqi512_mask: | |||
| 13177 | case X86::BI__builtin_ia32_loaddquhi128_mask: | |||
| 13178 | case X86::BI__builtin_ia32_loaddquhi256_mask: | |||
| 13179 | case X86::BI__builtin_ia32_loaddquhi512_mask: | |||
| 13180 | case X86::BI__builtin_ia32_loaddqusi128_mask: | |||
| 13181 | case X86::BI__builtin_ia32_loaddqusi256_mask: | |||
| 13182 | case X86::BI__builtin_ia32_loaddqusi512_mask: | |||
| 13183 | case X86::BI__builtin_ia32_loaddqudi128_mask: | |||
| 13184 | case X86::BI__builtin_ia32_loaddqudi256_mask: | |||
| 13185 | case X86::BI__builtin_ia32_loaddqudi512_mask: | |||
| 13186 | return EmitX86MaskedLoad(*this, Ops, Align(1)); | |||
| 13187 | ||||
| 13188 | case X86::BI__builtin_ia32_loadsh128_mask: | |||
| 13189 | case X86::BI__builtin_ia32_loadss128_mask: | |||
| 13190 | case X86::BI__builtin_ia32_loadsd128_mask: | |||
| 13191 | return EmitX86MaskedLoad(*this, Ops, Align(1)); | |||
| 13192 | ||||
| 13193 | case X86::BI__builtin_ia32_loadaps128_mask: | |||
| 13194 | case X86::BI__builtin_ia32_loadaps256_mask: | |||
| 13195 | case X86::BI__builtin_ia32_loadaps512_mask: | |||
| 13196 | case X86::BI__builtin_ia32_loadapd128_mask: | |||
| 13197 | case X86::BI__builtin_ia32_loadapd256_mask: | |||
| 13198 | case X86::BI__builtin_ia32_loadapd512_mask: | |||
| 13199 | case X86::BI__builtin_ia32_movdqa32load128_mask: | |||
| 13200 | case X86::BI__builtin_ia32_movdqa32load256_mask: | |||
| 13201 | case X86::BI__builtin_ia32_movdqa32load512_mask: | |||
| 13202 | case X86::BI__builtin_ia32_movdqa64load128_mask: | |||
| 13203 | case X86::BI__builtin_ia32_movdqa64load256_mask: | |||
| 13204 | case X86::BI__builtin_ia32_movdqa64load512_mask: | |||
| 13205 | return EmitX86MaskedLoad( | |||
| 13206 | *this, Ops, | |||
| 13207 | getContext().getTypeAlignInChars(E->getArg(1)->getType()).getAsAlign()); | |||
| 13208 | ||||
| 13209 | case X86::BI__builtin_ia32_expandloaddf128_mask: | |||
| 13210 | case X86::BI__builtin_ia32_expandloaddf256_mask: | |||
| 13211 | case X86::BI__builtin_ia32_expandloaddf512_mask: | |||
| 13212 | case X86::BI__builtin_ia32_expandloadsf128_mask: | |||
| 13213 | case X86::BI__builtin_ia32_expandloadsf256_mask: | |||
| 13214 | case X86::BI__builtin_ia32_expandloadsf512_mask: | |||
| 13215 | case X86::BI__builtin_ia32_expandloaddi128_mask: | |||
| 13216 | case X86::BI__builtin_ia32_expandloaddi256_mask: | |||
| 13217 | case X86::BI__builtin_ia32_expandloaddi512_mask: | |||
| 13218 | case X86::BI__builtin_ia32_expandloadsi128_mask: | |||
| 13219 | case X86::BI__builtin_ia32_expandloadsi256_mask: | |||
| 13220 | case X86::BI__builtin_ia32_expandloadsi512_mask: | |||
| 13221 | case X86::BI__builtin_ia32_expandloadhi128_mask: | |||
| 13222 | case X86::BI__builtin_ia32_expandloadhi256_mask: | |||
| 13223 | case X86::BI__builtin_ia32_expandloadhi512_mask: | |||
| 13224 | case X86::BI__builtin_ia32_expandloadqi128_mask: | |||
| 13225 | case X86::BI__builtin_ia32_expandloadqi256_mask: | |||
| 13226 | case X86::BI__builtin_ia32_expandloadqi512_mask: | |||
| 13227 | return EmitX86ExpandLoad(*this, Ops); | |||
| 13228 | ||||
| 13229 | case X86::BI__builtin_ia32_compressstoredf128_mask: | |||
| 13230 | case X86::BI__builtin_ia32_compressstoredf256_mask: | |||
| 13231 | case X86::BI__builtin_ia32_compressstoredf512_mask: | |||
| 13232 | case X86::BI__builtin_ia32_compressstoresf128_mask: | |||
| 13233 | case X86::BI__builtin_ia32_compressstoresf256_mask: | |||
| 13234 | case X86::BI__builtin_ia32_compressstoresf512_mask: | |||
| 13235 | case X86::BI__builtin_ia32_compressstoredi128_mask: | |||
| 13236 | case X86::BI__builtin_ia32_compressstoredi256_mask: | |||
| 13237 | case X86::BI__builtin_ia32_compressstoredi512_mask: | |||
| 13238 | case X86::BI__builtin_ia32_compressstoresi128_mask: | |||
| 13239 | case X86::BI__builtin_ia32_compressstoresi256_mask: | |||
| 13240 | case X86::BI__builtin_ia32_compressstoresi512_mask: | |||
| 13241 | case X86::BI__builtin_ia32_compressstorehi128_mask: | |||
| 13242 | case X86::BI__builtin_ia32_compressstorehi256_mask: | |||
| 13243 | case X86::BI__builtin_ia32_compressstorehi512_mask: | |||
| 13244 | case X86::BI__builtin_ia32_compressstoreqi128_mask: | |||
| 13245 | case X86::BI__builtin_ia32_compressstoreqi256_mask: | |||
| 13246 | case X86::BI__builtin_ia32_compressstoreqi512_mask: | |||
| 13247 | return EmitX86CompressStore(*this, Ops); | |||
| 13248 | ||||
| 13249 | case X86::BI__builtin_ia32_expanddf128_mask: | |||
| 13250 | case X86::BI__builtin_ia32_expanddf256_mask: | |||
| 13251 | case X86::BI__builtin_ia32_expanddf512_mask: | |||
| 13252 | case X86::BI__builtin_ia32_expandsf128_mask: | |||
| 13253 | case X86::BI__builtin_ia32_expandsf256_mask: | |||
| 13254 | case X86::BI__builtin_ia32_expandsf512_mask: | |||
| 13255 | case X86::BI__builtin_ia32_expanddi128_mask: | |||
| 13256 | case X86::BI__builtin_ia32_expanddi256_mask: | |||
| 13257 | case X86::BI__builtin_ia32_expanddi512_mask: | |||
| 13258 | case X86::BI__builtin_ia32_expandsi128_mask: | |||
| 13259 | case X86::BI__builtin_ia32_expandsi256_mask: | |||
| 13260 | case X86::BI__builtin_ia32_expandsi512_mask: | |||
| 13261 | case X86::BI__builtin_ia32_expandhi128_mask: | |||
| 13262 | case X86::BI__builtin_ia32_expandhi256_mask: | |||
| 13263 | case X86::BI__builtin_ia32_expandhi512_mask: | |||
| 13264 | case X86::BI__builtin_ia32_expandqi128_mask: | |||
| 13265 | case X86::BI__builtin_ia32_expandqi256_mask: | |||
| 13266 | case X86::BI__builtin_ia32_expandqi512_mask: | |||
| 13267 | return EmitX86CompressExpand(*this, Ops, /*IsCompress*/false); | |||
| 13268 | ||||
| 13269 | case X86::BI__builtin_ia32_compressdf128_mask: | |||
| 13270 | case X86::BI__builtin_ia32_compressdf256_mask: | |||
| 13271 | case X86::BI__builtin_ia32_compressdf512_mask: | |||
| 13272 | case X86::BI__builtin_ia32_compresssf128_mask: | |||
| 13273 | case X86::BI__builtin_ia32_compresssf256_mask: | |||
| 13274 | case X86::BI__builtin_ia32_compresssf512_mask: | |||
| 13275 | case X86::BI__builtin_ia32_compressdi128_mask: | |||
| 13276 | case X86::BI__builtin_ia32_compressdi256_mask: | |||
| 13277 | case X86::BI__builtin_ia32_compressdi512_mask: | |||
| 13278 | case X86::BI__builtin_ia32_compresssi128_mask: | |||
| 13279 | case X86::BI__builtin_ia32_compresssi256_mask: | |||
| 13280 | case X86::BI__builtin_ia32_compresssi512_mask: | |||
| 13281 | case X86::BI__builtin_ia32_compresshi128_mask: | |||
| 13282 | case X86::BI__builtin_ia32_compresshi256_mask: | |||
| 13283 | case X86::BI__builtin_ia32_compresshi512_mask: | |||
| 13284 | case X86::BI__builtin_ia32_compressqi128_mask: | |||
| 13285 | case X86::BI__builtin_ia32_compressqi256_mask: | |||
| 13286 | case X86::BI__builtin_ia32_compressqi512_mask: | |||
| 13287 | return EmitX86CompressExpand(*this, Ops, /*IsCompress*/true); | |||
| 13288 | ||||
| 13289 | case X86::BI__builtin_ia32_gather3div2df: | |||
| 13290 | case X86::BI__builtin_ia32_gather3div2di: | |||
| 13291 | case X86::BI__builtin_ia32_gather3div4df: | |||
| 13292 | case X86::BI__builtin_ia32_gather3div4di: | |||
| 13293 | case X86::BI__builtin_ia32_gather3div4sf: | |||
| 13294 | case X86::BI__builtin_ia32_gather3div4si: | |||
| 13295 | case X86::BI__builtin_ia32_gather3div8sf: | |||
| 13296 | case X86::BI__builtin_ia32_gather3div8si: | |||
| 13297 | case X86::BI__builtin_ia32_gather3siv2df: | |||
| 13298 | case X86::BI__builtin_ia32_gather3siv2di: | |||
| 13299 | case X86::BI__builtin_ia32_gather3siv4df: | |||
| 13300 | case X86::BI__builtin_ia32_gather3siv4di: | |||
| 13301 | case X86::BI__builtin_ia32_gather3siv4sf: | |||
| 13302 | case X86::BI__builtin_ia32_gather3siv4si: | |||
| 13303 | case X86::BI__builtin_ia32_gather3siv8sf: | |||
| 13304 | case X86::BI__builtin_ia32_gather3siv8si: | |||
| 13305 | case X86::BI__builtin_ia32_gathersiv8df: | |||
| 13306 | case X86::BI__builtin_ia32_gathersiv16sf: | |||
| 13307 | case X86::BI__builtin_ia32_gatherdiv8df: | |||
| 13308 | case X86::BI__builtin_ia32_gatherdiv16sf: | |||
| 13309 | case X86::BI__builtin_ia32_gathersiv8di: | |||
| 13310 | case X86::BI__builtin_ia32_gathersiv16si: | |||
| 13311 | case X86::BI__builtin_ia32_gatherdiv8di: | |||
| 13312 | case X86::BI__builtin_ia32_gatherdiv16si: { | |||
| 13313 | Intrinsic::ID IID; | |||
| 13314 | switch (BuiltinID) { | |||
| 13315 | default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 13315); | |||
| 13316 | case X86::BI__builtin_ia32_gather3div2df: | |||
| 13317 | IID = Intrinsic::x86_avx512_mask_gather3div2_df; | |||
| 13318 | break; | |||
| 13319 | case X86::BI__builtin_ia32_gather3div2di: | |||
| 13320 | IID = Intrinsic::x86_avx512_mask_gather3div2_di; | |||
| 13321 | break; | |||
| 13322 | case X86::BI__builtin_ia32_gather3div4df: | |||
| 13323 | IID = Intrinsic::x86_avx512_mask_gather3div4_df; | |||
| 13324 | break; | |||
| 13325 | case X86::BI__builtin_ia32_gather3div4di: | |||
| 13326 | IID = Intrinsic::x86_avx512_mask_gather3div4_di; | |||
| 13327 | break; | |||
| 13328 | case X86::BI__builtin_ia32_gather3div4sf: | |||
| 13329 | IID = Intrinsic::x86_avx512_mask_gather3div4_sf; | |||
| 13330 | break; | |||
| 13331 | case X86::BI__builtin_ia32_gather3div4si: | |||
| 13332 | IID = Intrinsic::x86_avx512_mask_gather3div4_si; | |||
| 13333 | break; | |||
| 13334 | case X86::BI__builtin_ia32_gather3div8sf: | |||
| 13335 | IID = Intrinsic::x86_avx512_mask_gather3div8_sf; | |||
| 13336 | break; | |||
| 13337 | case X86::BI__builtin_ia32_gather3div8si: | |||
| 13338 | IID = Intrinsic::x86_avx512_mask_gather3div8_si; | |||
| 13339 | break; | |||
| 13340 | case X86::BI__builtin_ia32_gather3siv2df: | |||
| 13341 | IID = Intrinsic::x86_avx512_mask_gather3siv2_df; | |||
| 13342 | break; | |||
| 13343 | case X86::BI__builtin_ia32_gather3siv2di: | |||
| 13344 | IID = Intrinsic::x86_avx512_mask_gather3siv2_di; | |||
| 13345 | break; | |||
| 13346 | case X86::BI__builtin_ia32_gather3siv4df: | |||
| 13347 | IID = Intrinsic::x86_avx512_mask_gather3siv4_df; | |||
| 13348 | break; | |||
| 13349 | case X86::BI__builtin_ia32_gather3siv4di: | |||
| 13350 | IID = Intrinsic::x86_avx512_mask_gather3siv4_di; | |||
| 13351 | break; | |||
| 13352 | case X86::BI__builtin_ia32_gather3siv4sf: | |||
| 13353 | IID = Intrinsic::x86_avx512_mask_gather3siv4_sf; | |||
| 13354 | break; | |||
| 13355 | case X86::BI__builtin_ia32_gather3siv4si: | |||
| 13356 | IID = Intrinsic::x86_avx512_mask_gather3siv4_si; | |||
| 13357 | break; | |||
| 13358 | case X86::BI__builtin_ia32_gather3siv8sf: | |||
| 13359 | IID = Intrinsic::x86_avx512_mask_gather3siv8_sf; | |||
| 13360 | break; | |||
| 13361 | case X86::BI__builtin_ia32_gather3siv8si: | |||
| 13362 | IID = Intrinsic::x86_avx512_mask_gather3siv8_si; | |||
| 13363 | break; | |||
| 13364 | case X86::BI__builtin_ia32_gathersiv8df: | |||
| 13365 | IID = Intrinsic::x86_avx512_mask_gather_dpd_512; | |||
| 13366 | break; | |||
| 13367 | case X86::BI__builtin_ia32_gathersiv16sf: | |||
| 13368 | IID = Intrinsic::x86_avx512_mask_gather_dps_512; | |||
| 13369 | break; | |||
| 13370 | case X86::BI__builtin_ia32_gatherdiv8df: | |||
| 13371 | IID = Intrinsic::x86_avx512_mask_gather_qpd_512; | |||
| 13372 | break; | |||
| 13373 | case X86::BI__builtin_ia32_gatherdiv16sf: | |||
| 13374 | IID = Intrinsic::x86_avx512_mask_gather_qps_512; | |||
| 13375 | break; | |||
| 13376 | case X86::BI__builtin_ia32_gathersiv8di: | |||
| 13377 | IID = Intrinsic::x86_avx512_mask_gather_dpq_512; | |||
| 13378 | break; | |||
| 13379 | case X86::BI__builtin_ia32_gathersiv16si: | |||
| 13380 | IID = Intrinsic::x86_avx512_mask_gather_dpi_512; | |||
| 13381 | break; | |||
| 13382 | case X86::BI__builtin_ia32_gatherdiv8di: | |||
| 13383 | IID = Intrinsic::x86_avx512_mask_gather_qpq_512; | |||
| 13384 | break; | |||
| 13385 | case X86::BI__builtin_ia32_gatherdiv16si: | |||
| 13386 | IID = Intrinsic::x86_avx512_mask_gather_qpi_512; | |||
| 13387 | break; | |||
| 13388 | } | |||
| 13389 | ||||
| 13390 | unsigned MinElts = std::min( | |||
| 13391 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(), | |||
| 13392 | cast<llvm::FixedVectorType>(Ops[2]->getType())->getNumElements()); | |||
| 13393 | Ops[3] = getMaskVecValue(*this, Ops[3], MinElts); | |||
| 13394 | Function *Intr = CGM.getIntrinsic(IID); | |||
| 13395 | return Builder.CreateCall(Intr, Ops); | |||
| 13396 | } | |||
| 13397 | ||||
| 13398 | case X86::BI__builtin_ia32_scattersiv8df: | |||
| 13399 | case X86::BI__builtin_ia32_scattersiv16sf: | |||
| 13400 | case X86::BI__builtin_ia32_scatterdiv8df: | |||
| 13401 | case X86::BI__builtin_ia32_scatterdiv16sf: | |||
| 13402 | case X86::BI__builtin_ia32_scattersiv8di: | |||
| 13403 | case X86::BI__builtin_ia32_scattersiv16si: | |||
| 13404 | case X86::BI__builtin_ia32_scatterdiv8di: | |||
| 13405 | case X86::BI__builtin_ia32_scatterdiv16si: | |||
| 13406 | case X86::BI__builtin_ia32_scatterdiv2df: | |||
| 13407 | case X86::BI__builtin_ia32_scatterdiv2di: | |||
| 13408 | case X86::BI__builtin_ia32_scatterdiv4df: | |||
| 13409 | case X86::BI__builtin_ia32_scatterdiv4di: | |||
| 13410 | case X86::BI__builtin_ia32_scatterdiv4sf: | |||
| 13411 | case X86::BI__builtin_ia32_scatterdiv4si: | |||
| 13412 | case X86::BI__builtin_ia32_scatterdiv8sf: | |||
| 13413 | case X86::BI__builtin_ia32_scatterdiv8si: | |||
| 13414 | case X86::BI__builtin_ia32_scattersiv2df: | |||
| 13415 | case X86::BI__builtin_ia32_scattersiv2di: | |||
| 13416 | case X86::BI__builtin_ia32_scattersiv4df: | |||
| 13417 | case X86::BI__builtin_ia32_scattersiv4di: | |||
| 13418 | case X86::BI__builtin_ia32_scattersiv4sf: | |||
| 13419 | case X86::BI__builtin_ia32_scattersiv4si: | |||
| 13420 | case X86::BI__builtin_ia32_scattersiv8sf: | |||
| 13421 | case X86::BI__builtin_ia32_scattersiv8si: { | |||
| 13422 | Intrinsic::ID IID; | |||
| 13423 | switch (BuiltinID) { | |||
| 13424 | default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 13424); | |||
| 13425 | case X86::BI__builtin_ia32_scattersiv8df: | |||
| 13426 | IID = Intrinsic::x86_avx512_mask_scatter_dpd_512; | |||
| 13427 | break; | |||
| 13428 | case X86::BI__builtin_ia32_scattersiv16sf: | |||
| 13429 | IID = Intrinsic::x86_avx512_mask_scatter_dps_512; | |||
| 13430 | break; | |||
| 13431 | case X86::BI__builtin_ia32_scatterdiv8df: | |||
| 13432 | IID = Intrinsic::x86_avx512_mask_scatter_qpd_512; | |||
| 13433 | break; | |||
| 13434 | case X86::BI__builtin_ia32_scatterdiv16sf: | |||
| 13435 | IID = Intrinsic::x86_avx512_mask_scatter_qps_512; | |||
| 13436 | break; | |||
| 13437 | case X86::BI__builtin_ia32_scattersiv8di: | |||
| 13438 | IID = Intrinsic::x86_avx512_mask_scatter_dpq_512; | |||
| 13439 | break; | |||
| 13440 | case X86::BI__builtin_ia32_scattersiv16si: | |||
| 13441 | IID = Intrinsic::x86_avx512_mask_scatter_dpi_512; | |||
| 13442 | break; | |||
| 13443 | case X86::BI__builtin_ia32_scatterdiv8di: | |||
| 13444 | IID = Intrinsic::x86_avx512_mask_scatter_qpq_512; | |||
| 13445 | break; | |||
| 13446 | case X86::BI__builtin_ia32_scatterdiv16si: | |||
| 13447 | IID = Intrinsic::x86_avx512_mask_scatter_qpi_512; | |||
| 13448 | break; | |||
| 13449 | case X86::BI__builtin_ia32_scatterdiv2df: | |||
| 13450 | IID = Intrinsic::x86_avx512_mask_scatterdiv2_df; | |||
| 13451 | break; | |||
| 13452 | case X86::BI__builtin_ia32_scatterdiv2di: | |||
| 13453 | IID = Intrinsic::x86_avx512_mask_scatterdiv2_di; | |||
| 13454 | break; | |||
| 13455 | case X86::BI__builtin_ia32_scatterdiv4df: | |||
| 13456 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_df; | |||
| 13457 | break; | |||
| 13458 | case X86::BI__builtin_ia32_scatterdiv4di: | |||
| 13459 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_di; | |||
| 13460 | break; | |||
| 13461 | case X86::BI__builtin_ia32_scatterdiv4sf: | |||
| 13462 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_sf; | |||
| 13463 | break; | |||
| 13464 | case X86::BI__builtin_ia32_scatterdiv4si: | |||
| 13465 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_si; | |||
| 13466 | break; | |||
| 13467 | case X86::BI__builtin_ia32_scatterdiv8sf: | |||
| 13468 | IID = Intrinsic::x86_avx512_mask_scatterdiv8_sf; | |||
| 13469 | break; | |||
| 13470 | case X86::BI__builtin_ia32_scatterdiv8si: | |||
| 13471 | IID = Intrinsic::x86_avx512_mask_scatterdiv8_si; | |||
| 13472 | break; | |||
| 13473 | case X86::BI__builtin_ia32_scattersiv2df: | |||
| 13474 | IID = Intrinsic::x86_avx512_mask_scattersiv2_df; | |||
| 13475 | break; | |||
| 13476 | case X86::BI__builtin_ia32_scattersiv2di: | |||
| 13477 | IID = Intrinsic::x86_avx512_mask_scattersiv2_di; | |||
| 13478 | break; | |||
| 13479 | case X86::BI__builtin_ia32_scattersiv4df: | |||
| 13480 | IID = Intrinsic::x86_avx512_mask_scattersiv4_df; | |||
| 13481 | break; | |||
| 13482 | case X86::BI__builtin_ia32_scattersiv4di: | |||
| 13483 | IID = Intrinsic::x86_avx512_mask_scattersiv4_di; | |||
| 13484 | break; | |||
| 13485 | case X86::BI__builtin_ia32_scattersiv4sf: | |||
| 13486 | IID = Intrinsic::x86_avx512_mask_scattersiv4_sf; | |||
| 13487 | break; | |||
| 13488 | case X86::BI__builtin_ia32_scattersiv4si: | |||
| 13489 | IID = Intrinsic::x86_avx512_mask_scattersiv4_si; | |||
| 13490 | break; | |||
| 13491 | case X86::BI__builtin_ia32_scattersiv8sf: | |||
| 13492 | IID = Intrinsic::x86_avx512_mask_scattersiv8_sf; | |||
| 13493 | break; | |||
| 13494 | case X86::BI__builtin_ia32_scattersiv8si: | |||
| 13495 | IID = Intrinsic::x86_avx512_mask_scattersiv8_si; | |||
| 13496 | break; | |||
| 13497 | } | |||
| 13498 | ||||
| 13499 | unsigned MinElts = std::min( | |||
| 13500 | cast<llvm::FixedVectorType>(Ops[2]->getType())->getNumElements(), | |||
| 13501 | cast<llvm::FixedVectorType>(Ops[3]->getType())->getNumElements()); | |||
| 13502 | Ops[1] = getMaskVecValue(*this, Ops[1], MinElts); | |||
| 13503 | Function *Intr = CGM.getIntrinsic(IID); | |||
| 13504 | return Builder.CreateCall(Intr, Ops); | |||
| 13505 | } | |||
| 13506 | ||||
| 13507 | case X86::BI__builtin_ia32_vextractf128_pd256: | |||
| 13508 | case X86::BI__builtin_ia32_vextractf128_ps256: | |||
| 13509 | case X86::BI__builtin_ia32_vextractf128_si256: | |||
| 13510 | case X86::BI__builtin_ia32_extract128i256: | |||
| 13511 | case X86::BI__builtin_ia32_extractf64x4_mask: | |||
| 13512 | case X86::BI__builtin_ia32_extractf32x4_mask: | |||
| 13513 | case X86::BI__builtin_ia32_extracti64x4_mask: | |||
| 13514 | case X86::BI__builtin_ia32_extracti32x4_mask: | |||
| 13515 | case X86::BI__builtin_ia32_extractf32x8_mask: | |||
| 13516 | case X86::BI__builtin_ia32_extracti32x8_mask: | |||
| 13517 | case X86::BI__builtin_ia32_extractf32x4_256_mask: | |||
| 13518 | case X86::BI__builtin_ia32_extracti32x4_256_mask: | |||
| 13519 | case X86::BI__builtin_ia32_extractf64x2_256_mask: | |||
| 13520 | case X86::BI__builtin_ia32_extracti64x2_256_mask: | |||
| 13521 | case X86::BI__builtin_ia32_extractf64x2_512_mask: | |||
| 13522 | case X86::BI__builtin_ia32_extracti64x2_512_mask: { | |||
| 13523 | auto *DstTy = cast<llvm::FixedVectorType>(ConvertType(E->getType())); | |||
| 13524 | unsigned NumElts = DstTy->getNumElements(); | |||
| 13525 | unsigned SrcNumElts = | |||
| 13526 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 13527 | unsigned SubVectors = SrcNumElts / NumElts; | |||
| 13528 | unsigned Index = cast<ConstantInt>(Ops[1])->getZExtValue(); | |||
| 13529 | assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")(static_cast <bool> (llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors") ? void (0) : __assert_fail ("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 13529, __extension__ __PRETTY_FUNCTION__ )); | |||
| 13530 | Index &= SubVectors - 1; // Remove any extra bits. | |||
| 13531 | Index *= NumElts; | |||
| 13532 | ||||
| 13533 | int Indices[16]; | |||
| 13534 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 13535 | Indices[i] = i + Index; | |||
| 13536 | ||||
| 13537 | Value *Res = Builder.CreateShuffleVector(Ops[0], | |||
| 13538 | makeArrayRef(Indices, NumElts), | |||
| 13539 | "extract"); | |||
| 13540 | ||||
| 13541 | if (Ops.size() == 4) | |||
| 13542 | Res = EmitX86Select(*this, Ops[3], Res, Ops[2]); | |||
| 13543 | ||||
| 13544 | return Res; | |||
| 13545 | } | |||
| 13546 | case X86::BI__builtin_ia32_vinsertf128_pd256: | |||
| 13547 | case X86::BI__builtin_ia32_vinsertf128_ps256: | |||
| 13548 | case X86::BI__builtin_ia32_vinsertf128_si256: | |||
| 13549 | case X86::BI__builtin_ia32_insert128i256: | |||
| 13550 | case X86::BI__builtin_ia32_insertf64x4: | |||
| 13551 | case X86::BI__builtin_ia32_insertf32x4: | |||
| 13552 | case X86::BI__builtin_ia32_inserti64x4: | |||
| 13553 | case X86::BI__builtin_ia32_inserti32x4: | |||
| 13554 | case X86::BI__builtin_ia32_insertf32x8: | |||
| 13555 | case X86::BI__builtin_ia32_inserti32x8: | |||
| 13556 | case X86::BI__builtin_ia32_insertf32x4_256: | |||
| 13557 | case X86::BI__builtin_ia32_inserti32x4_256: | |||
| 13558 | case X86::BI__builtin_ia32_insertf64x2_256: | |||
| 13559 | case X86::BI__builtin_ia32_inserti64x2_256: | |||
| 13560 | case X86::BI__builtin_ia32_insertf64x2_512: | |||
| 13561 | case X86::BI__builtin_ia32_inserti64x2_512: { | |||
| 13562 | unsigned DstNumElts = | |||
| 13563 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 13564 | unsigned SrcNumElts = | |||
| 13565 | cast<llvm::FixedVectorType>(Ops[1]->getType())->getNumElements(); | |||
| 13566 | unsigned SubVectors = DstNumElts / SrcNumElts; | |||
| 13567 | unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue(); | |||
| 13568 | assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")(static_cast <bool> (llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors") ? void (0) : __assert_fail ("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 13568, __extension__ __PRETTY_FUNCTION__ )); | |||
| 13569 | Index &= SubVectors - 1; // Remove any extra bits. | |||
| 13570 | Index *= SrcNumElts; | |||
| 13571 | ||||
| 13572 | int Indices[16]; | |||
| 13573 | for (unsigned i = 0; i != DstNumElts; ++i) | |||
| 13574 | Indices[i] = (i >= SrcNumElts) ? SrcNumElts + (i % SrcNumElts) : i; | |||
| 13575 | ||||
| 13576 | Value *Op1 = Builder.CreateShuffleVector(Ops[1], | |||
| 13577 | makeArrayRef(Indices, DstNumElts), | |||
| 13578 | "widen"); | |||
| 13579 | ||||
| 13580 | for (unsigned i = 0; i != DstNumElts; ++i) { | |||
| 13581 | if (i >= Index && i < (Index + SrcNumElts)) | |||
| 13582 | Indices[i] = (i - Index) + DstNumElts; | |||
| 13583 | else | |||
| 13584 | Indices[i] = i; | |||
| 13585 | } | |||
| 13586 | ||||
| 13587 | return Builder.CreateShuffleVector(Ops[0], Op1, | |||
| 13588 | makeArrayRef(Indices, DstNumElts), | |||
| 13589 | "insert"); | |||
| 13590 | } | |||
| 13591 | case X86::BI__builtin_ia32_pmovqd512_mask: | |||
| 13592 | case X86::BI__builtin_ia32_pmovwb512_mask: { | |||
| 13593 | Value *Res = Builder.CreateTrunc(Ops[0], Ops[1]->getType()); | |||
| 13594 | return EmitX86Select(*this, Ops[2], Res, Ops[1]); | |||
| 13595 | } | |||
| 13596 | case X86::BI__builtin_ia32_pmovdb512_mask: | |||
| 13597 | case X86::BI__builtin_ia32_pmovdw512_mask: | |||
| 13598 | case X86::BI__builtin_ia32_pmovqw512_mask: { | |||
| 13599 | if (const auto *C = dyn_cast<Constant>(Ops[2])) | |||
| 13600 | if (C->isAllOnesValue()) | |||
| 13601 | return Builder.CreateTrunc(Ops[0], Ops[1]->getType()); | |||
| 13602 | ||||
| 13603 | Intrinsic::ID IID; | |||
| 13604 | switch (BuiltinID) { | |||
| 13605 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 13605); | |||
| 13606 | case X86::BI__builtin_ia32_pmovdb512_mask: | |||
| 13607 | IID = Intrinsic::x86_avx512_mask_pmov_db_512; | |||
| 13608 | break; | |||
| 13609 | case X86::BI__builtin_ia32_pmovdw512_mask: | |||
| 13610 | IID = Intrinsic::x86_avx512_mask_pmov_dw_512; | |||
| 13611 | break; | |||
| 13612 | case X86::BI__builtin_ia32_pmovqw512_mask: | |||
| 13613 | IID = Intrinsic::x86_avx512_mask_pmov_qw_512; | |||
| 13614 | break; | |||
| 13615 | } | |||
| 13616 | ||||
| 13617 | Function *Intr = CGM.getIntrinsic(IID); | |||
| 13618 | return Builder.CreateCall(Intr, Ops); | |||
| 13619 | } | |||
| 13620 | case X86::BI__builtin_ia32_pblendw128: | |||
| 13621 | case X86::BI__builtin_ia32_blendpd: | |||
| 13622 | case X86::BI__builtin_ia32_blendps: | |||
| 13623 | case X86::BI__builtin_ia32_blendpd256: | |||
| 13624 | case X86::BI__builtin_ia32_blendps256: | |||
| 13625 | case X86::BI__builtin_ia32_pblendw256: | |||
| 13626 | case X86::BI__builtin_ia32_pblendd128: | |||
| 13627 | case X86::BI__builtin_ia32_pblendd256: { | |||
| 13628 | unsigned NumElts = | |||
| 13629 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 13630 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
| 13631 | ||||
| 13632 | int Indices[16]; | |||
| 13633 | // If there are more than 8 elements, the immediate is used twice so make | |||
| 13634 | // sure we handle that. | |||
| 13635 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 13636 | Indices[i] = ((Imm >> (i % 8)) & 0x1) ? NumElts + i : i; | |||
| 13637 | ||||
| 13638 | return Builder.CreateShuffleVector(Ops[0], Ops[1], | |||
| 13639 | makeArrayRef(Indices, NumElts), | |||
| 13640 | "blend"); | |||
| 13641 | } | |||
| 13642 | case X86::BI__builtin_ia32_pshuflw: | |||
| 13643 | case X86::BI__builtin_ia32_pshuflw256: | |||
| 13644 | case X86::BI__builtin_ia32_pshuflw512: { | |||
| 13645 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
| 13646 | auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13647 | unsigned NumElts = Ty->getNumElements(); | |||
| 13648 | ||||
| 13649 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
| 13650 | Imm = (Imm & 0xff) * 0x01010101; | |||
| 13651 | ||||
| 13652 | int Indices[32]; | |||
| 13653 | for (unsigned l = 0; l != NumElts; l += 8) { | |||
| 13654 | for (unsigned i = 0; i != 4; ++i) { | |||
| 13655 | Indices[l + i] = l + (Imm & 3); | |||
| 13656 | Imm >>= 2; | |||
| 13657 | } | |||
| 13658 | for (unsigned i = 4; i != 8; ++i) | |||
| 13659 | Indices[l + i] = l + i; | |||
| 13660 | } | |||
| 13661 | ||||
| 13662 | return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts), | |||
| 13663 | "pshuflw"); | |||
| 13664 | } | |||
| 13665 | case X86::BI__builtin_ia32_pshufhw: | |||
| 13666 | case X86::BI__builtin_ia32_pshufhw256: | |||
| 13667 | case X86::BI__builtin_ia32_pshufhw512: { | |||
| 13668 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
| 13669 | auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13670 | unsigned NumElts = Ty->getNumElements(); | |||
| 13671 | ||||
| 13672 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
| 13673 | Imm = (Imm & 0xff) * 0x01010101; | |||
| 13674 | ||||
| 13675 | int Indices[32]; | |||
| 13676 | for (unsigned l = 0; l != NumElts; l += 8) { | |||
| 13677 | for (unsigned i = 0; i != 4; ++i) | |||
| 13678 | Indices[l + i] = l + i; | |||
| 13679 | for (unsigned i = 4; i != 8; ++i) { | |||
| 13680 | Indices[l + i] = l + 4 + (Imm & 3); | |||
| 13681 | Imm >>= 2; | |||
| 13682 | } | |||
| 13683 | } | |||
| 13684 | ||||
| 13685 | return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts), | |||
| 13686 | "pshufhw"); | |||
| 13687 | } | |||
| 13688 | case X86::BI__builtin_ia32_pshufd: | |||
| 13689 | case X86::BI__builtin_ia32_pshufd256: | |||
| 13690 | case X86::BI__builtin_ia32_pshufd512: | |||
| 13691 | case X86::BI__builtin_ia32_vpermilpd: | |||
| 13692 | case X86::BI__builtin_ia32_vpermilps: | |||
| 13693 | case X86::BI__builtin_ia32_vpermilpd256: | |||
| 13694 | case X86::BI__builtin_ia32_vpermilps256: | |||
| 13695 | case X86::BI__builtin_ia32_vpermilpd512: | |||
| 13696 | case X86::BI__builtin_ia32_vpermilps512: { | |||
| 13697 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
| 13698 | auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13699 | unsigned NumElts = Ty->getNumElements(); | |||
| 13700 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128; | |||
| 13701 | unsigned NumLaneElts = NumElts / NumLanes; | |||
| 13702 | ||||
| 13703 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
| 13704 | Imm = (Imm & 0xff) * 0x01010101; | |||
| 13705 | ||||
| 13706 | int Indices[16]; | |||
| 13707 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { | |||
| 13708 | for (unsigned i = 0; i != NumLaneElts; ++i) { | |||
| 13709 | Indices[i + l] = (Imm % NumLaneElts) + l; | |||
| 13710 | Imm /= NumLaneElts; | |||
| 13711 | } | |||
| 13712 | } | |||
| 13713 | ||||
| 13714 | return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts), | |||
| 13715 | "permil"); | |||
| 13716 | } | |||
| 13717 | case X86::BI__builtin_ia32_shufpd: | |||
| 13718 | case X86::BI__builtin_ia32_shufpd256: | |||
| 13719 | case X86::BI__builtin_ia32_shufpd512: | |||
| 13720 | case X86::BI__builtin_ia32_shufps: | |||
| 13721 | case X86::BI__builtin_ia32_shufps256: | |||
| 13722 | case X86::BI__builtin_ia32_shufps512: { | |||
| 13723 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
| 13724 | auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13725 | unsigned NumElts = Ty->getNumElements(); | |||
| 13726 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128; | |||
| 13727 | unsigned NumLaneElts = NumElts / NumLanes; | |||
| 13728 | ||||
| 13729 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. | |||
| 13730 | Imm = (Imm & 0xff) * 0x01010101; | |||
| 13731 | ||||
| 13732 | int Indices[16]; | |||
| 13733 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { | |||
| 13734 | for (unsigned i = 0; i != NumLaneElts; ++i) { | |||
| 13735 | unsigned Index = Imm % NumLaneElts; | |||
| 13736 | Imm /= NumLaneElts; | |||
| 13737 | if (i >= (NumLaneElts / 2)) | |||
| 13738 | Index += NumElts; | |||
| 13739 | Indices[l + i] = l + Index; | |||
| 13740 | } | |||
| 13741 | } | |||
| 13742 | ||||
| 13743 | return Builder.CreateShuffleVector(Ops[0], Ops[1], | |||
| 13744 | makeArrayRef(Indices, NumElts), | |||
| 13745 | "shufp"); | |||
| 13746 | } | |||
| 13747 | case X86::BI__builtin_ia32_permdi256: | |||
| 13748 | case X86::BI__builtin_ia32_permdf256: | |||
| 13749 | case X86::BI__builtin_ia32_permdi512: | |||
| 13750 | case X86::BI__builtin_ia32_permdf512: { | |||
| 13751 | unsigned Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
| 13752 | auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13753 | unsigned NumElts = Ty->getNumElements(); | |||
| 13754 | ||||
| 13755 | // These intrinsics operate on 256-bit lanes of four 64-bit elements. | |||
| 13756 | int Indices[8]; | |||
| 13757 | for (unsigned l = 0; l != NumElts; l += 4) | |||
| 13758 | for (unsigned i = 0; i != 4; ++i) | |||
| 13759 | Indices[l + i] = l + ((Imm >> (2 * i)) & 0x3); | |||
| 13760 | ||||
| 13761 | return Builder.CreateShuffleVector(Ops[0], makeArrayRef(Indices, NumElts), | |||
| 13762 | "perm"); | |||
| 13763 | } | |||
| 13764 | case X86::BI__builtin_ia32_palignr128: | |||
| 13765 | case X86::BI__builtin_ia32_palignr256: | |||
| 13766 | case X86::BI__builtin_ia32_palignr512: { | |||
| 13767 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff; | |||
| 13768 | ||||
| 13769 | unsigned NumElts = | |||
| 13770 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 13771 | assert(NumElts % 16 == 0)(static_cast <bool> (NumElts % 16 == 0) ? void (0) : __assert_fail ("NumElts % 16 == 0", "clang/lib/CodeGen/CGBuiltin.cpp", 13771 , __extension__ __PRETTY_FUNCTION__)); | |||
| 13772 | ||||
| 13773 | // If palignr is shifting the pair of vectors more than the size of two | |||
| 13774 | // lanes, emit zero. | |||
| 13775 | if (ShiftVal >= 32) | |||
| 13776 | return llvm::Constant::getNullValue(ConvertType(E->getType())); | |||
| 13777 | ||||
| 13778 | // If palignr is shifting the pair of input vectors more than one lane, | |||
| 13779 | // but less than two lanes, convert to shifting in zeroes. | |||
| 13780 | if (ShiftVal > 16) { | |||
| 13781 | ShiftVal -= 16; | |||
| 13782 | Ops[1] = Ops[0]; | |||
| 13783 | Ops[0] = llvm::Constant::getNullValue(Ops[0]->getType()); | |||
| 13784 | } | |||
| 13785 | ||||
| 13786 | int Indices[64]; | |||
| 13787 | // 256-bit palignr operates on 128-bit lanes so we need to handle that | |||
| 13788 | for (unsigned l = 0; l != NumElts; l += 16) { | |||
| 13789 | for (unsigned i = 0; i != 16; ++i) { | |||
| 13790 | unsigned Idx = ShiftVal + i; | |||
| 13791 | if (Idx >= 16) | |||
| 13792 | Idx += NumElts - 16; // End of lane, switch operand. | |||
| 13793 | Indices[l + i] = Idx + l; | |||
| 13794 | } | |||
| 13795 | } | |||
| 13796 | ||||
| 13797 | return Builder.CreateShuffleVector(Ops[1], Ops[0], | |||
| 13798 | makeArrayRef(Indices, NumElts), | |||
| 13799 | "palignr"); | |||
| 13800 | } | |||
| 13801 | case X86::BI__builtin_ia32_alignd128: | |||
| 13802 | case X86::BI__builtin_ia32_alignd256: | |||
| 13803 | case X86::BI__builtin_ia32_alignd512: | |||
| 13804 | case X86::BI__builtin_ia32_alignq128: | |||
| 13805 | case X86::BI__builtin_ia32_alignq256: | |||
| 13806 | case X86::BI__builtin_ia32_alignq512: { | |||
| 13807 | unsigned NumElts = | |||
| 13808 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 13809 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff; | |||
| 13810 | ||||
| 13811 | // Mask the shift amount to width of a vector. | |||
| 13812 | ShiftVal &= NumElts - 1; | |||
| 13813 | ||||
| 13814 | int Indices[16]; | |||
| 13815 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 13816 | Indices[i] = i + ShiftVal; | |||
| 13817 | ||||
| 13818 | return Builder.CreateShuffleVector(Ops[1], Ops[0], | |||
| 13819 | makeArrayRef(Indices, NumElts), | |||
| 13820 | "valign"); | |||
| 13821 | } | |||
| 13822 | case X86::BI__builtin_ia32_shuf_f32x4_256: | |||
| 13823 | case X86::BI__builtin_ia32_shuf_f64x2_256: | |||
| 13824 | case X86::BI__builtin_ia32_shuf_i32x4_256: | |||
| 13825 | case X86::BI__builtin_ia32_shuf_i64x2_256: | |||
| 13826 | case X86::BI__builtin_ia32_shuf_f32x4: | |||
| 13827 | case X86::BI__builtin_ia32_shuf_f64x2: | |||
| 13828 | case X86::BI__builtin_ia32_shuf_i32x4: | |||
| 13829 | case X86::BI__builtin_ia32_shuf_i64x2: { | |||
| 13830 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
| 13831 | auto *Ty = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13832 | unsigned NumElts = Ty->getNumElements(); | |||
| 13833 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() == 512 ? 4 : 2; | |||
| 13834 | unsigned NumLaneElts = NumElts / NumLanes; | |||
| 13835 | ||||
| 13836 | int Indices[16]; | |||
| 13837 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { | |||
| 13838 | unsigned Index = (Imm % NumLanes) * NumLaneElts; | |||
| 13839 | Imm /= NumLanes; // Discard the bits we just used. | |||
| 13840 | if (l >= (NumElts / 2)) | |||
| 13841 | Index += NumElts; // Switch to other source. | |||
| 13842 | for (unsigned i = 0; i != NumLaneElts; ++i) { | |||
| 13843 | Indices[l + i] = Index + i; | |||
| 13844 | } | |||
| 13845 | } | |||
| 13846 | ||||
| 13847 | return Builder.CreateShuffleVector(Ops[0], Ops[1], | |||
| 13848 | makeArrayRef(Indices, NumElts), | |||
| 13849 | "shuf"); | |||
| 13850 | } | |||
| 13851 | ||||
| 13852 | case X86::BI__builtin_ia32_vperm2f128_pd256: | |||
| 13853 | case X86::BI__builtin_ia32_vperm2f128_ps256: | |||
| 13854 | case X86::BI__builtin_ia32_vperm2f128_si256: | |||
| 13855 | case X86::BI__builtin_ia32_permti256: { | |||
| 13856 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); | |||
| 13857 | unsigned NumElts = | |||
| 13858 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 13859 | ||||
| 13860 | // This takes a very simple approach since there are two lanes and a | |||
| 13861 | // shuffle can have 2 inputs. So we reserve the first input for the first | |||
| 13862 | // lane and the second input for the second lane. This may result in | |||
| 13863 | // duplicate sources, but this can be dealt with in the backend. | |||
| 13864 | ||||
| 13865 | Value *OutOps[2]; | |||
| 13866 | int Indices[8]; | |||
| 13867 | for (unsigned l = 0; l != 2; ++l) { | |||
| 13868 | // Determine the source for this lane. | |||
| 13869 | if (Imm & (1 << ((l * 4) + 3))) | |||
| 13870 | OutOps[l] = llvm::ConstantAggregateZero::get(Ops[0]->getType()); | |||
| 13871 | else if (Imm & (1 << ((l * 4) + 1))) | |||
| 13872 | OutOps[l] = Ops[1]; | |||
| 13873 | else | |||
| 13874 | OutOps[l] = Ops[0]; | |||
| 13875 | ||||
| 13876 | for (unsigned i = 0; i != NumElts/2; ++i) { | |||
| 13877 | // Start with ith element of the source for this lane. | |||
| 13878 | unsigned Idx = (l * NumElts) + i; | |||
| 13879 | // If bit 0 of the immediate half is set, switch to the high half of | |||
| 13880 | // the source. | |||
| 13881 | if (Imm & (1 << (l * 4))) | |||
| 13882 | Idx += NumElts/2; | |||
| 13883 | Indices[(l * (NumElts/2)) + i] = Idx; | |||
| 13884 | } | |||
| 13885 | } | |||
| 13886 | ||||
| 13887 | return Builder.CreateShuffleVector(OutOps[0], OutOps[1], | |||
| 13888 | makeArrayRef(Indices, NumElts), | |||
| 13889 | "vperm"); | |||
| 13890 | } | |||
| 13891 | ||||
| 13892 | case X86::BI__builtin_ia32_pslldqi128_byteshift: | |||
| 13893 | case X86::BI__builtin_ia32_pslldqi256_byteshift: | |||
| 13894 | case X86::BI__builtin_ia32_pslldqi512_byteshift: { | |||
| 13895 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
| 13896 | auto *ResultType = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13897 | // Builtin type is vXi64 so multiply by 8 to get bytes. | |||
| 13898 | unsigned NumElts = ResultType->getNumElements() * 8; | |||
| 13899 | ||||
| 13900 | // If pslldq is shifting the vector more than 15 bytes, emit zero. | |||
| 13901 | if (ShiftVal >= 16) | |||
| 13902 | return llvm::Constant::getNullValue(ResultType); | |||
| 13903 | ||||
| 13904 | int Indices[64]; | |||
| 13905 | // 256/512-bit pslldq operates on 128-bit lanes so we need to handle that | |||
| 13906 | for (unsigned l = 0; l != NumElts; l += 16) { | |||
| 13907 | for (unsigned i = 0; i != 16; ++i) { | |||
| 13908 | unsigned Idx = NumElts + i - ShiftVal; | |||
| 13909 | if (Idx < NumElts) Idx -= NumElts - 16; // end of lane, switch operand. | |||
| 13910 | Indices[l + i] = Idx + l; | |||
| 13911 | } | |||
| 13912 | } | |||
| 13913 | ||||
| 13914 | auto *VecTy = llvm::FixedVectorType::get(Int8Ty, NumElts); | |||
| 13915 | Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast"); | |||
| 13916 | Value *Zero = llvm::Constant::getNullValue(VecTy); | |||
| 13917 | Value *SV = Builder.CreateShuffleVector(Zero, Cast, | |||
| 13918 | makeArrayRef(Indices, NumElts), | |||
| 13919 | "pslldq"); | |||
| 13920 | return Builder.CreateBitCast(SV, Ops[0]->getType(), "cast"); | |||
| 13921 | } | |||
| 13922 | case X86::BI__builtin_ia32_psrldqi128_byteshift: | |||
| 13923 | case X86::BI__builtin_ia32_psrldqi256_byteshift: | |||
| 13924 | case X86::BI__builtin_ia32_psrldqi512_byteshift: { | |||
| 13925 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
| 13926 | auto *ResultType = cast<llvm::FixedVectorType>(Ops[0]->getType()); | |||
| 13927 | // Builtin type is vXi64 so multiply by 8 to get bytes. | |||
| 13928 | unsigned NumElts = ResultType->getNumElements() * 8; | |||
| 13929 | ||||
| 13930 | // If psrldq is shifting the vector more than 15 bytes, emit zero. | |||
| 13931 | if (ShiftVal >= 16) | |||
| 13932 | return llvm::Constant::getNullValue(ResultType); | |||
| 13933 | ||||
| 13934 | int Indices[64]; | |||
| 13935 | // 256/512-bit psrldq operates on 128-bit lanes so we need to handle that | |||
| 13936 | for (unsigned l = 0; l != NumElts; l += 16) { | |||
| 13937 | for (unsigned i = 0; i != 16; ++i) { | |||
| 13938 | unsigned Idx = i + ShiftVal; | |||
| 13939 | if (Idx >= 16) Idx += NumElts - 16; // end of lane, switch operand. | |||
| 13940 | Indices[l + i] = Idx + l; | |||
| 13941 | } | |||
| 13942 | } | |||
| 13943 | ||||
| 13944 | auto *VecTy = llvm::FixedVectorType::get(Int8Ty, NumElts); | |||
| 13945 | Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast"); | |||
| 13946 | Value *Zero = llvm::Constant::getNullValue(VecTy); | |||
| 13947 | Value *SV = Builder.CreateShuffleVector(Cast, Zero, | |||
| 13948 | makeArrayRef(Indices, NumElts), | |||
| 13949 | "psrldq"); | |||
| 13950 | return Builder.CreateBitCast(SV, ResultType, "cast"); | |||
| 13951 | } | |||
| 13952 | case X86::BI__builtin_ia32_kshiftliqi: | |||
| 13953 | case X86::BI__builtin_ia32_kshiftlihi: | |||
| 13954 | case X86::BI__builtin_ia32_kshiftlisi: | |||
| 13955 | case X86::BI__builtin_ia32_kshiftlidi: { | |||
| 13956 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
| 13957 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 13958 | ||||
| 13959 | if (ShiftVal >= NumElts) | |||
| 13960 | return llvm::Constant::getNullValue(Ops[0]->getType()); | |||
| 13961 | ||||
| 13962 | Value *In = getMaskVecValue(*this, Ops[0], NumElts); | |||
| 13963 | ||||
| 13964 | int Indices[64]; | |||
| 13965 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 13966 | Indices[i] = NumElts + i - ShiftVal; | |||
| 13967 | ||||
| 13968 | Value *Zero = llvm::Constant::getNullValue(In->getType()); | |||
| 13969 | Value *SV = Builder.CreateShuffleVector(Zero, In, | |||
| 13970 | makeArrayRef(Indices, NumElts), | |||
| 13971 | "kshiftl"); | |||
| 13972 | return Builder.CreateBitCast(SV, Ops[0]->getType()); | |||
| 13973 | } | |||
| 13974 | case X86::BI__builtin_ia32_kshiftriqi: | |||
| 13975 | case X86::BI__builtin_ia32_kshiftrihi: | |||
| 13976 | case X86::BI__builtin_ia32_kshiftrisi: | |||
| 13977 | case X86::BI__builtin_ia32_kshiftridi: { | |||
| 13978 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; | |||
| 13979 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 13980 | ||||
| 13981 | if (ShiftVal >= NumElts) | |||
| 13982 | return llvm::Constant::getNullValue(Ops[0]->getType()); | |||
| 13983 | ||||
| 13984 | Value *In = getMaskVecValue(*this, Ops[0], NumElts); | |||
| 13985 | ||||
| 13986 | int Indices[64]; | |||
| 13987 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 13988 | Indices[i] = i + ShiftVal; | |||
| 13989 | ||||
| 13990 | Value *Zero = llvm::Constant::getNullValue(In->getType()); | |||
| 13991 | Value *SV = Builder.CreateShuffleVector(In, Zero, | |||
| 13992 | makeArrayRef(Indices, NumElts), | |||
| 13993 | "kshiftr"); | |||
| 13994 | return Builder.CreateBitCast(SV, Ops[0]->getType()); | |||
| 13995 | } | |||
| 13996 | case X86::BI__builtin_ia32_movnti: | |||
| 13997 | case X86::BI__builtin_ia32_movnti64: | |||
| 13998 | case X86::BI__builtin_ia32_movntsd: | |||
| 13999 | case X86::BI__builtin_ia32_movntss: { | |||
| 14000 | llvm::MDNode *Node = llvm::MDNode::get( | |||
| 14001 | getLLVMContext(), llvm::ConstantAsMetadata::get(Builder.getInt32(1))); | |||
| 14002 | ||||
| 14003 | Value *Ptr = Ops[0]; | |||
| 14004 | Value *Src = Ops[1]; | |||
| 14005 | ||||
| 14006 | // Extract the 0'th element of the source vector. | |||
| 14007 | if (BuiltinID == X86::BI__builtin_ia32_movntsd || | |||
| 14008 | BuiltinID == X86::BI__builtin_ia32_movntss) | |||
| 14009 | Src = Builder.CreateExtractElement(Src, (uint64_t)0, "extract"); | |||
| 14010 | ||||
| 14011 | // Convert the type of the pointer to a pointer to the stored type. | |||
| 14012 | Value *BC = Builder.CreateBitCast( | |||
| 14013 | Ptr, llvm::PointerType::getUnqual(Src->getType()), "cast"); | |||
| 14014 | ||||
| 14015 | // Unaligned nontemporal store of the scalar value. | |||
| 14016 | StoreInst *SI = Builder.CreateDefaultAlignedStore(Src, BC); | |||
| 14017 | SI->setMetadata(CGM.getModule().getMDKindID("nontemporal"), Node); | |||
| 14018 | SI->setAlignment(llvm::Align(1)); | |||
| 14019 | return SI; | |||
| 14020 | } | |||
| 14021 | // Rotate is a special case of funnel shift - 1st 2 args are the same. | |||
| 14022 | case X86::BI__builtin_ia32_vprotb: | |||
| 14023 | case X86::BI__builtin_ia32_vprotw: | |||
| 14024 | case X86::BI__builtin_ia32_vprotd: | |||
| 14025 | case X86::BI__builtin_ia32_vprotq: | |||
| 14026 | case X86::BI__builtin_ia32_vprotbi: | |||
| 14027 | case X86::BI__builtin_ia32_vprotwi: | |||
| 14028 | case X86::BI__builtin_ia32_vprotdi: | |||
| 14029 | case X86::BI__builtin_ia32_vprotqi: | |||
| 14030 | case X86::BI__builtin_ia32_prold128: | |||
| 14031 | case X86::BI__builtin_ia32_prold256: | |||
| 14032 | case X86::BI__builtin_ia32_prold512: | |||
| 14033 | case X86::BI__builtin_ia32_prolq128: | |||
| 14034 | case X86::BI__builtin_ia32_prolq256: | |||
| 14035 | case X86::BI__builtin_ia32_prolq512: | |||
| 14036 | case X86::BI__builtin_ia32_prolvd128: | |||
| 14037 | case X86::BI__builtin_ia32_prolvd256: | |||
| 14038 | case X86::BI__builtin_ia32_prolvd512: | |||
| 14039 | case X86::BI__builtin_ia32_prolvq128: | |||
| 14040 | case X86::BI__builtin_ia32_prolvq256: | |||
| 14041 | case X86::BI__builtin_ia32_prolvq512: | |||
| 14042 | return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], false); | |||
| 14043 | case X86::BI__builtin_ia32_prord128: | |||
| 14044 | case X86::BI__builtin_ia32_prord256: | |||
| 14045 | case X86::BI__builtin_ia32_prord512: | |||
| 14046 | case X86::BI__builtin_ia32_prorq128: | |||
| 14047 | case X86::BI__builtin_ia32_prorq256: | |||
| 14048 | case X86::BI__builtin_ia32_prorq512: | |||
| 14049 | case X86::BI__builtin_ia32_prorvd128: | |||
| 14050 | case X86::BI__builtin_ia32_prorvd256: | |||
| 14051 | case X86::BI__builtin_ia32_prorvd512: | |||
| 14052 | case X86::BI__builtin_ia32_prorvq128: | |||
| 14053 | case X86::BI__builtin_ia32_prorvq256: | |||
| 14054 | case X86::BI__builtin_ia32_prorvq512: | |||
| 14055 | return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], true); | |||
| 14056 | case X86::BI__builtin_ia32_selectb_128: | |||
| 14057 | case X86::BI__builtin_ia32_selectb_256: | |||
| 14058 | case X86::BI__builtin_ia32_selectb_512: | |||
| 14059 | case X86::BI__builtin_ia32_selectw_128: | |||
| 14060 | case X86::BI__builtin_ia32_selectw_256: | |||
| 14061 | case X86::BI__builtin_ia32_selectw_512: | |||
| 14062 | case X86::BI__builtin_ia32_selectd_128: | |||
| 14063 | case X86::BI__builtin_ia32_selectd_256: | |||
| 14064 | case X86::BI__builtin_ia32_selectd_512: | |||
| 14065 | case X86::BI__builtin_ia32_selectq_128: | |||
| 14066 | case X86::BI__builtin_ia32_selectq_256: | |||
| 14067 | case X86::BI__builtin_ia32_selectq_512: | |||
| 14068 | case X86::BI__builtin_ia32_selectph_128: | |||
| 14069 | case X86::BI__builtin_ia32_selectph_256: | |||
| 14070 | case X86::BI__builtin_ia32_selectph_512: | |||
| 14071 | case X86::BI__builtin_ia32_selectps_128: | |||
| 14072 | case X86::BI__builtin_ia32_selectps_256: | |||
| 14073 | case X86::BI__builtin_ia32_selectps_512: | |||
| 14074 | case X86::BI__builtin_ia32_selectpd_128: | |||
| 14075 | case X86::BI__builtin_ia32_selectpd_256: | |||
| 14076 | case X86::BI__builtin_ia32_selectpd_512: | |||
| 14077 | return EmitX86Select(*this, Ops[0], Ops[1], Ops[2]); | |||
| 14078 | case X86::BI__builtin_ia32_selectsh_128: | |||
| 14079 | case X86::BI__builtin_ia32_selectss_128: | |||
| 14080 | case X86::BI__builtin_ia32_selectsd_128: { | |||
| 14081 | Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0); | |||
| 14082 | Value *B = Builder.CreateExtractElement(Ops[2], (uint64_t)0); | |||
| 14083 | A = EmitX86ScalarSelect(*this, Ops[0], A, B); | |||
| 14084 | return Builder.CreateInsertElement(Ops[1], A, (uint64_t)0); | |||
| 14085 | } | |||
| 14086 | case X86::BI__builtin_ia32_cmpb128_mask: | |||
| 14087 | case X86::BI__builtin_ia32_cmpb256_mask: | |||
| 14088 | case X86::BI__builtin_ia32_cmpb512_mask: | |||
| 14089 | case X86::BI__builtin_ia32_cmpw128_mask: | |||
| 14090 | case X86::BI__builtin_ia32_cmpw256_mask: | |||
| 14091 | case X86::BI__builtin_ia32_cmpw512_mask: | |||
| 14092 | case X86::BI__builtin_ia32_cmpd128_mask: | |||
| 14093 | case X86::BI__builtin_ia32_cmpd256_mask: | |||
| 14094 | case X86::BI__builtin_ia32_cmpd512_mask: | |||
| 14095 | case X86::BI__builtin_ia32_cmpq128_mask: | |||
| 14096 | case X86::BI__builtin_ia32_cmpq256_mask: | |||
| 14097 | case X86::BI__builtin_ia32_cmpq512_mask: { | |||
| 14098 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; | |||
| 14099 | return EmitX86MaskedCompare(*this, CC, true, Ops); | |||
| 14100 | } | |||
| 14101 | case X86::BI__builtin_ia32_ucmpb128_mask: | |||
| 14102 | case X86::BI__builtin_ia32_ucmpb256_mask: | |||
| 14103 | case X86::BI__builtin_ia32_ucmpb512_mask: | |||
| 14104 | case X86::BI__builtin_ia32_ucmpw128_mask: | |||
| 14105 | case X86::BI__builtin_ia32_ucmpw256_mask: | |||
| 14106 | case X86::BI__builtin_ia32_ucmpw512_mask: | |||
| 14107 | case X86::BI__builtin_ia32_ucmpd128_mask: | |||
| 14108 | case X86::BI__builtin_ia32_ucmpd256_mask: | |||
| 14109 | case X86::BI__builtin_ia32_ucmpd512_mask: | |||
| 14110 | case X86::BI__builtin_ia32_ucmpq128_mask: | |||
| 14111 | case X86::BI__builtin_ia32_ucmpq256_mask: | |||
| 14112 | case X86::BI__builtin_ia32_ucmpq512_mask: { | |||
| 14113 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; | |||
| 14114 | return EmitX86MaskedCompare(*this, CC, false, Ops); | |||
| 14115 | } | |||
| 14116 | case X86::BI__builtin_ia32_vpcomb: | |||
| 14117 | case X86::BI__builtin_ia32_vpcomw: | |||
| 14118 | case X86::BI__builtin_ia32_vpcomd: | |||
| 14119 | case X86::BI__builtin_ia32_vpcomq: | |||
| 14120 | return EmitX86vpcom(*this, Ops, true); | |||
| 14121 | case X86::BI__builtin_ia32_vpcomub: | |||
| 14122 | case X86::BI__builtin_ia32_vpcomuw: | |||
| 14123 | case X86::BI__builtin_ia32_vpcomud: | |||
| 14124 | case X86::BI__builtin_ia32_vpcomuq: | |||
| 14125 | return EmitX86vpcom(*this, Ops, false); | |||
| 14126 | ||||
| 14127 | case X86::BI__builtin_ia32_kortestcqi: | |||
| 14128 | case X86::BI__builtin_ia32_kortestchi: | |||
| 14129 | case X86::BI__builtin_ia32_kortestcsi: | |||
| 14130 | case X86::BI__builtin_ia32_kortestcdi: { | |||
| 14131 | Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops); | |||
| 14132 | Value *C = llvm::Constant::getAllOnesValue(Ops[0]->getType()); | |||
| 14133 | Value *Cmp = Builder.CreateICmpEQ(Or, C); | |||
| 14134 | return Builder.CreateZExt(Cmp, ConvertType(E->getType())); | |||
| 14135 | } | |||
| 14136 | case X86::BI__builtin_ia32_kortestzqi: | |||
| 14137 | case X86::BI__builtin_ia32_kortestzhi: | |||
| 14138 | case X86::BI__builtin_ia32_kortestzsi: | |||
| 14139 | case X86::BI__builtin_ia32_kortestzdi: { | |||
| 14140 | Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops); | |||
| 14141 | Value *C = llvm::Constant::getNullValue(Ops[0]->getType()); | |||
| 14142 | Value *Cmp = Builder.CreateICmpEQ(Or, C); | |||
| 14143 | return Builder.CreateZExt(Cmp, ConvertType(E->getType())); | |||
| 14144 | } | |||
| 14145 | ||||
| 14146 | case X86::BI__builtin_ia32_ktestcqi: | |||
| 14147 | case X86::BI__builtin_ia32_ktestzqi: | |||
| 14148 | case X86::BI__builtin_ia32_ktestchi: | |||
| 14149 | case X86::BI__builtin_ia32_ktestzhi: | |||
| 14150 | case X86::BI__builtin_ia32_ktestcsi: | |||
| 14151 | case X86::BI__builtin_ia32_ktestzsi: | |||
| 14152 | case X86::BI__builtin_ia32_ktestcdi: | |||
| 14153 | case X86::BI__builtin_ia32_ktestzdi: { | |||
| 14154 | Intrinsic::ID IID; | |||
| 14155 | switch (BuiltinID) { | |||
| 14156 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14156); | |||
| 14157 | case X86::BI__builtin_ia32_ktestcqi: | |||
| 14158 | IID = Intrinsic::x86_avx512_ktestc_b; | |||
| 14159 | break; | |||
| 14160 | case X86::BI__builtin_ia32_ktestzqi: | |||
| 14161 | IID = Intrinsic::x86_avx512_ktestz_b; | |||
| 14162 | break; | |||
| 14163 | case X86::BI__builtin_ia32_ktestchi: | |||
| 14164 | IID = Intrinsic::x86_avx512_ktestc_w; | |||
| 14165 | break; | |||
| 14166 | case X86::BI__builtin_ia32_ktestzhi: | |||
| 14167 | IID = Intrinsic::x86_avx512_ktestz_w; | |||
| 14168 | break; | |||
| 14169 | case X86::BI__builtin_ia32_ktestcsi: | |||
| 14170 | IID = Intrinsic::x86_avx512_ktestc_d; | |||
| 14171 | break; | |||
| 14172 | case X86::BI__builtin_ia32_ktestzsi: | |||
| 14173 | IID = Intrinsic::x86_avx512_ktestz_d; | |||
| 14174 | break; | |||
| 14175 | case X86::BI__builtin_ia32_ktestcdi: | |||
| 14176 | IID = Intrinsic::x86_avx512_ktestc_q; | |||
| 14177 | break; | |||
| 14178 | case X86::BI__builtin_ia32_ktestzdi: | |||
| 14179 | IID = Intrinsic::x86_avx512_ktestz_q; | |||
| 14180 | break; | |||
| 14181 | } | |||
| 14182 | ||||
| 14183 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 14184 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); | |||
| 14185 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); | |||
| 14186 | Function *Intr = CGM.getIntrinsic(IID); | |||
| 14187 | return Builder.CreateCall(Intr, {LHS, RHS}); | |||
| 14188 | } | |||
| 14189 | ||||
| 14190 | case X86::BI__builtin_ia32_kaddqi: | |||
| 14191 | case X86::BI__builtin_ia32_kaddhi: | |||
| 14192 | case X86::BI__builtin_ia32_kaddsi: | |||
| 14193 | case X86::BI__builtin_ia32_kadddi: { | |||
| 14194 | Intrinsic::ID IID; | |||
| 14195 | switch (BuiltinID) { | |||
| 14196 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14196); | |||
| 14197 | case X86::BI__builtin_ia32_kaddqi: | |||
| 14198 | IID = Intrinsic::x86_avx512_kadd_b; | |||
| 14199 | break; | |||
| 14200 | case X86::BI__builtin_ia32_kaddhi: | |||
| 14201 | IID = Intrinsic::x86_avx512_kadd_w; | |||
| 14202 | break; | |||
| 14203 | case X86::BI__builtin_ia32_kaddsi: | |||
| 14204 | IID = Intrinsic::x86_avx512_kadd_d; | |||
| 14205 | break; | |||
| 14206 | case X86::BI__builtin_ia32_kadddi: | |||
| 14207 | IID = Intrinsic::x86_avx512_kadd_q; | |||
| 14208 | break; | |||
| 14209 | } | |||
| 14210 | ||||
| 14211 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 14212 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); | |||
| 14213 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); | |||
| 14214 | Function *Intr = CGM.getIntrinsic(IID); | |||
| 14215 | Value *Res = Builder.CreateCall(Intr, {LHS, RHS}); | |||
| 14216 | return Builder.CreateBitCast(Res, Ops[0]->getType()); | |||
| 14217 | } | |||
| 14218 | case X86::BI__builtin_ia32_kandqi: | |||
| 14219 | case X86::BI__builtin_ia32_kandhi: | |||
| 14220 | case X86::BI__builtin_ia32_kandsi: | |||
| 14221 | case X86::BI__builtin_ia32_kanddi: | |||
| 14222 | return EmitX86MaskLogic(*this, Instruction::And, Ops); | |||
| 14223 | case X86::BI__builtin_ia32_kandnqi: | |||
| 14224 | case X86::BI__builtin_ia32_kandnhi: | |||
| 14225 | case X86::BI__builtin_ia32_kandnsi: | |||
| 14226 | case X86::BI__builtin_ia32_kandndi: | |||
| 14227 | return EmitX86MaskLogic(*this, Instruction::And, Ops, true); | |||
| 14228 | case X86::BI__builtin_ia32_korqi: | |||
| 14229 | case X86::BI__builtin_ia32_korhi: | |||
| 14230 | case X86::BI__builtin_ia32_korsi: | |||
| 14231 | case X86::BI__builtin_ia32_kordi: | |||
| 14232 | return EmitX86MaskLogic(*this, Instruction::Or, Ops); | |||
| 14233 | case X86::BI__builtin_ia32_kxnorqi: | |||
| 14234 | case X86::BI__builtin_ia32_kxnorhi: | |||
| 14235 | case X86::BI__builtin_ia32_kxnorsi: | |||
| 14236 | case X86::BI__builtin_ia32_kxnordi: | |||
| 14237 | return EmitX86MaskLogic(*this, Instruction::Xor, Ops, true); | |||
| 14238 | case X86::BI__builtin_ia32_kxorqi: | |||
| 14239 | case X86::BI__builtin_ia32_kxorhi: | |||
| 14240 | case X86::BI__builtin_ia32_kxorsi: | |||
| 14241 | case X86::BI__builtin_ia32_kxordi: | |||
| 14242 | return EmitX86MaskLogic(*this, Instruction::Xor, Ops); | |||
| 14243 | case X86::BI__builtin_ia32_knotqi: | |||
| 14244 | case X86::BI__builtin_ia32_knothi: | |||
| 14245 | case X86::BI__builtin_ia32_knotsi: | |||
| 14246 | case X86::BI__builtin_ia32_knotdi: { | |||
| 14247 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 14248 | Value *Res = getMaskVecValue(*this, Ops[0], NumElts); | |||
| 14249 | return Builder.CreateBitCast(Builder.CreateNot(Res), | |||
| 14250 | Ops[0]->getType()); | |||
| 14251 | } | |||
| 14252 | case X86::BI__builtin_ia32_kmovb: | |||
| 14253 | case X86::BI__builtin_ia32_kmovw: | |||
| 14254 | case X86::BI__builtin_ia32_kmovd: | |||
| 14255 | case X86::BI__builtin_ia32_kmovq: { | |||
| 14256 | // Bitcast to vXi1 type and then back to integer. This gets the mask | |||
| 14257 | // register type into the IR, but might be optimized out depending on | |||
| 14258 | // what's around it. | |||
| 14259 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 14260 | Value *Res = getMaskVecValue(*this, Ops[0], NumElts); | |||
| 14261 | return Builder.CreateBitCast(Res, Ops[0]->getType()); | |||
| 14262 | } | |||
| 14263 | ||||
| 14264 | case X86::BI__builtin_ia32_kunpckdi: | |||
| 14265 | case X86::BI__builtin_ia32_kunpcksi: | |||
| 14266 | case X86::BI__builtin_ia32_kunpckhi: { | |||
| 14267 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); | |||
| 14268 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); | |||
| 14269 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); | |||
| 14270 | int Indices[64]; | |||
| 14271 | for (unsigned i = 0; i != NumElts; ++i) | |||
| 14272 | Indices[i] = i; | |||
| 14273 | ||||
| 14274 | // First extract half of each vector. This gives better codegen than | |||
| 14275 | // doing it in a single shuffle. | |||
| 14276 | LHS = Builder.CreateShuffleVector(LHS, LHS, | |||
| 14277 | makeArrayRef(Indices, NumElts / 2)); | |||
| 14278 | RHS = Builder.CreateShuffleVector(RHS, RHS, | |||
| 14279 | makeArrayRef(Indices, NumElts / 2)); | |||
| 14280 | // Concat the vectors. | |||
| 14281 | // NOTE: Operands are swapped to match the intrinsic definition. | |||
| 14282 | Value *Res = Builder.CreateShuffleVector(RHS, LHS, | |||
| 14283 | makeArrayRef(Indices, NumElts)); | |||
| 14284 | return Builder.CreateBitCast(Res, Ops[0]->getType()); | |||
| 14285 | } | |||
| 14286 | ||||
| 14287 | case X86::BI__builtin_ia32_vplzcntd_128: | |||
| 14288 | case X86::BI__builtin_ia32_vplzcntd_256: | |||
| 14289 | case X86::BI__builtin_ia32_vplzcntd_512: | |||
| 14290 | case X86::BI__builtin_ia32_vplzcntq_128: | |||
| 14291 | case X86::BI__builtin_ia32_vplzcntq_256: | |||
| 14292 | case X86::BI__builtin_ia32_vplzcntq_512: { | |||
| 14293 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType()); | |||
| 14294 | return Builder.CreateCall(F, {Ops[0],Builder.getInt1(false)}); | |||
| 14295 | } | |||
| 14296 | case X86::BI__builtin_ia32_sqrtss: | |||
| 14297 | case X86::BI__builtin_ia32_sqrtsd: { | |||
| 14298 | Value *A = Builder.CreateExtractElement(Ops[0], (uint64_t)0); | |||
| 14299 | Function *F; | |||
| 14300 | if (Builder.getIsFPConstrained()) { | |||
| 14301 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 14302 | F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt, | |||
| 14303 | A->getType()); | |||
| 14304 | A = Builder.CreateConstrainedFPCall(F, {A}); | |||
| 14305 | } else { | |||
| 14306 | F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType()); | |||
| 14307 | A = Builder.CreateCall(F, {A}); | |||
| 14308 | } | |||
| 14309 | return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0); | |||
| 14310 | } | |||
| 14311 | case X86::BI__builtin_ia32_sqrtsh_round_mask: | |||
| 14312 | case X86::BI__builtin_ia32_sqrtsd_round_mask: | |||
| 14313 | case X86::BI__builtin_ia32_sqrtss_round_mask: { | |||
| 14314 | unsigned CC = cast<llvm::ConstantInt>(Ops[4])->getZExtValue(); | |||
| 14315 | // Support only if the rounding mode is 4 (AKA CUR_DIRECTION), | |||
| 14316 | // otherwise keep the intrinsic. | |||
| 14317 | if (CC != 4) { | |||
| 14318 | Intrinsic::ID IID; | |||
| 14319 | ||||
| 14320 | switch (BuiltinID) { | |||
| 14321 | default: | |||
| 14322 | llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14322); | |||
| 14323 | case X86::BI__builtin_ia32_sqrtsh_round_mask: | |||
| 14324 | IID = Intrinsic::x86_avx512fp16_mask_sqrt_sh; | |||
| 14325 | break; | |||
| 14326 | case X86::BI__builtin_ia32_sqrtsd_round_mask: | |||
| 14327 | IID = Intrinsic::x86_avx512_mask_sqrt_sd; | |||
| 14328 | break; | |||
| 14329 | case X86::BI__builtin_ia32_sqrtss_round_mask: | |||
| 14330 | IID = Intrinsic::x86_avx512_mask_sqrt_ss; | |||
| 14331 | break; | |||
| 14332 | } | |||
| 14333 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
| 14334 | } | |||
| 14335 | Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0); | |||
| 14336 | Function *F; | |||
| 14337 | if (Builder.getIsFPConstrained()) { | |||
| 14338 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 14339 | F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt, | |||
| 14340 | A->getType()); | |||
| 14341 | A = Builder.CreateConstrainedFPCall(F, A); | |||
| 14342 | } else { | |||
| 14343 | F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType()); | |||
| 14344 | A = Builder.CreateCall(F, A); | |||
| 14345 | } | |||
| 14346 | Value *Src = Builder.CreateExtractElement(Ops[2], (uint64_t)0); | |||
| 14347 | A = EmitX86ScalarSelect(*this, Ops[3], A, Src); | |||
| 14348 | return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0); | |||
| 14349 | } | |||
| 14350 | case X86::BI__builtin_ia32_sqrtpd256: | |||
| 14351 | case X86::BI__builtin_ia32_sqrtpd: | |||
| 14352 | case X86::BI__builtin_ia32_sqrtps256: | |||
| 14353 | case X86::BI__builtin_ia32_sqrtps: | |||
| 14354 | case X86::BI__builtin_ia32_sqrtph256: | |||
| 14355 | case X86::BI__builtin_ia32_sqrtph: | |||
| 14356 | case X86::BI__builtin_ia32_sqrtph512: | |||
| 14357 | case X86::BI__builtin_ia32_sqrtps512: | |||
| 14358 | case X86::BI__builtin_ia32_sqrtpd512: { | |||
| 14359 | if (Ops.size() == 2) { | |||
| 14360 | unsigned CC = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); | |||
| 14361 | // Support only if the rounding mode is 4 (AKA CUR_DIRECTION), | |||
| 14362 | // otherwise keep the intrinsic. | |||
| 14363 | if (CC != 4) { | |||
| 14364 | Intrinsic::ID IID; | |||
| 14365 | ||||
| 14366 | switch (BuiltinID) { | |||
| 14367 | default: | |||
| 14368 | llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14368); | |||
| 14369 | case X86::BI__builtin_ia32_sqrtph512: | |||
| 14370 | IID = Intrinsic::x86_avx512fp16_sqrt_ph_512; | |||
| 14371 | break; | |||
| 14372 | case X86::BI__builtin_ia32_sqrtps512: | |||
| 14373 | IID = Intrinsic::x86_avx512_sqrt_ps_512; | |||
| 14374 | break; | |||
| 14375 | case X86::BI__builtin_ia32_sqrtpd512: | |||
| 14376 | IID = Intrinsic::x86_avx512_sqrt_pd_512; | |||
| 14377 | break; | |||
| 14378 | } | |||
| 14379 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
| 14380 | } | |||
| 14381 | } | |||
| 14382 | if (Builder.getIsFPConstrained()) { | |||
| 14383 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 14384 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt, | |||
| 14385 | Ops[0]->getType()); | |||
| 14386 | return Builder.CreateConstrainedFPCall(F, Ops[0]); | |||
| 14387 | } else { | |||
| 14388 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, Ops[0]->getType()); | |||
| 14389 | return Builder.CreateCall(F, Ops[0]); | |||
| 14390 | } | |||
| 14391 | } | |||
| 14392 | ||||
| 14393 | case X86::BI__builtin_ia32_pmuludq128: | |||
| 14394 | case X86::BI__builtin_ia32_pmuludq256: | |||
| 14395 | case X86::BI__builtin_ia32_pmuludq512: | |||
| 14396 | return EmitX86Muldq(*this, /*IsSigned*/false, Ops); | |||
| 14397 | ||||
| 14398 | case X86::BI__builtin_ia32_pmuldq128: | |||
| 14399 | case X86::BI__builtin_ia32_pmuldq256: | |||
| 14400 | case X86::BI__builtin_ia32_pmuldq512: | |||
| 14401 | return EmitX86Muldq(*this, /*IsSigned*/true, Ops); | |||
| 14402 | ||||
| 14403 | case X86::BI__builtin_ia32_pternlogd512_mask: | |||
| 14404 | case X86::BI__builtin_ia32_pternlogq512_mask: | |||
| 14405 | case X86::BI__builtin_ia32_pternlogd128_mask: | |||
| 14406 | case X86::BI__builtin_ia32_pternlogd256_mask: | |||
| 14407 | case X86::BI__builtin_ia32_pternlogq128_mask: | |||
| 14408 | case X86::BI__builtin_ia32_pternlogq256_mask: | |||
| 14409 | return EmitX86Ternlog(*this, /*ZeroMask*/false, Ops); | |||
| 14410 | ||||
| 14411 | case X86::BI__builtin_ia32_pternlogd512_maskz: | |||
| 14412 | case X86::BI__builtin_ia32_pternlogq512_maskz: | |||
| 14413 | case X86::BI__builtin_ia32_pternlogd128_maskz: | |||
| 14414 | case X86::BI__builtin_ia32_pternlogd256_maskz: | |||
| 14415 | case X86::BI__builtin_ia32_pternlogq128_maskz: | |||
| 14416 | case X86::BI__builtin_ia32_pternlogq256_maskz: | |||
| 14417 | return EmitX86Ternlog(*this, /*ZeroMask*/true, Ops); | |||
| 14418 | ||||
| 14419 | case X86::BI__builtin_ia32_vpshldd128: | |||
| 14420 | case X86::BI__builtin_ia32_vpshldd256: | |||
| 14421 | case X86::BI__builtin_ia32_vpshldd512: | |||
| 14422 | case X86::BI__builtin_ia32_vpshldq128: | |||
| 14423 | case X86::BI__builtin_ia32_vpshldq256: | |||
| 14424 | case X86::BI__builtin_ia32_vpshldq512: | |||
| 14425 | case X86::BI__builtin_ia32_vpshldw128: | |||
| 14426 | case X86::BI__builtin_ia32_vpshldw256: | |||
| 14427 | case X86::BI__builtin_ia32_vpshldw512: | |||
| 14428 | return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); | |||
| 14429 | ||||
| 14430 | case X86::BI__builtin_ia32_vpshrdd128: | |||
| 14431 | case X86::BI__builtin_ia32_vpshrdd256: | |||
| 14432 | case X86::BI__builtin_ia32_vpshrdd512: | |||
| 14433 | case X86::BI__builtin_ia32_vpshrdq128: | |||
| 14434 | case X86::BI__builtin_ia32_vpshrdq256: | |||
| 14435 | case X86::BI__builtin_ia32_vpshrdq512: | |||
| 14436 | case X86::BI__builtin_ia32_vpshrdw128: | |||
| 14437 | case X86::BI__builtin_ia32_vpshrdw256: | |||
| 14438 | case X86::BI__builtin_ia32_vpshrdw512: | |||
| 14439 | // Ops 0 and 1 are swapped. | |||
| 14440 | return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); | |||
| 14441 | ||||
| 14442 | case X86::BI__builtin_ia32_vpshldvd128: | |||
| 14443 | case X86::BI__builtin_ia32_vpshldvd256: | |||
| 14444 | case X86::BI__builtin_ia32_vpshldvd512: | |||
| 14445 | case X86::BI__builtin_ia32_vpshldvq128: | |||
| 14446 | case X86::BI__builtin_ia32_vpshldvq256: | |||
| 14447 | case X86::BI__builtin_ia32_vpshldvq512: | |||
| 14448 | case X86::BI__builtin_ia32_vpshldvw128: | |||
| 14449 | case X86::BI__builtin_ia32_vpshldvw256: | |||
| 14450 | case X86::BI__builtin_ia32_vpshldvw512: | |||
| 14451 | return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); | |||
| 14452 | ||||
| 14453 | case X86::BI__builtin_ia32_vpshrdvd128: | |||
| 14454 | case X86::BI__builtin_ia32_vpshrdvd256: | |||
| 14455 | case X86::BI__builtin_ia32_vpshrdvd512: | |||
| 14456 | case X86::BI__builtin_ia32_vpshrdvq128: | |||
| 14457 | case X86::BI__builtin_ia32_vpshrdvq256: | |||
| 14458 | case X86::BI__builtin_ia32_vpshrdvq512: | |||
| 14459 | case X86::BI__builtin_ia32_vpshrdvw128: | |||
| 14460 | case X86::BI__builtin_ia32_vpshrdvw256: | |||
| 14461 | case X86::BI__builtin_ia32_vpshrdvw512: | |||
| 14462 | // Ops 0 and 1 are swapped. | |||
| 14463 | return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); | |||
| 14464 | ||||
| 14465 | // Reductions | |||
| 14466 | case X86::BI__builtin_ia32_reduce_add_d512: | |||
| 14467 | case X86::BI__builtin_ia32_reduce_add_q512: { | |||
| 14468 | Function *F = | |||
| 14469 | CGM.getIntrinsic(Intrinsic::vector_reduce_add, Ops[0]->getType()); | |||
| 14470 | return Builder.CreateCall(F, {Ops[0]}); | |||
| 14471 | } | |||
| 14472 | case X86::BI__builtin_ia32_reduce_fadd_pd512: | |||
| 14473 | case X86::BI__builtin_ia32_reduce_fadd_ps512: | |||
| 14474 | case X86::BI__builtin_ia32_reduce_fadd_ph512: | |||
| 14475 | case X86::BI__builtin_ia32_reduce_fadd_ph256: | |||
| 14476 | case X86::BI__builtin_ia32_reduce_fadd_ph128: { | |||
| 14477 | Function *F = | |||
| 14478 | CGM.getIntrinsic(Intrinsic::vector_reduce_fadd, Ops[1]->getType()); | |||
| 14479 | Builder.getFastMathFlags().setAllowReassoc(); | |||
| 14480 | return Builder.CreateCall(F, {Ops[0], Ops[1]}); | |||
| 14481 | } | |||
| 14482 | case X86::BI__builtin_ia32_reduce_fmul_pd512: | |||
| 14483 | case X86::BI__builtin_ia32_reduce_fmul_ps512: | |||
| 14484 | case X86::BI__builtin_ia32_reduce_fmul_ph512: | |||
| 14485 | case X86::BI__builtin_ia32_reduce_fmul_ph256: | |||
| 14486 | case X86::BI__builtin_ia32_reduce_fmul_ph128: { | |||
| 14487 | Function *F = | |||
| 14488 | CGM.getIntrinsic(Intrinsic::vector_reduce_fmul, Ops[1]->getType()); | |||
| 14489 | Builder.getFastMathFlags().setAllowReassoc(); | |||
| 14490 | return Builder.CreateCall(F, {Ops[0], Ops[1]}); | |||
| 14491 | } | |||
| 14492 | case X86::BI__builtin_ia32_reduce_fmax_pd512: | |||
| 14493 | case X86::BI__builtin_ia32_reduce_fmax_ps512: | |||
| 14494 | case X86::BI__builtin_ia32_reduce_fmax_ph512: | |||
| 14495 | case X86::BI__builtin_ia32_reduce_fmax_ph256: | |||
| 14496 | case X86::BI__builtin_ia32_reduce_fmax_ph128: { | |||
| 14497 | Function *F = | |||
| 14498 | CGM.getIntrinsic(Intrinsic::vector_reduce_fmax, Ops[0]->getType()); | |||
| 14499 | Builder.getFastMathFlags().setNoNaNs(); | |||
| 14500 | return Builder.CreateCall(F, {Ops[0]}); | |||
| 14501 | } | |||
| 14502 | case X86::BI__builtin_ia32_reduce_fmin_pd512: | |||
| 14503 | case X86::BI__builtin_ia32_reduce_fmin_ps512: | |||
| 14504 | case X86::BI__builtin_ia32_reduce_fmin_ph512: | |||
| 14505 | case X86::BI__builtin_ia32_reduce_fmin_ph256: | |||
| 14506 | case X86::BI__builtin_ia32_reduce_fmin_ph128: { | |||
| 14507 | Function *F = | |||
| 14508 | CGM.getIntrinsic(Intrinsic::vector_reduce_fmin, Ops[0]->getType()); | |||
| 14509 | Builder.getFastMathFlags().setNoNaNs(); | |||
| 14510 | return Builder.CreateCall(F, {Ops[0]}); | |||
| 14511 | } | |||
| 14512 | case X86::BI__builtin_ia32_reduce_mul_d512: | |||
| 14513 | case X86::BI__builtin_ia32_reduce_mul_q512: { | |||
| 14514 | Function *F = | |||
| 14515 | CGM.getIntrinsic(Intrinsic::vector_reduce_mul, Ops[0]->getType()); | |||
| 14516 | return Builder.CreateCall(F, {Ops[0]}); | |||
| 14517 | } | |||
| 14518 | ||||
| 14519 | // 3DNow! | |||
| 14520 | case X86::BI__builtin_ia32_pswapdsf: | |||
| 14521 | case X86::BI__builtin_ia32_pswapdsi: { | |||
| 14522 | llvm::Type *MMXTy = llvm::Type::getX86_MMXTy(getLLVMContext()); | |||
| 14523 | Ops[0] = Builder.CreateBitCast(Ops[0], MMXTy, "cast"); | |||
| 14524 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_3dnowa_pswapd); | |||
| 14525 | return Builder.CreateCall(F, Ops, "pswapd"); | |||
| 14526 | } | |||
| 14527 | case X86::BI__builtin_ia32_rdrand16_step: | |||
| 14528 | case X86::BI__builtin_ia32_rdrand32_step: | |||
| 14529 | case X86::BI__builtin_ia32_rdrand64_step: | |||
| 14530 | case X86::BI__builtin_ia32_rdseed16_step: | |||
| 14531 | case X86::BI__builtin_ia32_rdseed32_step: | |||
| 14532 | case X86::BI__builtin_ia32_rdseed64_step: { | |||
| 14533 | Intrinsic::ID ID; | |||
| 14534 | switch (BuiltinID) { | |||
| 14535 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14535); | |||
| 14536 | case X86::BI__builtin_ia32_rdrand16_step: | |||
| 14537 | ID = Intrinsic::x86_rdrand_16; | |||
| 14538 | break; | |||
| 14539 | case X86::BI__builtin_ia32_rdrand32_step: | |||
| 14540 | ID = Intrinsic::x86_rdrand_32; | |||
| 14541 | break; | |||
| 14542 | case X86::BI__builtin_ia32_rdrand64_step: | |||
| 14543 | ID = Intrinsic::x86_rdrand_64; | |||
| 14544 | break; | |||
| 14545 | case X86::BI__builtin_ia32_rdseed16_step: | |||
| 14546 | ID = Intrinsic::x86_rdseed_16; | |||
| 14547 | break; | |||
| 14548 | case X86::BI__builtin_ia32_rdseed32_step: | |||
| 14549 | ID = Intrinsic::x86_rdseed_32; | |||
| 14550 | break; | |||
| 14551 | case X86::BI__builtin_ia32_rdseed64_step: | |||
| 14552 | ID = Intrinsic::x86_rdseed_64; | |||
| 14553 | break; | |||
| 14554 | } | |||
| 14555 | ||||
| 14556 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID)); | |||
| 14557 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 0), | |||
| 14558 | Ops[0]); | |||
| 14559 | return Builder.CreateExtractValue(Call, 1); | |||
| 14560 | } | |||
| 14561 | case X86::BI__builtin_ia32_addcarryx_u32: | |||
| 14562 | case X86::BI__builtin_ia32_addcarryx_u64: | |||
| 14563 | case X86::BI__builtin_ia32_subborrow_u32: | |||
| 14564 | case X86::BI__builtin_ia32_subborrow_u64: { | |||
| 14565 | Intrinsic::ID IID; | |||
| 14566 | switch (BuiltinID) { | |||
| 14567 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14567); | |||
| 14568 | case X86::BI__builtin_ia32_addcarryx_u32: | |||
| 14569 | IID = Intrinsic::x86_addcarry_32; | |||
| 14570 | break; | |||
| 14571 | case X86::BI__builtin_ia32_addcarryx_u64: | |||
| 14572 | IID = Intrinsic::x86_addcarry_64; | |||
| 14573 | break; | |||
| 14574 | case X86::BI__builtin_ia32_subborrow_u32: | |||
| 14575 | IID = Intrinsic::x86_subborrow_32; | |||
| 14576 | break; | |||
| 14577 | case X86::BI__builtin_ia32_subborrow_u64: | |||
| 14578 | IID = Intrinsic::x86_subborrow_64; | |||
| 14579 | break; | |||
| 14580 | } | |||
| 14581 | ||||
| 14582 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), | |||
| 14583 | { Ops[0], Ops[1], Ops[2] }); | |||
| 14584 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1), | |||
| 14585 | Ops[3]); | |||
| 14586 | return Builder.CreateExtractValue(Call, 0); | |||
| 14587 | } | |||
| 14588 | ||||
| 14589 | case X86::BI__builtin_ia32_fpclassps128_mask: | |||
| 14590 | case X86::BI__builtin_ia32_fpclassps256_mask: | |||
| 14591 | case X86::BI__builtin_ia32_fpclassps512_mask: | |||
| 14592 | case X86::BI__builtin_ia32_fpclassph128_mask: | |||
| 14593 | case X86::BI__builtin_ia32_fpclassph256_mask: | |||
| 14594 | case X86::BI__builtin_ia32_fpclassph512_mask: | |||
| 14595 | case X86::BI__builtin_ia32_fpclasspd128_mask: | |||
| 14596 | case X86::BI__builtin_ia32_fpclasspd256_mask: | |||
| 14597 | case X86::BI__builtin_ia32_fpclasspd512_mask: { | |||
| 14598 | unsigned NumElts = | |||
| 14599 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 14600 | Value *MaskIn = Ops[2]; | |||
| 14601 | Ops.erase(&Ops[2]); | |||
| 14602 | ||||
| 14603 | Intrinsic::ID ID; | |||
| 14604 | switch (BuiltinID) { | |||
| 14605 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14605); | |||
| 14606 | case X86::BI__builtin_ia32_fpclassph128_mask: | |||
| 14607 | ID = Intrinsic::x86_avx512fp16_fpclass_ph_128; | |||
| 14608 | break; | |||
| 14609 | case X86::BI__builtin_ia32_fpclassph256_mask: | |||
| 14610 | ID = Intrinsic::x86_avx512fp16_fpclass_ph_256; | |||
| 14611 | break; | |||
| 14612 | case X86::BI__builtin_ia32_fpclassph512_mask: | |||
| 14613 | ID = Intrinsic::x86_avx512fp16_fpclass_ph_512; | |||
| 14614 | break; | |||
| 14615 | case X86::BI__builtin_ia32_fpclassps128_mask: | |||
| 14616 | ID = Intrinsic::x86_avx512_fpclass_ps_128; | |||
| 14617 | break; | |||
| 14618 | case X86::BI__builtin_ia32_fpclassps256_mask: | |||
| 14619 | ID = Intrinsic::x86_avx512_fpclass_ps_256; | |||
| 14620 | break; | |||
| 14621 | case X86::BI__builtin_ia32_fpclassps512_mask: | |||
| 14622 | ID = Intrinsic::x86_avx512_fpclass_ps_512; | |||
| 14623 | break; | |||
| 14624 | case X86::BI__builtin_ia32_fpclasspd128_mask: | |||
| 14625 | ID = Intrinsic::x86_avx512_fpclass_pd_128; | |||
| 14626 | break; | |||
| 14627 | case X86::BI__builtin_ia32_fpclasspd256_mask: | |||
| 14628 | ID = Intrinsic::x86_avx512_fpclass_pd_256; | |||
| 14629 | break; | |||
| 14630 | case X86::BI__builtin_ia32_fpclasspd512_mask: | |||
| 14631 | ID = Intrinsic::x86_avx512_fpclass_pd_512; | |||
| 14632 | break; | |||
| 14633 | } | |||
| 14634 | ||||
| 14635 | Value *Fpclass = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
| 14636 | return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn); | |||
| 14637 | } | |||
| 14638 | ||||
| 14639 | case X86::BI__builtin_ia32_vp2intersect_q_512: | |||
| 14640 | case X86::BI__builtin_ia32_vp2intersect_q_256: | |||
| 14641 | case X86::BI__builtin_ia32_vp2intersect_q_128: | |||
| 14642 | case X86::BI__builtin_ia32_vp2intersect_d_512: | |||
| 14643 | case X86::BI__builtin_ia32_vp2intersect_d_256: | |||
| 14644 | case X86::BI__builtin_ia32_vp2intersect_d_128: { | |||
| 14645 | unsigned NumElts = | |||
| 14646 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 14647 | Intrinsic::ID ID; | |||
| 14648 | ||||
| 14649 | switch (BuiltinID) { | |||
| 14650 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14650); | |||
| 14651 | case X86::BI__builtin_ia32_vp2intersect_q_512: | |||
| 14652 | ID = Intrinsic::x86_avx512_vp2intersect_q_512; | |||
| 14653 | break; | |||
| 14654 | case X86::BI__builtin_ia32_vp2intersect_q_256: | |||
| 14655 | ID = Intrinsic::x86_avx512_vp2intersect_q_256; | |||
| 14656 | break; | |||
| 14657 | case X86::BI__builtin_ia32_vp2intersect_q_128: | |||
| 14658 | ID = Intrinsic::x86_avx512_vp2intersect_q_128; | |||
| 14659 | break; | |||
| 14660 | case X86::BI__builtin_ia32_vp2intersect_d_512: | |||
| 14661 | ID = Intrinsic::x86_avx512_vp2intersect_d_512; | |||
| 14662 | break; | |||
| 14663 | case X86::BI__builtin_ia32_vp2intersect_d_256: | |||
| 14664 | ID = Intrinsic::x86_avx512_vp2intersect_d_256; | |||
| 14665 | break; | |||
| 14666 | case X86::BI__builtin_ia32_vp2intersect_d_128: | |||
| 14667 | ID = Intrinsic::x86_avx512_vp2intersect_d_128; | |||
| 14668 | break; | |||
| 14669 | } | |||
| 14670 | ||||
| 14671 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID), {Ops[0], Ops[1]}); | |||
| 14672 | Value *Result = Builder.CreateExtractValue(Call, 0); | |||
| 14673 | Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr); | |||
| 14674 | Builder.CreateDefaultAlignedStore(Result, Ops[2]); | |||
| 14675 | ||||
| 14676 | Result = Builder.CreateExtractValue(Call, 1); | |||
| 14677 | Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr); | |||
| 14678 | return Builder.CreateDefaultAlignedStore(Result, Ops[3]); | |||
| 14679 | } | |||
| 14680 | ||||
| 14681 | case X86::BI__builtin_ia32_vpmultishiftqb128: | |||
| 14682 | case X86::BI__builtin_ia32_vpmultishiftqb256: | |||
| 14683 | case X86::BI__builtin_ia32_vpmultishiftqb512: { | |||
| 14684 | Intrinsic::ID ID; | |||
| 14685 | switch (BuiltinID) { | |||
| 14686 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14686); | |||
| 14687 | case X86::BI__builtin_ia32_vpmultishiftqb128: | |||
| 14688 | ID = Intrinsic::x86_avx512_pmultishift_qb_128; | |||
| 14689 | break; | |||
| 14690 | case X86::BI__builtin_ia32_vpmultishiftqb256: | |||
| 14691 | ID = Intrinsic::x86_avx512_pmultishift_qb_256; | |||
| 14692 | break; | |||
| 14693 | case X86::BI__builtin_ia32_vpmultishiftqb512: | |||
| 14694 | ID = Intrinsic::x86_avx512_pmultishift_qb_512; | |||
| 14695 | break; | |||
| 14696 | } | |||
| 14697 | ||||
| 14698 | return Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
| 14699 | } | |||
| 14700 | ||||
| 14701 | case X86::BI__builtin_ia32_vpshufbitqmb128_mask: | |||
| 14702 | case X86::BI__builtin_ia32_vpshufbitqmb256_mask: | |||
| 14703 | case X86::BI__builtin_ia32_vpshufbitqmb512_mask: { | |||
| 14704 | unsigned NumElts = | |||
| 14705 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 14706 | Value *MaskIn = Ops[2]; | |||
| 14707 | Ops.erase(&Ops[2]); | |||
| 14708 | ||||
| 14709 | Intrinsic::ID ID; | |||
| 14710 | switch (BuiltinID) { | |||
| 14711 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14711); | |||
| 14712 | case X86::BI__builtin_ia32_vpshufbitqmb128_mask: | |||
| 14713 | ID = Intrinsic::x86_avx512_vpshufbitqmb_128; | |||
| 14714 | break; | |||
| 14715 | case X86::BI__builtin_ia32_vpshufbitqmb256_mask: | |||
| 14716 | ID = Intrinsic::x86_avx512_vpshufbitqmb_256; | |||
| 14717 | break; | |||
| 14718 | case X86::BI__builtin_ia32_vpshufbitqmb512_mask: | |||
| 14719 | ID = Intrinsic::x86_avx512_vpshufbitqmb_512; | |||
| 14720 | break; | |||
| 14721 | } | |||
| 14722 | ||||
| 14723 | Value *Shufbit = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
| 14724 | return EmitX86MaskedCompareResult(*this, Shufbit, NumElts, MaskIn); | |||
| 14725 | } | |||
| 14726 | ||||
| 14727 | // packed comparison intrinsics | |||
| 14728 | case X86::BI__builtin_ia32_cmpeqps: | |||
| 14729 | case X86::BI__builtin_ia32_cmpeqpd: | |||
| 14730 | return getVectorFCmpIR(CmpInst::FCMP_OEQ, /*IsSignaling*/false); | |||
| 14731 | case X86::BI__builtin_ia32_cmpltps: | |||
| 14732 | case X86::BI__builtin_ia32_cmpltpd: | |||
| 14733 | return getVectorFCmpIR(CmpInst::FCMP_OLT, /*IsSignaling*/true); | |||
| 14734 | case X86::BI__builtin_ia32_cmpleps: | |||
| 14735 | case X86::BI__builtin_ia32_cmplepd: | |||
| 14736 | return getVectorFCmpIR(CmpInst::FCMP_OLE, /*IsSignaling*/true); | |||
| 14737 | case X86::BI__builtin_ia32_cmpunordps: | |||
| 14738 | case X86::BI__builtin_ia32_cmpunordpd: | |||
| 14739 | return getVectorFCmpIR(CmpInst::FCMP_UNO, /*IsSignaling*/false); | |||
| 14740 | case X86::BI__builtin_ia32_cmpneqps: | |||
| 14741 | case X86::BI__builtin_ia32_cmpneqpd: | |||
| 14742 | return getVectorFCmpIR(CmpInst::FCMP_UNE, /*IsSignaling*/false); | |||
| 14743 | case X86::BI__builtin_ia32_cmpnltps: | |||
| 14744 | case X86::BI__builtin_ia32_cmpnltpd: | |||
| 14745 | return getVectorFCmpIR(CmpInst::FCMP_UGE, /*IsSignaling*/true); | |||
| 14746 | case X86::BI__builtin_ia32_cmpnleps: | |||
| 14747 | case X86::BI__builtin_ia32_cmpnlepd: | |||
| 14748 | return getVectorFCmpIR(CmpInst::FCMP_UGT, /*IsSignaling*/true); | |||
| 14749 | case X86::BI__builtin_ia32_cmpordps: | |||
| 14750 | case X86::BI__builtin_ia32_cmpordpd: | |||
| 14751 | return getVectorFCmpIR(CmpInst::FCMP_ORD, /*IsSignaling*/false); | |||
| 14752 | case X86::BI__builtin_ia32_cmpph128_mask: | |||
| 14753 | case X86::BI__builtin_ia32_cmpph256_mask: | |||
| 14754 | case X86::BI__builtin_ia32_cmpph512_mask: | |||
| 14755 | case X86::BI__builtin_ia32_cmpps128_mask: | |||
| 14756 | case X86::BI__builtin_ia32_cmpps256_mask: | |||
| 14757 | case X86::BI__builtin_ia32_cmpps512_mask: | |||
| 14758 | case X86::BI__builtin_ia32_cmppd128_mask: | |||
| 14759 | case X86::BI__builtin_ia32_cmppd256_mask: | |||
| 14760 | case X86::BI__builtin_ia32_cmppd512_mask: | |||
| 14761 | IsMaskFCmp = true; | |||
| 14762 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 14763 | case X86::BI__builtin_ia32_cmpps: | |||
| 14764 | case X86::BI__builtin_ia32_cmpps256: | |||
| 14765 | case X86::BI__builtin_ia32_cmppd: | |||
| 14766 | case X86::BI__builtin_ia32_cmppd256: { | |||
| 14767 | // Lowering vector comparisons to fcmp instructions, while | |||
| 14768 | // ignoring signalling behaviour requested | |||
| 14769 | // ignoring rounding mode requested | |||
| 14770 | // This is only possible if fp-model is not strict and FENV_ACCESS is off. | |||
| 14771 | ||||
| 14772 | // The third argument is the comparison condition, and integer in the | |||
| 14773 | // range [0, 31] | |||
| 14774 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x1f; | |||
| 14775 | ||||
| 14776 | // Lowering to IR fcmp instruction. | |||
| 14777 | // Ignoring requested signaling behaviour, | |||
| 14778 | // e.g. both _CMP_GT_OS & _CMP_GT_OQ are translated to FCMP_OGT. | |||
| 14779 | FCmpInst::Predicate Pred; | |||
| 14780 | bool IsSignaling; | |||
| 14781 | // Predicates for 16-31 repeat the 0-15 predicates. Only the signalling | |||
| 14782 | // behavior is inverted. We'll handle that after the switch. | |||
| 14783 | switch (CC & 0xf) { | |||
| 14784 | case 0x00: Pred = FCmpInst::FCMP_OEQ; IsSignaling = false; break; | |||
| 14785 | case 0x01: Pred = FCmpInst::FCMP_OLT; IsSignaling = true; break; | |||
| 14786 | case 0x02: Pred = FCmpInst::FCMP_OLE; IsSignaling = true; break; | |||
| 14787 | case 0x03: Pred = FCmpInst::FCMP_UNO; IsSignaling = false; break; | |||
| 14788 | case 0x04: Pred = FCmpInst::FCMP_UNE; IsSignaling = false; break; | |||
| 14789 | case 0x05: Pred = FCmpInst::FCMP_UGE; IsSignaling = true; break; | |||
| 14790 | case 0x06: Pred = FCmpInst::FCMP_UGT; IsSignaling = true; break; | |||
| 14791 | case 0x07: Pred = FCmpInst::FCMP_ORD; IsSignaling = false; break; | |||
| 14792 | case 0x08: Pred = FCmpInst::FCMP_UEQ; IsSignaling = false; break; | |||
| 14793 | case 0x09: Pred = FCmpInst::FCMP_ULT; IsSignaling = true; break; | |||
| 14794 | case 0x0a: Pred = FCmpInst::FCMP_ULE; IsSignaling = true; break; | |||
| 14795 | case 0x0b: Pred = FCmpInst::FCMP_FALSE; IsSignaling = false; break; | |||
| 14796 | case 0x0c: Pred = FCmpInst::FCMP_ONE; IsSignaling = false; break; | |||
| 14797 | case 0x0d: Pred = FCmpInst::FCMP_OGE; IsSignaling = true; break; | |||
| 14798 | case 0x0e: Pred = FCmpInst::FCMP_OGT; IsSignaling = true; break; | |||
| 14799 | case 0x0f: Pred = FCmpInst::FCMP_TRUE; IsSignaling = false; break; | |||
| 14800 | default: llvm_unreachable("Unhandled CC")::llvm::llvm_unreachable_internal("Unhandled CC", "clang/lib/CodeGen/CGBuiltin.cpp" , 14800); | |||
| 14801 | } | |||
| 14802 | ||||
| 14803 | // Invert the signalling behavior for 16-31. | |||
| 14804 | if (CC & 0x10) | |||
| 14805 | IsSignaling = !IsSignaling; | |||
| 14806 | ||||
| 14807 | // If the predicate is true or false and we're using constrained intrinsics, | |||
| 14808 | // we don't have a compare intrinsic we can use. Just use the legacy X86 | |||
| 14809 | // specific intrinsic. | |||
| 14810 | // If the intrinsic is mask enabled and we're using constrained intrinsics, | |||
| 14811 | // use the legacy X86 specific intrinsic. | |||
| 14812 | if (Builder.getIsFPConstrained() && | |||
| 14813 | (Pred == FCmpInst::FCMP_TRUE || Pred == FCmpInst::FCMP_FALSE || | |||
| 14814 | IsMaskFCmp)) { | |||
| 14815 | ||||
| 14816 | Intrinsic::ID IID; | |||
| 14817 | switch (BuiltinID) { | |||
| 14818 | default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 14818); | |||
| 14819 | case X86::BI__builtin_ia32_cmpps: | |||
| 14820 | IID = Intrinsic::x86_sse_cmp_ps; | |||
| 14821 | break; | |||
| 14822 | case X86::BI__builtin_ia32_cmpps256: | |||
| 14823 | IID = Intrinsic::x86_avx_cmp_ps_256; | |||
| 14824 | break; | |||
| 14825 | case X86::BI__builtin_ia32_cmppd: | |||
| 14826 | IID = Intrinsic::x86_sse2_cmp_pd; | |||
| 14827 | break; | |||
| 14828 | case X86::BI__builtin_ia32_cmppd256: | |||
| 14829 | IID = Intrinsic::x86_avx_cmp_pd_256; | |||
| 14830 | break; | |||
| 14831 | case X86::BI__builtin_ia32_cmpps512_mask: | |||
| 14832 | IID = Intrinsic::x86_avx512_mask_cmp_ps_512; | |||
| 14833 | break; | |||
| 14834 | case X86::BI__builtin_ia32_cmppd512_mask: | |||
| 14835 | IID = Intrinsic::x86_avx512_mask_cmp_pd_512; | |||
| 14836 | break; | |||
| 14837 | case X86::BI__builtin_ia32_cmpps128_mask: | |||
| 14838 | IID = Intrinsic::x86_avx512_mask_cmp_ps_128; | |||
| 14839 | break; | |||
| 14840 | case X86::BI__builtin_ia32_cmpps256_mask: | |||
| 14841 | IID = Intrinsic::x86_avx512_mask_cmp_ps_256; | |||
| 14842 | break; | |||
| 14843 | case X86::BI__builtin_ia32_cmppd128_mask: | |||
| 14844 | IID = Intrinsic::x86_avx512_mask_cmp_pd_128; | |||
| 14845 | break; | |||
| 14846 | case X86::BI__builtin_ia32_cmppd256_mask: | |||
| 14847 | IID = Intrinsic::x86_avx512_mask_cmp_pd_256; | |||
| 14848 | break; | |||
| 14849 | } | |||
| 14850 | ||||
| 14851 | Function *Intr = CGM.getIntrinsic(IID); | |||
| 14852 | if (IsMaskFCmp) { | |||
| 14853 | unsigned NumElts = | |||
| 14854 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 14855 | Ops[3] = getMaskVecValue(*this, Ops[3], NumElts); | |||
| 14856 | Value *Cmp = Builder.CreateCall(Intr, Ops); | |||
| 14857 | return EmitX86MaskedCompareResult(*this, Cmp, NumElts, nullptr); | |||
| 14858 | } | |||
| 14859 | ||||
| 14860 | return Builder.CreateCall(Intr, Ops); | |||
| 14861 | } | |||
| 14862 | ||||
| 14863 | // Builtins without the _mask suffix return a vector of integers | |||
| 14864 | // of the same width as the input vectors | |||
| 14865 | if (IsMaskFCmp) { | |||
| 14866 | // We ignore SAE if strict FP is disabled. We only keep precise | |||
| 14867 | // exception behavior under strict FP. | |||
| 14868 | // NOTE: If strict FP does ever go through here a CGFPOptionsRAII | |||
| 14869 | // object will be required. | |||
| 14870 | unsigned NumElts = | |||
| 14871 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements(); | |||
| 14872 | Value *Cmp; | |||
| 14873 | if (IsSignaling) | |||
| 14874 | Cmp = Builder.CreateFCmpS(Pred, Ops[0], Ops[1]); | |||
| 14875 | else | |||
| 14876 | Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]); | |||
| 14877 | return EmitX86MaskedCompareResult(*this, Cmp, NumElts, Ops[3]); | |||
| 14878 | } | |||
| 14879 | ||||
| 14880 | return getVectorFCmpIR(Pred, IsSignaling); | |||
| 14881 | } | |||
| 14882 | ||||
| 14883 | // SSE scalar comparison intrinsics | |||
| 14884 | case X86::BI__builtin_ia32_cmpeqss: | |||
| 14885 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 0); | |||
| 14886 | case X86::BI__builtin_ia32_cmpltss: | |||
| 14887 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 1); | |||
| 14888 | case X86::BI__builtin_ia32_cmpless: | |||
| 14889 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 2); | |||
| 14890 | case X86::BI__builtin_ia32_cmpunordss: | |||
| 14891 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 3); | |||
| 14892 | case X86::BI__builtin_ia32_cmpneqss: | |||
| 14893 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 4); | |||
| 14894 | case X86::BI__builtin_ia32_cmpnltss: | |||
| 14895 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 5); | |||
| 14896 | case X86::BI__builtin_ia32_cmpnless: | |||
| 14897 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 6); | |||
| 14898 | case X86::BI__builtin_ia32_cmpordss: | |||
| 14899 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 7); | |||
| 14900 | case X86::BI__builtin_ia32_cmpeqsd: | |||
| 14901 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 0); | |||
| 14902 | case X86::BI__builtin_ia32_cmpltsd: | |||
| 14903 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 1); | |||
| 14904 | case X86::BI__builtin_ia32_cmplesd: | |||
| 14905 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 2); | |||
| 14906 | case X86::BI__builtin_ia32_cmpunordsd: | |||
| 14907 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 3); | |||
| 14908 | case X86::BI__builtin_ia32_cmpneqsd: | |||
| 14909 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 4); | |||
| 14910 | case X86::BI__builtin_ia32_cmpnltsd: | |||
| 14911 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 5); | |||
| 14912 | case X86::BI__builtin_ia32_cmpnlesd: | |||
| 14913 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 6); | |||
| 14914 | case X86::BI__builtin_ia32_cmpordsd: | |||
| 14915 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 7); | |||
| 14916 | ||||
| 14917 | // f16c half2float intrinsics | |||
| 14918 | case X86::BI__builtin_ia32_vcvtph2ps: | |||
| 14919 | case X86::BI__builtin_ia32_vcvtph2ps256: | |||
| 14920 | case X86::BI__builtin_ia32_vcvtph2ps_mask: | |||
| 14921 | case X86::BI__builtin_ia32_vcvtph2ps256_mask: | |||
| 14922 | case X86::BI__builtin_ia32_vcvtph2ps512_mask: { | |||
| 14923 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); | |||
| 14924 | return EmitX86CvtF16ToFloatExpr(*this, Ops, ConvertType(E->getType())); | |||
| 14925 | } | |||
| 14926 | ||||
| 14927 | // AVX512 bf16 intrinsics | |||
| 14928 | case X86::BI__builtin_ia32_cvtneps2bf16_128_mask: { | |||
| 14929 | Ops[2] = getMaskVecValue( | |||
| 14930 | *this, Ops[2], | |||
| 14931 | cast<llvm::FixedVectorType>(Ops[0]->getType())->getNumElements()); | |||
| 14932 | Intrinsic::ID IID = Intrinsic::x86_avx512bf16_mask_cvtneps2bf16_128; | |||
| 14933 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
| 14934 | } | |||
| 14935 | case X86::BI__builtin_ia32_cvtsbf162ss_32: | |||
| 14936 | return EmitX86CvtBF16ToFloatExpr(*this, E, Ops); | |||
| 14937 | ||||
| 14938 | case X86::BI__builtin_ia32_cvtneps2bf16_256_mask: | |||
| 14939 | case X86::BI__builtin_ia32_cvtneps2bf16_512_mask: { | |||
| 14940 | Intrinsic::ID IID; | |||
| 14941 | switch (BuiltinID) { | |||
| 14942 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "clang/lib/CodeGen/CGBuiltin.cpp" , 14942); | |||
| 14943 | case X86::BI__builtin_ia32_cvtneps2bf16_256_mask: | |||
| 14944 | IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_256; | |||
| 14945 | break; | |||
| 14946 | case X86::BI__builtin_ia32_cvtneps2bf16_512_mask: | |||
| 14947 | IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_512; | |||
| 14948 | break; | |||
| 14949 | } | |||
| 14950 | Value *Res = Builder.CreateCall(CGM.getIntrinsic(IID), Ops[0]); | |||
| 14951 | return EmitX86Select(*this, Ops[2], Res, Ops[1]); | |||
| 14952 | } | |||
| 14953 | ||||
| 14954 | case X86::BI__cpuid: | |||
| 14955 | case X86::BI__cpuidex: { | |||
| 14956 | Value *FuncId = EmitScalarExpr(E->getArg(1)); | |||
| 14957 | Value *SubFuncId = BuiltinID == X86::BI__cpuidex | |||
| 14958 | ? EmitScalarExpr(E->getArg(2)) | |||
| 14959 | : llvm::ConstantInt::get(Int32Ty, 0); | |||
| 14960 | ||||
| 14961 | llvm::StructType *CpuidRetTy = | |||
| 14962 | llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty, Int32Ty); | |||
| 14963 | llvm::FunctionType *FTy = | |||
| 14964 | llvm::FunctionType::get(CpuidRetTy, {Int32Ty, Int32Ty}, false); | |||
| 14965 | ||||
| 14966 | StringRef Asm, Constraints; | |||
| 14967 | if (getTarget().getTriple().getArch() == llvm::Triple::x86) { | |||
| 14968 | Asm = "cpuid"; | |||
| 14969 | Constraints = "={ax},={bx},={cx},={dx},{ax},{cx}"; | |||
| 14970 | } else { | |||
| 14971 | // x86-64 uses %rbx as the base register, so preserve it. | |||
| 14972 | Asm = "xchgq %rbx, ${1:q}\n" | |||
| 14973 | "cpuid\n" | |||
| 14974 | "xchgq %rbx, ${1:q}"; | |||
| 14975 | Constraints = "={ax},=r,={cx},={dx},0,2"; | |||
| 14976 | } | |||
| 14977 | ||||
| 14978 | llvm::InlineAsm *IA = llvm::InlineAsm::get(FTy, Asm, Constraints, | |||
| 14979 | /*hasSideEffects=*/false); | |||
| 14980 | Value *IACall = Builder.CreateCall(IA, {FuncId, SubFuncId}); | |||
| 14981 | Value *BasePtr = EmitScalarExpr(E->getArg(0)); | |||
| 14982 | Value *Store = nullptr; | |||
| 14983 | for (unsigned i = 0; i < 4; i++) { | |||
| 14984 | Value *Extracted = Builder.CreateExtractValue(IACall, i); | |||
| 14985 | Value *StorePtr = Builder.CreateConstInBoundsGEP1_32(Int32Ty, BasePtr, i); | |||
| 14986 | Store = Builder.CreateAlignedStore(Extracted, StorePtr, getIntAlign()); | |||
| 14987 | } | |||
| 14988 | ||||
| 14989 | // Return the last store instruction to signal that we have emitted the | |||
| 14990 | // the intrinsic. | |||
| 14991 | return Store; | |||
| 14992 | } | |||
| 14993 | ||||
| 14994 | case X86::BI__emul: | |||
| 14995 | case X86::BI__emulu: { | |||
| 14996 | llvm::Type *Int64Ty = llvm::IntegerType::get(getLLVMContext(), 64); | |||
| 14997 | bool isSigned = (BuiltinID == X86::BI__emul); | |||
| 14998 | Value *LHS = Builder.CreateIntCast(Ops[0], Int64Ty, isSigned); | |||
| 14999 | Value *RHS = Builder.CreateIntCast(Ops[1], Int64Ty, isSigned); | |||
| 15000 | return Builder.CreateMul(LHS, RHS, "", !isSigned, isSigned); | |||
| 15001 | } | |||
| 15002 | case X86::BI__mulh: | |||
| 15003 | case X86::BI__umulh: | |||
| 15004 | case X86::BI_mul128: | |||
| 15005 | case X86::BI_umul128: { | |||
| 15006 | llvm::Type *ResType = ConvertType(E->getType()); | |||
| 15007 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); | |||
| 15008 | ||||
| 15009 | bool IsSigned = (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI_mul128); | |||
| 15010 | Value *LHS = Builder.CreateIntCast(Ops[0], Int128Ty, IsSigned); | |||
| 15011 | Value *RHS = Builder.CreateIntCast(Ops[1], Int128Ty, IsSigned); | |||
| 15012 | ||||
| 15013 | Value *MulResult, *HigherBits; | |||
| 15014 | if (IsSigned) { | |||
| 15015 | MulResult = Builder.CreateNSWMul(LHS, RHS); | |||
| 15016 | HigherBits = Builder.CreateAShr(MulResult, 64); | |||
| 15017 | } else { | |||
| 15018 | MulResult = Builder.CreateNUWMul(LHS, RHS); | |||
| 15019 | HigherBits = Builder.CreateLShr(MulResult, 64); | |||
| 15020 | } | |||
| 15021 | HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned); | |||
| 15022 | ||||
| 15023 | if (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI__umulh) | |||
| 15024 | return HigherBits; | |||
| 15025 | ||||
| 15026 | Address HighBitsAddress = EmitPointerWithAlignment(E->getArg(2)); | |||
| 15027 | Builder.CreateStore(HigherBits, HighBitsAddress); | |||
| 15028 | return Builder.CreateIntCast(MulResult, ResType, IsSigned); | |||
| 15029 | } | |||
| 15030 | ||||
| 15031 | case X86::BI__faststorefence: { | |||
| 15032 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, | |||
| 15033 | llvm::SyncScope::System); | |||
| 15034 | } | |||
| 15035 | case X86::BI__shiftleft128: | |||
| 15036 | case X86::BI__shiftright128: { | |||
| 15037 | llvm::Function *F = CGM.getIntrinsic( | |||
| 15038 | BuiltinID == X86::BI__shiftleft128 ? Intrinsic::fshl : Intrinsic::fshr, | |||
| 15039 | Int64Ty); | |||
| 15040 | // Flip low/high ops and zero-extend amount to matching type. | |||
| 15041 | // shiftleft128(Low, High, Amt) -> fshl(High, Low, Amt) | |||
| 15042 | // shiftright128(Low, High, Amt) -> fshr(High, Low, Amt) | |||
| 15043 | std::swap(Ops[0], Ops[1]); | |||
| 15044 | Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); | |||
| 15045 | return Builder.CreateCall(F, Ops); | |||
| 15046 | } | |||
| 15047 | case X86::BI_ReadWriteBarrier: | |||
| 15048 | case X86::BI_ReadBarrier: | |||
| 15049 | case X86::BI_WriteBarrier: { | |||
| 15050 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, | |||
| 15051 | llvm::SyncScope::SingleThread); | |||
| 15052 | } | |||
| 15053 | ||||
| 15054 | case X86::BI_AddressOfReturnAddress: { | |||
| 15055 | Function *F = | |||
| 15056 | CGM.getIntrinsic(Intrinsic::addressofreturnaddress, AllocaInt8PtrTy); | |||
| 15057 | return Builder.CreateCall(F); | |||
| 15058 | } | |||
| 15059 | case X86::BI__stosb: { | |||
| 15060 | // We treat __stosb as a volatile memset - it may not generate "rep stosb" | |||
| 15061 | // instruction, but it will create a memset that won't be optimized away. | |||
| 15062 | return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], Align(1), true); | |||
| 15063 | } | |||
| 15064 | case X86::BI__ud2: | |||
| 15065 | // llvm.trap makes a ud2a instruction on x86. | |||
| 15066 | return EmitTrapCall(Intrinsic::trap); | |||
| 15067 | case X86::BI__int2c: { | |||
| 15068 | // This syscall signals a driver assertion failure in x86 NT kernels. | |||
| 15069 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false); | |||
| 15070 | llvm::InlineAsm *IA = | |||
| 15071 | llvm::InlineAsm::get(FTy, "int $$0x2c", "", /*hasSideEffects=*/true); | |||
| 15072 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( | |||
| 15073 | getLLVMContext(), llvm::AttributeList::FunctionIndex, | |||
| 15074 | llvm::Attribute::NoReturn); | |||
| 15075 | llvm::CallInst *CI = Builder.CreateCall(IA); | |||
| 15076 | CI->setAttributes(NoReturnAttr); | |||
| 15077 | return CI; | |||
| 15078 | } | |||
| 15079 | case X86::BI__readfsbyte: | |||
| 15080 | case X86::BI__readfsword: | |||
| 15081 | case X86::BI__readfsdword: | |||
| 15082 | case X86::BI__readfsqword: { | |||
| 15083 | llvm::Type *IntTy = ConvertType(E->getType()); | |||
| 15084 | Value *Ptr = | |||
| 15085 | Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 257)); | |||
| 15086 | LoadInst *Load = Builder.CreateAlignedLoad( | |||
| 15087 | IntTy, Ptr, getContext().getTypeAlignInChars(E->getType())); | |||
| 15088 | Load->setVolatile(true); | |||
| 15089 | return Load; | |||
| 15090 | } | |||
| 15091 | case X86::BI__readgsbyte: | |||
| 15092 | case X86::BI__readgsword: | |||
| 15093 | case X86::BI__readgsdword: | |||
| 15094 | case X86::BI__readgsqword: { | |||
| 15095 | llvm::Type *IntTy = ConvertType(E->getType()); | |||
| 15096 | Value *Ptr = | |||
| 15097 | Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 256)); | |||
| 15098 | LoadInst *Load = Builder.CreateAlignedLoad( | |||
| 15099 | IntTy, Ptr, getContext().getTypeAlignInChars(E->getType())); | |||
| 15100 | Load->setVolatile(true); | |||
| 15101 | return Load; | |||
| 15102 | } | |||
| 15103 | case X86::BI__builtin_ia32_encodekey128_u32: { | |||
| 15104 | Intrinsic::ID IID = Intrinsic::x86_encodekey128; | |||
| 15105 | ||||
| 15106 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), {Ops[0], Ops[1]}); | |||
| 15107 | ||||
| 15108 | for (int i = 0; i < 3; ++i) { | |||
| 15109 | Value *Extract = Builder.CreateExtractValue(Call, i + 1); | |||
| 15110 | Value *Ptr = Builder.CreateConstGEP1_32(Int8Ty, Ops[2], i * 16); | |||
| 15111 | Ptr = Builder.CreateBitCast( | |||
| 15112 | Ptr, llvm::PointerType::getUnqual(Extract->getType())); | |||
| 15113 | Builder.CreateAlignedStore(Extract, Ptr, Align(1)); | |||
| 15114 | } | |||
| 15115 | ||||
| 15116 | return Builder.CreateExtractValue(Call, 0); | |||
| 15117 | } | |||
| 15118 | case X86::BI__builtin_ia32_encodekey256_u32: { | |||
| 15119 | Intrinsic::ID IID = Intrinsic::x86_encodekey256; | |||
| 15120 | ||||
| 15121 | Value *Call = | |||
| 15122 | Builder.CreateCall(CGM.getIntrinsic(IID), {Ops[0], Ops[1], Ops[2]}); | |||
| 15123 | ||||
| 15124 | for (int i = 0; i < 4; ++i) { | |||
| 15125 | Value *Extract = Builder.CreateExtractValue(Call, i + 1); | |||
| 15126 | Value *Ptr = Builder.CreateConstGEP1_32(Int8Ty, Ops[3], i * 16); | |||
| 15127 | Ptr = Builder.CreateBitCast( | |||
| 15128 | Ptr, llvm::PointerType::getUnqual(Extract->getType())); | |||
| 15129 | Builder.CreateAlignedStore(Extract, Ptr, Align(1)); | |||
| 15130 | } | |||
| 15131 | ||||
| 15132 | return Builder.CreateExtractValue(Call, 0); | |||
| 15133 | } | |||
| 15134 | case X86::BI__builtin_ia32_aesenc128kl_u8: | |||
| 15135 | case X86::BI__builtin_ia32_aesdec128kl_u8: | |||
| 15136 | case X86::BI__builtin_ia32_aesenc256kl_u8: | |||
| 15137 | case X86::BI__builtin_ia32_aesdec256kl_u8: { | |||
| 15138 | Intrinsic::ID IID; | |||
| 15139 | StringRef BlockName; | |||
| 15140 | switch (BuiltinID) { | |||
| 15141 | default: | |||
| 15142 | llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 15142); | |||
| 15143 | case X86::BI__builtin_ia32_aesenc128kl_u8: | |||
| 15144 | IID = Intrinsic::x86_aesenc128kl; | |||
| 15145 | BlockName = "aesenc128kl"; | |||
| 15146 | break; | |||
| 15147 | case X86::BI__builtin_ia32_aesdec128kl_u8: | |||
| 15148 | IID = Intrinsic::x86_aesdec128kl; | |||
| 15149 | BlockName = "aesdec128kl"; | |||
| 15150 | break; | |||
| 15151 | case X86::BI__builtin_ia32_aesenc256kl_u8: | |||
| 15152 | IID = Intrinsic::x86_aesenc256kl; | |||
| 15153 | BlockName = "aesenc256kl"; | |||
| 15154 | break; | |||
| 15155 | case X86::BI__builtin_ia32_aesdec256kl_u8: | |||
| 15156 | IID = Intrinsic::x86_aesdec256kl; | |||
| 15157 | BlockName = "aesdec256kl"; | |||
| 15158 | break; | |||
| 15159 | } | |||
| 15160 | ||||
| 15161 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), {Ops[1], Ops[2]}); | |||
| 15162 | ||||
| 15163 | BasicBlock *NoError = | |||
| 15164 | createBasicBlock(BlockName + "_no_error", this->CurFn); | |||
| 15165 | BasicBlock *Error = createBasicBlock(BlockName + "_error", this->CurFn); | |||
| 15166 | BasicBlock *End = createBasicBlock(BlockName + "_end", this->CurFn); | |||
| 15167 | ||||
| 15168 | Value *Ret = Builder.CreateExtractValue(Call, 0); | |||
| 15169 | Value *Succ = Builder.CreateTrunc(Ret, Builder.getInt1Ty()); | |||
| 15170 | Value *Out = Builder.CreateExtractValue(Call, 1); | |||
| 15171 | Builder.CreateCondBr(Succ, NoError, Error); | |||
| 15172 | ||||
| 15173 | Builder.SetInsertPoint(NoError); | |||
| 15174 | Builder.CreateDefaultAlignedStore(Out, Ops[0]); | |||
| 15175 | Builder.CreateBr(End); | |||
| 15176 | ||||
| 15177 | Builder.SetInsertPoint(Error); | |||
| 15178 | Constant *Zero = llvm::Constant::getNullValue(Out->getType()); | |||
| 15179 | Builder.CreateDefaultAlignedStore(Zero, Ops[0]); | |||
| 15180 | Builder.CreateBr(End); | |||
| 15181 | ||||
| 15182 | Builder.SetInsertPoint(End); | |||
| 15183 | return Builder.CreateExtractValue(Call, 0); | |||
| 15184 | } | |||
| 15185 | case X86::BI__builtin_ia32_aesencwide128kl_u8: | |||
| 15186 | case X86::BI__builtin_ia32_aesdecwide128kl_u8: | |||
| 15187 | case X86::BI__builtin_ia32_aesencwide256kl_u8: | |||
| 15188 | case X86::BI__builtin_ia32_aesdecwide256kl_u8: { | |||
| 15189 | Intrinsic::ID IID; | |||
| 15190 | StringRef BlockName; | |||
| 15191 | switch (BuiltinID) { | |||
| 15192 | case X86::BI__builtin_ia32_aesencwide128kl_u8: | |||
| 15193 | IID = Intrinsic::x86_aesencwide128kl; | |||
| 15194 | BlockName = "aesencwide128kl"; | |||
| 15195 | break; | |||
| 15196 | case X86::BI__builtin_ia32_aesdecwide128kl_u8: | |||
| 15197 | IID = Intrinsic::x86_aesdecwide128kl; | |||
| 15198 | BlockName = "aesdecwide128kl"; | |||
| 15199 | break; | |||
| 15200 | case X86::BI__builtin_ia32_aesencwide256kl_u8: | |||
| 15201 | IID = Intrinsic::x86_aesencwide256kl; | |||
| 15202 | BlockName = "aesencwide256kl"; | |||
| 15203 | break; | |||
| 15204 | case X86::BI__builtin_ia32_aesdecwide256kl_u8: | |||
| 15205 | IID = Intrinsic::x86_aesdecwide256kl; | |||
| 15206 | BlockName = "aesdecwide256kl"; | |||
| 15207 | break; | |||
| 15208 | } | |||
| 15209 | ||||
| 15210 | llvm::Type *Ty = FixedVectorType::get(Builder.getInt64Ty(), 2); | |||
| 15211 | Value *InOps[9]; | |||
| 15212 | InOps[0] = Ops[2]; | |||
| 15213 | for (int i = 0; i != 8; ++i) { | |||
| 15214 | Value *Ptr = Builder.CreateConstGEP1_32(Ty, Ops[1], i); | |||
| 15215 | InOps[i + 1] = Builder.CreateAlignedLoad(Ty, Ptr, Align(16)); | |||
| 15216 | } | |||
| 15217 | ||||
| 15218 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), InOps); | |||
| 15219 | ||||
| 15220 | BasicBlock *NoError = | |||
| 15221 | createBasicBlock(BlockName + "_no_error", this->CurFn); | |||
| 15222 | BasicBlock *Error = createBasicBlock(BlockName + "_error", this->CurFn); | |||
| 15223 | BasicBlock *End = createBasicBlock(BlockName + "_end", this->CurFn); | |||
| 15224 | ||||
| 15225 | Value *Ret = Builder.CreateExtractValue(Call, 0); | |||
| 15226 | Value *Succ = Builder.CreateTrunc(Ret, Builder.getInt1Ty()); | |||
| 15227 | Builder.CreateCondBr(Succ, NoError, Error); | |||
| 15228 | ||||
| 15229 | Builder.SetInsertPoint(NoError); | |||
| 15230 | for (int i = 0; i != 8; ++i) { | |||
| 15231 | Value *Extract = Builder.CreateExtractValue(Call, i + 1); | |||
| 15232 | Value *Ptr = Builder.CreateConstGEP1_32(Extract->getType(), Ops[0], i); | |||
| 15233 | Builder.CreateAlignedStore(Extract, Ptr, Align(16)); | |||
| 15234 | } | |||
| 15235 | Builder.CreateBr(End); | |||
| 15236 | ||||
| 15237 | Builder.SetInsertPoint(Error); | |||
| 15238 | for (int i = 0; i != 8; ++i) { | |||
| 15239 | Value *Out = Builder.CreateExtractValue(Call, i + 1); | |||
| 15240 | Constant *Zero = llvm::Constant::getNullValue(Out->getType()); | |||
| 15241 | Value *Ptr = Builder.CreateConstGEP1_32(Out->getType(), Ops[0], i); | |||
| 15242 | Builder.CreateAlignedStore(Zero, Ptr, Align(16)); | |||
| 15243 | } | |||
| 15244 | Builder.CreateBr(End); | |||
| 15245 | ||||
| 15246 | Builder.SetInsertPoint(End); | |||
| 15247 | return Builder.CreateExtractValue(Call, 0); | |||
| 15248 | } | |||
| 15249 | case X86::BI__builtin_ia32_vfcmaddcph512_mask: | |||
| 15250 | IsConjFMA = true; | |||
| 15251 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 15252 | case X86::BI__builtin_ia32_vfmaddcph512_mask: { | |||
| 15253 | Intrinsic::ID IID = IsConjFMA | |||
| 15254 | ? Intrinsic::x86_avx512fp16_mask_vfcmadd_cph_512 | |||
| 15255 | : Intrinsic::x86_avx512fp16_mask_vfmadd_cph_512; | |||
| 15256 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
| 15257 | return EmitX86Select(*this, Ops[3], Call, Ops[0]); | |||
| 15258 | } | |||
| 15259 | case X86::BI__builtin_ia32_vfcmaddcsh_round_mask: | |||
| 15260 | IsConjFMA = true; | |||
| 15261 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 15262 | case X86::BI__builtin_ia32_vfmaddcsh_round_mask: { | |||
| 15263 | Intrinsic::ID IID = IsConjFMA ? Intrinsic::x86_avx512fp16_mask_vfcmadd_csh | |||
| 15264 | : Intrinsic::x86_avx512fp16_mask_vfmadd_csh; | |||
| 15265 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
| 15266 | Value *And = Builder.CreateAnd(Ops[3], llvm::ConstantInt::get(Int8Ty, 1)); | |||
| 15267 | return EmitX86Select(*this, And, Call, Ops[0]); | |||
| 15268 | } | |||
| 15269 | case X86::BI__builtin_ia32_vfcmaddcsh_round_mask3: | |||
| 15270 | IsConjFMA = true; | |||
| 15271 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 15272 | case X86::BI__builtin_ia32_vfmaddcsh_round_mask3: { | |||
| 15273 | Intrinsic::ID IID = IsConjFMA ? Intrinsic::x86_avx512fp16_mask_vfcmadd_csh | |||
| 15274 | : Intrinsic::x86_avx512fp16_mask_vfmadd_csh; | |||
| 15275 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), Ops); | |||
| 15276 | static constexpr int Mask[] = {0, 5, 6, 7}; | |||
| 15277 | return Builder.CreateShuffleVector(Call, Ops[2], Mask); | |||
| 15278 | } | |||
| 15279 | } | |||
| 15280 | } | |||
| 15281 | ||||
| 15282 | Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID, | |||
| 15283 | const CallExpr *E) { | |||
| 15284 | SmallVector<Value*, 4> Ops; | |||
| 15285 | ||||
| 15286 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | |||
| ||||
| 15287 | if (E->getArg(i)->getType()->isArrayType()) | |||
| 15288 | Ops.push_back(EmitArrayToPointerDecay(E->getArg(i)).getPointer()); | |||
| 15289 | else | |||
| 15290 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 15291 | } | |||
| 15292 | ||||
| 15293 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
| 15294 | ||||
| 15295 | switch (BuiltinID) { | |||
| 15296 | default: return nullptr; | |||
| 15297 | ||||
| 15298 | // __builtin_ppc_get_timebase is GCC 4.8+'s PowerPC-specific name for what we | |||
| 15299 | // call __builtin_readcyclecounter. | |||
| 15300 | case PPC::BI__builtin_ppc_get_timebase: | |||
| 15301 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::readcyclecounter)); | |||
| 15302 | ||||
| 15303 | // vec_ld, vec_xl_be, vec_lvsl, vec_lvsr | |||
| 15304 | case PPC::BI__builtin_altivec_lvx: | |||
| 15305 | case PPC::BI__builtin_altivec_lvxl: | |||
| 15306 | case PPC::BI__builtin_altivec_lvebx: | |||
| 15307 | case PPC::BI__builtin_altivec_lvehx: | |||
| 15308 | case PPC::BI__builtin_altivec_lvewx: | |||
| 15309 | case PPC::BI__builtin_altivec_lvsl: | |||
| 15310 | case PPC::BI__builtin_altivec_lvsr: | |||
| 15311 | case PPC::BI__builtin_vsx_lxvd2x: | |||
| 15312 | case PPC::BI__builtin_vsx_lxvw4x: | |||
| 15313 | case PPC::BI__builtin_vsx_lxvd2x_be: | |||
| 15314 | case PPC::BI__builtin_vsx_lxvw4x_be: | |||
| 15315 | case PPC::BI__builtin_vsx_lxvl: | |||
| 15316 | case PPC::BI__builtin_vsx_lxvll: | |||
| 15317 | { | |||
| 15318 | if(BuiltinID == PPC::BI__builtin_vsx_lxvl || | |||
| 15319 | BuiltinID == PPC::BI__builtin_vsx_lxvll){ | |||
| 15320 | Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy); | |||
| 15321 | }else { | |||
| 15322 | Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy); | |||
| 15323 | Ops[0] = Builder.CreateGEP(Int8Ty, Ops[1], Ops[0]); | |||
| 15324 | Ops.pop_back(); | |||
| 15325 | } | |||
| 15326 | ||||
| 15327 | switch (BuiltinID) { | |||
| 15328 | default: llvm_unreachable("Unsupported ld/lvsl/lvsr intrinsic!")::llvm::llvm_unreachable_internal("Unsupported ld/lvsl/lvsr intrinsic!" , "clang/lib/CodeGen/CGBuiltin.cpp", 15328); | |||
| 15329 | case PPC::BI__builtin_altivec_lvx: | |||
| 15330 | ID = Intrinsic::ppc_altivec_lvx; | |||
| 15331 | break; | |||
| 15332 | case PPC::BI__builtin_altivec_lvxl: | |||
| 15333 | ID = Intrinsic::ppc_altivec_lvxl; | |||
| 15334 | break; | |||
| 15335 | case PPC::BI__builtin_altivec_lvebx: | |||
| 15336 | ID = Intrinsic::ppc_altivec_lvebx; | |||
| 15337 | break; | |||
| 15338 | case PPC::BI__builtin_altivec_lvehx: | |||
| 15339 | ID = Intrinsic::ppc_altivec_lvehx; | |||
| 15340 | break; | |||
| 15341 | case PPC::BI__builtin_altivec_lvewx: | |||
| 15342 | ID = Intrinsic::ppc_altivec_lvewx; | |||
| 15343 | break; | |||
| 15344 | case PPC::BI__builtin_altivec_lvsl: | |||
| 15345 | ID = Intrinsic::ppc_altivec_lvsl; | |||
| 15346 | break; | |||
| 15347 | case PPC::BI__builtin_altivec_lvsr: | |||
| 15348 | ID = Intrinsic::ppc_altivec_lvsr; | |||
| 15349 | break; | |||
| 15350 | case PPC::BI__builtin_vsx_lxvd2x: | |||
| 15351 | ID = Intrinsic::ppc_vsx_lxvd2x; | |||
| 15352 | break; | |||
| 15353 | case PPC::BI__builtin_vsx_lxvw4x: | |||
| 15354 | ID = Intrinsic::ppc_vsx_lxvw4x; | |||
| 15355 | break; | |||
| 15356 | case PPC::BI__builtin_vsx_lxvd2x_be: | |||
| 15357 | ID = Intrinsic::ppc_vsx_lxvd2x_be; | |||
| 15358 | break; | |||
| 15359 | case PPC::BI__builtin_vsx_lxvw4x_be: | |||
| 15360 | ID = Intrinsic::ppc_vsx_lxvw4x_be; | |||
| 15361 | break; | |||
| 15362 | case PPC::BI__builtin_vsx_lxvl: | |||
| 15363 | ID = Intrinsic::ppc_vsx_lxvl; | |||
| 15364 | break; | |||
| 15365 | case PPC::BI__builtin_vsx_lxvll: | |||
| 15366 | ID = Intrinsic::ppc_vsx_lxvll; | |||
| 15367 | break; | |||
| 15368 | } | |||
| 15369 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
| 15370 | return Builder.CreateCall(F, Ops, ""); | |||
| 15371 | } | |||
| 15372 | ||||
| 15373 | // vec_st, vec_xst_be | |||
| 15374 | case PPC::BI__builtin_altivec_stvx: | |||
| 15375 | case PPC::BI__builtin_altivec_stvxl: | |||
| 15376 | case PPC::BI__builtin_altivec_stvebx: | |||
| 15377 | case PPC::BI__builtin_altivec_stvehx: | |||
| 15378 | case PPC::BI__builtin_altivec_stvewx: | |||
| 15379 | case PPC::BI__builtin_vsx_stxvd2x: | |||
| 15380 | case PPC::BI__builtin_vsx_stxvw4x: | |||
| 15381 | case PPC::BI__builtin_vsx_stxvd2x_be: | |||
| 15382 | case PPC::BI__builtin_vsx_stxvw4x_be: | |||
| 15383 | case PPC::BI__builtin_vsx_stxvl: | |||
| 15384 | case PPC::BI__builtin_vsx_stxvll: | |||
| 15385 | { | |||
| 15386 | if(BuiltinID == PPC::BI__builtin_vsx_stxvl || | |||
| 15387 | BuiltinID == PPC::BI__builtin_vsx_stxvll ){ | |||
| 15388 | Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy); | |||
| 15389 | }else { | |||
| 15390 | Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy); | |||
| 15391 | Ops[1] = Builder.CreateGEP(Int8Ty, Ops[2], Ops[1]); | |||
| 15392 | Ops.pop_back(); | |||
| 15393 | } | |||
| 15394 | ||||
| 15395 | switch (BuiltinID) { | |||
| 15396 | default: llvm_unreachable("Unsupported st intrinsic!")::llvm::llvm_unreachable_internal("Unsupported st intrinsic!" , "clang/lib/CodeGen/CGBuiltin.cpp", 15396); | |||
| 15397 | case PPC::BI__builtin_altivec_stvx: | |||
| 15398 | ID = Intrinsic::ppc_altivec_stvx; | |||
| 15399 | break; | |||
| 15400 | case PPC::BI__builtin_altivec_stvxl: | |||
| 15401 | ID = Intrinsic::ppc_altivec_stvxl; | |||
| 15402 | break; | |||
| 15403 | case PPC::BI__builtin_altivec_stvebx: | |||
| 15404 | ID = Intrinsic::ppc_altivec_stvebx; | |||
| 15405 | break; | |||
| 15406 | case PPC::BI__builtin_altivec_stvehx: | |||
| 15407 | ID = Intrinsic::ppc_altivec_stvehx; | |||
| 15408 | break; | |||
| 15409 | case PPC::BI__builtin_altivec_stvewx: | |||
| 15410 | ID = Intrinsic::ppc_altivec_stvewx; | |||
| 15411 | break; | |||
| 15412 | case PPC::BI__builtin_vsx_stxvd2x: | |||
| 15413 | ID = Intrinsic::ppc_vsx_stxvd2x; | |||
| 15414 | break; | |||
| 15415 | case PPC::BI__builtin_vsx_stxvw4x: | |||
| 15416 | ID = Intrinsic::ppc_vsx_stxvw4x; | |||
| 15417 | break; | |||
| 15418 | case PPC::BI__builtin_vsx_stxvd2x_be: | |||
| 15419 | ID = Intrinsic::ppc_vsx_stxvd2x_be; | |||
| 15420 | break; | |||
| 15421 | case PPC::BI__builtin_vsx_stxvw4x_be: | |||
| 15422 | ID = Intrinsic::ppc_vsx_stxvw4x_be; | |||
| 15423 | break; | |||
| 15424 | case PPC::BI__builtin_vsx_stxvl: | |||
| 15425 | ID = Intrinsic::ppc_vsx_stxvl; | |||
| 15426 | break; | |||
| 15427 | case PPC::BI__builtin_vsx_stxvll: | |||
| 15428 | ID = Intrinsic::ppc_vsx_stxvll; | |||
| 15429 | break; | |||
| 15430 | } | |||
| 15431 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
| 15432 | return Builder.CreateCall(F, Ops, ""); | |||
| 15433 | } | |||
| 15434 | case PPC::BI__builtin_vsx_ldrmb: { | |||
| 15435 | // Essentially boils down to performing an unaligned VMX load sequence so | |||
| 15436 | // as to avoid crossing a page boundary and then shuffling the elements | |||
| 15437 | // into the right side of the vector register. | |||
| 15438 | int64_t NumBytes = cast<ConstantInt>(Ops[1])->getZExtValue(); | |||
| 15439 | llvm::Type *ResTy = ConvertType(E->getType()); | |||
| 15440 | bool IsLE = getTarget().isLittleEndian(); | |||
| 15441 | ||||
| 15442 | // If the user wants the entire vector, just load the entire vector. | |||
| 15443 | if (NumBytes == 16) { | |||
| 15444 | Value *BC = Builder.CreateBitCast(Ops[0], ResTy->getPointerTo()); | |||
| 15445 | Value *LD = | |||
| 15446 | Builder.CreateLoad(Address(BC, ResTy, CharUnits::fromQuantity(1))); | |||
| 15447 | if (!IsLE) | |||
| 15448 | return LD; | |||
| 15449 | ||||
| 15450 | // Reverse the bytes on LE. | |||
| 15451 | SmallVector<int, 16> RevMask; | |||
| 15452 | for (int Idx = 0; Idx < 16; Idx++) | |||
| 15453 | RevMask.push_back(15 - Idx); | |||
| 15454 | return Builder.CreateShuffleVector(LD, LD, RevMask); | |||
| 15455 | } | |||
| 15456 | ||||
| 15457 | llvm::Function *Lvx = CGM.getIntrinsic(Intrinsic::ppc_altivec_lvx); | |||
| 15458 | llvm::Function *Lvs = CGM.getIntrinsic(IsLE ? Intrinsic::ppc_altivec_lvsr | |||
| 15459 | : Intrinsic::ppc_altivec_lvsl); | |||
| 15460 | llvm::Function *Vperm = CGM.getIntrinsic(Intrinsic::ppc_altivec_vperm); | |||
| 15461 | Value *HiMem = Builder.CreateGEP( | |||
| 15462 | Int8Ty, Ops[0], ConstantInt::get(Ops[1]->getType(), NumBytes - 1)); | |||
| 15463 | Value *LoLd = Builder.CreateCall(Lvx, Ops[0], "ld.lo"); | |||
| 15464 | Value *HiLd = Builder.CreateCall(Lvx, HiMem, "ld.hi"); | |||
| 15465 | Value *Mask1 = Builder.CreateCall(Lvs, Ops[0], "mask1"); | |||
| 15466 | ||||
| 15467 | Ops.clear(); | |||
| 15468 | Ops.push_back(IsLE ? HiLd : LoLd); | |||
| 15469 | Ops.push_back(IsLE ? LoLd : HiLd); | |||
| 15470 | Ops.push_back(Mask1); | |||
| 15471 | Value *AllElts = Builder.CreateCall(Vperm, Ops, "shuffle1"); | |||
| 15472 | Constant *Zero = llvm::Constant::getNullValue(IsLE ? ResTy : AllElts->getType()); | |||
| 15473 | ||||
| 15474 | if (IsLE) { | |||
| 15475 | SmallVector<int, 16> Consts; | |||
| 15476 | for (int Idx = 0; Idx < 16; Idx++) { | |||
| 15477 | int Val = (NumBytes - Idx - 1 >= 0) ? (NumBytes - Idx - 1) | |||
| 15478 | : 16 - (NumBytes - Idx); | |||
| 15479 | Consts.push_back(Val); | |||
| 15480 | } | |||
| 15481 | return Builder.CreateShuffleVector(Builder.CreateBitCast(AllElts, ResTy), | |||
| 15482 | Zero, Consts); | |||
| 15483 | } | |||
| 15484 | SmallVector<Constant *, 16> Consts; | |||
| 15485 | for (int Idx = 0; Idx < 16; Idx++) | |||
| 15486 | Consts.push_back(Builder.getInt8(NumBytes + Idx)); | |||
| 15487 | Value *Mask2 = ConstantVector::get(Consts); | |||
| 15488 | return Builder.CreateBitCast( | |||
| 15489 | Builder.CreateCall(Vperm, {Zero, AllElts, Mask2}, "shuffle2"), ResTy); | |||
| 15490 | } | |||
| 15491 | case PPC::BI__builtin_vsx_strmb: { | |||
| 15492 | int64_t NumBytes = cast<ConstantInt>(Ops[1])->getZExtValue(); | |||
| 15493 | bool IsLE = getTarget().isLittleEndian(); | |||
| 15494 | auto StoreSubVec = [&](unsigned Width, unsigned Offset, unsigned EltNo) { | |||
| 15495 | // Storing the whole vector, simply store it on BE and reverse bytes and | |||
| 15496 | // store on LE. | |||
| 15497 | if (Width == 16) { | |||
| 15498 | Value *BC = | |||
| 15499 | Builder.CreateBitCast(Ops[0], Ops[2]->getType()->getPointerTo()); | |||
| 15500 | Value *StVec = Ops[2]; | |||
| 15501 | if (IsLE) { | |||
| 15502 | SmallVector<int, 16> RevMask; | |||
| 15503 | for (int Idx = 0; Idx < 16; Idx++) | |||
| 15504 | RevMask.push_back(15 - Idx); | |||
| 15505 | StVec = Builder.CreateShuffleVector(Ops[2], Ops[2], RevMask); | |||
| 15506 | } | |||
| 15507 | return Builder.CreateStore( | |||
| 15508 | StVec, Address(BC, Ops[2]->getType(), CharUnits::fromQuantity(1))); | |||
| 15509 | } | |||
| 15510 | auto *ConvTy = Int64Ty; | |||
| 15511 | unsigned NumElts = 0; | |||
| 15512 | switch (Width) { | |||
| 15513 | default: | |||
| 15514 | llvm_unreachable("width for stores must be a power of 2")::llvm::llvm_unreachable_internal("width for stores must be a power of 2" , "clang/lib/CodeGen/CGBuiltin.cpp", 15514); | |||
| 15515 | case 8: | |||
| 15516 | ConvTy = Int64Ty; | |||
| 15517 | NumElts = 2; | |||
| 15518 | break; | |||
| 15519 | case 4: | |||
| 15520 | ConvTy = Int32Ty; | |||
| 15521 | NumElts = 4; | |||
| 15522 | break; | |||
| 15523 | case 2: | |||
| 15524 | ConvTy = Int16Ty; | |||
| 15525 | NumElts = 8; | |||
| 15526 | break; | |||
| 15527 | case 1: | |||
| 15528 | ConvTy = Int8Ty; | |||
| 15529 | NumElts = 16; | |||
| 15530 | break; | |||
| 15531 | } | |||
| 15532 | Value *Vec = Builder.CreateBitCast( | |||
| 15533 | Ops[2], llvm::FixedVectorType::get(ConvTy, NumElts)); | |||
| 15534 | Value *Ptr = Builder.CreateGEP(Int8Ty, Ops[0], | |||
| 15535 | ConstantInt::get(Int64Ty, Offset)); | |||
| 15536 | Value *PtrBC = Builder.CreateBitCast(Ptr, ConvTy->getPointerTo()); | |||
| 15537 | Value *Elt = Builder.CreateExtractElement(Vec, EltNo); | |||
| 15538 | if (IsLE && Width > 1) { | |||
| 15539 | Function *F = CGM.getIntrinsic(Intrinsic::bswap, ConvTy); | |||
| 15540 | Elt = Builder.CreateCall(F, Elt); | |||
| 15541 | } | |||
| 15542 | return Builder.CreateStore( | |||
| 15543 | Elt, Address(PtrBC, ConvTy, CharUnits::fromQuantity(1))); | |||
| 15544 | }; | |||
| 15545 | unsigned Stored = 0; | |||
| 15546 | unsigned RemainingBytes = NumBytes; | |||
| 15547 | Value *Result; | |||
| 15548 | if (NumBytes == 16) | |||
| 15549 | return StoreSubVec(16, 0, 0); | |||
| 15550 | if (NumBytes >= 8) { | |||
| 15551 | Result = StoreSubVec(8, NumBytes - 8, IsLE ? 0 : 1); | |||
| 15552 | RemainingBytes -= 8; | |||
| 15553 | Stored += 8; | |||
| 15554 | } | |||
| 15555 | if (RemainingBytes >= 4) { | |||
| 15556 | Result = StoreSubVec(4, NumBytes - Stored - 4, | |||
| 15557 | IsLE ? (Stored >> 2) : 3 - (Stored >> 2)); | |||
| 15558 | RemainingBytes -= 4; | |||
| 15559 | Stored += 4; | |||
| 15560 | } | |||
| 15561 | if (RemainingBytes >= 2) { | |||
| 15562 | Result = StoreSubVec(2, NumBytes - Stored - 2, | |||
| 15563 | IsLE ? (Stored >> 1) : 7 - (Stored >> 1)); | |||
| 15564 | RemainingBytes -= 2; | |||
| 15565 | Stored += 2; | |||
| 15566 | } | |||
| 15567 | if (RemainingBytes) | |||
| 15568 | Result = | |||
| 15569 | StoreSubVec(1, NumBytes - Stored - 1, IsLE ? Stored : 15 - Stored); | |||
| 15570 | return Result; | |||
| ||||
| 15571 | } | |||
| 15572 | // Square root | |||
| 15573 | case PPC::BI__builtin_vsx_xvsqrtsp: | |||
| 15574 | case PPC::BI__builtin_vsx_xvsqrtdp: { | |||
| 15575 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15576 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15577 | if (Builder.getIsFPConstrained()) { | |||
| 15578 | llvm::Function *F = CGM.getIntrinsic( | |||
| 15579 | Intrinsic::experimental_constrained_sqrt, ResultType); | |||
| 15580 | return Builder.CreateConstrainedFPCall(F, X); | |||
| 15581 | } else { | |||
| 15582 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType); | |||
| 15583 | return Builder.CreateCall(F, X); | |||
| 15584 | } | |||
| 15585 | } | |||
| 15586 | // Count leading zeros | |||
| 15587 | case PPC::BI__builtin_altivec_vclzb: | |||
| 15588 | case PPC::BI__builtin_altivec_vclzh: | |||
| 15589 | case PPC::BI__builtin_altivec_vclzw: | |||
| 15590 | case PPC::BI__builtin_altivec_vclzd: { | |||
| 15591 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15592 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15593 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
| 15594 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType); | |||
| 15595 | return Builder.CreateCall(F, {X, Undef}); | |||
| 15596 | } | |||
| 15597 | case PPC::BI__builtin_altivec_vctzb: | |||
| 15598 | case PPC::BI__builtin_altivec_vctzh: | |||
| 15599 | case PPC::BI__builtin_altivec_vctzw: | |||
| 15600 | case PPC::BI__builtin_altivec_vctzd: { | |||
| 15601 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15602 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15603 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
| 15604 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType); | |||
| 15605 | return Builder.CreateCall(F, {X, Undef}); | |||
| 15606 | } | |||
| 15607 | case PPC::BI__builtin_altivec_vec_replace_elt: | |||
| 15608 | case PPC::BI__builtin_altivec_vec_replace_unaligned: { | |||
| 15609 | // The third argument of vec_replace_elt and vec_replace_unaligned must | |||
| 15610 | // be a compile time constant and will be emitted either to the vinsw | |||
| 15611 | // or vinsd instruction. | |||
| 15612 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); | |||
| 15613 | assert(ArgCI &&(static_cast <bool> (ArgCI && "Third Arg to vinsw/vinsd intrinsic must be a constant integer!" ) ? void (0) : __assert_fail ("ArgCI && \"Third Arg to vinsw/vinsd intrinsic must be a constant integer!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 15614, __extension__ __PRETTY_FUNCTION__ )) | |||
| 15614 | "Third Arg to vinsw/vinsd intrinsic must be a constant integer!")(static_cast <bool> (ArgCI && "Third Arg to vinsw/vinsd intrinsic must be a constant integer!" ) ? void (0) : __assert_fail ("ArgCI && \"Third Arg to vinsw/vinsd intrinsic must be a constant integer!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 15614, __extension__ __PRETTY_FUNCTION__ )); | |||
| 15615 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15616 | llvm::Function *F = nullptr; | |||
| 15617 | Value *Call = nullptr; | |||
| 15618 | int64_t ConstArg = ArgCI->getSExtValue(); | |||
| 15619 | unsigned ArgWidth = Ops[1]->getType()->getPrimitiveSizeInBits(); | |||
| 15620 | bool Is32Bit = false; | |||
| 15621 | assert((ArgWidth == 32 || ArgWidth == 64) && "Invalid argument width")(static_cast <bool> ((ArgWidth == 32 || ArgWidth == 64) && "Invalid argument width") ? void (0) : __assert_fail ("(ArgWidth == 32 || ArgWidth == 64) && \"Invalid argument width\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 15621, __extension__ __PRETTY_FUNCTION__ )); | |||
| 15622 | // The input to vec_replace_elt is an element index, not a byte index. | |||
| 15623 | if (BuiltinID == PPC::BI__builtin_altivec_vec_replace_elt) | |||
| 15624 | ConstArg *= ArgWidth / 8; | |||
| 15625 | if (ArgWidth == 32) { | |||
| 15626 | Is32Bit = true; | |||
| 15627 | // When the second argument is 32 bits, it can either be an integer or | |||
| 15628 | // a float. The vinsw intrinsic is used in this case. | |||
| 15629 | F = CGM.getIntrinsic(Intrinsic::ppc_altivec_vinsw); | |||
| 15630 | // Fix the constant according to endianess. | |||
| 15631 | if (getTarget().isLittleEndian()) | |||
| 15632 | ConstArg = 12 - ConstArg; | |||
| 15633 | } else { | |||
| 15634 | // When the second argument is 64 bits, it can either be a long long or | |||
| 15635 | // a double. The vinsd intrinsic is used in this case. | |||
| 15636 | F = CGM.getIntrinsic(Intrinsic::ppc_altivec_vinsd); | |||
| 15637 | // Fix the constant for little endian. | |||
| 15638 | if (getTarget().isLittleEndian()) | |||
| 15639 | ConstArg = 8 - ConstArg; | |||
| 15640 | } | |||
| 15641 | Ops[2] = ConstantInt::getSigned(Int32Ty, ConstArg); | |||
| 15642 | // Depending on ArgWidth, the input vector could be a float or a double. | |||
| 15643 | // If the input vector is a float type, bitcast the inputs to integers. Or, | |||
| 15644 | // if the input vector is a double, bitcast the inputs to 64-bit integers. | |||
| 15645 | if (!Ops[1]->getType()->isIntegerTy(ArgWidth)) { | |||
| 15646 | Ops[0] = Builder.CreateBitCast( | |||
| 15647 | Ops[0], Is32Bit ? llvm::FixedVectorType::get(Int32Ty, 4) | |||
| 15648 | : llvm::FixedVectorType::get(Int64Ty, 2)); | |||
| 15649 | Ops[1] = Builder.CreateBitCast(Ops[1], Is32Bit ? Int32Ty : Int64Ty); | |||
| 15650 | } | |||
| 15651 | // Emit the call to vinsw or vinsd. | |||
| 15652 | Call = Builder.CreateCall(F, Ops); | |||
| 15653 | // Depending on the builtin, bitcast to the approriate result type. | |||
| 15654 | if (BuiltinID == PPC::BI__builtin_altivec_vec_replace_elt && | |||
| 15655 | !Ops[1]->getType()->isIntegerTy()) | |||
| 15656 | return Builder.CreateBitCast(Call, ResultType); | |||
| 15657 | else if (BuiltinID == PPC::BI__builtin_altivec_vec_replace_elt && | |||
| 15658 | Ops[1]->getType()->isIntegerTy()) | |||
| 15659 | return Call; | |||
| 15660 | else | |||
| 15661 | return Builder.CreateBitCast(Call, | |||
| 15662 | llvm::FixedVectorType::get(Int8Ty, 16)); | |||
| 15663 | } | |||
| 15664 | case PPC::BI__builtin_altivec_vpopcntb: | |||
| 15665 | case PPC::BI__builtin_altivec_vpopcnth: | |||
| 15666 | case PPC::BI__builtin_altivec_vpopcntw: | |||
| 15667 | case PPC::BI__builtin_altivec_vpopcntd: { | |||
| 15668 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15669 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15670 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); | |||
| 15671 | return Builder.CreateCall(F, X); | |||
| 15672 | } | |||
| 15673 | case PPC::BI__builtin_altivec_vadduqm: | |||
| 15674 | case PPC::BI__builtin_altivec_vsubuqm: { | |||
| 15675 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); | |||
| 15676 | Ops[0] = | |||
| 15677 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int128Ty, 1)); | |||
| 15678 | Ops[1] = | |||
| 15679 | Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int128Ty, 1)); | |||
| 15680 | if (BuiltinID == PPC::BI__builtin_altivec_vadduqm) | |||
| 15681 | return Builder.CreateAdd(Ops[0], Ops[1], "vadduqm"); | |||
| 15682 | else | |||
| 15683 | return Builder.CreateSub(Ops[0], Ops[1], "vsubuqm"); | |||
| 15684 | } | |||
| 15685 | // Rotate and insert under mask operation. | |||
| 15686 | // __rldimi(rs, is, shift, mask) | |||
| 15687 | // (rotl64(rs, shift) & mask) | (is & ~mask) | |||
| 15688 | // __rlwimi(rs, is, shift, mask) | |||
| 15689 | // (rotl(rs, shift) & mask) | (is & ~mask) | |||
| 15690 | case PPC::BI__builtin_ppc_rldimi: | |||
| 15691 | case PPC::BI__builtin_ppc_rlwimi: { | |||
| 15692 | llvm::Type *Ty = Ops[0]->getType(); | |||
| 15693 | Function *F = CGM.getIntrinsic(Intrinsic::fshl, Ty); | |||
| 15694 | if (BuiltinID == PPC::BI__builtin_ppc_rldimi) | |||
| 15695 | Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); | |||
| 15696 | Value *Shift = Builder.CreateCall(F, {Ops[0], Ops[0], Ops[2]}); | |||
| 15697 | Value *X = Builder.CreateAnd(Shift, Ops[3]); | |||
| 15698 | Value *Y = Builder.CreateAnd(Ops[1], Builder.CreateNot(Ops[3])); | |||
| 15699 | return Builder.CreateOr(X, Y); | |||
| 15700 | } | |||
| 15701 | // Rotate and insert under mask operation. | |||
| 15702 | // __rlwnm(rs, shift, mask) | |||
| 15703 | // rotl(rs, shift) & mask | |||
| 15704 | case PPC::BI__builtin_ppc_rlwnm: { | |||
| 15705 | llvm::Type *Ty = Ops[0]->getType(); | |||
| 15706 | Function *F = CGM.getIntrinsic(Intrinsic::fshl, Ty); | |||
| 15707 | Value *Shift = Builder.CreateCall(F, {Ops[0], Ops[0], Ops[1]}); | |||
| 15708 | return Builder.CreateAnd(Shift, Ops[2]); | |||
| 15709 | } | |||
| 15710 | case PPC::BI__builtin_ppc_poppar4: | |||
| 15711 | case PPC::BI__builtin_ppc_poppar8: { | |||
| 15712 | llvm::Type *ArgType = Ops[0]->getType(); | |||
| 15713 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); | |||
| 15714 | Value *Tmp = Builder.CreateCall(F, Ops[0]); | |||
| 15715 | ||||
| 15716 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15717 | Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1)); | |||
| 15718 | if (Result->getType() != ResultType) | |||
| 15719 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, | |||
| 15720 | "cast"); | |||
| 15721 | return Result; | |||
| 15722 | } | |||
| 15723 | case PPC::BI__builtin_ppc_cmpb: { | |||
| 15724 | if (getTarget().getTriple().isPPC64()) { | |||
| 15725 | Function *F = | |||
| 15726 | CGM.getIntrinsic(Intrinsic::ppc_cmpb, {Int64Ty, Int64Ty, Int64Ty}); | |||
| 15727 | return Builder.CreateCall(F, Ops, "cmpb"); | |||
| 15728 | } | |||
| 15729 | // For 32 bit, emit the code as below: | |||
| 15730 | // %conv = trunc i64 %a to i32 | |||
| 15731 | // %conv1 = trunc i64 %b to i32 | |||
| 15732 | // %shr = lshr i64 %a, 32 | |||
| 15733 | // %conv2 = trunc i64 %shr to i32 | |||
| 15734 | // %shr3 = lshr i64 %b, 32 | |||
| 15735 | // %conv4 = trunc i64 %shr3 to i32 | |||
| 15736 | // %0 = tail call i32 @llvm.ppc.cmpb32(i32 %conv, i32 %conv1) | |||
| 15737 | // %conv5 = zext i32 %0 to i64 | |||
| 15738 | // %1 = tail call i32 @llvm.ppc.cmpb32(i32 %conv2, i32 %conv4) | |||
| 15739 | // %conv614 = zext i32 %1 to i64 | |||
| 15740 | // %shl = shl nuw i64 %conv614, 32 | |||
| 15741 | // %or = or i64 %shl, %conv5 | |||
| 15742 | // ret i64 %or | |||
| 15743 | Function *F = | |||
| 15744 | CGM.getIntrinsic(Intrinsic::ppc_cmpb, {Int32Ty, Int32Ty, Int32Ty}); | |||
| 15745 | Value *ArgOneLo = Builder.CreateTrunc(Ops[0], Int32Ty); | |||
| 15746 | Value *ArgTwoLo = Builder.CreateTrunc(Ops[1], Int32Ty); | |||
| 15747 | Constant *ShiftAmt = ConstantInt::get(Int64Ty, 32); | |||
| 15748 | Value *ArgOneHi = | |||
| 15749 | Builder.CreateTrunc(Builder.CreateLShr(Ops[0], ShiftAmt), Int32Ty); | |||
| 15750 | Value *ArgTwoHi = | |||
| 15751 | Builder.CreateTrunc(Builder.CreateLShr(Ops[1], ShiftAmt), Int32Ty); | |||
| 15752 | Value *ResLo = Builder.CreateZExt( | |||
| 15753 | Builder.CreateCall(F, {ArgOneLo, ArgTwoLo}, "cmpb"), Int64Ty); | |||
| 15754 | Value *ResHiShift = Builder.CreateZExt( | |||
| 15755 | Builder.CreateCall(F, {ArgOneHi, ArgTwoHi}, "cmpb"), Int64Ty); | |||
| 15756 | Value *ResHi = Builder.CreateShl(ResHiShift, ShiftAmt); | |||
| 15757 | return Builder.CreateOr(ResLo, ResHi); | |||
| 15758 | } | |||
| 15759 | // Copy sign | |||
| 15760 | case PPC::BI__builtin_vsx_xvcpsgnsp: | |||
| 15761 | case PPC::BI__builtin_vsx_xvcpsgndp: { | |||
| 15762 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15763 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15764 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 15765 | ID = Intrinsic::copysign; | |||
| 15766 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); | |||
| 15767 | return Builder.CreateCall(F, {X, Y}); | |||
| 15768 | } | |||
| 15769 | // Rounding/truncation | |||
| 15770 | case PPC::BI__builtin_vsx_xvrspip: | |||
| 15771 | case PPC::BI__builtin_vsx_xvrdpip: | |||
| 15772 | case PPC::BI__builtin_vsx_xvrdpim: | |||
| 15773 | case PPC::BI__builtin_vsx_xvrspim: | |||
| 15774 | case PPC::BI__builtin_vsx_xvrdpi: | |||
| 15775 | case PPC::BI__builtin_vsx_xvrspi: | |||
| 15776 | case PPC::BI__builtin_vsx_xvrdpic: | |||
| 15777 | case PPC::BI__builtin_vsx_xvrspic: | |||
| 15778 | case PPC::BI__builtin_vsx_xvrdpiz: | |||
| 15779 | case PPC::BI__builtin_vsx_xvrspiz: { | |||
| 15780 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15781 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15782 | if (BuiltinID == PPC::BI__builtin_vsx_xvrdpim || | |||
| 15783 | BuiltinID == PPC::BI__builtin_vsx_xvrspim) | |||
| 15784 | ID = Builder.getIsFPConstrained() | |||
| 15785 | ? Intrinsic::experimental_constrained_floor | |||
| 15786 | : Intrinsic::floor; | |||
| 15787 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpi || | |||
| 15788 | BuiltinID == PPC::BI__builtin_vsx_xvrspi) | |||
| 15789 | ID = Builder.getIsFPConstrained() | |||
| 15790 | ? Intrinsic::experimental_constrained_round | |||
| 15791 | : Intrinsic::round; | |||
| 15792 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpic || | |||
| 15793 | BuiltinID == PPC::BI__builtin_vsx_xvrspic) | |||
| 15794 | ID = Builder.getIsFPConstrained() | |||
| 15795 | ? Intrinsic::experimental_constrained_rint | |||
| 15796 | : Intrinsic::rint; | |||
| 15797 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpip || | |||
| 15798 | BuiltinID == PPC::BI__builtin_vsx_xvrspip) | |||
| 15799 | ID = Builder.getIsFPConstrained() | |||
| 15800 | ? Intrinsic::experimental_constrained_ceil | |||
| 15801 | : Intrinsic::ceil; | |||
| 15802 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpiz || | |||
| 15803 | BuiltinID == PPC::BI__builtin_vsx_xvrspiz) | |||
| 15804 | ID = Builder.getIsFPConstrained() | |||
| 15805 | ? Intrinsic::experimental_constrained_trunc | |||
| 15806 | : Intrinsic::trunc; | |||
| 15807 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); | |||
| 15808 | return Builder.getIsFPConstrained() ? Builder.CreateConstrainedFPCall(F, X) | |||
| 15809 | : Builder.CreateCall(F, X); | |||
| 15810 | } | |||
| 15811 | ||||
| 15812 | // Absolute value | |||
| 15813 | case PPC::BI__builtin_vsx_xvabsdp: | |||
| 15814 | case PPC::BI__builtin_vsx_xvabssp: { | |||
| 15815 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15816 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15817 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); | |||
| 15818 | return Builder.CreateCall(F, X); | |||
| 15819 | } | |||
| 15820 | ||||
| 15821 | // Fastmath by default | |||
| 15822 | case PPC::BI__builtin_ppc_recipdivf: | |||
| 15823 | case PPC::BI__builtin_ppc_recipdivd: | |||
| 15824 | case PPC::BI__builtin_ppc_rsqrtf: | |||
| 15825 | case PPC::BI__builtin_ppc_rsqrtd: { | |||
| 15826 | FastMathFlags FMF = Builder.getFastMathFlags(); | |||
| 15827 | Builder.getFastMathFlags().setFast(); | |||
| 15828 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15829 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15830 | ||||
| 15831 | if (BuiltinID == PPC::BI__builtin_ppc_recipdivf || | |||
| 15832 | BuiltinID == PPC::BI__builtin_ppc_recipdivd) { | |||
| 15833 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 15834 | Value *FDiv = Builder.CreateFDiv(X, Y, "recipdiv"); | |||
| 15835 | Builder.getFastMathFlags() &= (FMF); | |||
| 15836 | return FDiv; | |||
| 15837 | } | |||
| 15838 | auto *One = ConstantFP::get(ResultType, 1.0); | |||
| 15839 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType); | |||
| 15840 | Value *FDiv = Builder.CreateFDiv(One, Builder.CreateCall(F, X), "rsqrt"); | |||
| 15841 | Builder.getFastMathFlags() &= (FMF); | |||
| 15842 | return FDiv; | |||
| 15843 | } | |||
| 15844 | case PPC::BI__builtin_ppc_alignx: { | |||
| 15845 | ConstantInt *AlignmentCI = cast<ConstantInt>(Ops[0]); | |||
| 15846 | if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment)) | |||
| 15847 | AlignmentCI = ConstantInt::get(AlignmentCI->getType(), | |||
| 15848 | llvm::Value::MaximumAlignment); | |||
| 15849 | ||||
| 15850 | emitAlignmentAssumption(Ops[1], E->getArg(1), | |||
| 15851 | /*The expr loc is sufficient.*/ SourceLocation(), | |||
| 15852 | AlignmentCI, nullptr); | |||
| 15853 | return Ops[1]; | |||
| 15854 | } | |||
| 15855 | case PPC::BI__builtin_ppc_rdlam: { | |||
| 15856 | llvm::Type *Ty = Ops[0]->getType(); | |||
| 15857 | Value *ShiftAmt = Builder.CreateIntCast(Ops[1], Ty, false); | |||
| 15858 | Function *F = CGM.getIntrinsic(Intrinsic::fshl, Ty); | |||
| 15859 | Value *Rotate = Builder.CreateCall(F, {Ops[0], Ops[0], ShiftAmt}); | |||
| 15860 | return Builder.CreateAnd(Rotate, Ops[2]); | |||
| 15861 | } | |||
| 15862 | case PPC::BI__builtin_ppc_load2r: { | |||
| 15863 | Function *F = CGM.getIntrinsic(Intrinsic::ppc_load2r); | |||
| 15864 | Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy); | |||
| 15865 | Value *LoadIntrinsic = Builder.CreateCall(F, Ops); | |||
| 15866 | return Builder.CreateTrunc(LoadIntrinsic, Int16Ty); | |||
| 15867 | } | |||
| 15868 | // FMA variations | |||
| 15869 | case PPC::BI__builtin_ppc_fnmsub: | |||
| 15870 | case PPC::BI__builtin_ppc_fnmsubs: | |||
| 15871 | case PPC::BI__builtin_vsx_xvmaddadp: | |||
| 15872 | case PPC::BI__builtin_vsx_xvmaddasp: | |||
| 15873 | case PPC::BI__builtin_vsx_xvnmaddadp: | |||
| 15874 | case PPC::BI__builtin_vsx_xvnmaddasp: | |||
| 15875 | case PPC::BI__builtin_vsx_xvmsubadp: | |||
| 15876 | case PPC::BI__builtin_vsx_xvmsubasp: | |||
| 15877 | case PPC::BI__builtin_vsx_xvnmsubadp: | |||
| 15878 | case PPC::BI__builtin_vsx_xvnmsubasp: { | |||
| 15879 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 15880 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 15881 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 15882 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
| 15883 | llvm::Function *F; | |||
| 15884 | if (Builder.getIsFPConstrained()) | |||
| 15885 | F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType); | |||
| 15886 | else | |||
| 15887 | F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
| 15888 | switch (BuiltinID) { | |||
| 15889 | case PPC::BI__builtin_vsx_xvmaddadp: | |||
| 15890 | case PPC::BI__builtin_vsx_xvmaddasp: | |||
| 15891 | if (Builder.getIsFPConstrained()) | |||
| 15892 | return Builder.CreateConstrainedFPCall(F, {X, Y, Z}); | |||
| 15893 | else | |||
| 15894 | return Builder.CreateCall(F, {X, Y, Z}); | |||
| 15895 | case PPC::BI__builtin_vsx_xvnmaddadp: | |||
| 15896 | case PPC::BI__builtin_vsx_xvnmaddasp: | |||
| 15897 | if (Builder.getIsFPConstrained()) | |||
| 15898 | return Builder.CreateFNeg( | |||
| 15899 | Builder.CreateConstrainedFPCall(F, {X, Y, Z}), "neg"); | |||
| 15900 | else | |||
| 15901 | return Builder.CreateFNeg(Builder.CreateCall(F, {X, Y, Z}), "neg"); | |||
| 15902 | case PPC::BI__builtin_vsx_xvmsubadp: | |||
| 15903 | case PPC::BI__builtin_vsx_xvmsubasp: | |||
| 15904 | if (Builder.getIsFPConstrained()) | |||
| 15905 | return Builder.CreateConstrainedFPCall( | |||
| 15906 | F, {X, Y, Builder.CreateFNeg(Z, "neg")}); | |||
| 15907 | else | |||
| 15908 | return Builder.CreateCall(F, {X, Y, Builder.CreateFNeg(Z, "neg")}); | |||
| 15909 | case PPC::BI__builtin_ppc_fnmsub: | |||
| 15910 | case PPC::BI__builtin_ppc_fnmsubs: | |||
| 15911 | case PPC::BI__builtin_vsx_xvnmsubadp: | |||
| 15912 | case PPC::BI__builtin_vsx_xvnmsubasp: | |||
| 15913 | if (Builder.getIsFPConstrained()) | |||
| 15914 | return Builder.CreateFNeg( | |||
| 15915 | Builder.CreateConstrainedFPCall( | |||
| 15916 | F, {X, Y, Builder.CreateFNeg(Z, "neg")}), | |||
| 15917 | "neg"); | |||
| 15918 | else | |||
| 15919 | return Builder.CreateCall( | |||
| 15920 | CGM.getIntrinsic(Intrinsic::ppc_fnmsub, ResultType), {X, Y, Z}); | |||
| 15921 | } | |||
| 15922 | llvm_unreachable("Unknown FMA operation")::llvm::llvm_unreachable_internal("Unknown FMA operation", "clang/lib/CodeGen/CGBuiltin.cpp" , 15922); | |||
| 15923 | return nullptr; // Suppress no-return warning | |||
| 15924 | } | |||
| 15925 | ||||
| 15926 | case PPC::BI__builtin_vsx_insertword: { | |||
| 15927 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxinsertw); | |||
| 15928 | ||||
| 15929 | // Third argument is a compile time constant int. It must be clamped to | |||
| 15930 | // to the range [0, 12]. | |||
| 15931 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); | |||
| 15932 | assert(ArgCI &&(static_cast <bool> (ArgCI && "Third arg to xxinsertw intrinsic must be constant integer" ) ? void (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 15933, __extension__ __PRETTY_FUNCTION__ )) | |||
| 15933 | "Third arg to xxinsertw intrinsic must be constant integer")(static_cast <bool> (ArgCI && "Third arg to xxinsertw intrinsic must be constant integer" ) ? void (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 15933, __extension__ __PRETTY_FUNCTION__ )); | |||
| 15934 | const int64_t MaxIndex = 12; | |||
| 15935 | int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex); | |||
| 15936 | ||||
| 15937 | // The builtin semantics don't exactly match the xxinsertw instructions | |||
| 15938 | // semantics (which ppc_vsx_xxinsertw follows). The builtin extracts the | |||
| 15939 | // word from the first argument, and inserts it in the second argument. The | |||
| 15940 | // instruction extracts the word from its second input register and inserts | |||
| 15941 | // it into its first input register, so swap the first and second arguments. | |||
| 15942 | std::swap(Ops[0], Ops[1]); | |||
| 15943 | ||||
| 15944 | // Need to cast the second argument from a vector of unsigned int to a | |||
| 15945 | // vector of long long. | |||
| 15946 | Ops[1] = | |||
| 15947 | Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int64Ty, 2)); | |||
| 15948 | ||||
| 15949 | if (getTarget().isLittleEndian()) { | |||
| 15950 | // Reverse the double words in the vector we will extract from. | |||
| 15951 | Ops[0] = | |||
| 15952 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2)); | |||
| 15953 | Ops[0] = Builder.CreateShuffleVector(Ops[0], Ops[0], ArrayRef<int>{1, 0}); | |||
| 15954 | ||||
| 15955 | // Reverse the index. | |||
| 15956 | Index = MaxIndex - Index; | |||
| 15957 | } | |||
| 15958 | ||||
| 15959 | // Intrinsic expects the first arg to be a vector of int. | |||
| 15960 | Ops[0] = | |||
| 15961 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 4)); | |||
| 15962 | Ops[2] = ConstantInt::getSigned(Int32Ty, Index); | |||
| 15963 | return Builder.CreateCall(F, Ops); | |||
| 15964 | } | |||
| 15965 | ||||
| 15966 | case PPC::BI__builtin_vsx_extractuword: { | |||
| 15967 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxextractuw); | |||
| 15968 | ||||
| 15969 | // Intrinsic expects the first argument to be a vector of doublewords. | |||
| 15970 | Ops[0] = | |||
| 15971 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2)); | |||
| 15972 | ||||
| 15973 | // The second argument is a compile time constant int that needs to | |||
| 15974 | // be clamped to the range [0, 12]. | |||
| 15975 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[1]); | |||
| 15976 | assert(ArgCI &&(static_cast <bool> (ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!" ) ? void (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 15977, __extension__ __PRETTY_FUNCTION__ )) | |||
| 15977 | "Second Arg to xxextractuw intrinsic must be a constant integer!")(static_cast <bool> (ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!" ) ? void (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 15977, __extension__ __PRETTY_FUNCTION__ )); | |||
| 15978 | const int64_t MaxIndex = 12; | |||
| 15979 | int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex); | |||
| 15980 | ||||
| 15981 | if (getTarget().isLittleEndian()) { | |||
| 15982 | // Reverse the index. | |||
| 15983 | Index = MaxIndex - Index; | |||
| 15984 | Ops[1] = ConstantInt::getSigned(Int32Ty, Index); | |||
| 15985 | ||||
| 15986 | // Emit the call, then reverse the double words of the results vector. | |||
| 15987 | Value *Call = Builder.CreateCall(F, Ops); | |||
| 15988 | ||||
| 15989 | Value *ShuffleCall = | |||
| 15990 | Builder.CreateShuffleVector(Call, Call, ArrayRef<int>{1, 0}); | |||
| 15991 | return ShuffleCall; | |||
| 15992 | } else { | |||
| 15993 | Ops[1] = ConstantInt::getSigned(Int32Ty, Index); | |||
| 15994 | return Builder.CreateCall(F, Ops); | |||
| 15995 | } | |||
| 15996 | } | |||
| 15997 | ||||
| 15998 | case PPC::BI__builtin_vsx_xxpermdi: { | |||
| 15999 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); | |||
| 16000 | assert(ArgCI && "Third arg must be constant integer!")(static_cast <bool> (ArgCI && "Third arg must be constant integer!" ) ? void (0) : __assert_fail ("ArgCI && \"Third arg must be constant integer!\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 16000, __extension__ __PRETTY_FUNCTION__ )); | |||
| 16001 | ||||
| 16002 | unsigned Index = ArgCI->getZExtValue(); | |||
| 16003 | Ops[0] = | |||
| 16004 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int64Ty, 2)); | |||
| 16005 | Ops[1] = | |||
| 16006 | Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int64Ty, 2)); | |||
| 16007 | ||||
| 16008 | // Account for endianness by treating this as just a shuffle. So we use the | |||
| 16009 | // same indices for both LE and BE in order to produce expected results in | |||
| 16010 | // both cases. | |||
| 16011 | int ElemIdx0 = (Index & 2) >> 1; | |||
| 16012 | int ElemIdx1 = 2 + (Index & 1); | |||
| 16013 | ||||
| 16014 | int ShuffleElts[2] = {ElemIdx0, ElemIdx1}; | |||
| 16015 | Value *ShuffleCall = | |||
| 16016 | Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleElts); | |||
| 16017 | QualType BIRetType = E->getType(); | |||
| 16018 | auto RetTy = ConvertType(BIRetType); | |||
| 16019 | return Builder.CreateBitCast(ShuffleCall, RetTy); | |||
| 16020 | } | |||
| 16021 | ||||
| 16022 | case PPC::BI__builtin_vsx_xxsldwi: { | |||
| 16023 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); | |||
| 16024 | assert(ArgCI && "Third argument must be a compile time constant")(static_cast <bool> (ArgCI && "Third argument must be a compile time constant" ) ? void (0) : __assert_fail ("ArgCI && \"Third argument must be a compile time constant\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 16024, __extension__ __PRETTY_FUNCTION__ )); | |||
| 16025 | unsigned Index = ArgCI->getZExtValue() & 0x3; | |||
| 16026 | Ops[0] = | |||
| 16027 | Builder.CreateBitCast(Ops[0], llvm::FixedVectorType::get(Int32Ty, 4)); | |||
| 16028 | Ops[1] = | |||
| 16029 | Builder.CreateBitCast(Ops[1], llvm::FixedVectorType::get(Int32Ty, 4)); | |||
| 16030 | ||||
| 16031 | // Create a shuffle mask | |||
| 16032 | int ElemIdx0; | |||
| 16033 | int ElemIdx1; | |||
| 16034 | int ElemIdx2; | |||
| 16035 | int ElemIdx3; | |||
| 16036 | if (getTarget().isLittleEndian()) { | |||
| 16037 | // Little endian element N comes from element 8+N-Index of the | |||
| 16038 | // concatenated wide vector (of course, using modulo arithmetic on | |||
| 16039 | // the total number of elements). | |||
| 16040 | ElemIdx0 = (8 - Index) % 8; | |||
| 16041 | ElemIdx1 = (9 - Index) % 8; | |||
| 16042 | ElemIdx2 = (10 - Index) % 8; | |||
| 16043 | ElemIdx3 = (11 - Index) % 8; | |||
| 16044 | } else { | |||
| 16045 | // Big endian ElemIdx<N> = Index + N | |||
| 16046 | ElemIdx0 = Index; | |||
| 16047 | ElemIdx1 = Index + 1; | |||
| 16048 | ElemIdx2 = Index + 2; | |||
| 16049 | ElemIdx3 = Index + 3; | |||
| 16050 | } | |||
| 16051 | ||||
| 16052 | int ShuffleElts[4] = {ElemIdx0, ElemIdx1, ElemIdx2, ElemIdx3}; | |||
| 16053 | Value *ShuffleCall = | |||
| 16054 | Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleElts); | |||
| 16055 | QualType BIRetType = E->getType(); | |||
| 16056 | auto RetTy = ConvertType(BIRetType); | |||
| 16057 | return Builder.CreateBitCast(ShuffleCall, RetTy); | |||
| 16058 | } | |||
| 16059 | ||||
| 16060 | case PPC::BI__builtin_pack_vector_int128: { | |||
| 16061 | bool isLittleEndian = getTarget().isLittleEndian(); | |||
| 16062 | Value *UndefValue = | |||
| 16063 | llvm::UndefValue::get(llvm::FixedVectorType::get(Ops[0]->getType(), 2)); | |||
| 16064 | Value *Res = Builder.CreateInsertElement( | |||
| 16065 | UndefValue, Ops[0], (uint64_t)(isLittleEndian ? 1 : 0)); | |||
| 16066 | Res = Builder.CreateInsertElement(Res, Ops[1], | |||
| 16067 | (uint64_t)(isLittleEndian ? 0 : 1)); | |||
| 16068 | return Builder.CreateBitCast(Res, ConvertType(E->getType())); | |||
| 16069 | } | |||
| 16070 | ||||
| 16071 | case PPC::BI__builtin_unpack_vector_int128: { | |||
| 16072 | ConstantInt *Index = cast<ConstantInt>(Ops[1]); | |||
| 16073 | Value *Unpacked = Builder.CreateBitCast( | |||
| 16074 | Ops[0], llvm::FixedVectorType::get(ConvertType(E->getType()), 2)); | |||
| 16075 | ||||
| 16076 | if (getTarget().isLittleEndian()) | |||
| 16077 | Index = ConstantInt::get(Index->getType(), 1 - Index->getZExtValue()); | |||
| 16078 | ||||
| 16079 | return Builder.CreateExtractElement(Unpacked, Index); | |||
| 16080 | } | |||
| 16081 | ||||
| 16082 | case PPC::BI__builtin_ppc_sthcx: { | |||
| 16083 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_sthcx); | |||
| 16084 | Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy); | |||
| 16085 | Ops[1] = Builder.CreateSExt(Ops[1], Int32Ty); | |||
| 16086 | return Builder.CreateCall(F, Ops); | |||
| 16087 | } | |||
| 16088 | ||||
| 16089 | // The PPC MMA builtins take a pointer to a __vector_quad as an argument. | |||
| 16090 | // Some of the MMA instructions accumulate their result into an existing | |||
| 16091 | // accumulator whereas the others generate a new accumulator. So we need to | |||
| 16092 | // use custom code generation to expand a builtin call with a pointer to a | |||
| 16093 | // load (if the corresponding instruction accumulates its result) followed by | |||
| 16094 | // the call to the intrinsic and a store of the result. | |||
| 16095 | #define CUSTOM_BUILTIN(Name, Intr, Types, Accumulate) \ | |||
| 16096 | case PPC::BI__builtin_##Name: | |||
| 16097 | #include "clang/Basic/BuiltinsPPC.def" | |||
| 16098 | { | |||
| 16099 | // The first argument of these two builtins is a pointer used to store their | |||
| 16100 | // result. However, the llvm intrinsics return their result in multiple | |||
| 16101 | // return values. So, here we emit code extracting these values from the | |||
| 16102 | // intrinsic results and storing them using that pointer. | |||
| 16103 | if (BuiltinID == PPC::BI__builtin_mma_disassemble_acc || | |||
| 16104 | BuiltinID == PPC::BI__builtin_vsx_disassemble_pair || | |||
| 16105 | BuiltinID == PPC::BI__builtin_mma_disassemble_pair) { | |||
| 16106 | unsigned NumVecs = 2; | |||
| 16107 | auto Intrinsic = Intrinsic::ppc_vsx_disassemble_pair; | |||
| 16108 | if (BuiltinID == PPC::BI__builtin_mma_disassemble_acc) { | |||
| 16109 | NumVecs = 4; | |||
| 16110 | Intrinsic = Intrinsic::ppc_mma_disassemble_acc; | |||
| 16111 | } | |||
| 16112 | llvm::Function *F = CGM.getIntrinsic(Intrinsic); | |||
| 16113 | Address Addr = EmitPointerWithAlignment(E->getArg(1)); | |||
| 16114 | Value *Vec = Builder.CreateLoad(Addr); | |||
| 16115 | Value *Call = Builder.CreateCall(F, {Vec}); | |||
| 16116 | llvm::Type *VTy = llvm::FixedVectorType::get(Int8Ty, 16); | |||
| 16117 | Value *Ptr = Builder.CreateBitCast(Ops[0], VTy->getPointerTo()); | |||
| 16118 | for (unsigned i=0; i<NumVecs; i++) { | |||
| 16119 | Value *Vec = Builder.CreateExtractValue(Call, i); | |||
| 16120 | llvm::ConstantInt* Index = llvm::ConstantInt::get(IntTy, i); | |||
| 16121 | Value *GEP = Builder.CreateInBoundsGEP(VTy, Ptr, Index); | |||
| 16122 | Builder.CreateAlignedStore(Vec, GEP, MaybeAlign(16)); | |||
| 16123 | } | |||
| 16124 | return Call; | |||
| 16125 | } | |||
| 16126 | if (BuiltinID == PPC::BI__builtin_vsx_build_pair || | |||
| 16127 | BuiltinID == PPC::BI__builtin_mma_build_acc) { | |||
| 16128 | // Reverse the order of the operands for LE, so the | |||
| 16129 | // same builtin call can be used on both LE and BE | |||
| 16130 | // without the need for the programmer to swap operands. | |||
| 16131 | // The operands are reversed starting from the second argument, | |||
| 16132 | // the first operand is the pointer to the pair/accumulator | |||
| 16133 | // that is being built. | |||
| 16134 | if (getTarget().isLittleEndian()) | |||
| 16135 | std::reverse(Ops.begin() + 1, Ops.end()); | |||
| 16136 | } | |||
| 16137 | bool Accumulate; | |||
| 16138 | switch (BuiltinID) { | |||
| 16139 | #define CUSTOM_BUILTIN(Name, Intr, Types, Acc) \ | |||
| 16140 | case PPC::BI__builtin_##Name: \ | |||
| 16141 | ID = Intrinsic::ppc_##Intr; \ | |||
| 16142 | Accumulate = Acc; \ | |||
| 16143 | break; | |||
| 16144 | #include "clang/Basic/BuiltinsPPC.def" | |||
| 16145 | } | |||
| 16146 | if (BuiltinID == PPC::BI__builtin_vsx_lxvp || | |||
| 16147 | BuiltinID == PPC::BI__builtin_vsx_stxvp || | |||
| 16148 | BuiltinID == PPC::BI__builtin_mma_lxvp || | |||
| 16149 | BuiltinID == PPC::BI__builtin_mma_stxvp) { | |||
| 16150 | if (BuiltinID == PPC::BI__builtin_vsx_lxvp || | |||
| 16151 | BuiltinID == PPC::BI__builtin_mma_lxvp) { | |||
| 16152 | Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy); | |||
| 16153 | Ops[0] = Builder.CreateGEP(Int8Ty, Ops[1], Ops[0]); | |||
| 16154 | } else { | |||
| 16155 | Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy); | |||
| 16156 | Ops[1] = Builder.CreateGEP(Int8Ty, Ops[2], Ops[1]); | |||
| 16157 | } | |||
| 16158 | Ops.pop_back(); | |||
| 16159 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
| 16160 | return Builder.CreateCall(F, Ops, ""); | |||
| 16161 | } | |||
| 16162 | SmallVector<Value*, 4> CallOps; | |||
| 16163 | if (Accumulate) { | |||
| 16164 | Address Addr = EmitPointerWithAlignment(E->getArg(0)); | |||
| 16165 | Value *Acc = Builder.CreateLoad(Addr); | |||
| 16166 | CallOps.push_back(Acc); | |||
| 16167 | } | |||
| 16168 | for (unsigned i=1; i<Ops.size(); i++) | |||
| 16169 | CallOps.push_back(Ops[i]); | |||
| 16170 | llvm::Function *F = CGM.getIntrinsic(ID); | |||
| 16171 | Value *Call = Builder.CreateCall(F, CallOps); | |||
| 16172 | return Builder.CreateAlignedStore(Call, Ops[0], MaybeAlign(64)); | |||
| 16173 | } | |||
| 16174 | ||||
| 16175 | case PPC::BI__builtin_ppc_compare_and_swap: | |||
| 16176 | case PPC::BI__builtin_ppc_compare_and_swaplp: { | |||
| 16177 | Address Addr = EmitPointerWithAlignment(E->getArg(0)); | |||
| 16178 | Address OldValAddr = EmitPointerWithAlignment(E->getArg(1)); | |||
| 16179 | Value *OldVal = Builder.CreateLoad(OldValAddr); | |||
| 16180 | QualType AtomicTy = E->getArg(0)->getType()->getPointeeType(); | |||
| 16181 | LValue LV = MakeAddrLValue(Addr, AtomicTy); | |||
| 16182 | auto Pair = EmitAtomicCompareExchange( | |||
| 16183 | LV, RValue::get(OldVal), RValue::get(Ops[2]), E->getExprLoc(), | |||
| 16184 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Monotonic, true); | |||
| 16185 | // Unlike c11's atomic_compare_exchange, accroding to | |||
| 16186 | // https://www.ibm.com/docs/en/xl-c-and-cpp-aix/16.1?topic=functions-compare-swap-compare-swaplp | |||
| 16187 | // > In either case, the contents of the memory location specified by addr | |||
| 16188 | // > are copied into the memory location specified by old_val_addr. | |||
| 16189 | // But it hasn't specified storing to OldValAddr is atomic or not and | |||
| 16190 | // which order to use. Now following XL's codegen, treat it as a normal | |||
| 16191 | // store. | |||
| 16192 | Value *LoadedVal = Pair.first.getScalarVal(); | |||
| 16193 | Builder.CreateStore(LoadedVal, OldValAddr); | |||
| 16194 | return Builder.CreateZExt(Pair.second, Builder.getInt32Ty()); | |||
| 16195 | } | |||
| 16196 | case PPC::BI__builtin_ppc_fetch_and_add: | |||
| 16197 | case PPC::BI__builtin_ppc_fetch_and_addlp: { | |||
| 16198 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, | |||
| 16199 | llvm::AtomicOrdering::Monotonic); | |||
| 16200 | } | |||
| 16201 | case PPC::BI__builtin_ppc_fetch_and_and: | |||
| 16202 | case PPC::BI__builtin_ppc_fetch_and_andlp: { | |||
| 16203 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, | |||
| 16204 | llvm::AtomicOrdering::Monotonic); | |||
| 16205 | } | |||
| 16206 | ||||
| 16207 | case PPC::BI__builtin_ppc_fetch_and_or: | |||
| 16208 | case PPC::BI__builtin_ppc_fetch_and_orlp: { | |||
| 16209 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, | |||
| 16210 | llvm::AtomicOrdering::Monotonic); | |||
| 16211 | } | |||
| 16212 | case PPC::BI__builtin_ppc_fetch_and_swap: | |||
| 16213 | case PPC::BI__builtin_ppc_fetch_and_swaplp: { | |||
| 16214 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, | |||
| 16215 | llvm::AtomicOrdering::Monotonic); | |||
| 16216 | } | |||
| 16217 | case PPC::BI__builtin_ppc_ldarx: | |||
| 16218 | case PPC::BI__builtin_ppc_lwarx: | |||
| 16219 | case PPC::BI__builtin_ppc_lharx: | |||
| 16220 | case PPC::BI__builtin_ppc_lbarx: | |||
| 16221 | return emitPPCLoadReserveIntrinsic(*this, BuiltinID, E); | |||
| 16222 | case PPC::BI__builtin_ppc_mfspr: { | |||
| 16223 | llvm::Type *RetType = CGM.getDataLayout().getTypeSizeInBits(VoidPtrTy) == 32 | |||
| 16224 | ? Int32Ty | |||
| 16225 | : Int64Ty; | |||
| 16226 | Function *F = CGM.getIntrinsic(Intrinsic::ppc_mfspr, RetType); | |||
| 16227 | return Builder.CreateCall(F, Ops); | |||
| 16228 | } | |||
| 16229 | case PPC::BI__builtin_ppc_mtspr: { | |||
| 16230 | llvm::Type *RetType = CGM.getDataLayout().getTypeSizeInBits(VoidPtrTy) == 32 | |||
| 16231 | ? Int32Ty | |||
| 16232 | : Int64Ty; | |||
| 16233 | Function *F = CGM.getIntrinsic(Intrinsic::ppc_mtspr, RetType); | |||
| 16234 | return Builder.CreateCall(F, Ops); | |||
| 16235 | } | |||
| 16236 | case PPC::BI__builtin_ppc_popcntb: { | |||
| 16237 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); | |||
| 16238 | llvm::Type *ArgType = ArgValue->getType(); | |||
| 16239 | Function *F = CGM.getIntrinsic(Intrinsic::ppc_popcntb, {ArgType, ArgType}); | |||
| 16240 | return Builder.CreateCall(F, Ops, "popcntb"); | |||
| 16241 | } | |||
| 16242 | case PPC::BI__builtin_ppc_mtfsf: { | |||
| 16243 | // The builtin takes a uint32 that needs to be cast to an | |||
| 16244 | // f64 to be passed to the intrinsic. | |||
| 16245 | Value *Cast = Builder.CreateUIToFP(Ops[1], DoubleTy); | |||
| 16246 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_mtfsf); | |||
| 16247 | return Builder.CreateCall(F, {Ops[0], Cast}, ""); | |||
| 16248 | } | |||
| 16249 | ||||
| 16250 | case PPC::BI__builtin_ppc_swdiv_nochk: | |||
| 16251 | case PPC::BI__builtin_ppc_swdivs_nochk: { | |||
| 16252 | FastMathFlags FMF = Builder.getFastMathFlags(); | |||
| 16253 | Builder.getFastMathFlags().setFast(); | |||
| 16254 | Value *FDiv = Builder.CreateFDiv(Ops[0], Ops[1], "swdiv_nochk"); | |||
| 16255 | Builder.getFastMathFlags() &= (FMF); | |||
| 16256 | return FDiv; | |||
| 16257 | } | |||
| 16258 | case PPC::BI__builtin_ppc_fric: | |||
| 16259 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin( | |||
| 16260 | *this, E, Intrinsic::rint, | |||
| 16261 | Intrinsic::experimental_constrained_rint)) | |||
| 16262 | .getScalarVal(); | |||
| 16263 | case PPC::BI__builtin_ppc_frim: | |||
| 16264 | case PPC::BI__builtin_ppc_frims: | |||
| 16265 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin( | |||
| 16266 | *this, E, Intrinsic::floor, | |||
| 16267 | Intrinsic::experimental_constrained_floor)) | |||
| 16268 | .getScalarVal(); | |||
| 16269 | case PPC::BI__builtin_ppc_frin: | |||
| 16270 | case PPC::BI__builtin_ppc_frins: | |||
| 16271 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin( | |||
| 16272 | *this, E, Intrinsic::round, | |||
| 16273 | Intrinsic::experimental_constrained_round)) | |||
| 16274 | .getScalarVal(); | |||
| 16275 | case PPC::BI__builtin_ppc_frip: | |||
| 16276 | case PPC::BI__builtin_ppc_frips: | |||
| 16277 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin( | |||
| 16278 | *this, E, Intrinsic::ceil, | |||
| 16279 | Intrinsic::experimental_constrained_ceil)) | |||
| 16280 | .getScalarVal(); | |||
| 16281 | case PPC::BI__builtin_ppc_friz: | |||
| 16282 | case PPC::BI__builtin_ppc_frizs: | |||
| 16283 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin( | |||
| 16284 | *this, E, Intrinsic::trunc, | |||
| 16285 | Intrinsic::experimental_constrained_trunc)) | |||
| 16286 | .getScalarVal(); | |||
| 16287 | case PPC::BI__builtin_ppc_fsqrt: | |||
| 16288 | case PPC::BI__builtin_ppc_fsqrts: | |||
| 16289 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin( | |||
| 16290 | *this, E, Intrinsic::sqrt, | |||
| 16291 | Intrinsic::experimental_constrained_sqrt)) | |||
| 16292 | .getScalarVal(); | |||
| 16293 | case PPC::BI__builtin_ppc_test_data_class: { | |||
| 16294 | llvm::Type *ArgType = EmitScalarExpr(E->getArg(0))->getType(); | |||
| 16295 | unsigned IntrinsicID; | |||
| 16296 | if (ArgType->isDoubleTy()) | |||
| 16297 | IntrinsicID = Intrinsic::ppc_test_data_class_d; | |||
| 16298 | else if (ArgType->isFloatTy()) | |||
| 16299 | IntrinsicID = Intrinsic::ppc_test_data_class_f; | |||
| 16300 | else | |||
| 16301 | llvm_unreachable("Invalid Argument Type")::llvm::llvm_unreachable_internal("Invalid Argument Type", "clang/lib/CodeGen/CGBuiltin.cpp" , 16301); | |||
| 16302 | return Builder.CreateCall(CGM.getIntrinsic(IntrinsicID), Ops, | |||
| 16303 | "test_data_class"); | |||
| 16304 | } | |||
| 16305 | case PPC::BI__builtin_ppc_maxfe: | |||
| 16306 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::ppc_maxfe), Ops); | |||
| 16307 | case PPC::BI__builtin_ppc_maxfl: | |||
| 16308 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::ppc_maxfl), Ops); | |||
| 16309 | case PPC::BI__builtin_ppc_maxfs: | |||
| 16310 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::ppc_maxfs), Ops); | |||
| 16311 | case PPC::BI__builtin_ppc_minfe: | |||
| 16312 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::ppc_minfe), Ops); | |||
| 16313 | case PPC::BI__builtin_ppc_minfl: | |||
| 16314 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::ppc_minfl), Ops); | |||
| 16315 | case PPC::BI__builtin_ppc_minfs: | |||
| 16316 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::ppc_minfs), Ops); | |||
| 16317 | case PPC::BI__builtin_ppc_swdiv: | |||
| 16318 | case PPC::BI__builtin_ppc_swdivs: | |||
| 16319 | return Builder.CreateFDiv(Ops[0], Ops[1], "swdiv"); | |||
| 16320 | } | |||
| 16321 | } | |||
| 16322 | ||||
| 16323 | namespace { | |||
| 16324 | // If \p E is not null pointer, insert address space cast to match return | |||
| 16325 | // type of \p E if necessary. | |||
| 16326 | Value *EmitAMDGPUDispatchPtr(CodeGenFunction &CGF, | |||
| 16327 | const CallExpr *E = nullptr) { | |||
| 16328 | auto *F = CGF.CGM.getIntrinsic(Intrinsic::amdgcn_dispatch_ptr); | |||
| 16329 | auto *Call = CGF.Builder.CreateCall(F); | |||
| 16330 | Call->addRetAttr( | |||
| 16331 | Attribute::getWithDereferenceableBytes(Call->getContext(), 64)); | |||
| 16332 | Call->addRetAttr(Attribute::getWithAlignment(Call->getContext(), Align(4))); | |||
| 16333 | if (!E) | |||
| 16334 | return Call; | |||
| 16335 | QualType BuiltinRetType = E->getType(); | |||
| 16336 | auto *RetTy = cast<llvm::PointerType>(CGF.ConvertType(BuiltinRetType)); | |||
| 16337 | if (RetTy == Call->getType()) | |||
| 16338 | return Call; | |||
| 16339 | return CGF.Builder.CreateAddrSpaceCast(Call, RetTy); | |||
| 16340 | } | |||
| 16341 | ||||
| 16342 | Value *EmitAMDGPUImplicitArgPtr(CodeGenFunction &CGF) { | |||
| 16343 | auto *F = CGF.CGM.getIntrinsic(Intrinsic::amdgcn_implicitarg_ptr); | |||
| 16344 | auto *Call = CGF.Builder.CreateCall(F); | |||
| 16345 | Call->addRetAttr( | |||
| 16346 | Attribute::getWithDereferenceableBytes(Call->getContext(), 256)); | |||
| 16347 | Call->addRetAttr(Attribute::getWithAlignment(Call->getContext(), Align(8))); | |||
| 16348 | return Call; | |||
| 16349 | } | |||
| 16350 | ||||
| 16351 | // \p Index is 0, 1, and 2 for x, y, and z dimension, respectively. | |||
| 16352 | Value *EmitAMDGPUWorkGroupSize(CodeGenFunction &CGF, unsigned Index) { | |||
| 16353 | bool IsCOV_5 = CGF.getTarget().getTargetOpts().CodeObjectVersion == | |||
| 16354 | clang::TargetOptions::COV_5; | |||
| 16355 | Constant *Offset; | |||
| 16356 | Value *DP; | |||
| 16357 | if (IsCOV_5) { | |||
| 16358 | // Indexing the implicit kernarg segment. | |||
| 16359 | Offset = llvm::ConstantInt::get(CGF.Int32Ty, 12 + Index * 2); | |||
| 16360 | DP = EmitAMDGPUImplicitArgPtr(CGF); | |||
| 16361 | } else { | |||
| 16362 | // Indexing the HSA kernel_dispatch_packet struct. | |||
| 16363 | Offset = llvm::ConstantInt::get(CGF.Int32Ty, 4 + Index * 2); | |||
| 16364 | DP = EmitAMDGPUDispatchPtr(CGF); | |||
| 16365 | } | |||
| 16366 | ||||
| 16367 | auto *GEP = CGF.Builder.CreateGEP(CGF.Int8Ty, DP, Offset); | |||
| 16368 | auto *DstTy = | |||
| 16369 | CGF.Int16Ty->getPointerTo(GEP->getType()->getPointerAddressSpace()); | |||
| 16370 | auto *Cast = CGF.Builder.CreateBitCast(GEP, DstTy); | |||
| 16371 | auto *LD = CGF.Builder.CreateLoad( | |||
| 16372 | Address(Cast, CGF.Int16Ty, CharUnits::fromQuantity(2))); | |||
| 16373 | llvm::MDBuilder MDHelper(CGF.getLLVMContext()); | |||
| 16374 | llvm::MDNode *RNode = MDHelper.createRange(APInt(16, 1), | |||
| 16375 | APInt(16, CGF.getTarget().getMaxOpenCLWorkGroupSize() + 1)); | |||
| 16376 | LD->setMetadata(llvm::LLVMContext::MD_range, RNode); | |||
| 16377 | LD->setMetadata(llvm::LLVMContext::MD_invariant_load, | |||
| 16378 | llvm::MDNode::get(CGF.getLLVMContext(), None)); | |||
| 16379 | return LD; | |||
| 16380 | } | |||
| 16381 | ||||
| 16382 | // \p Index is 0, 1, and 2 for x, y, and z dimension, respectively. | |||
| 16383 | Value *EmitAMDGPUGridSize(CodeGenFunction &CGF, unsigned Index) { | |||
| 16384 | const unsigned XOffset = 12; | |||
| 16385 | auto *DP = EmitAMDGPUDispatchPtr(CGF); | |||
| 16386 | // Indexing the HSA kernel_dispatch_packet struct. | |||
| 16387 | auto *Offset = llvm::ConstantInt::get(CGF.Int32Ty, XOffset + Index * 4); | |||
| 16388 | auto *GEP = CGF.Builder.CreateGEP(CGF.Int8Ty, DP, Offset); | |||
| 16389 | auto *DstTy = | |||
| 16390 | CGF.Int32Ty->getPointerTo(GEP->getType()->getPointerAddressSpace()); | |||
| 16391 | auto *Cast = CGF.Builder.CreateBitCast(GEP, DstTy); | |||
| 16392 | auto *LD = CGF.Builder.CreateLoad( | |||
| 16393 | Address(Cast, CGF.Int32Ty, CharUnits::fromQuantity(4))); | |||
| 16394 | LD->setMetadata(llvm::LLVMContext::MD_invariant_load, | |||
| 16395 | llvm::MDNode::get(CGF.getLLVMContext(), None)); | |||
| 16396 | return LD; | |||
| 16397 | } | |||
| 16398 | } // namespace | |||
| 16399 | ||||
| 16400 | // For processing memory ordering and memory scope arguments of various | |||
| 16401 | // amdgcn builtins. | |||
| 16402 | // \p Order takes a C++11 comptabile memory-ordering specifier and converts | |||
| 16403 | // it into LLVM's memory ordering specifier using atomic C ABI, and writes | |||
| 16404 | // to \p AO. \p Scope takes a const char * and converts it into AMDGCN | |||
| 16405 | // specific SyncScopeID and writes it to \p SSID. | |||
| 16406 | bool CodeGenFunction::ProcessOrderScopeAMDGCN(Value *Order, Value *Scope, | |||
| 16407 | llvm::AtomicOrdering &AO, | |||
| 16408 | llvm::SyncScope::ID &SSID) { | |||
| 16409 | if (isa<llvm::ConstantInt>(Order)) { | |||
| 16410 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); | |||
| 16411 | ||||
| 16412 | // Map C11/C++11 memory ordering to LLVM memory ordering | |||
| 16413 | assert(llvm::isValidAtomicOrderingCABI(ord))(static_cast <bool> (llvm::isValidAtomicOrderingCABI(ord )) ? void (0) : __assert_fail ("llvm::isValidAtomicOrderingCABI(ord)" , "clang/lib/CodeGen/CGBuiltin.cpp", 16413, __extension__ __PRETTY_FUNCTION__ )); | |||
| 16414 | switch (static_cast<llvm::AtomicOrderingCABI>(ord)) { | |||
| 16415 | case llvm::AtomicOrderingCABI::acquire: | |||
| 16416 | case llvm::AtomicOrderingCABI::consume: | |||
| 16417 | AO = llvm::AtomicOrdering::Acquire; | |||
| 16418 | break; | |||
| 16419 | case llvm::AtomicOrderingCABI::release: | |||
| 16420 | AO = llvm::AtomicOrdering::Release; | |||
| 16421 | break; | |||
| 16422 | case llvm::AtomicOrderingCABI::acq_rel: | |||
| 16423 | AO = llvm::AtomicOrdering::AcquireRelease; | |||
| 16424 | break; | |||
| 16425 | case llvm::AtomicOrderingCABI::seq_cst: | |||
| 16426 | AO = llvm::AtomicOrdering::SequentiallyConsistent; | |||
| 16427 | break; | |||
| 16428 | case llvm::AtomicOrderingCABI::relaxed: | |||
| 16429 | AO = llvm::AtomicOrdering::Monotonic; | |||
| 16430 | break; | |||
| 16431 | } | |||
| 16432 | ||||
| 16433 | StringRef scp; | |||
| 16434 | llvm::getConstantStringInfo(Scope, scp); | |||
| 16435 | SSID = getLLVMContext().getOrInsertSyncScopeID(scp); | |||
| 16436 | return true; | |||
| 16437 | } | |||
| 16438 | return false; | |||
| 16439 | } | |||
| 16440 | ||||
| 16441 | Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID, | |||
| 16442 | const CallExpr *E) { | |||
| 16443 | llvm::AtomicOrdering AO = llvm::AtomicOrdering::SequentiallyConsistent; | |||
| 16444 | llvm::SyncScope::ID SSID; | |||
| 16445 | switch (BuiltinID) { | |||
| 16446 | case AMDGPU::BI__builtin_amdgcn_div_scale: | |||
| 16447 | case AMDGPU::BI__builtin_amdgcn_div_scalef: { | |||
| 16448 | // Translate from the intrinsics's struct return to the builtin's out | |||
| 16449 | // argument. | |||
| 16450 | ||||
| 16451 | Address FlagOutPtr = EmitPointerWithAlignment(E->getArg(3)); | |||
| 16452 | ||||
| 16453 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16454 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 16455 | llvm::Value *Z = EmitScalarExpr(E->getArg(2)); | |||
| 16456 | ||||
| 16457 | llvm::Function *Callee = CGM.getIntrinsic(Intrinsic::amdgcn_div_scale, | |||
| 16458 | X->getType()); | |||
| 16459 | ||||
| 16460 | llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z}); | |||
| 16461 | ||||
| 16462 | llvm::Value *Result = Builder.CreateExtractValue(Tmp, 0); | |||
| 16463 | llvm::Value *Flag = Builder.CreateExtractValue(Tmp, 1); | |||
| 16464 | ||||
| 16465 | llvm::Type *RealFlagType = FlagOutPtr.getElementType(); | |||
| 16466 | ||||
| 16467 | llvm::Value *FlagExt = Builder.CreateZExt(Flag, RealFlagType); | |||
| 16468 | Builder.CreateStore(FlagExt, FlagOutPtr); | |||
| 16469 | return Result; | |||
| 16470 | } | |||
| 16471 | case AMDGPU::BI__builtin_amdgcn_div_fmas: | |||
| 16472 | case AMDGPU::BI__builtin_amdgcn_div_fmasf: { | |||
| 16473 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16474 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
| 16475 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
| 16476 | llvm::Value *Src3 = EmitScalarExpr(E->getArg(3)); | |||
| 16477 | ||||
| 16478 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_div_fmas, | |||
| 16479 | Src0->getType()); | |||
| 16480 | llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3); | |||
| 16481 | return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool}); | |||
| 16482 | } | |||
| 16483 | ||||
| 16484 | case AMDGPU::BI__builtin_amdgcn_ds_swizzle: | |||
| 16485 | return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_ds_swizzle); | |||
| 16486 | case AMDGPU::BI__builtin_amdgcn_mov_dpp8: | |||
| 16487 | return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_mov_dpp8); | |||
| 16488 | case AMDGPU::BI__builtin_amdgcn_mov_dpp: | |||
| 16489 | case AMDGPU::BI__builtin_amdgcn_update_dpp: { | |||
| 16490 | llvm::SmallVector<llvm::Value *, 6> Args; | |||
| 16491 | for (unsigned I = 0; I != E->getNumArgs(); ++I) | |||
| 16492 | Args.push_back(EmitScalarExpr(E->getArg(I))); | |||
| 16493 | assert(Args.size() == 5 || Args.size() == 6)(static_cast <bool> (Args.size() == 5 || Args.size() == 6) ? void (0) : __assert_fail ("Args.size() == 5 || Args.size() == 6" , "clang/lib/CodeGen/CGBuiltin.cpp", 16493, __extension__ __PRETTY_FUNCTION__ )); | |||
| 16494 | if (Args.size() == 5) | |||
| 16495 | Args.insert(Args.begin(), llvm::UndefValue::get(Args[0]->getType())); | |||
| 16496 | Function *F = | |||
| 16497 | CGM.getIntrinsic(Intrinsic::amdgcn_update_dpp, Args[0]->getType()); | |||
| 16498 | return Builder.CreateCall(F, Args); | |||
| 16499 | } | |||
| 16500 | case AMDGPU::BI__builtin_amdgcn_div_fixup: | |||
| 16501 | case AMDGPU::BI__builtin_amdgcn_div_fixupf: | |||
| 16502 | case AMDGPU::BI__builtin_amdgcn_div_fixuph: | |||
| 16503 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_div_fixup); | |||
| 16504 | case AMDGPU::BI__builtin_amdgcn_trig_preop: | |||
| 16505 | case AMDGPU::BI__builtin_amdgcn_trig_preopf: | |||
| 16506 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_trig_preop); | |||
| 16507 | case AMDGPU::BI__builtin_amdgcn_rcp: | |||
| 16508 | case AMDGPU::BI__builtin_amdgcn_rcpf: | |||
| 16509 | case AMDGPU::BI__builtin_amdgcn_rcph: | |||
| 16510 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rcp); | |||
| 16511 | case AMDGPU::BI__builtin_amdgcn_sqrt: | |||
| 16512 | case AMDGPU::BI__builtin_amdgcn_sqrtf: | |||
| 16513 | case AMDGPU::BI__builtin_amdgcn_sqrth: | |||
| 16514 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sqrt); | |||
| 16515 | case AMDGPU::BI__builtin_amdgcn_rsq: | |||
| 16516 | case AMDGPU::BI__builtin_amdgcn_rsqf: | |||
| 16517 | case AMDGPU::BI__builtin_amdgcn_rsqh: | |||
| 16518 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq); | |||
| 16519 | case AMDGPU::BI__builtin_amdgcn_rsq_clamp: | |||
| 16520 | case AMDGPU::BI__builtin_amdgcn_rsq_clampf: | |||
| 16521 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq_clamp); | |||
| 16522 | case AMDGPU::BI__builtin_amdgcn_sinf: | |||
| 16523 | case AMDGPU::BI__builtin_amdgcn_sinh: | |||
| 16524 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sin); | |||
| 16525 | case AMDGPU::BI__builtin_amdgcn_cosf: | |||
| 16526 | case AMDGPU::BI__builtin_amdgcn_cosh: | |||
| 16527 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_cos); | |||
| 16528 | case AMDGPU::BI__builtin_amdgcn_dispatch_ptr: | |||
| 16529 | return EmitAMDGPUDispatchPtr(*this, E); | |||
| 16530 | case AMDGPU::BI__builtin_amdgcn_log_clampf: | |||
| 16531 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_log_clamp); | |||
| 16532 | case AMDGPU::BI__builtin_amdgcn_ldexp: | |||
| 16533 | case AMDGPU::BI__builtin_amdgcn_ldexpf: | |||
| 16534 | case AMDGPU::BI__builtin_amdgcn_ldexph: | |||
| 16535 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_ldexp); | |||
| 16536 | case AMDGPU::BI__builtin_amdgcn_frexp_mant: | |||
| 16537 | case AMDGPU::BI__builtin_amdgcn_frexp_mantf: | |||
| 16538 | case AMDGPU::BI__builtin_amdgcn_frexp_manth: | |||
| 16539 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_frexp_mant); | |||
| 16540 | case AMDGPU::BI__builtin_amdgcn_frexp_exp: | |||
| 16541 | case AMDGPU::BI__builtin_amdgcn_frexp_expf: { | |||
| 16542 | Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16543 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp, | |||
| 16544 | { Builder.getInt32Ty(), Src0->getType() }); | |||
| 16545 | return Builder.CreateCall(F, Src0); | |||
| 16546 | } | |||
| 16547 | case AMDGPU::BI__builtin_amdgcn_frexp_exph: { | |||
| 16548 | Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16549 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp, | |||
| 16550 | { Builder.getInt16Ty(), Src0->getType() }); | |||
| 16551 | return Builder.CreateCall(F, Src0); | |||
| 16552 | } | |||
| 16553 | case AMDGPU::BI__builtin_amdgcn_fract: | |||
| 16554 | case AMDGPU::BI__builtin_amdgcn_fractf: | |||
| 16555 | case AMDGPU::BI__builtin_amdgcn_fracth: | |||
| 16556 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_fract); | |||
| 16557 | case AMDGPU::BI__builtin_amdgcn_lerp: | |||
| 16558 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_lerp); | |||
| 16559 | case AMDGPU::BI__builtin_amdgcn_ubfe: | |||
| 16560 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_ubfe); | |||
| 16561 | case AMDGPU::BI__builtin_amdgcn_sbfe: | |||
| 16562 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_sbfe); | |||
| 16563 | case AMDGPU::BI__builtin_amdgcn_uicmp: | |||
| 16564 | case AMDGPU::BI__builtin_amdgcn_uicmpl: | |||
| 16565 | case AMDGPU::BI__builtin_amdgcn_sicmp: | |||
| 16566 | case AMDGPU::BI__builtin_amdgcn_sicmpl: { | |||
| 16567 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16568 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
| 16569 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
| 16570 | ||||
| 16571 | // FIXME-GFX10: How should 32 bit mask be handled? | |||
| 16572 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_icmp, | |||
| 16573 | { Builder.getInt64Ty(), Src0->getType() }); | |||
| 16574 | return Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
| 16575 | } | |||
| 16576 | case AMDGPU::BI__builtin_amdgcn_fcmp: | |||
| 16577 | case AMDGPU::BI__builtin_amdgcn_fcmpf: { | |||
| 16578 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16579 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
| 16580 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
| 16581 | ||||
| 16582 | // FIXME-GFX10: How should 32 bit mask be handled? | |||
| 16583 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_fcmp, | |||
| 16584 | { Builder.getInt64Ty(), Src0->getType() }); | |||
| 16585 | return Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
| 16586 | } | |||
| 16587 | case AMDGPU::BI__builtin_amdgcn_class: | |||
| 16588 | case AMDGPU::BI__builtin_amdgcn_classf: | |||
| 16589 | case AMDGPU::BI__builtin_amdgcn_classh: | |||
| 16590 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_class); | |||
| 16591 | case AMDGPU::BI__builtin_amdgcn_fmed3f: | |||
| 16592 | case AMDGPU::BI__builtin_amdgcn_fmed3h: | |||
| 16593 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fmed3); | |||
| 16594 | case AMDGPU::BI__builtin_amdgcn_ds_append: | |||
| 16595 | case AMDGPU::BI__builtin_amdgcn_ds_consume: { | |||
| 16596 | Intrinsic::ID Intrin = BuiltinID == AMDGPU::BI__builtin_amdgcn_ds_append ? | |||
| 16597 | Intrinsic::amdgcn_ds_append : Intrinsic::amdgcn_ds_consume; | |||
| 16598 | Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16599 | Function *F = CGM.getIntrinsic(Intrin, { Src0->getType() }); | |||
| 16600 | return Builder.CreateCall(F, { Src0, Builder.getFalse() }); | |||
| 16601 | } | |||
| 16602 | case AMDGPU::BI__builtin_amdgcn_ds_faddf: | |||
| 16603 | case AMDGPU::BI__builtin_amdgcn_ds_fminf: | |||
| 16604 | case AMDGPU::BI__builtin_amdgcn_ds_fmaxf: { | |||
| 16605 | Intrinsic::ID Intrin; | |||
| 16606 | switch (BuiltinID) { | |||
| 16607 | case AMDGPU::BI__builtin_amdgcn_ds_faddf: | |||
| 16608 | Intrin = Intrinsic::amdgcn_ds_fadd; | |||
| 16609 | break; | |||
| 16610 | case AMDGPU::BI__builtin_amdgcn_ds_fminf: | |||
| 16611 | Intrin = Intrinsic::amdgcn_ds_fmin; | |||
| 16612 | break; | |||
| 16613 | case AMDGPU::BI__builtin_amdgcn_ds_fmaxf: | |||
| 16614 | Intrin = Intrinsic::amdgcn_ds_fmax; | |||
| 16615 | break; | |||
| 16616 | } | |||
| 16617 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16618 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
| 16619 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
| 16620 | llvm::Value *Src3 = EmitScalarExpr(E->getArg(3)); | |||
| 16621 | llvm::Value *Src4 = EmitScalarExpr(E->getArg(4)); | |||
| 16622 | llvm::Function *F = CGM.getIntrinsic(Intrin, { Src1->getType() }); | |||
| 16623 | llvm::FunctionType *FTy = F->getFunctionType(); | |||
| 16624 | llvm::Type *PTy = FTy->getParamType(0); | |||
| 16625 | Src0 = Builder.CreatePointerBitCastOrAddrSpaceCast(Src0, PTy); | |||
| 16626 | return Builder.CreateCall(F, { Src0, Src1, Src2, Src3, Src4 }); | |||
| 16627 | } | |||
| 16628 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64: | |||
| 16629 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32: | |||
| 16630 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16: | |||
| 16631 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64: | |||
| 16632 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64: | |||
| 16633 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64: | |||
| 16634 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64: | |||
| 16635 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: | |||
| 16636 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f32: | |||
| 16637 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2f16: { | |||
| 16638 | Intrinsic::ID IID; | |||
| 16639 | llvm::Type *ArgTy = llvm::Type::getDoubleTy(getLLVMContext()); | |||
| 16640 | switch (BuiltinID) { | |||
| 16641 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32: | |||
| 16642 | ArgTy = llvm::Type::getFloatTy(getLLVMContext()); | |||
| 16643 | IID = Intrinsic::amdgcn_global_atomic_fadd; | |||
| 16644 | break; | |||
| 16645 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16: | |||
| 16646 | ArgTy = llvm::FixedVectorType::get( | |||
| 16647 | llvm::Type::getHalfTy(getLLVMContext()), 2); | |||
| 16648 | IID = Intrinsic::amdgcn_global_atomic_fadd; | |||
| 16649 | break; | |||
| 16650 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64: | |||
| 16651 | IID = Intrinsic::amdgcn_global_atomic_fadd; | |||
| 16652 | break; | |||
| 16653 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64: | |||
| 16654 | IID = Intrinsic::amdgcn_global_atomic_fmin; | |||
| 16655 | break; | |||
| 16656 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64: | |||
| 16657 | IID = Intrinsic::amdgcn_global_atomic_fmax; | |||
| 16658 | break; | |||
| 16659 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64: | |||
| 16660 | IID = Intrinsic::amdgcn_flat_atomic_fadd; | |||
| 16661 | break; | |||
| 16662 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64: | |||
| 16663 | IID = Intrinsic::amdgcn_flat_atomic_fmin; | |||
| 16664 | break; | |||
| 16665 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: | |||
| 16666 | IID = Intrinsic::amdgcn_flat_atomic_fmax; | |||
| 16667 | break; | |||
| 16668 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f32: | |||
| 16669 | ArgTy = llvm::Type::getFloatTy(getLLVMContext()); | |||
| 16670 | IID = Intrinsic::amdgcn_flat_atomic_fadd; | |||
| 16671 | break; | |||
| 16672 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2f16: | |||
| 16673 | ArgTy = llvm::FixedVectorType::get( | |||
| 16674 | llvm::Type::getHalfTy(getLLVMContext()), 2); | |||
| 16675 | IID = Intrinsic::amdgcn_flat_atomic_fadd; | |||
| 16676 | break; | |||
| 16677 | } | |||
| 16678 | llvm::Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
| 16679 | llvm::Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 16680 | llvm::Function *F = | |||
| 16681 | CGM.getIntrinsic(IID, {ArgTy, Addr->getType(), Val->getType()}); | |||
| 16682 | return Builder.CreateCall(F, {Addr, Val}); | |||
| 16683 | } | |||
| 16684 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2bf16: | |||
| 16685 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2bf16: { | |||
| 16686 | Intrinsic::ID IID; | |||
| 16687 | switch (BuiltinID) { | |||
| 16688 | case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2bf16: | |||
| 16689 | IID = Intrinsic::amdgcn_global_atomic_fadd_v2bf16; | |||
| 16690 | break; | |||
| 16691 | case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2bf16: | |||
| 16692 | IID = Intrinsic::amdgcn_flat_atomic_fadd_v2bf16; | |||
| 16693 | break; | |||
| 16694 | } | |||
| 16695 | llvm::Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
| 16696 | llvm::Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 16697 | llvm::Function *F = CGM.getIntrinsic(IID, {Addr->getType()}); | |||
| 16698 | return Builder.CreateCall(F, {Addr, Val}); | |||
| 16699 | } | |||
| 16700 | case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f64: | |||
| 16701 | case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f32: { | |||
| 16702 | Intrinsic::ID IID; | |||
| 16703 | llvm::Type *ArgTy; | |||
| 16704 | switch (BuiltinID) { | |||
| 16705 | case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f32: | |||
| 16706 | ArgTy = llvm::Type::getFloatTy(getLLVMContext()); | |||
| 16707 | IID = Intrinsic::amdgcn_ds_fadd; | |||
| 16708 | break; | |||
| 16709 | case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_f64: | |||
| 16710 | ArgTy = llvm::Type::getDoubleTy(getLLVMContext()); | |||
| 16711 | IID = Intrinsic::amdgcn_ds_fadd; | |||
| 16712 | break; | |||
| 16713 | } | |||
| 16714 | llvm::Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
| 16715 | llvm::Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 16716 | llvm::Constant *ZeroI32 = llvm::ConstantInt::getIntegerValue( | |||
| 16717 | llvm::Type::getInt32Ty(getLLVMContext()), APInt(32, 0, true)); | |||
| 16718 | llvm::Constant *ZeroI1 = llvm::ConstantInt::getIntegerValue( | |||
| 16719 | llvm::Type::getInt1Ty(getLLVMContext()), APInt(1, 0)); | |||
| 16720 | llvm::Function *F = CGM.getIntrinsic(IID, {ArgTy}); | |||
| 16721 | return Builder.CreateCall(F, {Addr, Val, ZeroI32, ZeroI32, ZeroI1}); | |||
| 16722 | } | |||
| 16723 | case AMDGPU::BI__builtin_amdgcn_read_exec: { | |||
| 16724 | CallInst *CI = cast<CallInst>( | |||
| 16725 | EmitSpecialRegisterBuiltin(*this, E, Int64Ty, Int64Ty, NormalRead, "exec")); | |||
| 16726 | CI->setConvergent(); | |||
| 16727 | return CI; | |||
| 16728 | } | |||
| 16729 | case AMDGPU::BI__builtin_amdgcn_read_exec_lo: | |||
| 16730 | case AMDGPU::BI__builtin_amdgcn_read_exec_hi: { | |||
| 16731 | StringRef RegName = BuiltinID == AMDGPU::BI__builtin_amdgcn_read_exec_lo ? | |||
| 16732 | "exec_lo" : "exec_hi"; | |||
| 16733 | CallInst *CI = cast<CallInst>( | |||
| 16734 | EmitSpecialRegisterBuiltin(*this, E, Int32Ty, Int32Ty, NormalRead, RegName)); | |||
| 16735 | CI->setConvergent(); | |||
| 16736 | return CI; | |||
| 16737 | } | |||
| 16738 | case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray: | |||
| 16739 | case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_h: | |||
| 16740 | case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_l: | |||
| 16741 | case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_lh: { | |||
| 16742 | llvm::Value *NodePtr = EmitScalarExpr(E->getArg(0)); | |||
| 16743 | llvm::Value *RayExtent = EmitScalarExpr(E->getArg(1)); | |||
| 16744 | llvm::Value *RayOrigin = EmitScalarExpr(E->getArg(2)); | |||
| 16745 | llvm::Value *RayDir = EmitScalarExpr(E->getArg(3)); | |||
| 16746 | llvm::Value *RayInverseDir = EmitScalarExpr(E->getArg(4)); | |||
| 16747 | llvm::Value *TextureDescr = EmitScalarExpr(E->getArg(5)); | |||
| 16748 | ||||
| 16749 | // The builtins take these arguments as vec4 where the last element is | |||
| 16750 | // ignored. The intrinsic takes them as vec3. | |||
| 16751 | RayOrigin = Builder.CreateShuffleVector(RayOrigin, RayOrigin, | |||
| 16752 | ArrayRef<int>{0, 1, 2}); | |||
| 16753 | RayDir = | |||
| 16754 | Builder.CreateShuffleVector(RayDir, RayDir, ArrayRef<int>{0, 1, 2}); | |||
| 16755 | RayInverseDir = Builder.CreateShuffleVector(RayInverseDir, RayInverseDir, | |||
| 16756 | ArrayRef<int>{0, 1, 2}); | |||
| 16757 | ||||
| 16758 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_image_bvh_intersect_ray, | |||
| 16759 | {NodePtr->getType(), RayDir->getType()}); | |||
| 16760 | return Builder.CreateCall(F, {NodePtr, RayExtent, RayOrigin, RayDir, | |||
| 16761 | RayInverseDir, TextureDescr}); | |||
| 16762 | } | |||
| 16763 | ||||
| 16764 | // amdgcn workitem | |||
| 16765 | case AMDGPU::BI__builtin_amdgcn_workitem_id_x: | |||
| 16766 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_x, 0, 1024); | |||
| 16767 | case AMDGPU::BI__builtin_amdgcn_workitem_id_y: | |||
| 16768 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_y, 0, 1024); | |||
| 16769 | case AMDGPU::BI__builtin_amdgcn_workitem_id_z: | |||
| 16770 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_z, 0, 1024); | |||
| 16771 | ||||
| 16772 | // amdgcn workgroup size | |||
| 16773 | case AMDGPU::BI__builtin_amdgcn_workgroup_size_x: | |||
| 16774 | return EmitAMDGPUWorkGroupSize(*this, 0); | |||
| 16775 | case AMDGPU::BI__builtin_amdgcn_workgroup_size_y: | |||
| 16776 | return EmitAMDGPUWorkGroupSize(*this, 1); | |||
| 16777 | case AMDGPU::BI__builtin_amdgcn_workgroup_size_z: | |||
| 16778 | return EmitAMDGPUWorkGroupSize(*this, 2); | |||
| 16779 | ||||
| 16780 | // amdgcn grid size | |||
| 16781 | case AMDGPU::BI__builtin_amdgcn_grid_size_x: | |||
| 16782 | return EmitAMDGPUGridSize(*this, 0); | |||
| 16783 | case AMDGPU::BI__builtin_amdgcn_grid_size_y: | |||
| 16784 | return EmitAMDGPUGridSize(*this, 1); | |||
| 16785 | case AMDGPU::BI__builtin_amdgcn_grid_size_z: | |||
| 16786 | return EmitAMDGPUGridSize(*this, 2); | |||
| 16787 | ||||
| 16788 | // r600 intrinsics | |||
| 16789 | case AMDGPU::BI__builtin_r600_recipsqrt_ieee: | |||
| 16790 | case AMDGPU::BI__builtin_r600_recipsqrt_ieeef: | |||
| 16791 | return emitUnaryBuiltin(*this, E, Intrinsic::r600_recipsqrt_ieee); | |||
| 16792 | case AMDGPU::BI__builtin_r600_read_tidig_x: | |||
| 16793 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_x, 0, 1024); | |||
| 16794 | case AMDGPU::BI__builtin_r600_read_tidig_y: | |||
| 16795 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_y, 0, 1024); | |||
| 16796 | case AMDGPU::BI__builtin_r600_read_tidig_z: | |||
| 16797 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_z, 0, 1024); | |||
| 16798 | case AMDGPU::BI__builtin_amdgcn_alignbit: { | |||
| 16799 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); | |||
| 16800 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); | |||
| 16801 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); | |||
| 16802 | Function *F = CGM.getIntrinsic(Intrinsic::fshr, Src0->getType()); | |||
| 16803 | return Builder.CreateCall(F, { Src0, Src1, Src2 }); | |||
| 16804 | } | |||
| 16805 | ||||
| 16806 | case AMDGPU::BI__builtin_amdgcn_fence: { | |||
| 16807 | if (ProcessOrderScopeAMDGCN(EmitScalarExpr(E->getArg(0)), | |||
| 16808 | EmitScalarExpr(E->getArg(1)), AO, SSID)) | |||
| 16809 | return Builder.CreateFence(AO, SSID); | |||
| 16810 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 16811 | } | |||
| 16812 | case AMDGPU::BI__builtin_amdgcn_atomic_inc32: | |||
| 16813 | case AMDGPU::BI__builtin_amdgcn_atomic_inc64: | |||
| 16814 | case AMDGPU::BI__builtin_amdgcn_atomic_dec32: | |||
| 16815 | case AMDGPU::BI__builtin_amdgcn_atomic_dec64: { | |||
| 16816 | unsigned BuiltinAtomicOp; | |||
| 16817 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16818 | ||||
| 16819 | switch (BuiltinID) { | |||
| 16820 | case AMDGPU::BI__builtin_amdgcn_atomic_inc32: | |||
| 16821 | case AMDGPU::BI__builtin_amdgcn_atomic_inc64: | |||
| 16822 | BuiltinAtomicOp = Intrinsic::amdgcn_atomic_inc; | |||
| 16823 | break; | |||
| 16824 | case AMDGPU::BI__builtin_amdgcn_atomic_dec32: | |||
| 16825 | case AMDGPU::BI__builtin_amdgcn_atomic_dec64: | |||
| 16826 | BuiltinAtomicOp = Intrinsic::amdgcn_atomic_dec; | |||
| 16827 | break; | |||
| 16828 | } | |||
| 16829 | ||||
| 16830 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 16831 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 16832 | ||||
| 16833 | llvm::Function *F = | |||
| 16834 | CGM.getIntrinsic(BuiltinAtomicOp, {ResultType, Ptr->getType()}); | |||
| 16835 | ||||
| 16836 | if (ProcessOrderScopeAMDGCN(EmitScalarExpr(E->getArg(2)), | |||
| 16837 | EmitScalarExpr(E->getArg(3)), AO, SSID)) { | |||
| 16838 | ||||
| 16839 | // llvm.amdgcn.atomic.inc and llvm.amdgcn.atomic.dec expects ordering and | |||
| 16840 | // scope as unsigned values | |||
| 16841 | Value *MemOrder = Builder.getInt32(static_cast<int>(AO)); | |||
| 16842 | Value *MemScope = Builder.getInt32(static_cast<int>(SSID)); | |||
| 16843 | ||||
| 16844 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); | |||
| 16845 | bool Volatile = | |||
| 16846 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); | |||
| 16847 | Value *IsVolatile = Builder.getInt1(static_cast<bool>(Volatile)); | |||
| 16848 | ||||
| 16849 | return Builder.CreateCall(F, {Ptr, Val, MemOrder, MemScope, IsVolatile}); | |||
| 16850 | } | |||
| 16851 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; | |||
| 16852 | } | |||
| 16853 | default: | |||
| 16854 | return nullptr; | |||
| 16855 | } | |||
| 16856 | } | |||
| 16857 | ||||
| 16858 | /// Handle a SystemZ function in which the final argument is a pointer | |||
| 16859 | /// to an int that receives the post-instruction CC value. At the LLVM level | |||
| 16860 | /// this is represented as a function that returns a {result, cc} pair. | |||
| 16861 | static Value *EmitSystemZIntrinsicWithCC(CodeGenFunction &CGF, | |||
| 16862 | unsigned IntrinsicID, | |||
| 16863 | const CallExpr *E) { | |||
| 16864 | unsigned NumArgs = E->getNumArgs() - 1; | |||
| 16865 | SmallVector<Value *, 8> Args(NumArgs); | |||
| 16866 | for (unsigned I = 0; I < NumArgs; ++I) | |||
| 16867 | Args[I] = CGF.EmitScalarExpr(E->getArg(I)); | |||
| 16868 | Address CCPtr = CGF.EmitPointerWithAlignment(E->getArg(NumArgs)); | |||
| 16869 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID); | |||
| 16870 | Value *Call = CGF.Builder.CreateCall(F, Args); | |||
| 16871 | Value *CC = CGF.Builder.CreateExtractValue(Call, 1); | |||
| 16872 | CGF.Builder.CreateStore(CC, CCPtr); | |||
| 16873 | return CGF.Builder.CreateExtractValue(Call, 0); | |||
| 16874 | } | |||
| 16875 | ||||
| 16876 | Value *CodeGenFunction::EmitSystemZBuiltinExpr(unsigned BuiltinID, | |||
| 16877 | const CallExpr *E) { | |||
| 16878 | switch (BuiltinID) { | |||
| 16879 | case SystemZ::BI__builtin_tbegin: { | |||
| 16880 | Value *TDB = EmitScalarExpr(E->getArg(0)); | |||
| 16881 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c); | |||
| 16882 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin); | |||
| 16883 | return Builder.CreateCall(F, {TDB, Control}); | |||
| 16884 | } | |||
| 16885 | case SystemZ::BI__builtin_tbegin_nofloat: { | |||
| 16886 | Value *TDB = EmitScalarExpr(E->getArg(0)); | |||
| 16887 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c); | |||
| 16888 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin_nofloat); | |||
| 16889 | return Builder.CreateCall(F, {TDB, Control}); | |||
| 16890 | } | |||
| 16891 | case SystemZ::BI__builtin_tbeginc: { | |||
| 16892 | Value *TDB = llvm::ConstantPointerNull::get(Int8PtrTy); | |||
| 16893 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff08); | |||
| 16894 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbeginc); | |||
| 16895 | return Builder.CreateCall(F, {TDB, Control}); | |||
| 16896 | } | |||
| 16897 | case SystemZ::BI__builtin_tabort: { | |||
| 16898 | Value *Data = EmitScalarExpr(E->getArg(0)); | |||
| 16899 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tabort); | |||
| 16900 | return Builder.CreateCall(F, Builder.CreateSExt(Data, Int64Ty, "tabort")); | |||
| 16901 | } | |||
| 16902 | case SystemZ::BI__builtin_non_tx_store: { | |||
| 16903 | Value *Address = EmitScalarExpr(E->getArg(0)); | |||
| 16904 | Value *Data = EmitScalarExpr(E->getArg(1)); | |||
| 16905 | Function *F = CGM.getIntrinsic(Intrinsic::s390_ntstg); | |||
| 16906 | return Builder.CreateCall(F, {Data, Address}); | |||
| 16907 | } | |||
| 16908 | ||||
| 16909 | // Vector builtins. Note that most vector builtins are mapped automatically | |||
| 16910 | // to target-specific LLVM intrinsics. The ones handled specially here can | |||
| 16911 | // be represented via standard LLVM IR, which is preferable to enable common | |||
| 16912 | // LLVM optimizations. | |||
| 16913 | ||||
| 16914 | case SystemZ::BI__builtin_s390_vpopctb: | |||
| 16915 | case SystemZ::BI__builtin_s390_vpopcth: | |||
| 16916 | case SystemZ::BI__builtin_s390_vpopctf: | |||
| 16917 | case SystemZ::BI__builtin_s390_vpopctg: { | |||
| 16918 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16919 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16920 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); | |||
| 16921 | return Builder.CreateCall(F, X); | |||
| 16922 | } | |||
| 16923 | ||||
| 16924 | case SystemZ::BI__builtin_s390_vclzb: | |||
| 16925 | case SystemZ::BI__builtin_s390_vclzh: | |||
| 16926 | case SystemZ::BI__builtin_s390_vclzf: | |||
| 16927 | case SystemZ::BI__builtin_s390_vclzg: { | |||
| 16928 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16929 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16930 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
| 16931 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType); | |||
| 16932 | return Builder.CreateCall(F, {X, Undef}); | |||
| 16933 | } | |||
| 16934 | ||||
| 16935 | case SystemZ::BI__builtin_s390_vctzb: | |||
| 16936 | case SystemZ::BI__builtin_s390_vctzh: | |||
| 16937 | case SystemZ::BI__builtin_s390_vctzf: | |||
| 16938 | case SystemZ::BI__builtin_s390_vctzg: { | |||
| 16939 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16940 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16941 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); | |||
| 16942 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType); | |||
| 16943 | return Builder.CreateCall(F, {X, Undef}); | |||
| 16944 | } | |||
| 16945 | ||||
| 16946 | case SystemZ::BI__builtin_s390_vfsqsb: | |||
| 16947 | case SystemZ::BI__builtin_s390_vfsqdb: { | |||
| 16948 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16949 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16950 | if (Builder.getIsFPConstrained()) { | |||
| 16951 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_sqrt, ResultType); | |||
| 16952 | return Builder.CreateConstrainedFPCall(F, { X }); | |||
| 16953 | } else { | |||
| 16954 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType); | |||
| 16955 | return Builder.CreateCall(F, X); | |||
| 16956 | } | |||
| 16957 | } | |||
| 16958 | case SystemZ::BI__builtin_s390_vfmasb: | |||
| 16959 | case SystemZ::BI__builtin_s390_vfmadb: { | |||
| 16960 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16961 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16962 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 16963 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
| 16964 | if (Builder.getIsFPConstrained()) { | |||
| 16965 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType); | |||
| 16966 | return Builder.CreateConstrainedFPCall(F, {X, Y, Z}); | |||
| 16967 | } else { | |||
| 16968 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
| 16969 | return Builder.CreateCall(F, {X, Y, Z}); | |||
| 16970 | } | |||
| 16971 | } | |||
| 16972 | case SystemZ::BI__builtin_s390_vfmssb: | |||
| 16973 | case SystemZ::BI__builtin_s390_vfmsdb: { | |||
| 16974 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16975 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16976 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 16977 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
| 16978 | if (Builder.getIsFPConstrained()) { | |||
| 16979 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType); | |||
| 16980 | return Builder.CreateConstrainedFPCall(F, {X, Y, Builder.CreateFNeg(Z, "neg")}); | |||
| 16981 | } else { | |||
| 16982 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
| 16983 | return Builder.CreateCall(F, {X, Y, Builder.CreateFNeg(Z, "neg")}); | |||
| 16984 | } | |||
| 16985 | } | |||
| 16986 | case SystemZ::BI__builtin_s390_vfnmasb: | |||
| 16987 | case SystemZ::BI__builtin_s390_vfnmadb: { | |||
| 16988 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 16989 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 16990 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 16991 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
| 16992 | if (Builder.getIsFPConstrained()) { | |||
| 16993 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType); | |||
| 16994 | return Builder.CreateFNeg(Builder.CreateConstrainedFPCall(F, {X, Y, Z}), "neg"); | |||
| 16995 | } else { | |||
| 16996 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
| 16997 | return Builder.CreateFNeg(Builder.CreateCall(F, {X, Y, Z}), "neg"); | |||
| 16998 | } | |||
| 16999 | } | |||
| 17000 | case SystemZ::BI__builtin_s390_vfnmssb: | |||
| 17001 | case SystemZ::BI__builtin_s390_vfnmsdb: { | |||
| 17002 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 17003 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 17004 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 17005 | Value *Z = EmitScalarExpr(E->getArg(2)); | |||
| 17006 | if (Builder.getIsFPConstrained()) { | |||
| 17007 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_fma, ResultType); | |||
| 17008 | Value *NegZ = Builder.CreateFNeg(Z, "sub"); | |||
| 17009 | return Builder.CreateFNeg(Builder.CreateConstrainedFPCall(F, {X, Y, NegZ})); | |||
| 17010 | } else { | |||
| 17011 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); | |||
| 17012 | Value *NegZ = Builder.CreateFNeg(Z, "neg"); | |||
| 17013 | return Builder.CreateFNeg(Builder.CreateCall(F, {X, Y, NegZ})); | |||
| 17014 | } | |||
| 17015 | } | |||
| 17016 | case SystemZ::BI__builtin_s390_vflpsb: | |||
| 17017 | case SystemZ::BI__builtin_s390_vflpdb: { | |||
| 17018 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 17019 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 17020 | Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); | |||
| 17021 | return Builder.CreateCall(F, X); | |||
| 17022 | } | |||
| 17023 | case SystemZ::BI__builtin_s390_vflnsb: | |||
| 17024 | case SystemZ::BI__builtin_s390_vflndb: { | |||
| 17025 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 17026 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 17027 | Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); | |||
| 17028 | return Builder.CreateFNeg(Builder.CreateCall(F, X), "neg"); | |||
| 17029 | } | |||
| 17030 | case SystemZ::BI__builtin_s390_vfisb: | |||
| 17031 | case SystemZ::BI__builtin_s390_vfidb: { | |||
| 17032 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 17033 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 17034 | // Constant-fold the M4 and M5 mask arguments. | |||
| 17035 | llvm::APSInt M4 = *E->getArg(1)->getIntegerConstantExpr(getContext()); | |||
| 17036 | llvm::APSInt M5 = *E->getArg(2)->getIntegerConstantExpr(getContext()); | |||
| 17037 | // Check whether this instance can be represented via a LLVM standard | |||
| 17038 | // intrinsic. We only support some combinations of M4 and M5. | |||
| 17039 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
| 17040 | Intrinsic::ID CI; | |||
| 17041 | switch (M4.getZExtValue()) { | |||
| 17042 | default: break; | |||
| 17043 | case 0: // IEEE-inexact exception allowed | |||
| 17044 | switch (M5.getZExtValue()) { | |||
| 17045 | default: break; | |||
| 17046 | case 0: ID = Intrinsic::rint; | |||
| 17047 | CI = Intrinsic::experimental_constrained_rint; break; | |||
| 17048 | } | |||
| 17049 | break; | |||
| 17050 | case 4: // IEEE-inexact exception suppressed | |||
| 17051 | switch (M5.getZExtValue()) { | |||
| 17052 | default: break; | |||
| 17053 | case 0: ID = Intrinsic::nearbyint; | |||
| 17054 | CI = Intrinsic::experimental_constrained_nearbyint; break; | |||
| 17055 | case 1: ID = Intrinsic::round; | |||
| 17056 | CI = Intrinsic::experimental_constrained_round; break; | |||
| 17057 | case 5: ID = Intrinsic::trunc; | |||
| 17058 | CI = Intrinsic::experimental_constrained_trunc; break; | |||
| 17059 | case 6: ID = Intrinsic::ceil; | |||
| 17060 | CI = Intrinsic::experimental_constrained_ceil; break; | |||
| 17061 | case 7: ID = Intrinsic::floor; | |||
| 17062 | CI = Intrinsic::experimental_constrained_floor; break; | |||
| 17063 | } | |||
| 17064 | break; | |||
| 17065 | } | |||
| 17066 | if (ID != Intrinsic::not_intrinsic) { | |||
| 17067 | if (Builder.getIsFPConstrained()) { | |||
| 17068 | Function *F = CGM.getIntrinsic(CI, ResultType); | |||
| 17069 | return Builder.CreateConstrainedFPCall(F, X); | |||
| 17070 | } else { | |||
| 17071 | Function *F = CGM.getIntrinsic(ID, ResultType); | |||
| 17072 | return Builder.CreateCall(F, X); | |||
| 17073 | } | |||
| 17074 | } | |||
| 17075 | switch (BuiltinID) { // FIXME: constrained version? | |||
| 17076 | case SystemZ::BI__builtin_s390_vfisb: ID = Intrinsic::s390_vfisb; break; | |||
| 17077 | case SystemZ::BI__builtin_s390_vfidb: ID = Intrinsic::s390_vfidb; break; | |||
| 17078 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "clang/lib/CodeGen/CGBuiltin.cpp" , 17078); | |||
| 17079 | } | |||
| 17080 | Function *F = CGM.getIntrinsic(ID); | |||
| 17081 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); | |||
| 17082 | Value *M5Value = llvm::ConstantInt::get(getLLVMContext(), M5); | |||
| 17083 | return Builder.CreateCall(F, {X, M4Value, M5Value}); | |||
| 17084 | } | |||
| 17085 | case SystemZ::BI__builtin_s390_vfmaxsb: | |||
| 17086 | case SystemZ::BI__builtin_s390_vfmaxdb: { | |||
| 17087 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 17088 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 17089 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 17090 | // Constant-fold the M4 mask argument. | |||
| 17091 | llvm::APSInt M4 = *E->getArg(2)->getIntegerConstantExpr(getContext()); | |||
| 17092 | // Check whether this instance can be represented via a LLVM standard | |||
| 17093 | // intrinsic. We only support some values of M4. | |||
| 17094 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
| 17095 | Intrinsic::ID CI; | |||
| 17096 | switch (M4.getZExtValue()) { | |||
| 17097 | default: break; | |||
| 17098 | case 4: ID = Intrinsic::maxnum; | |||
| 17099 | CI = Intrinsic::experimental_constrained_maxnum; break; | |||
| 17100 | } | |||
| 17101 | if (ID != Intrinsic::not_intrinsic) { | |||
| 17102 | if (Builder.getIsFPConstrained()) { | |||
| 17103 | Function *F = CGM.getIntrinsic(CI, ResultType); | |||
| 17104 | return Builder.CreateConstrainedFPCall(F, {X, Y}); | |||
| 17105 | } else { | |||
| 17106 | Function *F = CGM.getIntrinsic(ID, ResultType); | |||
| 17107 | return Builder.CreateCall(F, {X, Y}); | |||
| 17108 | } | |||
| 17109 | } | |||
| 17110 | switch (BuiltinID) { | |||
| 17111 | case SystemZ::BI__builtin_s390_vfmaxsb: ID = Intrinsic::s390_vfmaxsb; break; | |||
| 17112 | case SystemZ::BI__builtin_s390_vfmaxdb: ID = Intrinsic::s390_vfmaxdb; break; | |||
| 17113 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "clang/lib/CodeGen/CGBuiltin.cpp" , 17113); | |||
| 17114 | } | |||
| 17115 | Function *F = CGM.getIntrinsic(ID); | |||
| 17116 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); | |||
| 17117 | return Builder.CreateCall(F, {X, Y, M4Value}); | |||
| 17118 | } | |||
| 17119 | case SystemZ::BI__builtin_s390_vfminsb: | |||
| 17120 | case SystemZ::BI__builtin_s390_vfmindb: { | |||
| 17121 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 17122 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 17123 | Value *Y = EmitScalarExpr(E->getArg(1)); | |||
| 17124 | // Constant-fold the M4 mask argument. | |||
| 17125 | llvm::APSInt M4 = *E->getArg(2)->getIntegerConstantExpr(getContext()); | |||
| 17126 | // Check whether this instance can be represented via a LLVM standard | |||
| 17127 | // intrinsic. We only support some values of M4. | |||
| 17128 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
| 17129 | Intrinsic::ID CI; | |||
| 17130 | switch (M4.getZExtValue()) { | |||
| 17131 | default: break; | |||
| 17132 | case 4: ID = Intrinsic::minnum; | |||
| 17133 | CI = Intrinsic::experimental_constrained_minnum; break; | |||
| 17134 | } | |||
| 17135 | if (ID != Intrinsic::not_intrinsic) { | |||
| 17136 | if (Builder.getIsFPConstrained()) { | |||
| 17137 | Function *F = CGM.getIntrinsic(CI, ResultType); | |||
| 17138 | return Builder.CreateConstrainedFPCall(F, {X, Y}); | |||
| 17139 | } else { | |||
| 17140 | Function *F = CGM.getIntrinsic(ID, ResultType); | |||
| 17141 | return Builder.CreateCall(F, {X, Y}); | |||
| 17142 | } | |||
| 17143 | } | |||
| 17144 | switch (BuiltinID) { | |||
| 17145 | case SystemZ::BI__builtin_s390_vfminsb: ID = Intrinsic::s390_vfminsb; break; | |||
| 17146 | case SystemZ::BI__builtin_s390_vfmindb: ID = Intrinsic::s390_vfmindb; break; | |||
| 17147 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "clang/lib/CodeGen/CGBuiltin.cpp" , 17147); | |||
| 17148 | } | |||
| 17149 | Function *F = CGM.getIntrinsic(ID); | |||
| 17150 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); | |||
| 17151 | return Builder.CreateCall(F, {X, Y, M4Value}); | |||
| 17152 | } | |||
| 17153 | ||||
| 17154 | case SystemZ::BI__builtin_s390_vlbrh: | |||
| 17155 | case SystemZ::BI__builtin_s390_vlbrf: | |||
| 17156 | case SystemZ::BI__builtin_s390_vlbrg: { | |||
| 17157 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 17158 | Value *X = EmitScalarExpr(E->getArg(0)); | |||
| 17159 | Function *F = CGM.getIntrinsic(Intrinsic::bswap, ResultType); | |||
| 17160 | return Builder.CreateCall(F, X); | |||
| 17161 | } | |||
| 17162 | ||||
| 17163 | // Vector intrinsics that output the post-instruction CC value. | |||
| 17164 | ||||
| 17165 | #define INTRINSIC_WITH_CC(NAME) \ | |||
| 17166 | case SystemZ::BI__builtin_##NAME: \ | |||
| 17167 | return EmitSystemZIntrinsicWithCC(*this, Intrinsic::NAME, E) | |||
| 17168 | ||||
| 17169 | INTRINSIC_WITH_CC(s390_vpkshs); | |||
| 17170 | INTRINSIC_WITH_CC(s390_vpksfs); | |||
| 17171 | INTRINSIC_WITH_CC(s390_vpksgs); | |||
| 17172 | ||||
| 17173 | INTRINSIC_WITH_CC(s390_vpklshs); | |||
| 17174 | INTRINSIC_WITH_CC(s390_vpklsfs); | |||
| 17175 | INTRINSIC_WITH_CC(s390_vpklsgs); | |||
| 17176 | ||||
| 17177 | INTRINSIC_WITH_CC(s390_vceqbs); | |||
| 17178 | INTRINSIC_WITH_CC(s390_vceqhs); | |||
| 17179 | INTRINSIC_WITH_CC(s390_vceqfs); | |||
| 17180 | INTRINSIC_WITH_CC(s390_vceqgs); | |||
| 17181 | ||||
| 17182 | INTRINSIC_WITH_CC(s390_vchbs); | |||
| 17183 | INTRINSIC_WITH_CC(s390_vchhs); | |||
| 17184 | INTRINSIC_WITH_CC(s390_vchfs); | |||
| 17185 | INTRINSIC_WITH_CC(s390_vchgs); | |||
| 17186 | ||||
| 17187 | INTRINSIC_WITH_CC(s390_vchlbs); | |||
| 17188 | INTRINSIC_WITH_CC(s390_vchlhs); | |||
| 17189 | INTRINSIC_WITH_CC(s390_vchlfs); | |||
| 17190 | INTRINSIC_WITH_CC(s390_vchlgs); | |||
| 17191 | ||||
| 17192 | INTRINSIC_WITH_CC(s390_vfaebs); | |||
| 17193 | INTRINSIC_WITH_CC(s390_vfaehs); | |||
| 17194 | INTRINSIC_WITH_CC(s390_vfaefs); | |||
| 17195 | ||||
| 17196 | INTRINSIC_WITH_CC(s390_vfaezbs); | |||
| 17197 | INTRINSIC_WITH_CC(s390_vfaezhs); | |||
| 17198 | INTRINSIC_WITH_CC(s390_vfaezfs); | |||
| 17199 | ||||
| 17200 | INTRINSIC_WITH_CC(s390_vfeebs); | |||
| 17201 | INTRINSIC_WITH_CC(s390_vfeehs); | |||
| 17202 | INTRINSIC_WITH_CC(s390_vfeefs); | |||
| 17203 | ||||
| 17204 | INTRINSIC_WITH_CC(s390_vfeezbs); | |||
| 17205 | INTRINSIC_WITH_CC(s390_vfeezhs); | |||
| 17206 | INTRINSIC_WITH_CC(s390_vfeezfs); | |||
| 17207 | ||||
| 17208 | INTRINSIC_WITH_CC(s390_vfenebs); | |||
| 17209 | INTRINSIC_WITH_CC(s390_vfenehs); | |||
| 17210 | INTRINSIC_WITH_CC(s390_vfenefs); | |||
| 17211 | ||||
| 17212 | INTRINSIC_WITH_CC(s390_vfenezbs); | |||
| 17213 | INTRINSIC_WITH_CC(s390_vfenezhs); | |||
| 17214 | INTRINSIC_WITH_CC(s390_vfenezfs); | |||
| 17215 | ||||
| 17216 | INTRINSIC_WITH_CC(s390_vistrbs); | |||
| 17217 | INTRINSIC_WITH_CC(s390_vistrhs); | |||
| 17218 | INTRINSIC_WITH_CC(s390_vistrfs); | |||
| 17219 | ||||
| 17220 | INTRINSIC_WITH_CC(s390_vstrcbs); | |||
| 17221 | INTRINSIC_WITH_CC(s390_vstrchs); | |||
| 17222 | INTRINSIC_WITH_CC(s390_vstrcfs); | |||
| 17223 | ||||
| 17224 | INTRINSIC_WITH_CC(s390_vstrczbs); | |||
| 17225 | INTRINSIC_WITH_CC(s390_vstrczhs); | |||
| 17226 | INTRINSIC_WITH_CC(s390_vstrczfs); | |||
| 17227 | ||||
| 17228 | INTRINSIC_WITH_CC(s390_vfcesbs); | |||
| 17229 | INTRINSIC_WITH_CC(s390_vfcedbs); | |||
| 17230 | INTRINSIC_WITH_CC(s390_vfchsbs); | |||
| 17231 | INTRINSIC_WITH_CC(s390_vfchdbs); | |||
| 17232 | INTRINSIC_WITH_CC(s390_vfchesbs); | |||
| 17233 | INTRINSIC_WITH_CC(s390_vfchedbs); | |||
| 17234 | ||||
| 17235 | INTRINSIC_WITH_CC(s390_vftcisb); | |||
| 17236 | INTRINSIC_WITH_CC(s390_vftcidb); | |||
| 17237 | ||||
| 17238 | INTRINSIC_WITH_CC(s390_vstrsb); | |||
| 17239 | INTRINSIC_WITH_CC(s390_vstrsh); | |||
| 17240 | INTRINSIC_WITH_CC(s390_vstrsf); | |||
| 17241 | ||||
| 17242 | INTRINSIC_WITH_CC(s390_vstrszb); | |||
| 17243 | INTRINSIC_WITH_CC(s390_vstrszh); | |||
| 17244 | INTRINSIC_WITH_CC(s390_vstrszf); | |||
| 17245 | ||||
| 17246 | #undef INTRINSIC_WITH_CC | |||
| 17247 | ||||
| 17248 | default: | |||
| 17249 | return nullptr; | |||
| 17250 | } | |||
| 17251 | } | |||
| 17252 | ||||
| 17253 | namespace { | |||
| 17254 | // Helper classes for mapping MMA builtins to particular LLVM intrinsic variant. | |||
| 17255 | struct NVPTXMmaLdstInfo { | |||
| 17256 | unsigned NumResults; // Number of elements to load/store | |||
| 17257 | // Intrinsic IDs for row/col variants. 0 if particular layout is unsupported. | |||
| 17258 | unsigned IID_col; | |||
| 17259 | unsigned IID_row; | |||
| 17260 | }; | |||
| 17261 | ||||
| 17262 | #define MMA_INTR(geom_op_type, layout) \ | |||
| 17263 | Intrinsic::nvvm_wmma_##geom_op_type##_##layout##_stride | |||
| 17264 | #define MMA_LDST(n, geom_op_type) \ | |||
| 17265 | { n, MMA_INTR(geom_op_type, col), MMA_INTR(geom_op_type, row) } | |||
| 17266 | ||||
| 17267 | static NVPTXMmaLdstInfo getNVPTXMmaLdstInfo(unsigned BuiltinID) { | |||
| 17268 | switch (BuiltinID) { | |||
| 17269 | // FP MMA loads | |||
| 17270 | case NVPTX::BI__hmma_m16n16k16_ld_a: | |||
| 17271 | return MMA_LDST(8, m16n16k16_load_a_f16); | |||
| 17272 | case NVPTX::BI__hmma_m16n16k16_ld_b: | |||
| 17273 | return MMA_LDST(8, m16n16k16_load_b_f16); | |||
| 17274 | case NVPTX::BI__hmma_m16n16k16_ld_c_f16: | |||
| 17275 | return MMA_LDST(4, m16n16k16_load_c_f16); | |||
| 17276 | case NVPTX::BI__hmma_m16n16k16_ld_c_f32: | |||
| 17277 | return MMA_LDST(8, m16n16k16_load_c_f32); | |||
| 17278 | case NVPTX::BI__hmma_m32n8k16_ld_a: | |||
| 17279 | return MMA_LDST(8, m32n8k16_load_a_f16); | |||
| 17280 | case NVPTX::BI__hmma_m32n8k16_ld_b: | |||
| 17281 | return MMA_LDST(8, m32n8k16_load_b_f16); | |||
| 17282 | case NVPTX::BI__hmma_m32n8k16_ld_c_f16: | |||
| 17283 | return MMA_LDST(4, m32n8k16_load_c_f16); | |||
| 17284 | case NVPTX::BI__hmma_m32n8k16_ld_c_f32: | |||
| 17285 | return MMA_LDST(8, m32n8k16_load_c_f32); | |||
| 17286 | case NVPTX::BI__hmma_m8n32k16_ld_a: | |||
| 17287 | return MMA_LDST(8, m8n32k16_load_a_f16); | |||
| 17288 | case NVPTX::BI__hmma_m8n32k16_ld_b: | |||
| 17289 | return MMA_LDST(8, m8n32k16_load_b_f16); | |||
| 17290 | case NVPTX::BI__hmma_m8n32k16_ld_c_f16: | |||
| 17291 | return MMA_LDST(4, m8n32k16_load_c_f16); | |||
| 17292 | case NVPTX::BI__hmma_m8n32k16_ld_c_f32: | |||
| 17293 | return MMA_LDST(8, m8n32k16_load_c_f32); | |||
| 17294 | ||||
| 17295 | // Integer MMA loads | |||
| 17296 | case NVPTX::BI__imma_m16n16k16_ld_a_s8: | |||
| 17297 | return MMA_LDST(2, m16n16k16_load_a_s8); | |||
| 17298 | case NVPTX::BI__imma_m16n16k16_ld_a_u8: | |||
| 17299 | return MMA_LDST(2, m16n16k16_load_a_u8); | |||
| 17300 | case NVPTX::BI__imma_m16n16k16_ld_b_s8: | |||
| 17301 | return MMA_LDST(2, m16n16k16_load_b_s8); | |||
| 17302 | case NVPTX::BI__imma_m16n16k16_ld_b_u8: | |||
| 17303 | return MMA_LDST(2, m16n16k16_load_b_u8); | |||
| 17304 | case NVPTX::BI__imma_m16n16k16_ld_c: | |||
| 17305 | return MMA_LDST(8, m16n16k16_load_c_s32); | |||
| 17306 | case NVPTX::BI__imma_m32n8k16_ld_a_s8: | |||
| 17307 | return MMA_LDST(4, m32n8k16_load_a_s8); | |||
| 17308 | case NVPTX::BI__imma_m32n8k16_ld_a_u8: | |||
| 17309 | return MMA_LDST(4, m32n8k16_load_a_u8); | |||
| 17310 | case NVPTX::BI__imma_m32n8k16_ld_b_s8: | |||
| 17311 | return MMA_LDST(1, m32n8k16_load_b_s8); | |||
| 17312 | case NVPTX::BI__imma_m32n8k16_ld_b_u8: | |||
| 17313 | return MMA_LDST(1, m32n8k16_load_b_u8); | |||
| 17314 | case NVPTX::BI__imma_m32n8k16_ld_c: | |||
| 17315 | return MMA_LDST(8, m32n8k16_load_c_s32); | |||
| 17316 | case NVPTX::BI__imma_m8n32k16_ld_a_s8: | |||
| 17317 | return MMA_LDST(1, m8n32k16_load_a_s8); | |||
| 17318 | case NVPTX::BI__imma_m8n32k16_ld_a_u8: | |||
| 17319 | return MMA_LDST(1, m8n32k16_load_a_u8); | |||
| 17320 | case NVPTX::BI__imma_m8n32k16_ld_b_s8: | |||
| 17321 | return MMA_LDST(4, m8n32k16_load_b_s8); | |||
| 17322 | case NVPTX::BI__imma_m8n32k16_ld_b_u8: | |||
| 17323 | return MMA_LDST(4, m8n32k16_load_b_u8); | |||
| 17324 | case NVPTX::BI__imma_m8n32k16_ld_c: | |||
| 17325 | return MMA_LDST(8, m8n32k16_load_c_s32); | |||
| 17326 | ||||
| 17327 | // Sub-integer MMA loads. | |||
| 17328 | // Only row/col layout is supported by A/B fragments. | |||
| 17329 | case NVPTX::BI__imma_m8n8k32_ld_a_s4: | |||
| 17330 | return {1, 0, MMA_INTR(m8n8k32_load_a_s4, row)}; | |||
| 17331 | case NVPTX::BI__imma_m8n8k32_ld_a_u4: | |||
| 17332 | return {1, 0, MMA_INTR(m8n8k32_load_a_u4, row)}; | |||
| 17333 | case NVPTX::BI__imma_m8n8k32_ld_b_s4: | |||
| 17334 | return {1, MMA_INTR(m8n8k32_load_b_s4, col), 0}; | |||
| 17335 | case NVPTX::BI__imma_m8n8k32_ld_b_u4: | |||
| 17336 | return {1, MMA_INTR(m8n8k32_load_b_u4, col), 0}; | |||
| 17337 | case NVPTX::BI__imma_m8n8k32_ld_c: | |||
| 17338 | return MMA_LDST(2, m8n8k32_load_c_s32); | |||
| 17339 | case NVPTX::BI__bmma_m8n8k128_ld_a_b1: | |||
| 17340 | return {1, 0, MMA_INTR(m8n8k128_load_a_b1, row)}; | |||
| 17341 | case NVPTX::BI__bmma_m8n8k128_ld_b_b1: | |||
| 17342 | return {1, MMA_INTR(m8n8k128_load_b_b1, col), 0}; | |||
| 17343 | case NVPTX::BI__bmma_m8n8k128_ld_c: | |||
| 17344 | return MMA_LDST(2, m8n8k128_load_c_s32); | |||
| 17345 | ||||
| 17346 | // Double MMA loads | |||
| 17347 | case NVPTX::BI__dmma_m8n8k4_ld_a: | |||
| 17348 | return MMA_LDST(1, m8n8k4_load_a_f64); | |||
| 17349 | case NVPTX::BI__dmma_m8n8k4_ld_b: | |||
| 17350 | return MMA_LDST(1, m8n8k4_load_b_f64); | |||
| 17351 | case NVPTX::BI__dmma_m8n8k4_ld_c: | |||
| 17352 | return MMA_LDST(2, m8n8k4_load_c_f64); | |||
| 17353 | ||||
| 17354 | // Alternate float MMA loads | |||
| 17355 | case NVPTX::BI__mma_bf16_m16n16k16_ld_a: | |||
| 17356 | return MMA_LDST(4, m16n16k16_load_a_bf16); | |||
| 17357 | case NVPTX::BI__mma_bf16_m16n16k16_ld_b: | |||
| 17358 | return MMA_LDST(4, m16n16k16_load_b_bf16); | |||
| 17359 | case NVPTX::BI__mma_bf16_m8n32k16_ld_a: | |||
| 17360 | return MMA_LDST(2, m8n32k16_load_a_bf16); | |||
| 17361 | case NVPTX::BI__mma_bf16_m8n32k16_ld_b: | |||
| 17362 | return MMA_LDST(8, m8n32k16_load_b_bf16); | |||
| 17363 | case NVPTX::BI__mma_bf16_m32n8k16_ld_a: | |||
| 17364 | return MMA_LDST(8, m32n8k16_load_a_bf16); | |||
| 17365 | case NVPTX::BI__mma_bf16_m32n8k16_ld_b: | |||
| 17366 | return MMA_LDST(2, m32n8k16_load_b_bf16); | |||
| 17367 | case NVPTX::BI__mma_tf32_m16n16k8_ld_a: | |||
| 17368 | return MMA_LDST(4, m16n16k8_load_a_tf32); | |||
| 17369 | case NVPTX::BI__mma_tf32_m16n16k8_ld_b: | |||
| 17370 | return MMA_LDST(4, m16n16k8_load_b_tf32); | |||
| 17371 | case NVPTX::BI__mma_tf32_m16n16k8_ld_c: | |||
| 17372 | return MMA_LDST(8, m16n16k8_load_c_f32); | |||
| 17373 | ||||
| 17374 | // NOTE: We need to follow inconsitent naming scheme used by NVCC. Unlike | |||
| 17375 | // PTX and LLVM IR where stores always use fragment D, NVCC builtins always | |||
| 17376 | // use fragment C for both loads and stores. | |||
| 17377 | // FP MMA stores. | |||
| 17378 | case NVPTX::BI__hmma_m16n16k16_st_c_f16: | |||
| 17379 | return MMA_LDST(4, m16n16k16_store_d_f16); | |||
| 17380 | case NVPTX::BI__hmma_m16n16k16_st_c_f32: | |||
| 17381 | return MMA_LDST(8, m16n16k16_store_d_f32); | |||
| 17382 | case NVPTX::BI__hmma_m32n8k16_st_c_f16: | |||
| 17383 | return MMA_LDST(4, m32n8k16_store_d_f16); | |||
| 17384 | case NVPTX::BI__hmma_m32n8k16_st_c_f32: | |||
| 17385 | return MMA_LDST(8, m32n8k16_store_d_f32); | |||
| 17386 | case NVPTX::BI__hmma_m8n32k16_st_c_f16: | |||
| 17387 | return MMA_LDST(4, m8n32k16_store_d_f16); | |||
| 17388 | case NVPTX::BI__hmma_m8n32k16_st_c_f32: | |||
| 17389 | return MMA_LDST(8, m8n32k16_store_d_f32); | |||
| 17390 | ||||
| 17391 | // Integer and sub-integer MMA stores. | |||
| 17392 | // Another naming quirk. Unlike other MMA builtins that use PTX types in the | |||
| 17393 | // name, integer loads/stores use LLVM's i32. | |||
| 17394 | case NVPTX::BI__imma_m16n16k16_st_c_i32: | |||
| 17395 | return MMA_LDST(8, m16n16k16_store_d_s32); | |||
| 17396 | case NVPTX::BI__imma_m32n8k16_st_c_i32: | |||
| 17397 | return MMA_LDST(8, m32n8k16_store_d_s32); | |||
| 17398 | case NVPTX::BI__imma_m8n32k16_st_c_i32: | |||
| 17399 | return MMA_LDST(8, m8n32k16_store_d_s32); | |||
| 17400 | case NVPTX::BI__imma_m8n8k32_st_c_i32: | |||
| 17401 | return MMA_LDST(2, m8n8k32_store_d_s32); | |||
| 17402 | case NVPTX::BI__bmma_m8n8k128_st_c_i32: | |||
| 17403 | return MMA_LDST(2, m8n8k128_store_d_s32); | |||
| 17404 | ||||
| 17405 | // Double MMA store | |||
| 17406 | case NVPTX::BI__dmma_m8n8k4_st_c_f64: | |||
| 17407 | return MMA_LDST(2, m8n8k4_store_d_f64); | |||
| 17408 | ||||
| 17409 | // Alternate float MMA store | |||
| 17410 | case NVPTX::BI__mma_m16n16k8_st_c_f32: | |||
| 17411 | return MMA_LDST(8, m16n16k8_store_d_f32); | |||
| 17412 | ||||
| 17413 | default: | |||
| 17414 | llvm_unreachable("Unknown MMA builtin")::llvm::llvm_unreachable_internal("Unknown MMA builtin", "clang/lib/CodeGen/CGBuiltin.cpp" , 17414); | |||
| 17415 | } | |||
| 17416 | } | |||
| 17417 | #undef MMA_LDST | |||
| 17418 | #undef MMA_INTR | |||
| 17419 | ||||
| 17420 | ||||
| 17421 | struct NVPTXMmaInfo { | |||
| 17422 | unsigned NumEltsA; | |||
| 17423 | unsigned NumEltsB; | |||
| 17424 | unsigned NumEltsC; | |||
| 17425 | unsigned NumEltsD; | |||
| 17426 | ||||
| 17427 | // Variants are ordered by layout-A/layout-B/satf, where 'row' has priority | |||
| 17428 | // over 'col' for layout. The index of non-satf variants is expected to match | |||
| 17429 | // the undocumented layout constants used by CUDA's mma.hpp. | |||
| 17430 | std::array<unsigned, 8> Variants; | |||
| 17431 | ||||
| 17432 | unsigned getMMAIntrinsic(int Layout, bool Satf) { | |||
| 17433 | unsigned Index = Layout + 4 * Satf; | |||
| 17434 | if (Index >= Variants.size()) | |||
| 17435 | return 0; | |||
| 17436 | return Variants[Index]; | |||
| 17437 | } | |||
| 17438 | }; | |||
| 17439 | ||||
| 17440 | // Returns an intrinsic that matches Layout and Satf for valid combinations of | |||
| 17441 | // Layout and Satf, 0 otherwise. | |||
| 17442 | static NVPTXMmaInfo getNVPTXMmaInfo(unsigned BuiltinID) { | |||
| 17443 | // clang-format off | |||
| 17444 | #define MMA_VARIANTS(geom, type) \ | |||
| 17445 | Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type, \ | |||
| 17446 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ | |||
| 17447 | Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type, \ | |||
| 17448 | Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type | |||
| 17449 | #define MMA_SATF_VARIANTS(geom, type) \ | |||
| 17450 | MMA_VARIANTS(geom, type), \ | |||
| 17451 | Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type##_satfinite, \ | |||
| 17452 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \ | |||
| 17453 | Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type##_satfinite, \ | |||
| 17454 | Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type##_satfinite | |||
| 17455 | // Sub-integer MMA only supports row.col layout. | |||
| 17456 | #define MMA_VARIANTS_I4(geom, type) \ | |||
| 17457 | 0, \ | |||
| 17458 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ | |||
| 17459 | 0, \ | |||
| 17460 | 0, \ | |||
| 17461 | 0, \ | |||
| 17462 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \ | |||
| 17463 | 0, \ | |||
| 17464 | 0 | |||
| 17465 | // b1 MMA does not support .satfinite. | |||
| 17466 | #define MMA_VARIANTS_B1_XOR(geom, type) \ | |||
| 17467 | 0, \ | |||
| 17468 | Intrinsic::nvvm_wmma_##geom##_mma_xor_popc_row_col_##type, \ | |||
| 17469 | 0, \ | |||
| 17470 | 0, \ | |||
| 17471 | 0, \ | |||
| 17472 | 0, \ | |||
| 17473 | 0, \ | |||
| 17474 | 0 | |||
| 17475 | #define MMA_VARIANTS_B1_AND(geom, type) \ | |||
| 17476 | 0, \ | |||
| 17477 | Intrinsic::nvvm_wmma_##geom##_mma_and_popc_row_col_##type, \ | |||
| 17478 | 0, \ | |||
| 17479 | 0, \ | |||
| 17480 | 0, \ | |||
| 17481 | 0, \ | |||
| 17482 | 0, \ | |||
| 17483 | 0 | |||
| 17484 | // clang-format on | |||
| 17485 | switch (BuiltinID) { | |||
| 17486 | // FP MMA | |||
| 17487 | // Note that 'type' argument of MMA_SATF_VARIANTS uses D_C notation, while | |||
| 17488 | // NumEltsN of return value are ordered as A,B,C,D. | |||
| 17489 | case NVPTX::BI__hmma_m16n16k16_mma_f16f16: | |||
| 17490 | return {8, 8, 4, 4, {{MMA_SATF_VARIANTS(m16n16k16, f16_f16)}}}; | |||
| 17491 | case NVPTX::BI__hmma_m16n16k16_mma_f32f16: | |||
| 17492 | return {8, 8, 4, 8, {{MMA_SATF_VARIANTS(m16n16k16, f32_f16)}}}; | |||
| 17493 | case NVPTX::BI__hmma_m16n16k16_mma_f16f32: | |||
| 17494 | return {8, 8, 8, 4, {{MMA_SATF_VARIANTS(m16n16k16, f16_f32)}}}; | |||
| 17495 | case NVPTX::BI__hmma_m16n16k16_mma_f32f32: | |||
| 17496 | return {8, 8, 8, 8, {{MMA_SATF_VARIANTS(m16n16k16, f32_f32)}}}; | |||
| 17497 | case NVPTX::BI__hmma_m32n8k16_mma_f16f16: | |||
| 17498 | return {8, 8, 4, 4, {{MMA_SATF_VARIANTS(m32n8k16, f16_f16)}}}; | |||
| 17499 | case NVPTX::BI__hmma_m32n8k16_mma_f32f16: | |||
| 17500 | return {8, 8, 4, 8, {{MMA_SATF_VARIANTS(m32n8k16, f32_f16)}}}; | |||
| 17501 | case NVPTX::BI__hmma_m32n8k16_mma_f16f32: | |||
| 17502 | return {8, 8, 8, 4, {{MMA_SATF_VARIANTS(m32n8k16, f16_f32)}}}; | |||
| 17503 | case NVPTX::BI__hmma_m32n8k16_mma_f32f32: | |||
| 17504 | return {8, 8, 8, 8, {{MMA_SATF_VARIANTS(m32n8k16, f32_f32)}}}; | |||
| 17505 | case NVPTX::BI__hmma_m8n32k16_mma_f16f16: | |||
| 17506 | return {8, 8, 4, 4, {{MMA_SATF_VARIANTS(m8n32k16, f16_f16)}}}; | |||
| 17507 | case NVPTX::BI__hmma_m8n32k16_mma_f32f16: | |||
| 17508 | return {8, 8, 4, 8, {{MMA_SATF_VARIANTS(m8n32k16, f32_f16)}}}; | |||
| 17509 | case NVPTX::BI__hmma_m8n32k16_mma_f16f32: | |||
| 17510 | return {8, 8, 8, 4, {{MMA_SATF_VARIANTS(m8n32k16, f16_f32)}}}; | |||
| 17511 | case NVPTX::BI__hmma_m8n32k16_mma_f32f32: | |||
| 17512 | return {8, 8, 8, 8, {{MMA_SATF_VARIANTS(m8n32k16, f32_f32)}}}; | |||
| 17513 | ||||
| 17514 | // Integer MMA | |||
| 17515 | case NVPTX::BI__imma_m16n16k16_mma_s8: | |||
| 17516 | return {2, 2, 8, 8, {{MMA_SATF_VARIANTS(m16n16k16, s8)}}}; | |||
| 17517 | case NVPTX::BI__imma_m16n16k16_mma_u8: | |||
| 17518 | return {2, 2, 8, 8, {{MMA_SATF_VARIANTS(m16n16k16, u8)}}}; | |||
| 17519 | case NVPTX::BI__imma_m32n8k16_mma_s8: | |||
| 17520 | return {4, 1, 8, 8, {{MMA_SATF_VARIANTS(m32n8k16, s8)}}}; | |||
| 17521 | case NVPTX::BI__imma_m32n8k16_mma_u8: | |||
| 17522 | return {4, 1, 8, 8, {{MMA_SATF_VARIANTS(m32n8k16, u8)}}}; | |||
| 17523 | case NVPTX::BI__imma_m8n32k16_mma_s8: | |||
| 17524 | return {1, 4, 8, 8, {{MMA_SATF_VARIANTS(m8n32k16, s8)}}}; | |||
| 17525 | case NVPTX::BI__imma_m8n32k16_mma_u8: | |||
| 17526 | return {1, 4, 8, 8, {{MMA_SATF_VARIANTS(m8n32k16, u8)}}}; | |||
| 17527 | ||||
| 17528 | // Sub-integer MMA | |||
| 17529 | case NVPTX::BI__imma_m8n8k32_mma_s4: | |||
| 17530 | return {1, 1, 2, 2, {{MMA_VARIANTS_I4(m8n8k32, s4)}}}; | |||
| 17531 | case NVPTX::BI__imma_m8n8k32_mma_u4: | |||
| 17532 | return {1, 1, 2, 2, {{MMA_VARIANTS_I4(m8n8k32, u4)}}}; | |||
| 17533 | case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1: | |||
| 17534 | return {1, 1, 2, 2, {{MMA_VARIANTS_B1_XOR(m8n8k128, b1)}}}; | |||
| 17535 | case NVPTX::BI__bmma_m8n8k128_mma_and_popc_b1: | |||
| 17536 | return {1, 1, 2, 2, {{MMA_VARIANTS_B1_AND(m8n8k128, b1)}}}; | |||
| 17537 | ||||
| 17538 | // Double MMA | |||
| 17539 | case NVPTX::BI__dmma_m8n8k4_mma_f64: | |||
| 17540 | return {1, 1, 2, 2, {{MMA_VARIANTS(m8n8k4, f64)}}}; | |||
| 17541 | ||||
| 17542 | // Alternate FP MMA | |||
| 17543 | case NVPTX::BI__mma_bf16_m16n16k16_mma_f32: | |||
| 17544 | return {4, 4, 8, 8, {{MMA_VARIANTS(m16n16k16, bf16)}}}; | |||
| 17545 | case NVPTX::BI__mma_bf16_m8n32k16_mma_f32: | |||
| 17546 | return {2, 8, 8, 8, {{MMA_VARIANTS(m8n32k16, bf16)}}}; | |||
| 17547 | case NVPTX::BI__mma_bf16_m32n8k16_mma_f32: | |||
| 17548 | return {8, 2, 8, 8, {{MMA_VARIANTS(m32n8k16, bf16)}}}; | |||
| 17549 | case NVPTX::BI__mma_tf32_m16n16k8_mma_f32: | |||
| 17550 | return {4, 4, 8, 8, {{MMA_VARIANTS(m16n16k8, tf32)}}}; | |||
| 17551 | default: | |||
| 17552 | llvm_unreachable("Unexpected builtin ID.")::llvm::llvm_unreachable_internal("Unexpected builtin ID.", "clang/lib/CodeGen/CGBuiltin.cpp" , 17552); | |||
| 17553 | } | |||
| 17554 | #undef MMA_VARIANTS | |||
| 17555 | #undef MMA_SATF_VARIANTS | |||
| 17556 | #undef MMA_VARIANTS_I4 | |||
| 17557 | #undef MMA_VARIANTS_B1_AND | |||
| 17558 | #undef MMA_VARIANTS_B1_XOR | |||
| 17559 | } | |||
| 17560 | ||||
| 17561 | } // namespace | |||
| 17562 | ||||
| 17563 | Value * | |||
| 17564 | CodeGenFunction::EmitNVPTXBuiltinExpr(unsigned BuiltinID, const CallExpr *E) { | |||
| 17565 | auto MakeLdg = [&](unsigned IntrinsicID) { | |||
| 17566 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 17567 | QualType ArgType = E->getArg(0)->getType(); | |||
| 17568 | clang::CharUnits Align = CGM.getNaturalPointeeTypeAlignment(ArgType); | |||
| 17569 | llvm::Type *ElemTy = ConvertTypeForMem(ArgType->getPointeeType()); | |||
| 17570 | return Builder.CreateCall( | |||
| 17571 | CGM.getIntrinsic(IntrinsicID, {ElemTy, Ptr->getType()}), | |||
| 17572 | {Ptr, ConstantInt::get(Builder.getInt32Ty(), Align.getQuantity())}); | |||
| 17573 | }; | |||
| 17574 | auto MakeScopedAtomic = [&](unsigned IntrinsicID) { | |||
| 17575 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 17576 | llvm::Type *ElemTy = | |||
| 17577 | ConvertTypeForMem(E->getArg(0)->getType()->getPointeeType()); | |||
| 17578 | return Builder.CreateCall( | |||
| 17579 | CGM.getIntrinsic(IntrinsicID, {ElemTy, Ptr->getType()}), | |||
| 17580 | {Ptr, EmitScalarExpr(E->getArg(1))}); | |||
| 17581 | }; | |||
| 17582 | switch (BuiltinID) { | |||
| 17583 | case NVPTX::BI__nvvm_atom_add_gen_i: | |||
| 17584 | case NVPTX::BI__nvvm_atom_add_gen_l: | |||
| 17585 | case NVPTX::BI__nvvm_atom_add_gen_ll: | |||
| 17586 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Add, E); | |||
| 17587 | ||||
| 17588 | case NVPTX::BI__nvvm_atom_sub_gen_i: | |||
| 17589 | case NVPTX::BI__nvvm_atom_sub_gen_l: | |||
| 17590 | case NVPTX::BI__nvvm_atom_sub_gen_ll: | |||
| 17591 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Sub, E); | |||
| 17592 | ||||
| 17593 | case NVPTX::BI__nvvm_atom_and_gen_i: | |||
| 17594 | case NVPTX::BI__nvvm_atom_and_gen_l: | |||
| 17595 | case NVPTX::BI__nvvm_atom_and_gen_ll: | |||
| 17596 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::And, E); | |||
| 17597 | ||||
| 17598 | case NVPTX::BI__nvvm_atom_or_gen_i: | |||
| 17599 | case NVPTX::BI__nvvm_atom_or_gen_l: | |||
| 17600 | case NVPTX::BI__nvvm_atom_or_gen_ll: | |||
| 17601 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Or, E); | |||
| 17602 | ||||
| 17603 | case NVPTX::BI__nvvm_atom_xor_gen_i: | |||
| 17604 | case NVPTX::BI__nvvm_atom_xor_gen_l: | |||
| 17605 | case NVPTX::BI__nvvm_atom_xor_gen_ll: | |||
| 17606 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xor, E); | |||
| 17607 | ||||
| 17608 | case NVPTX::BI__nvvm_atom_xchg_gen_i: | |||
| 17609 | case NVPTX::BI__nvvm_atom_xchg_gen_l: | |||
| 17610 | case NVPTX::BI__nvvm_atom_xchg_gen_ll: | |||
| 17611 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xchg, E); | |||
| 17612 | ||||
| 17613 | case NVPTX::BI__nvvm_atom_max_gen_i: | |||
| 17614 | case NVPTX::BI__nvvm_atom_max_gen_l: | |||
| 17615 | case NVPTX::BI__nvvm_atom_max_gen_ll: | |||
| 17616 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E); | |||
| 17617 | ||||
| 17618 | case NVPTX::BI__nvvm_atom_max_gen_ui: | |||
| 17619 | case NVPTX::BI__nvvm_atom_max_gen_ul: | |||
| 17620 | case NVPTX::BI__nvvm_atom_max_gen_ull: | |||
| 17621 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMax, E); | |||
| 17622 | ||||
| 17623 | case NVPTX::BI__nvvm_atom_min_gen_i: | |||
| 17624 | case NVPTX::BI__nvvm_atom_min_gen_l: | |||
| 17625 | case NVPTX::BI__nvvm_atom_min_gen_ll: | |||
| 17626 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E); | |||
| 17627 | ||||
| 17628 | case NVPTX::BI__nvvm_atom_min_gen_ui: | |||
| 17629 | case NVPTX::BI__nvvm_atom_min_gen_ul: | |||
| 17630 | case NVPTX::BI__nvvm_atom_min_gen_ull: | |||
| 17631 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMin, E); | |||
| 17632 | ||||
| 17633 | case NVPTX::BI__nvvm_atom_cas_gen_i: | |||
| 17634 | case NVPTX::BI__nvvm_atom_cas_gen_l: | |||
| 17635 | case NVPTX::BI__nvvm_atom_cas_gen_ll: | |||
| 17636 | // __nvvm_atom_cas_gen_* should return the old value rather than the | |||
| 17637 | // success flag. | |||
| 17638 | return MakeAtomicCmpXchgValue(*this, E, /*ReturnBool=*/false); | |||
| 17639 | ||||
| 17640 | case NVPTX::BI__nvvm_atom_add_gen_f: | |||
| 17641 | case NVPTX::BI__nvvm_atom_add_gen_d: { | |||
| 17642 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 17643 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 17644 | return Builder.CreateAtomicRMW(llvm::AtomicRMWInst::FAdd, Ptr, Val, | |||
| 17645 | AtomicOrdering::SequentiallyConsistent); | |||
| 17646 | } | |||
| 17647 | ||||
| 17648 | case NVPTX::BI__nvvm_atom_inc_gen_ui: { | |||
| 17649 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 17650 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 17651 | Function *FnALI32 = | |||
| 17652 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_inc_32, Ptr->getType()); | |||
| 17653 | return Builder.CreateCall(FnALI32, {Ptr, Val}); | |||
| 17654 | } | |||
| 17655 | ||||
| 17656 | case NVPTX::BI__nvvm_atom_dec_gen_ui: { | |||
| 17657 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 17658 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 17659 | Function *FnALD32 = | |||
| 17660 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_dec_32, Ptr->getType()); | |||
| 17661 | return Builder.CreateCall(FnALD32, {Ptr, Val}); | |||
| 17662 | } | |||
| 17663 | ||||
| 17664 | case NVPTX::BI__nvvm_ldg_c: | |||
| 17665 | case NVPTX::BI__nvvm_ldg_c2: | |||
| 17666 | case NVPTX::BI__nvvm_ldg_c4: | |||
| 17667 | case NVPTX::BI__nvvm_ldg_s: | |||
| 17668 | case NVPTX::BI__nvvm_ldg_s2: | |||
| 17669 | case NVPTX::BI__nvvm_ldg_s4: | |||
| 17670 | case NVPTX::BI__nvvm_ldg_i: | |||
| 17671 | case NVPTX::BI__nvvm_ldg_i2: | |||
| 17672 | case NVPTX::BI__nvvm_ldg_i4: | |||
| 17673 | case NVPTX::BI__nvvm_ldg_l: | |||
| 17674 | case NVPTX::BI__nvvm_ldg_ll: | |||
| 17675 | case NVPTX::BI__nvvm_ldg_ll2: | |||
| 17676 | case NVPTX::BI__nvvm_ldg_uc: | |||
| 17677 | case NVPTX::BI__nvvm_ldg_uc2: | |||
| 17678 | case NVPTX::BI__nvvm_ldg_uc4: | |||
| 17679 | case NVPTX::BI__nvvm_ldg_us: | |||
| 17680 | case NVPTX::BI__nvvm_ldg_us2: | |||
| 17681 | case NVPTX::BI__nvvm_ldg_us4: | |||
| 17682 | case NVPTX::BI__nvvm_ldg_ui: | |||
| 17683 | case NVPTX::BI__nvvm_ldg_ui2: | |||
| 17684 | case NVPTX::BI__nvvm_ldg_ui4: | |||
| 17685 | case NVPTX::BI__nvvm_ldg_ul: | |||
| 17686 | case NVPTX::BI__nvvm_ldg_ull: | |||
| 17687 | case NVPTX::BI__nvvm_ldg_ull2: | |||
| 17688 | // PTX Interoperability section 2.2: "For a vector with an even number of | |||
| 17689 | // elements, its alignment is set to number of elements times the alignment | |||
| 17690 | // of its member: n*alignof(t)." | |||
| 17691 | return MakeLdg(Intrinsic::nvvm_ldg_global_i); | |||
| 17692 | case NVPTX::BI__nvvm_ldg_f: | |||
| 17693 | case NVPTX::BI__nvvm_ldg_f2: | |||
| 17694 | case NVPTX::BI__nvvm_ldg_f4: | |||
| 17695 | case NVPTX::BI__nvvm_ldg_d: | |||
| 17696 | case NVPTX::BI__nvvm_ldg_d2: | |||
| 17697 | return MakeLdg(Intrinsic::nvvm_ldg_global_f); | |||
| 17698 | ||||
| 17699 | case NVPTX::BI__nvvm_atom_cta_add_gen_i: | |||
| 17700 | case NVPTX::BI__nvvm_atom_cta_add_gen_l: | |||
| 17701 | case NVPTX::BI__nvvm_atom_cta_add_gen_ll: | |||
| 17702 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_cta); | |||
| 17703 | case NVPTX::BI__nvvm_atom_sys_add_gen_i: | |||
| 17704 | case NVPTX::BI__nvvm_atom_sys_add_gen_l: | |||
| 17705 | case NVPTX::BI__nvvm_atom_sys_add_gen_ll: | |||
| 17706 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_sys); | |||
| 17707 | case NVPTX::BI__nvvm_atom_cta_add_gen_f: | |||
| 17708 | case NVPTX::BI__nvvm_atom_cta_add_gen_d: | |||
| 17709 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_cta); | |||
| 17710 | case NVPTX::BI__nvvm_atom_sys_add_gen_f: | |||
| 17711 | case NVPTX::BI__nvvm_atom_sys_add_gen_d: | |||
| 17712 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_sys); | |||
| 17713 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_i: | |||
| 17714 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_l: | |||
| 17715 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_ll: | |||
| 17716 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_cta); | |||
| 17717 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_i: | |||
| 17718 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_l: | |||
| 17719 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_ll: | |||
| 17720 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_sys); | |||
| 17721 | case NVPTX::BI__nvvm_atom_cta_max_gen_i: | |||
| 17722 | case NVPTX::BI__nvvm_atom_cta_max_gen_ui: | |||
| 17723 | case NVPTX::BI__nvvm_atom_cta_max_gen_l: | |||
| 17724 | case NVPTX::BI__nvvm_atom_cta_max_gen_ul: | |||
| 17725 | case NVPTX::BI__nvvm_atom_cta_max_gen_ll: | |||
| 17726 | case NVPTX::BI__nvvm_atom_cta_max_gen_ull: | |||
| 17727 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_cta); | |||
| 17728 | case NVPTX::BI__nvvm_atom_sys_max_gen_i: | |||
| 17729 | case NVPTX::BI__nvvm_atom_sys_max_gen_ui: | |||
| 17730 | case NVPTX::BI__nvvm_atom_sys_max_gen_l: | |||
| 17731 | case NVPTX::BI__nvvm_atom_sys_max_gen_ul: | |||
| 17732 | case NVPTX::BI__nvvm_atom_sys_max_gen_ll: | |||
| 17733 | case NVPTX::BI__nvvm_atom_sys_max_gen_ull: | |||
| 17734 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_sys); | |||
| 17735 | case NVPTX::BI__nvvm_atom_cta_min_gen_i: | |||
| 17736 | case NVPTX::BI__nvvm_atom_cta_min_gen_ui: | |||
| 17737 | case NVPTX::BI__nvvm_atom_cta_min_gen_l: | |||
| 17738 | case NVPTX::BI__nvvm_atom_cta_min_gen_ul: | |||
| 17739 | case NVPTX::BI__nvvm_atom_cta_min_gen_ll: | |||
| 17740 | case NVPTX::BI__nvvm_atom_cta_min_gen_ull: | |||
| 17741 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_cta); | |||
| 17742 | case NVPTX::BI__nvvm_atom_sys_min_gen_i: | |||
| 17743 | case NVPTX::BI__nvvm_atom_sys_min_gen_ui: | |||
| 17744 | case NVPTX::BI__nvvm_atom_sys_min_gen_l: | |||
| 17745 | case NVPTX::BI__nvvm_atom_sys_min_gen_ul: | |||
| 17746 | case NVPTX::BI__nvvm_atom_sys_min_gen_ll: | |||
| 17747 | case NVPTX::BI__nvvm_atom_sys_min_gen_ull: | |||
| 17748 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_sys); | |||
| 17749 | case NVPTX::BI__nvvm_atom_cta_inc_gen_ui: | |||
| 17750 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_cta); | |||
| 17751 | case NVPTX::BI__nvvm_atom_cta_dec_gen_ui: | |||
| 17752 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_cta); | |||
| 17753 | case NVPTX::BI__nvvm_atom_sys_inc_gen_ui: | |||
| 17754 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_sys); | |||
| 17755 | case NVPTX::BI__nvvm_atom_sys_dec_gen_ui: | |||
| 17756 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_sys); | |||
| 17757 | case NVPTX::BI__nvvm_atom_cta_and_gen_i: | |||
| 17758 | case NVPTX::BI__nvvm_atom_cta_and_gen_l: | |||
| 17759 | case NVPTX::BI__nvvm_atom_cta_and_gen_ll: | |||
| 17760 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_cta); | |||
| 17761 | case NVPTX::BI__nvvm_atom_sys_and_gen_i: | |||
| 17762 | case NVPTX::BI__nvvm_atom_sys_and_gen_l: | |||
| 17763 | case NVPTX::BI__nvvm_atom_sys_and_gen_ll: | |||
| 17764 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_sys); | |||
| 17765 | case NVPTX::BI__nvvm_atom_cta_or_gen_i: | |||
| 17766 | case NVPTX::BI__nvvm_atom_cta_or_gen_l: | |||
| 17767 | case NVPTX::BI__nvvm_atom_cta_or_gen_ll: | |||
| 17768 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_cta); | |||
| 17769 | case NVPTX::BI__nvvm_atom_sys_or_gen_i: | |||
| 17770 | case NVPTX::BI__nvvm_atom_sys_or_gen_l: | |||
| 17771 | case NVPTX::BI__nvvm_atom_sys_or_gen_ll: | |||
| 17772 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_sys); | |||
| 17773 | case NVPTX::BI__nvvm_atom_cta_xor_gen_i: | |||
| 17774 | case NVPTX::BI__nvvm_atom_cta_xor_gen_l: | |||
| 17775 | case NVPTX::BI__nvvm_atom_cta_xor_gen_ll: | |||
| 17776 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_cta); | |||
| 17777 | case NVPTX::BI__nvvm_atom_sys_xor_gen_i: | |||
| 17778 | case NVPTX::BI__nvvm_atom_sys_xor_gen_l: | |||
| 17779 | case NVPTX::BI__nvvm_atom_sys_xor_gen_ll: | |||
| 17780 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_sys); | |||
| 17781 | case NVPTX::BI__nvvm_atom_cta_cas_gen_i: | |||
| 17782 | case NVPTX::BI__nvvm_atom_cta_cas_gen_l: | |||
| 17783 | case NVPTX::BI__nvvm_atom_cta_cas_gen_ll: { | |||
| 17784 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 17785 | llvm::Type *ElemTy = | |||
| 17786 | ConvertTypeForMem(E->getArg(0)->getType()->getPointeeType()); | |||
| 17787 | return Builder.CreateCall( | |||
| 17788 | CGM.getIntrinsic( | |||
| 17789 | Intrinsic::nvvm_atomic_cas_gen_i_cta, {ElemTy, Ptr->getType()}), | |||
| 17790 | {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))}); | |||
| 17791 | } | |||
| 17792 | case NVPTX::BI__nvvm_atom_sys_cas_gen_i: | |||
| 17793 | case NVPTX::BI__nvvm_atom_sys_cas_gen_l: | |||
| 17794 | case NVPTX::BI__nvvm_atom_sys_cas_gen_ll: { | |||
| 17795 | Value *Ptr = EmitScalarExpr(E->getArg(0)); | |||
| 17796 | llvm::Type *ElemTy = | |||
| 17797 | ConvertTypeForMem(E->getArg(0)->getType()->getPointeeType()); | |||
| 17798 | return Builder.CreateCall( | |||
| 17799 | CGM.getIntrinsic( | |||
| 17800 | Intrinsic::nvvm_atomic_cas_gen_i_sys, {ElemTy, Ptr->getType()}), | |||
| 17801 | {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))}); | |||
| 17802 | } | |||
| 17803 | case NVPTX::BI__nvvm_match_all_sync_i32p: | |||
| 17804 | case NVPTX::BI__nvvm_match_all_sync_i64p: { | |||
| 17805 | Value *Mask = EmitScalarExpr(E->getArg(0)); | |||
| 17806 | Value *Val = EmitScalarExpr(E->getArg(1)); | |||
| 17807 | Address PredOutPtr = EmitPointerWithAlignment(E->getArg(2)); | |||
| 17808 | Value *ResultPair = Builder.CreateCall( | |||
| 17809 | CGM.getIntrinsic(BuiltinID == NVPTX::BI__nvvm_match_all_sync_i32p | |||
| 17810 | ? Intrinsic::nvvm_match_all_sync_i32p | |||
| 17811 | : Intrinsic::nvvm_match_all_sync_i64p), | |||
| 17812 | {Mask, Val}); | |||
| 17813 | Value *Pred = Builder.CreateZExt(Builder.CreateExtractValue(ResultPair, 1), | |||
| 17814 | PredOutPtr.getElementType()); | |||
| 17815 | Builder.CreateStore(Pred, PredOutPtr); | |||
| 17816 | return Builder.CreateExtractValue(ResultPair, 0); | |||
| 17817 | } | |||
| 17818 | ||||
| 17819 | // FP MMA loads | |||
| 17820 | case NVPTX::BI__hmma_m16n16k16_ld_a: | |||
| 17821 | case NVPTX::BI__hmma_m16n16k16_ld_b: | |||
| 17822 | case NVPTX::BI__hmma_m16n16k16_ld_c_f16: | |||
| 17823 | case NVPTX::BI__hmma_m16n16k16_ld_c_f32: | |||
| 17824 | case NVPTX::BI__hmma_m32n8k16_ld_a: | |||
| 17825 | case NVPTX::BI__hmma_m32n8k16_ld_b: | |||
| 17826 | case NVPTX::BI__hmma_m32n8k16_ld_c_f16: | |||
| 17827 | case NVPTX::BI__hmma_m32n8k16_ld_c_f32: | |||
| 17828 | case NVPTX::BI__hmma_m8n32k16_ld_a: | |||
| 17829 | case NVPTX::BI__hmma_m8n32k16_ld_b: | |||
| 17830 | case NVPTX::BI__hmma_m8n32k16_ld_c_f16: | |||
| 17831 | case NVPTX::BI__hmma_m8n32k16_ld_c_f32: | |||
| 17832 | // Integer MMA loads. | |||
| 17833 | case NVPTX::BI__imma_m16n16k16_ld_a_s8: | |||
| 17834 | case NVPTX::BI__imma_m16n16k16_ld_a_u8: | |||
| 17835 | case NVPTX::BI__imma_m16n16k16_ld_b_s8: | |||
| 17836 | case NVPTX::BI__imma_m16n16k16_ld_b_u8: | |||
| 17837 | case NVPTX::BI__imma_m16n16k16_ld_c: | |||
| 17838 | case NVPTX::BI__imma_m32n8k16_ld_a_s8: | |||
| 17839 | case NVPTX::BI__imma_m32n8k16_ld_a_u8: | |||
| 17840 | case NVPTX::BI__imma_m32n8k16_ld_b_s8: | |||
| 17841 | case NVPTX::BI__imma_m32n8k16_ld_b_u8: | |||
| 17842 | case NVPTX::BI__imma_m32n8k16_ld_c: | |||
| 17843 | case NVPTX::BI__imma_m8n32k16_ld_a_s8: | |||
| 17844 | case NVPTX::BI__imma_m8n32k16_ld_a_u8: | |||
| 17845 | case NVPTX::BI__imma_m8n32k16_ld_b_s8: | |||
| 17846 | case NVPTX::BI__imma_m8n32k16_ld_b_u8: | |||
| 17847 | case NVPTX::BI__imma_m8n32k16_ld_c: | |||
| 17848 | // Sub-integer MMA loads. | |||
| 17849 | case NVPTX::BI__imma_m8n8k32_ld_a_s4: | |||
| 17850 | case NVPTX::BI__imma_m8n8k32_ld_a_u4: | |||
| 17851 | case NVPTX::BI__imma_m8n8k32_ld_b_s4: | |||
| 17852 | case NVPTX::BI__imma_m8n8k32_ld_b_u4: | |||
| 17853 | case NVPTX::BI__imma_m8n8k32_ld_c: | |||
| 17854 | case NVPTX::BI__bmma_m8n8k128_ld_a_b1: | |||
| 17855 | case NVPTX::BI__bmma_m8n8k128_ld_b_b1: | |||
| 17856 | case NVPTX::BI__bmma_m8n8k128_ld_c: | |||
| 17857 | // Double MMA loads. | |||
| 17858 | case NVPTX::BI__dmma_m8n8k4_ld_a: | |||
| 17859 | case NVPTX::BI__dmma_m8n8k4_ld_b: | |||
| 17860 | case NVPTX::BI__dmma_m8n8k4_ld_c: | |||
| 17861 | // Alternate float MMA loads. | |||
| 17862 | case NVPTX::BI__mma_bf16_m16n16k16_ld_a: | |||
| 17863 | case NVPTX::BI__mma_bf16_m16n16k16_ld_b: | |||
| 17864 | case NVPTX::BI__mma_bf16_m8n32k16_ld_a: | |||
| 17865 | case NVPTX::BI__mma_bf16_m8n32k16_ld_b: | |||
| 17866 | case NVPTX::BI__mma_bf16_m32n8k16_ld_a: | |||
| 17867 | case NVPTX::BI__mma_bf16_m32n8k16_ld_b: | |||
| 17868 | case NVPTX::BI__mma_tf32_m16n16k8_ld_a: | |||
| 17869 | case NVPTX::BI__mma_tf32_m16n16k8_ld_b: | |||
| 17870 | case NVPTX::BI__mma_tf32_m16n16k8_ld_c: { | |||
| 17871 | Address Dst = EmitPointerWithAlignment(E->getArg(0)); | |||
| 17872 | Value *Src = EmitScalarExpr(E->getArg(1)); | |||
| 17873 | Value *Ldm = EmitScalarExpr(E->getArg(2)); | |||
| 17874 | Optional<llvm::APSInt> isColMajorArg = | |||
| 17875 | E->getArg(3)->getIntegerConstantExpr(getContext()); | |||
| 17876 | if (!isColMajorArg) | |||
| 17877 | return nullptr; | |||
| 17878 | bool isColMajor = isColMajorArg->getSExtValue(); | |||
| 17879 | NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID); | |||
| 17880 | unsigned IID = isColMajor ? II.IID_col : II.IID_row; | |||
| 17881 | if (IID == 0) | |||
| 17882 | return nullptr; | |||
| 17883 | ||||
| 17884 | Value *Result = | |||
| 17885 | Builder.CreateCall(CGM.getIntrinsic(IID, Src->getType()), {Src, Ldm}); | |||
| 17886 | ||||
| 17887 | // Save returned values. | |||
| 17888 | assert(II.NumResults)(static_cast <bool> (II.NumResults) ? void (0) : __assert_fail ("II.NumResults", "clang/lib/CodeGen/CGBuiltin.cpp", 17888, __extension__ __PRETTY_FUNCTION__)); | |||
| 17889 | if (II.NumResults == 1) { | |||
| 17890 | Builder.CreateAlignedStore(Result, Dst.getPointer(), | |||
| 17891 | CharUnits::fromQuantity(4)); | |||
| 17892 | } else { | |||
| 17893 | for (unsigned i = 0; i < II.NumResults; ++i) { | |||
| 17894 | Builder.CreateAlignedStore( | |||
| 17895 | Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), | |||
| 17896 | Dst.getElementType()), | |||
| 17897 | Builder.CreateGEP(Dst.getElementType(), Dst.getPointer(), | |||
| 17898 | llvm::ConstantInt::get(IntTy, i)), | |||
| 17899 | CharUnits::fromQuantity(4)); | |||
| 17900 | } | |||
| 17901 | } | |||
| 17902 | return Result; | |||
| 17903 | } | |||
| 17904 | ||||
| 17905 | case NVPTX::BI__hmma_m16n16k16_st_c_f16: | |||
| 17906 | case NVPTX::BI__hmma_m16n16k16_st_c_f32: | |||
| 17907 | case NVPTX::BI__hmma_m32n8k16_st_c_f16: | |||
| 17908 | case NVPTX::BI__hmma_m32n8k16_st_c_f32: | |||
| 17909 | case NVPTX::BI__hmma_m8n32k16_st_c_f16: | |||
| 17910 | case NVPTX::BI__hmma_m8n32k16_st_c_f32: | |||
| 17911 | case NVPTX::BI__imma_m16n16k16_st_c_i32: | |||
| 17912 | case NVPTX::BI__imma_m32n8k16_st_c_i32: | |||
| 17913 | case NVPTX::BI__imma_m8n32k16_st_c_i32: | |||
| 17914 | case NVPTX::BI__imma_m8n8k32_st_c_i32: | |||
| 17915 | case NVPTX::BI__bmma_m8n8k128_st_c_i32: | |||
| 17916 | case NVPTX::BI__dmma_m8n8k4_st_c_f64: | |||
| 17917 | case NVPTX::BI__mma_m16n16k8_st_c_f32: { | |||
| 17918 | Value *Dst = EmitScalarExpr(E->getArg(0)); | |||
| 17919 | Address Src = EmitPointerWithAlignment(E->getArg(1)); | |||
| 17920 | Value *Ldm = EmitScalarExpr(E->getArg(2)); | |||
| 17921 | Optional<llvm::APSInt> isColMajorArg = | |||
| 17922 | E->getArg(3)->getIntegerConstantExpr(getContext()); | |||
| 17923 | if (!isColMajorArg) | |||
| 17924 | return nullptr; | |||
| 17925 | bool isColMajor = isColMajorArg->getSExtValue(); | |||
| 17926 | NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID); | |||
| 17927 | unsigned IID = isColMajor ? II.IID_col : II.IID_row; | |||
| 17928 | if (IID == 0) | |||
| 17929 | return nullptr; | |||
| 17930 | Function *Intrinsic = | |||
| 17931 | CGM.getIntrinsic(IID, Dst->getType()); | |||
| 17932 | llvm::Type *ParamType = Intrinsic->getFunctionType()->getParamType(1); | |||
| 17933 | SmallVector<Value *, 10> Values = {Dst}; | |||
| 17934 | for (unsigned i = 0; i < II.NumResults; ++i) { | |||
| 17935 | Value *V = Builder.CreateAlignedLoad( | |||
| 17936 | Src.getElementType(), | |||
| 17937 | Builder.CreateGEP(Src.getElementType(), Src.getPointer(), | |||
| 17938 | llvm::ConstantInt::get(IntTy, i)), | |||
| 17939 | CharUnits::fromQuantity(4)); | |||
| 17940 | Values.push_back(Builder.CreateBitCast(V, ParamType)); | |||
| 17941 | } | |||
| 17942 | Values.push_back(Ldm); | |||
| 17943 | Value *Result = Builder.CreateCall(Intrinsic, Values); | |||
| 17944 | return Result; | |||
| 17945 | } | |||
| 17946 | ||||
| 17947 | // BI__hmma_m16n16k16_mma_<Dtype><CType>(d, a, b, c, layout, satf) --> | |||
| 17948 | // Intrinsic::nvvm_wmma_m16n16k16_mma_sync<layout A,B><DType><CType><Satf> | |||
| 17949 | case NVPTX::BI__hmma_m16n16k16_mma_f16f16: | |||
| 17950 | case NVPTX::BI__hmma_m16n16k16_mma_f32f16: | |||
| 17951 | case NVPTX::BI__hmma_m16n16k16_mma_f32f32: | |||
| 17952 | case NVPTX::BI__hmma_m16n16k16_mma_f16f32: | |||
| 17953 | case NVPTX::BI__hmma_m32n8k16_mma_f16f16: | |||
| 17954 | case NVPTX::BI__hmma_m32n8k16_mma_f32f16: | |||
| 17955 | case NVPTX::BI__hmma_m32n8k16_mma_f32f32: | |||
| 17956 | case NVPTX::BI__hmma_m32n8k16_mma_f16f32: | |||
| 17957 | case NVPTX::BI__hmma_m8n32k16_mma_f16f16: | |||
| 17958 | case NVPTX::BI__hmma_m8n32k16_mma_f32f16: | |||
| 17959 | case NVPTX::BI__hmma_m8n32k16_mma_f32f32: | |||
| 17960 | case NVPTX::BI__hmma_m8n32k16_mma_f16f32: | |||
| 17961 | case NVPTX::BI__imma_m16n16k16_mma_s8: | |||
| 17962 | case NVPTX::BI__imma_m16n16k16_mma_u8: | |||
| 17963 | case NVPTX::BI__imma_m32n8k16_mma_s8: | |||
| 17964 | case NVPTX::BI__imma_m32n8k16_mma_u8: | |||
| 17965 | case NVPTX::BI__imma_m8n32k16_mma_s8: | |||
| 17966 | case NVPTX::BI__imma_m8n32k16_mma_u8: | |||
| 17967 | case NVPTX::BI__imma_m8n8k32_mma_s4: | |||
| 17968 | case NVPTX::BI__imma_m8n8k32_mma_u4: | |||
| 17969 | case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1: | |||
| 17970 | case NVPTX::BI__bmma_m8n8k128_mma_and_popc_b1: | |||
| 17971 | case NVPTX::BI__dmma_m8n8k4_mma_f64: | |||
| 17972 | case NVPTX::BI__mma_bf16_m16n16k16_mma_f32: | |||
| 17973 | case NVPTX::BI__mma_bf16_m8n32k16_mma_f32: | |||
| 17974 | case NVPTX::BI__mma_bf16_m32n8k16_mma_f32: | |||
| 17975 | case NVPTX::BI__mma_tf32_m16n16k8_mma_f32: { | |||
| 17976 | Address Dst = EmitPointerWithAlignment(E->getArg(0)); | |||
| 17977 | Address SrcA = EmitPointerWithAlignment(E->getArg(1)); | |||
| 17978 | Address SrcB = EmitPointerWithAlignment(E->getArg(2)); | |||
| 17979 | Address SrcC = EmitPointerWithAlignment(E->getArg(3)); | |||
| 17980 | Optional<llvm::APSInt> LayoutArg = | |||
| 17981 | E->getArg(4)->getIntegerConstantExpr(getContext()); | |||
| 17982 | if (!LayoutArg) | |||
| 17983 | return nullptr; | |||
| 17984 | int Layout = LayoutArg->getSExtValue(); | |||
| 17985 | if (Layout < 0 || Layout > 3) | |||
| 17986 | return nullptr; | |||
| 17987 | llvm::APSInt SatfArg; | |||
| 17988 | if (BuiltinID == NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1 || | |||
| 17989 | BuiltinID == NVPTX::BI__bmma_m8n8k128_mma_and_popc_b1) | |||
| 17990 | SatfArg = 0; // .b1 does not have satf argument. | |||
| 17991 | else if (Optional<llvm::APSInt> OptSatfArg = | |||
| 17992 | E->getArg(5)->getIntegerConstantExpr(getContext())) | |||
| 17993 | SatfArg = *OptSatfArg; | |||
| 17994 | else | |||
| 17995 | return nullptr; | |||
| 17996 | bool Satf = SatfArg.getSExtValue(); | |||
| 17997 | NVPTXMmaInfo MI = getNVPTXMmaInfo(BuiltinID); | |||
| 17998 | unsigned IID = MI.getMMAIntrinsic(Layout, Satf); | |||
| 17999 | if (IID == 0) // Unsupported combination of Layout/Satf. | |||
| 18000 | return nullptr; | |||
| 18001 | ||||
| 18002 | SmallVector<Value *, 24> Values; | |||
| 18003 | Function *Intrinsic = CGM.getIntrinsic(IID); | |||
| 18004 | llvm::Type *AType = Intrinsic->getFunctionType()->getParamType(0); | |||
| 18005 | // Load A | |||
| 18006 | for (unsigned i = 0; i < MI.NumEltsA; ++i) { | |||
| 18007 | Value *V = Builder.CreateAlignedLoad( | |||
| 18008 | SrcA.getElementType(), | |||
| 18009 | Builder.CreateGEP(SrcA.getElementType(), SrcA.getPointer(), | |||
| 18010 | llvm::ConstantInt::get(IntTy, i)), | |||
| 18011 | CharUnits::fromQuantity(4)); | |||
| 18012 | Values.push_back(Builder.CreateBitCast(V, AType)); | |||
| 18013 | } | |||
| 18014 | // Load B | |||
| 18015 | llvm::Type *BType = Intrinsic->getFunctionType()->getParamType(MI.NumEltsA); | |||
| 18016 | for (unsigned i = 0; i < MI.NumEltsB; ++i) { | |||
| 18017 | Value *V = Builder.CreateAlignedLoad( | |||
| 18018 | SrcB.getElementType(), | |||
| 18019 | Builder.CreateGEP(SrcB.getElementType(), SrcB.getPointer(), | |||
| 18020 | llvm::ConstantInt::get(IntTy, i)), | |||
| 18021 | CharUnits::fromQuantity(4)); | |||
| 18022 | Values.push_back(Builder.CreateBitCast(V, BType)); | |||
| 18023 | } | |||
| 18024 | // Load C | |||
| 18025 | llvm::Type *CType = | |||
| 18026 | Intrinsic->getFunctionType()->getParamType(MI.NumEltsA + MI.NumEltsB); | |||
| 18027 | for (unsigned i = 0; i < MI.NumEltsC; ++i) { | |||
| 18028 | Value *V = Builder.CreateAlignedLoad( | |||
| 18029 | SrcC.getElementType(), | |||
| 18030 | Builder.CreateGEP(SrcC.getElementType(), SrcC.getPointer(), | |||
| 18031 | llvm::ConstantInt::get(IntTy, i)), | |||
| 18032 | CharUnits::fromQuantity(4)); | |||
| 18033 | Values.push_back(Builder.CreateBitCast(V, CType)); | |||
| 18034 | } | |||
| 18035 | Value *Result = Builder.CreateCall(Intrinsic, Values); | |||
| 18036 | llvm::Type *DType = Dst.getElementType(); | |||
| 18037 | for (unsigned i = 0; i < MI.NumEltsD; ++i) | |||
| 18038 | Builder.CreateAlignedStore( | |||
| 18039 | Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), DType), | |||
| 18040 | Builder.CreateGEP(Dst.getElementType(), Dst.getPointer(), | |||
| 18041 | llvm::ConstantInt::get(IntTy, i)), | |||
| 18042 | CharUnits::fromQuantity(4)); | |||
| 18043 | return Result; | |||
| 18044 | } | |||
| 18045 | default: | |||
| 18046 | return nullptr; | |||
| 18047 | } | |||
| 18048 | } | |||
| 18049 | ||||
| 18050 | namespace { | |||
| 18051 | struct BuiltinAlignArgs { | |||
| 18052 | llvm::Value *Src = nullptr; | |||
| 18053 | llvm::Type *SrcType = nullptr; | |||
| 18054 | llvm::Value *Alignment = nullptr; | |||
| 18055 | llvm::Value *Mask = nullptr; | |||
| 18056 | llvm::IntegerType *IntType = nullptr; | |||
| 18057 | ||||
| 18058 | BuiltinAlignArgs(const CallExpr *E, CodeGenFunction &CGF) { | |||
| 18059 | QualType AstType = E->getArg(0)->getType(); | |||
| 18060 | if (AstType->isArrayType()) | |||
| 18061 | Src = CGF.EmitArrayToPointerDecay(E->getArg(0)).getPointer(); | |||
| 18062 | else | |||
| 18063 | Src = CGF.EmitScalarExpr(E->getArg(0)); | |||
| 18064 | SrcType = Src->getType(); | |||
| 18065 | if (SrcType->isPointerTy()) { | |||
| 18066 | IntType = IntegerType::get( | |||
| 18067 | CGF.getLLVMContext(), | |||
| 18068 | CGF.CGM.getDataLayout().getIndexTypeSizeInBits(SrcType)); | |||
| 18069 | } else { | |||
| 18070 | assert(SrcType->isIntegerTy())(static_cast <bool> (SrcType->isIntegerTy()) ? void ( 0) : __assert_fail ("SrcType->isIntegerTy()", "clang/lib/CodeGen/CGBuiltin.cpp" , 18070, __extension__ __PRETTY_FUNCTION__)); | |||
| 18071 | IntType = cast<llvm::IntegerType>(SrcType); | |||
| 18072 | } | |||
| 18073 | Alignment = CGF.EmitScalarExpr(E->getArg(1)); | |||
| 18074 | Alignment = CGF.Builder.CreateZExtOrTrunc(Alignment, IntType, "alignment"); | |||
| 18075 | auto *One = llvm::ConstantInt::get(IntType, 1); | |||
| 18076 | Mask = CGF.Builder.CreateSub(Alignment, One, "mask"); | |||
| 18077 | } | |||
| 18078 | }; | |||
| 18079 | } // namespace | |||
| 18080 | ||||
| 18081 | /// Generate (x & (y-1)) == 0. | |||
| 18082 | RValue CodeGenFunction::EmitBuiltinIsAligned(const CallExpr *E) { | |||
| 18083 | BuiltinAlignArgs Args(E, *this); | |||
| 18084 | llvm::Value *SrcAddress = Args.Src; | |||
| 18085 | if (Args.SrcType->isPointerTy()) | |||
| 18086 | SrcAddress = | |||
| 18087 | Builder.CreateBitOrPointerCast(Args.Src, Args.IntType, "src_addr"); | |||
| 18088 | return RValue::get(Builder.CreateICmpEQ( | |||
| 18089 | Builder.CreateAnd(SrcAddress, Args.Mask, "set_bits"), | |||
| 18090 | llvm::Constant::getNullValue(Args.IntType), "is_aligned")); | |||
| 18091 | } | |||
| 18092 | ||||
| 18093 | /// Generate (x & ~(y-1)) to align down or ((x+(y-1)) & ~(y-1)) to align up. | |||
| 18094 | /// Note: For pointer types we can avoid ptrtoint/inttoptr pairs by using the | |||
| 18095 | /// llvm.ptrmask instrinsic (with a GEP before in the align_up case). | |||
| 18096 | /// TODO: actually use ptrmask once most optimization passes know about it. | |||
| 18097 | RValue CodeGenFunction::EmitBuiltinAlignTo(const CallExpr *E, bool AlignUp) { | |||
| 18098 | BuiltinAlignArgs Args(E, *this); | |||
| 18099 | llvm::Value *SrcAddr = Args.Src; | |||
| 18100 | if (Args.Src->getType()->isPointerTy()) | |||
| 18101 | SrcAddr = Builder.CreatePtrToInt(Args.Src, Args.IntType, "intptr"); | |||
| 18102 | llvm::Value *SrcForMask = SrcAddr; | |||
| 18103 | if (AlignUp) { | |||
| 18104 | // When aligning up we have to first add the mask to ensure we go over the | |||
| 18105 | // next alignment value and then align down to the next valid multiple. | |||
| 18106 | // By adding the mask, we ensure that align_up on an already aligned | |||
| 18107 | // value will not change the value. | |||
| 18108 | SrcForMask = Builder.CreateAdd(SrcForMask, Args.Mask, "over_boundary"); | |||
| 18109 | } | |||
| 18110 | // Invert the mask to only clear the lower bits. | |||
| 18111 | llvm::Value *InvertedMask = Builder.CreateNot(Args.Mask, "inverted_mask"); | |||
| 18112 | llvm::Value *Result = | |||
| 18113 | Builder.CreateAnd(SrcForMask, InvertedMask, "aligned_result"); | |||
| 18114 | if (Args.Src->getType()->isPointerTy()) { | |||
| 18115 | /// TODO: Use ptrmask instead of ptrtoint+gep once it is optimized well. | |||
| 18116 | // Result = Builder.CreateIntrinsic( | |||
| 18117 | // Intrinsic::ptrmask, {Args.SrcType, SrcForMask->getType(), Args.IntType}, | |||
| 18118 | // {SrcForMask, NegatedMask}, nullptr, "aligned_result"); | |||
| 18119 | Result->setName("aligned_intptr"); | |||
| 18120 | llvm::Value *Difference = Builder.CreateSub(Result, SrcAddr, "diff"); | |||
| 18121 | // The result must point to the same underlying allocation. This means we | |||
| 18122 | // can use an inbounds GEP to enable better optimization. | |||
| 18123 | Value *Base = EmitCastToVoidPtr(Args.Src); | |||
| 18124 | if (getLangOpts().isSignedOverflowDefined()) | |||
| 18125 | Result = Builder.CreateGEP(Int8Ty, Base, Difference, "aligned_result"); | |||
| 18126 | else | |||
| 18127 | Result = EmitCheckedInBoundsGEP(Int8Ty, Base, Difference, | |||
| 18128 | /*SignedIndices=*/true, | |||
| 18129 | /*isSubtraction=*/!AlignUp, | |||
| 18130 | E->getExprLoc(), "aligned_result"); | |||
| 18131 | Result = Builder.CreatePointerCast(Result, Args.SrcType); | |||
| 18132 | // Emit an alignment assumption to ensure that the new alignment is | |||
| 18133 | // propagated to loads/stores, etc. | |||
| 18134 | emitAlignmentAssumption(Result, E, E->getExprLoc(), Args.Alignment); | |||
| 18135 | } | |||
| 18136 | assert(Result->getType() == Args.SrcType)(static_cast <bool> (Result->getType() == Args.SrcType ) ? void (0) : __assert_fail ("Result->getType() == Args.SrcType" , "clang/lib/CodeGen/CGBuiltin.cpp", 18136, __extension__ __PRETTY_FUNCTION__ )); | |||
| 18137 | return RValue::get(Result); | |||
| 18138 | } | |||
| 18139 | ||||
| 18140 | Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID, | |||
| 18141 | const CallExpr *E) { | |||
| 18142 | switch (BuiltinID) { | |||
| 18143 | case WebAssembly::BI__builtin_wasm_memory_size: { | |||
| 18144 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 18145 | Value *I = EmitScalarExpr(E->getArg(0)); | |||
| 18146 | Function *Callee = | |||
| 18147 | CGM.getIntrinsic(Intrinsic::wasm_memory_size, ResultType); | |||
| 18148 | return Builder.CreateCall(Callee, I); | |||
| 18149 | } | |||
| 18150 | case WebAssembly::BI__builtin_wasm_memory_grow: { | |||
| 18151 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 18152 | Value *Args[] = {EmitScalarExpr(E->getArg(0)), | |||
| 18153 | EmitScalarExpr(E->getArg(1))}; | |||
| 18154 | Function *Callee = | |||
| 18155 | CGM.getIntrinsic(Intrinsic::wasm_memory_grow, ResultType); | |||
| 18156 | return Builder.CreateCall(Callee, Args); | |||
| 18157 | } | |||
| 18158 | case WebAssembly::BI__builtin_wasm_tls_size: { | |||
| 18159 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 18160 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_size, ResultType); | |||
| 18161 | return Builder.CreateCall(Callee); | |||
| 18162 | } | |||
| 18163 | case WebAssembly::BI__builtin_wasm_tls_align: { | |||
| 18164 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 18165 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_align, ResultType); | |||
| 18166 | return Builder.CreateCall(Callee); | |||
| 18167 | } | |||
| 18168 | case WebAssembly::BI__builtin_wasm_tls_base: { | |||
| 18169 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_base); | |||
| 18170 | return Builder.CreateCall(Callee); | |||
| 18171 | } | |||
| 18172 | case WebAssembly::BI__builtin_wasm_throw: { | |||
| 18173 | Value *Tag = EmitScalarExpr(E->getArg(0)); | |||
| 18174 | Value *Obj = EmitScalarExpr(E->getArg(1)); | |||
| 18175 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_throw); | |||
| 18176 | return Builder.CreateCall(Callee, {Tag, Obj}); | |||
| 18177 | } | |||
| 18178 | case WebAssembly::BI__builtin_wasm_rethrow: { | |||
| 18179 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_rethrow); | |||
| 18180 | return Builder.CreateCall(Callee); | |||
| 18181 | } | |||
| 18182 | case WebAssembly::BI__builtin_wasm_memory_atomic_wait32: { | |||
| 18183 | Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
| 18184 | Value *Expected = EmitScalarExpr(E->getArg(1)); | |||
| 18185 | Value *Timeout = EmitScalarExpr(E->getArg(2)); | |||
| 18186 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_atomic_wait32); | |||
| 18187 | return Builder.CreateCall(Callee, {Addr, Expected, Timeout}); | |||
| 18188 | } | |||
| 18189 | case WebAssembly::BI__builtin_wasm_memory_atomic_wait64: { | |||
| 18190 | Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
| 18191 | Value *Expected = EmitScalarExpr(E->getArg(1)); | |||
| 18192 | Value *Timeout = EmitScalarExpr(E->getArg(2)); | |||
| 18193 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_atomic_wait64); | |||
| 18194 | return Builder.CreateCall(Callee, {Addr, Expected, Timeout}); | |||
| 18195 | } | |||
| 18196 | case WebAssembly::BI__builtin_wasm_memory_atomic_notify: { | |||
| 18197 | Value *Addr = EmitScalarExpr(E->getArg(0)); | |||
| 18198 | Value *Count = EmitScalarExpr(E->getArg(1)); | |||
| 18199 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_atomic_notify); | |||
| 18200 | return Builder.CreateCall(Callee, {Addr, Count}); | |||
| 18201 | } | |||
| 18202 | case WebAssembly::BI__builtin_wasm_trunc_s_i32_f32: | |||
| 18203 | case WebAssembly::BI__builtin_wasm_trunc_s_i32_f64: | |||
| 18204 | case WebAssembly::BI__builtin_wasm_trunc_s_i64_f32: | |||
| 18205 | case WebAssembly::BI__builtin_wasm_trunc_s_i64_f64: { | |||
| 18206 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
| 18207 | llvm::Type *ResT = ConvertType(E->getType()); | |||
| 18208 | Function *Callee = | |||
| 18209 | CGM.getIntrinsic(Intrinsic::wasm_trunc_signed, {ResT, Src->getType()}); | |||
| 18210 | return Builder.CreateCall(Callee, {Src}); | |||
| 18211 | } | |||
| 18212 | case WebAssembly::BI__builtin_wasm_trunc_u_i32_f32: | |||
| 18213 | case WebAssembly::BI__builtin_wasm_trunc_u_i32_f64: | |||
| 18214 | case WebAssembly::BI__builtin_wasm_trunc_u_i64_f32: | |||
| 18215 | case WebAssembly::BI__builtin_wasm_trunc_u_i64_f64: { | |||
| 18216 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
| 18217 | llvm::Type *ResT = ConvertType(E->getType()); | |||
| 18218 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_unsigned, | |||
| 18219 | {ResT, Src->getType()}); | |||
| 18220 | return Builder.CreateCall(Callee, {Src}); | |||
| 18221 | } | |||
| 18222 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f32: | |||
| 18223 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f64: | |||
| 18224 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f32: | |||
| 18225 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f64: | |||
| 18226 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32x4_f32x4: { | |||
| 18227 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
| 18228 | llvm::Type *ResT = ConvertType(E->getType()); | |||
| 18229 | Function *Callee = | |||
| 18230 | CGM.getIntrinsic(Intrinsic::fptosi_sat, {ResT, Src->getType()}); | |||
| 18231 | return Builder.CreateCall(Callee, {Src}); | |||
| 18232 | } | |||
| 18233 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f32: | |||
| 18234 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f64: | |||
| 18235 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f32: | |||
| 18236 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f64: | |||
| 18237 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32x4_f32x4: { | |||
| 18238 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
| 18239 | llvm::Type *ResT = ConvertType(E->getType()); | |||
| 18240 | Function *Callee = | |||
| 18241 | CGM.getIntrinsic(Intrinsic::fptoui_sat, {ResT, Src->getType()}); | |||
| 18242 | return Builder.CreateCall(Callee, {Src}); | |||
| 18243 | } | |||
| 18244 | case WebAssembly::BI__builtin_wasm_min_f32: | |||
| 18245 | case WebAssembly::BI__builtin_wasm_min_f64: | |||
| 18246 | case WebAssembly::BI__builtin_wasm_min_f32x4: | |||
| 18247 | case WebAssembly::BI__builtin_wasm_min_f64x2: { | |||
| 18248 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18249 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18250 | Function *Callee = | |||
| 18251 | CGM.getIntrinsic(Intrinsic::minimum, ConvertType(E->getType())); | |||
| 18252 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18253 | } | |||
| 18254 | case WebAssembly::BI__builtin_wasm_max_f32: | |||
| 18255 | case WebAssembly::BI__builtin_wasm_max_f64: | |||
| 18256 | case WebAssembly::BI__builtin_wasm_max_f32x4: | |||
| 18257 | case WebAssembly::BI__builtin_wasm_max_f64x2: { | |||
| 18258 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18259 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18260 | Function *Callee = | |||
| 18261 | CGM.getIntrinsic(Intrinsic::maximum, ConvertType(E->getType())); | |||
| 18262 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18263 | } | |||
| 18264 | case WebAssembly::BI__builtin_wasm_pmin_f32x4: | |||
| 18265 | case WebAssembly::BI__builtin_wasm_pmin_f64x2: { | |||
| 18266 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18267 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18268 | Function *Callee = | |||
| 18269 | CGM.getIntrinsic(Intrinsic::wasm_pmin, ConvertType(E->getType())); | |||
| 18270 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18271 | } | |||
| 18272 | case WebAssembly::BI__builtin_wasm_pmax_f32x4: | |||
| 18273 | case WebAssembly::BI__builtin_wasm_pmax_f64x2: { | |||
| 18274 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18275 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18276 | Function *Callee = | |||
| 18277 | CGM.getIntrinsic(Intrinsic::wasm_pmax, ConvertType(E->getType())); | |||
| 18278 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18279 | } | |||
| 18280 | case WebAssembly::BI__builtin_wasm_ceil_f32x4: | |||
| 18281 | case WebAssembly::BI__builtin_wasm_floor_f32x4: | |||
| 18282 | case WebAssembly::BI__builtin_wasm_trunc_f32x4: | |||
| 18283 | case WebAssembly::BI__builtin_wasm_nearest_f32x4: | |||
| 18284 | case WebAssembly::BI__builtin_wasm_ceil_f64x2: | |||
| 18285 | case WebAssembly::BI__builtin_wasm_floor_f64x2: | |||
| 18286 | case WebAssembly::BI__builtin_wasm_trunc_f64x2: | |||
| 18287 | case WebAssembly::BI__builtin_wasm_nearest_f64x2: { | |||
| 18288 | unsigned IntNo; | |||
| 18289 | switch (BuiltinID) { | |||
| 18290 | case WebAssembly::BI__builtin_wasm_ceil_f32x4: | |||
| 18291 | case WebAssembly::BI__builtin_wasm_ceil_f64x2: | |||
| 18292 | IntNo = Intrinsic::ceil; | |||
| 18293 | break; | |||
| 18294 | case WebAssembly::BI__builtin_wasm_floor_f32x4: | |||
| 18295 | case WebAssembly::BI__builtin_wasm_floor_f64x2: | |||
| 18296 | IntNo = Intrinsic::floor; | |||
| 18297 | break; | |||
| 18298 | case WebAssembly::BI__builtin_wasm_trunc_f32x4: | |||
| 18299 | case WebAssembly::BI__builtin_wasm_trunc_f64x2: | |||
| 18300 | IntNo = Intrinsic::trunc; | |||
| 18301 | break; | |||
| 18302 | case WebAssembly::BI__builtin_wasm_nearest_f32x4: | |||
| 18303 | case WebAssembly::BI__builtin_wasm_nearest_f64x2: | |||
| 18304 | IntNo = Intrinsic::nearbyint; | |||
| 18305 | break; | |||
| 18306 | default: | |||
| 18307 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18307); | |||
| 18308 | } | |||
| 18309 | Value *Value = EmitScalarExpr(E->getArg(0)); | |||
| 18310 | Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType())); | |||
| 18311 | return Builder.CreateCall(Callee, Value); | |||
| 18312 | } | |||
| 18313 | case WebAssembly::BI__builtin_wasm_swizzle_i8x16: { | |||
| 18314 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
| 18315 | Value *Indices = EmitScalarExpr(E->getArg(1)); | |||
| 18316 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_swizzle); | |||
| 18317 | return Builder.CreateCall(Callee, {Src, Indices}); | |||
| 18318 | } | |||
| 18319 | case WebAssembly::BI__builtin_wasm_add_sat_s_i8x16: | |||
| 18320 | case WebAssembly::BI__builtin_wasm_add_sat_u_i8x16: | |||
| 18321 | case WebAssembly::BI__builtin_wasm_add_sat_s_i16x8: | |||
| 18322 | case WebAssembly::BI__builtin_wasm_add_sat_u_i16x8: | |||
| 18323 | case WebAssembly::BI__builtin_wasm_sub_sat_s_i8x16: | |||
| 18324 | case WebAssembly::BI__builtin_wasm_sub_sat_u_i8x16: | |||
| 18325 | case WebAssembly::BI__builtin_wasm_sub_sat_s_i16x8: | |||
| 18326 | case WebAssembly::BI__builtin_wasm_sub_sat_u_i16x8: { | |||
| 18327 | unsigned IntNo; | |||
| 18328 | switch (BuiltinID) { | |||
| 18329 | case WebAssembly::BI__builtin_wasm_add_sat_s_i8x16: | |||
| 18330 | case WebAssembly::BI__builtin_wasm_add_sat_s_i16x8: | |||
| 18331 | IntNo = Intrinsic::sadd_sat; | |||
| 18332 | break; | |||
| 18333 | case WebAssembly::BI__builtin_wasm_add_sat_u_i8x16: | |||
| 18334 | case WebAssembly::BI__builtin_wasm_add_sat_u_i16x8: | |||
| 18335 | IntNo = Intrinsic::uadd_sat; | |||
| 18336 | break; | |||
| 18337 | case WebAssembly::BI__builtin_wasm_sub_sat_s_i8x16: | |||
| 18338 | case WebAssembly::BI__builtin_wasm_sub_sat_s_i16x8: | |||
| 18339 | IntNo = Intrinsic::wasm_sub_sat_signed; | |||
| 18340 | break; | |||
| 18341 | case WebAssembly::BI__builtin_wasm_sub_sat_u_i8x16: | |||
| 18342 | case WebAssembly::BI__builtin_wasm_sub_sat_u_i16x8: | |||
| 18343 | IntNo = Intrinsic::wasm_sub_sat_unsigned; | |||
| 18344 | break; | |||
| 18345 | default: | |||
| 18346 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18346); | |||
| 18347 | } | |||
| 18348 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18349 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18350 | Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType())); | |||
| 18351 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18352 | } | |||
| 18353 | case WebAssembly::BI__builtin_wasm_abs_i8x16: | |||
| 18354 | case WebAssembly::BI__builtin_wasm_abs_i16x8: | |||
| 18355 | case WebAssembly::BI__builtin_wasm_abs_i32x4: | |||
| 18356 | case WebAssembly::BI__builtin_wasm_abs_i64x2: { | |||
| 18357 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18358 | Value *Neg = Builder.CreateNeg(Vec, "neg"); | |||
| 18359 | Constant *Zero = llvm::Constant::getNullValue(Vec->getType()); | |||
| 18360 | Value *ICmp = Builder.CreateICmpSLT(Vec, Zero, "abscond"); | |||
| 18361 | return Builder.CreateSelect(ICmp, Neg, Vec, "abs"); | |||
| 18362 | } | |||
| 18363 | case WebAssembly::BI__builtin_wasm_min_s_i8x16: | |||
| 18364 | case WebAssembly::BI__builtin_wasm_min_u_i8x16: | |||
| 18365 | case WebAssembly::BI__builtin_wasm_max_s_i8x16: | |||
| 18366 | case WebAssembly::BI__builtin_wasm_max_u_i8x16: | |||
| 18367 | case WebAssembly::BI__builtin_wasm_min_s_i16x8: | |||
| 18368 | case WebAssembly::BI__builtin_wasm_min_u_i16x8: | |||
| 18369 | case WebAssembly::BI__builtin_wasm_max_s_i16x8: | |||
| 18370 | case WebAssembly::BI__builtin_wasm_max_u_i16x8: | |||
| 18371 | case WebAssembly::BI__builtin_wasm_min_s_i32x4: | |||
| 18372 | case WebAssembly::BI__builtin_wasm_min_u_i32x4: | |||
| 18373 | case WebAssembly::BI__builtin_wasm_max_s_i32x4: | |||
| 18374 | case WebAssembly::BI__builtin_wasm_max_u_i32x4: { | |||
| 18375 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18376 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18377 | Value *ICmp; | |||
| 18378 | switch (BuiltinID) { | |||
| 18379 | case WebAssembly::BI__builtin_wasm_min_s_i8x16: | |||
| 18380 | case WebAssembly::BI__builtin_wasm_min_s_i16x8: | |||
| 18381 | case WebAssembly::BI__builtin_wasm_min_s_i32x4: | |||
| 18382 | ICmp = Builder.CreateICmpSLT(LHS, RHS); | |||
| 18383 | break; | |||
| 18384 | case WebAssembly::BI__builtin_wasm_min_u_i8x16: | |||
| 18385 | case WebAssembly::BI__builtin_wasm_min_u_i16x8: | |||
| 18386 | case WebAssembly::BI__builtin_wasm_min_u_i32x4: | |||
| 18387 | ICmp = Builder.CreateICmpULT(LHS, RHS); | |||
| 18388 | break; | |||
| 18389 | case WebAssembly::BI__builtin_wasm_max_s_i8x16: | |||
| 18390 | case WebAssembly::BI__builtin_wasm_max_s_i16x8: | |||
| 18391 | case WebAssembly::BI__builtin_wasm_max_s_i32x4: | |||
| 18392 | ICmp = Builder.CreateICmpSGT(LHS, RHS); | |||
| 18393 | break; | |||
| 18394 | case WebAssembly::BI__builtin_wasm_max_u_i8x16: | |||
| 18395 | case WebAssembly::BI__builtin_wasm_max_u_i16x8: | |||
| 18396 | case WebAssembly::BI__builtin_wasm_max_u_i32x4: | |||
| 18397 | ICmp = Builder.CreateICmpUGT(LHS, RHS); | |||
| 18398 | break; | |||
| 18399 | default: | |||
| 18400 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18400); | |||
| 18401 | } | |||
| 18402 | return Builder.CreateSelect(ICmp, LHS, RHS); | |||
| 18403 | } | |||
| 18404 | case WebAssembly::BI__builtin_wasm_avgr_u_i8x16: | |||
| 18405 | case WebAssembly::BI__builtin_wasm_avgr_u_i16x8: { | |||
| 18406 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18407 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18408 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_avgr_unsigned, | |||
| 18409 | ConvertType(E->getType())); | |||
| 18410 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18411 | } | |||
| 18412 | case WebAssembly::BI__builtin_wasm_q15mulr_sat_s_i16x8: { | |||
| 18413 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18414 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18415 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_q15mulr_sat_signed); | |||
| 18416 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18417 | } | |||
| 18418 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_s_i16x8: | |||
| 18419 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_u_i16x8: | |||
| 18420 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_s_i32x4: | |||
| 18421 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_u_i32x4: { | |||
| 18422 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18423 | unsigned IntNo; | |||
| 18424 | switch (BuiltinID) { | |||
| 18425 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_s_i16x8: | |||
| 18426 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_s_i32x4: | |||
| 18427 | IntNo = Intrinsic::wasm_extadd_pairwise_signed; | |||
| 18428 | break; | |||
| 18429 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_u_i16x8: | |||
| 18430 | case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_u_i32x4: | |||
| 18431 | IntNo = Intrinsic::wasm_extadd_pairwise_unsigned; | |||
| 18432 | break; | |||
| 18433 | default: | |||
| 18434 | llvm_unreachable("unexptected builtin ID")::llvm::llvm_unreachable_internal("unexptected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18434); | |||
| 18435 | } | |||
| 18436 | ||||
| 18437 | Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType())); | |||
| 18438 | return Builder.CreateCall(Callee, Vec); | |||
| 18439 | } | |||
| 18440 | case WebAssembly::BI__builtin_wasm_bitselect: { | |||
| 18441 | Value *V1 = EmitScalarExpr(E->getArg(0)); | |||
| 18442 | Value *V2 = EmitScalarExpr(E->getArg(1)); | |||
| 18443 | Value *C = EmitScalarExpr(E->getArg(2)); | |||
| 18444 | Function *Callee = | |||
| 18445 | CGM.getIntrinsic(Intrinsic::wasm_bitselect, ConvertType(E->getType())); | |||
| 18446 | return Builder.CreateCall(Callee, {V1, V2, C}); | |||
| 18447 | } | |||
| 18448 | case WebAssembly::BI__builtin_wasm_dot_s_i32x4_i16x8: { | |||
| 18449 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18450 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18451 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_dot); | |||
| 18452 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18453 | } | |||
| 18454 | case WebAssembly::BI__builtin_wasm_popcnt_i8x16: { | |||
| 18455 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18456 | Function *Callee = | |||
| 18457 | CGM.getIntrinsic(Intrinsic::ctpop, ConvertType(E->getType())); | |||
| 18458 | return Builder.CreateCall(Callee, {Vec}); | |||
| 18459 | } | |||
| 18460 | case WebAssembly::BI__builtin_wasm_any_true_v128: | |||
| 18461 | case WebAssembly::BI__builtin_wasm_all_true_i8x16: | |||
| 18462 | case WebAssembly::BI__builtin_wasm_all_true_i16x8: | |||
| 18463 | case WebAssembly::BI__builtin_wasm_all_true_i32x4: | |||
| 18464 | case WebAssembly::BI__builtin_wasm_all_true_i64x2: { | |||
| 18465 | unsigned IntNo; | |||
| 18466 | switch (BuiltinID) { | |||
| 18467 | case WebAssembly::BI__builtin_wasm_any_true_v128: | |||
| 18468 | IntNo = Intrinsic::wasm_anytrue; | |||
| 18469 | break; | |||
| 18470 | case WebAssembly::BI__builtin_wasm_all_true_i8x16: | |||
| 18471 | case WebAssembly::BI__builtin_wasm_all_true_i16x8: | |||
| 18472 | case WebAssembly::BI__builtin_wasm_all_true_i32x4: | |||
| 18473 | case WebAssembly::BI__builtin_wasm_all_true_i64x2: | |||
| 18474 | IntNo = Intrinsic::wasm_alltrue; | |||
| 18475 | break; | |||
| 18476 | default: | |||
| 18477 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18477); | |||
| 18478 | } | |||
| 18479 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18480 | Function *Callee = CGM.getIntrinsic(IntNo, Vec->getType()); | |||
| 18481 | return Builder.CreateCall(Callee, {Vec}); | |||
| 18482 | } | |||
| 18483 | case WebAssembly::BI__builtin_wasm_bitmask_i8x16: | |||
| 18484 | case WebAssembly::BI__builtin_wasm_bitmask_i16x8: | |||
| 18485 | case WebAssembly::BI__builtin_wasm_bitmask_i32x4: | |||
| 18486 | case WebAssembly::BI__builtin_wasm_bitmask_i64x2: { | |||
| 18487 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18488 | Function *Callee = | |||
| 18489 | CGM.getIntrinsic(Intrinsic::wasm_bitmask, Vec->getType()); | |||
| 18490 | return Builder.CreateCall(Callee, {Vec}); | |||
| 18491 | } | |||
| 18492 | case WebAssembly::BI__builtin_wasm_abs_f32x4: | |||
| 18493 | case WebAssembly::BI__builtin_wasm_abs_f64x2: { | |||
| 18494 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18495 | Function *Callee = CGM.getIntrinsic(Intrinsic::fabs, Vec->getType()); | |||
| 18496 | return Builder.CreateCall(Callee, {Vec}); | |||
| 18497 | } | |||
| 18498 | case WebAssembly::BI__builtin_wasm_sqrt_f32x4: | |||
| 18499 | case WebAssembly::BI__builtin_wasm_sqrt_f64x2: { | |||
| 18500 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18501 | Function *Callee = CGM.getIntrinsic(Intrinsic::sqrt, Vec->getType()); | |||
| 18502 | return Builder.CreateCall(Callee, {Vec}); | |||
| 18503 | } | |||
| 18504 | case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8: | |||
| 18505 | case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8: | |||
| 18506 | case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4: | |||
| 18507 | case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4: { | |||
| 18508 | Value *Low = EmitScalarExpr(E->getArg(0)); | |||
| 18509 | Value *High = EmitScalarExpr(E->getArg(1)); | |||
| 18510 | unsigned IntNo; | |||
| 18511 | switch (BuiltinID) { | |||
| 18512 | case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8: | |||
| 18513 | case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4: | |||
| 18514 | IntNo = Intrinsic::wasm_narrow_signed; | |||
| 18515 | break; | |||
| 18516 | case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8: | |||
| 18517 | case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4: | |||
| 18518 | IntNo = Intrinsic::wasm_narrow_unsigned; | |||
| 18519 | break; | |||
| 18520 | default: | |||
| 18521 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18521); | |||
| 18522 | } | |||
| 18523 | Function *Callee = | |||
| 18524 | CGM.getIntrinsic(IntNo, {ConvertType(E->getType()), Low->getType()}); | |||
| 18525 | return Builder.CreateCall(Callee, {Low, High}); | |||
| 18526 | } | |||
| 18527 | case WebAssembly::BI__builtin_wasm_trunc_sat_s_zero_f64x2_i32x4: | |||
| 18528 | case WebAssembly::BI__builtin_wasm_trunc_sat_u_zero_f64x2_i32x4: { | |||
| 18529 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18530 | unsigned IntNo; | |||
| 18531 | switch (BuiltinID) { | |||
| 18532 | case WebAssembly::BI__builtin_wasm_trunc_sat_s_zero_f64x2_i32x4: | |||
| 18533 | IntNo = Intrinsic::fptosi_sat; | |||
| 18534 | break; | |||
| 18535 | case WebAssembly::BI__builtin_wasm_trunc_sat_u_zero_f64x2_i32x4: | |||
| 18536 | IntNo = Intrinsic::fptoui_sat; | |||
| 18537 | break; | |||
| 18538 | default: | |||
| 18539 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18539); | |||
| 18540 | } | |||
| 18541 | llvm::Type *SrcT = Vec->getType(); | |||
| 18542 | llvm::Type *TruncT = SrcT->getWithNewType(Builder.getInt32Ty()); | |||
| 18543 | Function *Callee = CGM.getIntrinsic(IntNo, {TruncT, SrcT}); | |||
| 18544 | Value *Trunc = Builder.CreateCall(Callee, Vec); | |||
| 18545 | Value *Splat = Constant::getNullValue(TruncT); | |||
| 18546 | return Builder.CreateShuffleVector(Trunc, Splat, ArrayRef<int>{0, 1, 2, 3}); | |||
| 18547 | } | |||
| 18548 | case WebAssembly::BI__builtin_wasm_shuffle_i8x16: { | |||
| 18549 | Value *Ops[18]; | |||
| 18550 | size_t OpIdx = 0; | |||
| 18551 | Ops[OpIdx++] = EmitScalarExpr(E->getArg(0)); | |||
| 18552 | Ops[OpIdx++] = EmitScalarExpr(E->getArg(1)); | |||
| 18553 | while (OpIdx < 18) { | |||
| 18554 | Optional<llvm::APSInt> LaneConst = | |||
| 18555 | E->getArg(OpIdx)->getIntegerConstantExpr(getContext()); | |||
| 18556 | assert(LaneConst && "Constant arg isn't actually constant?")(static_cast <bool> (LaneConst && "Constant arg isn't actually constant?" ) ? void (0) : __assert_fail ("LaneConst && \"Constant arg isn't actually constant?\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 18556, __extension__ __PRETTY_FUNCTION__ )); | |||
| 18557 | Ops[OpIdx++] = llvm::ConstantInt::get(getLLVMContext(), *LaneConst); | |||
| 18558 | } | |||
| 18559 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_shuffle); | |||
| 18560 | return Builder.CreateCall(Callee, Ops); | |||
| 18561 | } | |||
| 18562 | case WebAssembly::BI__builtin_wasm_fma_f32x4: | |||
| 18563 | case WebAssembly::BI__builtin_wasm_fms_f32x4: | |||
| 18564 | case WebAssembly::BI__builtin_wasm_fma_f64x2: | |||
| 18565 | case WebAssembly::BI__builtin_wasm_fms_f64x2: { | |||
| 18566 | Value *A = EmitScalarExpr(E->getArg(0)); | |||
| 18567 | Value *B = EmitScalarExpr(E->getArg(1)); | |||
| 18568 | Value *C = EmitScalarExpr(E->getArg(2)); | |||
| 18569 | unsigned IntNo; | |||
| 18570 | switch (BuiltinID) { | |||
| 18571 | case WebAssembly::BI__builtin_wasm_fma_f32x4: | |||
| 18572 | case WebAssembly::BI__builtin_wasm_fma_f64x2: | |||
| 18573 | IntNo = Intrinsic::wasm_fma; | |||
| 18574 | break; | |||
| 18575 | case WebAssembly::BI__builtin_wasm_fms_f32x4: | |||
| 18576 | case WebAssembly::BI__builtin_wasm_fms_f64x2: | |||
| 18577 | IntNo = Intrinsic::wasm_fms; | |||
| 18578 | break; | |||
| 18579 | default: | |||
| 18580 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18580); | |||
| 18581 | } | |||
| 18582 | Function *Callee = CGM.getIntrinsic(IntNo, A->getType()); | |||
| 18583 | return Builder.CreateCall(Callee, {A, B, C}); | |||
| 18584 | } | |||
| 18585 | case WebAssembly::BI__builtin_wasm_laneselect_i8x16: | |||
| 18586 | case WebAssembly::BI__builtin_wasm_laneselect_i16x8: | |||
| 18587 | case WebAssembly::BI__builtin_wasm_laneselect_i32x4: | |||
| 18588 | case WebAssembly::BI__builtin_wasm_laneselect_i64x2: { | |||
| 18589 | Value *A = EmitScalarExpr(E->getArg(0)); | |||
| 18590 | Value *B = EmitScalarExpr(E->getArg(1)); | |||
| 18591 | Value *C = EmitScalarExpr(E->getArg(2)); | |||
| 18592 | Function *Callee = | |||
| 18593 | CGM.getIntrinsic(Intrinsic::wasm_laneselect, A->getType()); | |||
| 18594 | return Builder.CreateCall(Callee, {A, B, C}); | |||
| 18595 | } | |||
| 18596 | case WebAssembly::BI__builtin_wasm_relaxed_swizzle_i8x16: { | |||
| 18597 | Value *Src = EmitScalarExpr(E->getArg(0)); | |||
| 18598 | Value *Indices = EmitScalarExpr(E->getArg(1)); | |||
| 18599 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_relaxed_swizzle); | |||
| 18600 | return Builder.CreateCall(Callee, {Src, Indices}); | |||
| 18601 | } | |||
| 18602 | case WebAssembly::BI__builtin_wasm_relaxed_min_f32x4: | |||
| 18603 | case WebAssembly::BI__builtin_wasm_relaxed_max_f32x4: | |||
| 18604 | case WebAssembly::BI__builtin_wasm_relaxed_min_f64x2: | |||
| 18605 | case WebAssembly::BI__builtin_wasm_relaxed_max_f64x2: { | |||
| 18606 | Value *LHS = EmitScalarExpr(E->getArg(0)); | |||
| 18607 | Value *RHS = EmitScalarExpr(E->getArg(1)); | |||
| 18608 | unsigned IntNo; | |||
| 18609 | switch (BuiltinID) { | |||
| 18610 | case WebAssembly::BI__builtin_wasm_relaxed_min_f32x4: | |||
| 18611 | case WebAssembly::BI__builtin_wasm_relaxed_min_f64x2: | |||
| 18612 | IntNo = Intrinsic::wasm_relaxed_min; | |||
| 18613 | break; | |||
| 18614 | case WebAssembly::BI__builtin_wasm_relaxed_max_f32x4: | |||
| 18615 | case WebAssembly::BI__builtin_wasm_relaxed_max_f64x2: | |||
| 18616 | IntNo = Intrinsic::wasm_relaxed_max; | |||
| 18617 | break; | |||
| 18618 | default: | |||
| 18619 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18619); | |||
| 18620 | } | |||
| 18621 | Function *Callee = CGM.getIntrinsic(IntNo, LHS->getType()); | |||
| 18622 | return Builder.CreateCall(Callee, {LHS, RHS}); | |||
| 18623 | } | |||
| 18624 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_s_i32x4_f32x4: | |||
| 18625 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_u_i32x4_f32x4: | |||
| 18626 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_s_zero_i32x4_f64x2: | |||
| 18627 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_u_zero_i32x4_f64x2: { | |||
| 18628 | Value *Vec = EmitScalarExpr(E->getArg(0)); | |||
| 18629 | unsigned IntNo; | |||
| 18630 | switch (BuiltinID) { | |||
| 18631 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_s_i32x4_f32x4: | |||
| 18632 | IntNo = Intrinsic::wasm_relaxed_trunc_signed; | |||
| 18633 | break; | |||
| 18634 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_u_i32x4_f32x4: | |||
| 18635 | IntNo = Intrinsic::wasm_relaxed_trunc_unsigned; | |||
| 18636 | break; | |||
| 18637 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_s_zero_i32x4_f64x2: | |||
| 18638 | IntNo = Intrinsic::wasm_relaxed_trunc_signed_zero; | |||
| 18639 | break; | |||
| 18640 | case WebAssembly::BI__builtin_wasm_relaxed_trunc_u_zero_i32x4_f64x2: | |||
| 18641 | IntNo = Intrinsic::wasm_relaxed_trunc_unsigned_zero; | |||
| 18642 | break; | |||
| 18643 | default: | |||
| 18644 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18644); | |||
| 18645 | } | |||
| 18646 | Function *Callee = CGM.getIntrinsic(IntNo); | |||
| 18647 | return Builder.CreateCall(Callee, {Vec}); | |||
| 18648 | } | |||
| 18649 | default: | |||
| 18650 | return nullptr; | |||
| 18651 | } | |||
| 18652 | } | |||
| 18653 | ||||
| 18654 | static std::pair<Intrinsic::ID, unsigned> | |||
| 18655 | getIntrinsicForHexagonNonGCCBuiltin(unsigned BuiltinID) { | |||
| 18656 | struct Info { | |||
| 18657 | unsigned BuiltinID; | |||
| 18658 | Intrinsic::ID IntrinsicID; | |||
| 18659 | unsigned VecLen; | |||
| 18660 | }; | |||
| 18661 | Info Infos[] = { | |||
| 18662 | #define CUSTOM_BUILTIN_MAPPING(x,s) \ | |||
| 18663 | { Hexagon::BI__builtin_HEXAGON_##x, Intrinsic::hexagon_##x, s }, | |||
| 18664 | CUSTOM_BUILTIN_MAPPING(L2_loadrub_pci, 0) | |||
| 18665 | CUSTOM_BUILTIN_MAPPING(L2_loadrb_pci, 0) | |||
| 18666 | CUSTOM_BUILTIN_MAPPING(L2_loadruh_pci, 0) | |||
| 18667 | CUSTOM_BUILTIN_MAPPING(L2_loadrh_pci, 0) | |||
| 18668 | CUSTOM_BUILTIN_MAPPING(L2_loadri_pci, 0) | |||
| 18669 | CUSTOM_BUILTIN_MAPPING(L2_loadrd_pci, 0) | |||
| 18670 | CUSTOM_BUILTIN_MAPPING(L2_loadrub_pcr, 0) | |||
| 18671 | CUSTOM_BUILTIN_MAPPING(L2_loadrb_pcr, 0) | |||
| 18672 | CUSTOM_BUILTIN_MAPPING(L2_loadruh_pcr, 0) | |||
| 18673 | CUSTOM_BUILTIN_MAPPING(L2_loadrh_pcr, 0) | |||
| 18674 | CUSTOM_BUILTIN_MAPPING(L2_loadri_pcr, 0) | |||
| 18675 | CUSTOM_BUILTIN_MAPPING(L2_loadrd_pcr, 0) | |||
| 18676 | CUSTOM_BUILTIN_MAPPING(S2_storerb_pci, 0) | |||
| 18677 | CUSTOM_BUILTIN_MAPPING(S2_storerh_pci, 0) | |||
| 18678 | CUSTOM_BUILTIN_MAPPING(S2_storerf_pci, 0) | |||
| 18679 | CUSTOM_BUILTIN_MAPPING(S2_storeri_pci, 0) | |||
| 18680 | CUSTOM_BUILTIN_MAPPING(S2_storerd_pci, 0) | |||
| 18681 | CUSTOM_BUILTIN_MAPPING(S2_storerb_pcr, 0) | |||
| 18682 | CUSTOM_BUILTIN_MAPPING(S2_storerh_pcr, 0) | |||
| 18683 | CUSTOM_BUILTIN_MAPPING(S2_storerf_pcr, 0) | |||
| 18684 | CUSTOM_BUILTIN_MAPPING(S2_storeri_pcr, 0) | |||
| 18685 | CUSTOM_BUILTIN_MAPPING(S2_storerd_pcr, 0) | |||
| 18686 | // Legacy builtins that take a vector in place of a vector predicate. | |||
| 18687 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstoreq, 64) | |||
| 18688 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorenq, 64) | |||
| 18689 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentq, 64) | |||
| 18690 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentnq, 64) | |||
| 18691 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstoreq_128B, 128) | |||
| 18692 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorenq_128B, 128) | |||
| 18693 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentq_128B, 128) | |||
| 18694 | CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentnq_128B, 128) | |||
| 18695 | #include "clang/Basic/BuiltinsHexagonMapCustomDep.def" | |||
| 18696 | #undef CUSTOM_BUILTIN_MAPPING | |||
| 18697 | }; | |||
| 18698 | ||||
| 18699 | auto CmpInfo = [] (Info A, Info B) { return A.BuiltinID < B.BuiltinID; }; | |||
| 18700 | static const bool SortOnce = (llvm::sort(Infos, CmpInfo), true); | |||
| 18701 | (void)SortOnce; | |||
| 18702 | ||||
| 18703 | const Info *F = std::lower_bound(std::begin(Infos), std::end(Infos), | |||
| 18704 | Info{BuiltinID, 0, 0}, CmpInfo); | |||
| 18705 | if (F == std::end(Infos) || F->BuiltinID != BuiltinID) | |||
| 18706 | return {Intrinsic::not_intrinsic, 0}; | |||
| 18707 | ||||
| 18708 | return {F->IntrinsicID, F->VecLen}; | |||
| 18709 | } | |||
| 18710 | ||||
| 18711 | Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID, | |||
| 18712 | const CallExpr *E) { | |||
| 18713 | Intrinsic::ID ID; | |||
| 18714 | unsigned VecLen; | |||
| 18715 | std::tie(ID, VecLen) = getIntrinsicForHexagonNonGCCBuiltin(BuiltinID); | |||
| 18716 | ||||
| 18717 | auto MakeCircOp = [this, E](unsigned IntID, bool IsLoad) { | |||
| 18718 | // The base pointer is passed by address, so it needs to be loaded. | |||
| 18719 | Address A = EmitPointerWithAlignment(E->getArg(0)); | |||
| 18720 | Address BP = Address(Builder.CreateBitCast( | |||
| 18721 | A.getPointer(), Int8PtrPtrTy), Int8PtrTy, A.getAlignment()); | |||
| 18722 | llvm::Value *Base = Builder.CreateLoad(BP); | |||
| 18723 | // The treatment of both loads and stores is the same: the arguments for | |||
| 18724 | // the builtin are the same as the arguments for the intrinsic. | |||
| 18725 | // Load: | |||
| 18726 | // builtin(Base, Inc, Mod, Start) -> intr(Base, Inc, Mod, Start) | |||
| 18727 | // builtin(Base, Mod, Start) -> intr(Base, Mod, Start) | |||
| 18728 | // Store: | |||
| 18729 | // builtin(Base, Inc, Mod, Val, Start) -> intr(Base, Inc, Mod, Val, Start) | |||
| 18730 | // builtin(Base, Mod, Val, Start) -> intr(Base, Mod, Val, Start) | |||
| 18731 | SmallVector<llvm::Value*,5> Ops = { Base }; | |||
| 18732 | for (unsigned i = 1, e = E->getNumArgs(); i != e; ++i) | |||
| 18733 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 18734 | ||||
| 18735 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops); | |||
| 18736 | // The load intrinsics generate two results (Value, NewBase), stores | |||
| 18737 | // generate one (NewBase). The new base address needs to be stored. | |||
| 18738 | llvm::Value *NewBase = IsLoad ? Builder.CreateExtractValue(Result, 1) | |||
| 18739 | : Result; | |||
| 18740 | llvm::Value *LV = Builder.CreateBitCast( | |||
| 18741 | EmitScalarExpr(E->getArg(0)), NewBase->getType()->getPointerTo()); | |||
| 18742 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); | |||
| 18743 | llvm::Value *RetVal = | |||
| 18744 | Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment()); | |||
| 18745 | if (IsLoad) | |||
| 18746 | RetVal = Builder.CreateExtractValue(Result, 0); | |||
| 18747 | return RetVal; | |||
| 18748 | }; | |||
| 18749 | ||||
| 18750 | // Handle the conversion of bit-reverse load intrinsics to bit code. | |||
| 18751 | // The intrinsic call after this function only reads from memory and the | |||
| 18752 | // write to memory is dealt by the store instruction. | |||
| 18753 | auto MakeBrevLd = [this, E](unsigned IntID, llvm::Type *DestTy) { | |||
| 18754 | // The intrinsic generates one result, which is the new value for the base | |||
| 18755 | // pointer. It needs to be returned. The result of the load instruction is | |||
| 18756 | // passed to intrinsic by address, so the value needs to be stored. | |||
| 18757 | llvm::Value *BaseAddress = | |||
| 18758 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int8PtrTy); | |||
| 18759 | ||||
| 18760 | // Expressions like &(*pt++) will be incremented per evaluation. | |||
| 18761 | // EmitPointerWithAlignment and EmitScalarExpr evaluates the expression | |||
| 18762 | // per call. | |||
| 18763 | Address DestAddr = EmitPointerWithAlignment(E->getArg(1)); | |||
| 18764 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), Int8PtrTy), | |||
| 18765 | Int8Ty, DestAddr.getAlignment()); | |||
| 18766 | llvm::Value *DestAddress = DestAddr.getPointer(); | |||
| 18767 | ||||
| 18768 | // Operands are Base, Dest, Modifier. | |||
| 18769 | // The intrinsic format in LLVM IR is defined as | |||
| 18770 | // { ValueType, i8* } (i8*, i32). | |||
| 18771 | llvm::Value *Result = Builder.CreateCall( | |||
| 18772 | CGM.getIntrinsic(IntID), {BaseAddress, EmitScalarExpr(E->getArg(2))}); | |||
| 18773 | ||||
| 18774 | // The value needs to be stored as the variable is passed by reference. | |||
| 18775 | llvm::Value *DestVal = Builder.CreateExtractValue(Result, 0); | |||
| 18776 | ||||
| 18777 | // The store needs to be truncated to fit the destination type. | |||
| 18778 | // While i32 and i64 are natively supported on Hexagon, i8 and i16 needs | |||
| 18779 | // to be handled with stores of respective destination type. | |||
| 18780 | DestVal = Builder.CreateTrunc(DestVal, DestTy); | |||
| 18781 | ||||
| 18782 | llvm::Value *DestForStore = | |||
| 18783 | Builder.CreateBitCast(DestAddress, DestVal->getType()->getPointerTo()); | |||
| 18784 | Builder.CreateAlignedStore(DestVal, DestForStore, DestAddr.getAlignment()); | |||
| 18785 | // The updated value of the base pointer is returned. | |||
| 18786 | return Builder.CreateExtractValue(Result, 1); | |||
| 18787 | }; | |||
| 18788 | ||||
| 18789 | auto V2Q = [this, VecLen] (llvm::Value *Vec) { | |||
| 18790 | Intrinsic::ID ID = VecLen == 128 ? Intrinsic::hexagon_V6_vandvrt_128B | |||
| 18791 | : Intrinsic::hexagon_V6_vandvrt; | |||
| 18792 | return Builder.CreateCall(CGM.getIntrinsic(ID), | |||
| 18793 | {Vec, Builder.getInt32(-1)}); | |||
| 18794 | }; | |||
| 18795 | auto Q2V = [this, VecLen] (llvm::Value *Pred) { | |||
| 18796 | Intrinsic::ID ID = VecLen == 128 ? Intrinsic::hexagon_V6_vandqrt_128B | |||
| 18797 | : Intrinsic::hexagon_V6_vandqrt; | |||
| 18798 | return Builder.CreateCall(CGM.getIntrinsic(ID), | |||
| 18799 | {Pred, Builder.getInt32(-1)}); | |||
| 18800 | }; | |||
| 18801 | ||||
| 18802 | switch (BuiltinID) { | |||
| 18803 | // These intrinsics return a tuple {Vector, VectorPred} in LLVM IR, | |||
| 18804 | // and the corresponding C/C++ builtins use loads/stores to update | |||
| 18805 | // the predicate. | |||
| 18806 | case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry: | |||
| 18807 | case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry_128B: | |||
| 18808 | case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry: | |||
| 18809 | case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry_128B: { | |||
| 18810 | // Get the type from the 0-th argument. | |||
| 18811 | llvm::Type *VecType = ConvertType(E->getArg(0)->getType()); | |||
| 18812 | Address PredAddr = Builder.CreateElementBitCast( | |||
| 18813 | EmitPointerWithAlignment(E->getArg(2)), VecType); | |||
| 18814 | llvm::Value *PredIn = V2Q(Builder.CreateLoad(PredAddr)); | |||
| 18815 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), | |||
| 18816 | {EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), PredIn}); | |||
| 18817 | ||||
| 18818 | llvm::Value *PredOut = Builder.CreateExtractValue(Result, 1); | |||
| 18819 | Builder.CreateAlignedStore(Q2V(PredOut), PredAddr.getPointer(), | |||
| 18820 | PredAddr.getAlignment()); | |||
| 18821 | return Builder.CreateExtractValue(Result, 0); | |||
| 18822 | } | |||
| 18823 | ||||
| 18824 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstoreq: | |||
| 18825 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorenq: | |||
| 18826 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentq: | |||
| 18827 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentnq: | |||
| 18828 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstoreq_128B: | |||
| 18829 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorenq_128B: | |||
| 18830 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentq_128B: | |||
| 18831 | case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentnq_128B: { | |||
| 18832 | SmallVector<llvm::Value*,4> Ops; | |||
| 18833 | const Expr *PredOp = E->getArg(0); | |||
| 18834 | // There will be an implicit cast to a boolean vector. Strip it. | |||
| 18835 | if (auto *Cast = dyn_cast<ImplicitCastExpr>(PredOp)) { | |||
| 18836 | if (Cast->getCastKind() == CK_BitCast) | |||
| 18837 | PredOp = Cast->getSubExpr(); | |||
| 18838 | Ops.push_back(V2Q(EmitScalarExpr(PredOp))); | |||
| 18839 | } | |||
| 18840 | for (int i = 1, e = E->getNumArgs(); i != e; ++i) | |||
| 18841 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 18842 | return Builder.CreateCall(CGM.getIntrinsic(ID), Ops); | |||
| 18843 | } | |||
| 18844 | ||||
| 18845 | case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pci: | |||
| 18846 | case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pci: | |||
| 18847 | case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pci: | |||
| 18848 | case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pci: | |||
| 18849 | case Hexagon::BI__builtin_HEXAGON_L2_loadri_pci: | |||
| 18850 | case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pci: | |||
| 18851 | case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pcr: | |||
| 18852 | case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pcr: | |||
| 18853 | case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pcr: | |||
| 18854 | case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pcr: | |||
| 18855 | case Hexagon::BI__builtin_HEXAGON_L2_loadri_pcr: | |||
| 18856 | case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pcr: | |||
| 18857 | return MakeCircOp(ID, /*IsLoad=*/true); | |||
| 18858 | case Hexagon::BI__builtin_HEXAGON_S2_storerb_pci: | |||
| 18859 | case Hexagon::BI__builtin_HEXAGON_S2_storerh_pci: | |||
| 18860 | case Hexagon::BI__builtin_HEXAGON_S2_storerf_pci: | |||
| 18861 | case Hexagon::BI__builtin_HEXAGON_S2_storeri_pci: | |||
| 18862 | case Hexagon::BI__builtin_HEXAGON_S2_storerd_pci: | |||
| 18863 | case Hexagon::BI__builtin_HEXAGON_S2_storerb_pcr: | |||
| 18864 | case Hexagon::BI__builtin_HEXAGON_S2_storerh_pcr: | |||
| 18865 | case Hexagon::BI__builtin_HEXAGON_S2_storerf_pcr: | |||
| 18866 | case Hexagon::BI__builtin_HEXAGON_S2_storeri_pcr: | |||
| 18867 | case Hexagon::BI__builtin_HEXAGON_S2_storerd_pcr: | |||
| 18868 | return MakeCircOp(ID, /*IsLoad=*/false); | |||
| 18869 | case Hexagon::BI__builtin_brev_ldub: | |||
| 18870 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrub_pbr, Int8Ty); | |||
| 18871 | case Hexagon::BI__builtin_brev_ldb: | |||
| 18872 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrb_pbr, Int8Ty); | |||
| 18873 | case Hexagon::BI__builtin_brev_lduh: | |||
| 18874 | return MakeBrevLd(Intrinsic::hexagon_L2_loadruh_pbr, Int16Ty); | |||
| 18875 | case Hexagon::BI__builtin_brev_ldh: | |||
| 18876 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrh_pbr, Int16Ty); | |||
| 18877 | case Hexagon::BI__builtin_brev_ldw: | |||
| 18878 | return MakeBrevLd(Intrinsic::hexagon_L2_loadri_pbr, Int32Ty); | |||
| 18879 | case Hexagon::BI__builtin_brev_ldd: | |||
| 18880 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrd_pbr, Int64Ty); | |||
| 18881 | } // switch | |||
| 18882 | ||||
| 18883 | return nullptr; | |||
| 18884 | } | |||
| 18885 | ||||
| 18886 | Value *CodeGenFunction::EmitRISCVBuiltinExpr(unsigned BuiltinID, | |||
| 18887 | const CallExpr *E, | |||
| 18888 | ReturnValueSlot ReturnValue) { | |||
| 18889 | SmallVector<Value *, 4> Ops; | |||
| 18890 | llvm::Type *ResultType = ConvertType(E->getType()); | |||
| 18891 | ||||
| 18892 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) | |||
| 18893 | Ops.push_back(EmitScalarExpr(E->getArg(i))); | |||
| 18894 | ||||
| 18895 | Intrinsic::ID ID = Intrinsic::not_intrinsic; | |||
| 18896 | unsigned NF = 1; | |||
| 18897 | constexpr unsigned TAIL_UNDISTURBED = 0; | |||
| 18898 | ||||
| 18899 | // Required for overloaded intrinsics. | |||
| 18900 | llvm::SmallVector<llvm::Type *, 2> IntrinsicTypes; | |||
| 18901 | switch (BuiltinID) { | |||
| 18902 | default: llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18902); | |||
| 18903 | case RISCV::BI__builtin_riscv_orc_b_32: | |||
| 18904 | case RISCV::BI__builtin_riscv_orc_b_64: | |||
| 18905 | case RISCV::BI__builtin_riscv_clmul: | |||
| 18906 | case RISCV::BI__builtin_riscv_clmulh: | |||
| 18907 | case RISCV::BI__builtin_riscv_clmulr: | |||
| 18908 | case RISCV::BI__builtin_riscv_bcompress_32: | |||
| 18909 | case RISCV::BI__builtin_riscv_bcompress_64: | |||
| 18910 | case RISCV::BI__builtin_riscv_bdecompress_32: | |||
| 18911 | case RISCV::BI__builtin_riscv_bdecompress_64: | |||
| 18912 | case RISCV::BI__builtin_riscv_bfp_32: | |||
| 18913 | case RISCV::BI__builtin_riscv_bfp_64: | |||
| 18914 | case RISCV::BI__builtin_riscv_grev_32: | |||
| 18915 | case RISCV::BI__builtin_riscv_grev_64: | |||
| 18916 | case RISCV::BI__builtin_riscv_gorc_32: | |||
| 18917 | case RISCV::BI__builtin_riscv_gorc_64: | |||
| 18918 | case RISCV::BI__builtin_riscv_shfl_32: | |||
| 18919 | case RISCV::BI__builtin_riscv_shfl_64: | |||
| 18920 | case RISCV::BI__builtin_riscv_unshfl_32: | |||
| 18921 | case RISCV::BI__builtin_riscv_unshfl_64: | |||
| 18922 | case RISCV::BI__builtin_riscv_xperm4: | |||
| 18923 | case RISCV::BI__builtin_riscv_xperm8: | |||
| 18924 | case RISCV::BI__builtin_riscv_xperm_n: | |||
| 18925 | case RISCV::BI__builtin_riscv_xperm_b: | |||
| 18926 | case RISCV::BI__builtin_riscv_xperm_h: | |||
| 18927 | case RISCV::BI__builtin_riscv_xperm_w: | |||
| 18928 | case RISCV::BI__builtin_riscv_crc32_b: | |||
| 18929 | case RISCV::BI__builtin_riscv_crc32_h: | |||
| 18930 | case RISCV::BI__builtin_riscv_crc32_w: | |||
| 18931 | case RISCV::BI__builtin_riscv_crc32_d: | |||
| 18932 | case RISCV::BI__builtin_riscv_crc32c_b: | |||
| 18933 | case RISCV::BI__builtin_riscv_crc32c_h: | |||
| 18934 | case RISCV::BI__builtin_riscv_crc32c_w: | |||
| 18935 | case RISCV::BI__builtin_riscv_crc32c_d: | |||
| 18936 | case RISCV::BI__builtin_riscv_fsl_32: | |||
| 18937 | case RISCV::BI__builtin_riscv_fsr_32: | |||
| 18938 | case RISCV::BI__builtin_riscv_fsl_64: | |||
| 18939 | case RISCV::BI__builtin_riscv_fsr_64: | |||
| 18940 | case RISCV::BI__builtin_riscv_brev8: | |||
| 18941 | case RISCV::BI__builtin_riscv_zip_32: | |||
| 18942 | case RISCV::BI__builtin_riscv_unzip_32: { | |||
| 18943 | switch (BuiltinID) { | |||
| 18944 | default: llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "clang/lib/CodeGen/CGBuiltin.cpp" , 18944); | |||
| 18945 | // Zbb | |||
| 18946 | case RISCV::BI__builtin_riscv_orc_b_32: | |||
| 18947 | case RISCV::BI__builtin_riscv_orc_b_64: | |||
| 18948 | ID = Intrinsic::riscv_orc_b; | |||
| 18949 | break; | |||
| 18950 | ||||
| 18951 | // Zbc | |||
| 18952 | case RISCV::BI__builtin_riscv_clmul: | |||
| 18953 | ID = Intrinsic::riscv_clmul; | |||
| 18954 | break; | |||
| 18955 | case RISCV::BI__builtin_riscv_clmulh: | |||
| 18956 | ID = Intrinsic::riscv_clmulh; | |||
| 18957 | break; | |||
| 18958 | case RISCV::BI__builtin_riscv_clmulr: | |||
| 18959 | ID = Intrinsic::riscv_clmulr; | |||
| 18960 | break; | |||
| 18961 | ||||
| 18962 | // Zbe | |||
| 18963 | case RISCV::BI__builtin_riscv_bcompress_32: | |||
| 18964 | case RISCV::BI__builtin_riscv_bcompress_64: | |||
| 18965 | ID = Intrinsic::riscv_bcompress; | |||
| 18966 | break; | |||
| 18967 | case RISCV::BI__builtin_riscv_bdecompress_32: | |||
| 18968 | case RISCV::BI__builtin_riscv_bdecompress_64: | |||
| 18969 | ID = Intrinsic::riscv_bdecompress; | |||
| 18970 | break; | |||
| 18971 | ||||
| 18972 | // Zbf | |||
| 18973 | case RISCV::BI__builtin_riscv_bfp_32: | |||
| 18974 | case RISCV::BI__builtin_riscv_bfp_64: | |||
| 18975 | ID = Intrinsic::riscv_bfp; | |||
| 18976 | break; | |||
| 18977 | ||||
| 18978 | // Zbp | |||
| 18979 | case RISCV::BI__builtin_riscv_grev_32: | |||
| 18980 | case RISCV::BI__builtin_riscv_grev_64: | |||
| 18981 | ID = Intrinsic::riscv_grev; | |||
| 18982 | break; | |||
| 18983 | case RISCV::BI__builtin_riscv_gorc_32: | |||
| 18984 | case RISCV::BI__builtin_riscv_gorc_64: | |||
| 18985 | ID = Intrinsic::riscv_gorc; | |||
| 18986 | break; | |||
| 18987 | case RISCV::BI__builtin_riscv_shfl_32: | |||
| 18988 | case RISCV::BI__builtin_riscv_shfl_64: | |||
| 18989 | ID = Intrinsic::riscv_shfl; | |||
| 18990 | break; | |||
| 18991 | case RISCV::BI__builtin_riscv_unshfl_32: | |||
| 18992 | case RISCV::BI__builtin_riscv_unshfl_64: | |||
| 18993 | ID = Intrinsic::riscv_unshfl; | |||
| 18994 | break; | |||
| 18995 | case RISCV::BI__builtin_riscv_xperm_n: | |||
| 18996 | ID = Intrinsic::riscv_xperm_n; | |||
| 18997 | break; | |||
| 18998 | case RISCV::BI__builtin_riscv_xperm_b: | |||
| 18999 | ID = Intrinsic::riscv_xperm_b; | |||
| 19000 | break; | |||
| 19001 | case RISCV::BI__builtin_riscv_xperm_h: | |||
| 19002 | ID = Intrinsic::riscv_xperm_h; | |||
| 19003 | break; | |||
| 19004 | case RISCV::BI__builtin_riscv_xperm_w: | |||
| 19005 | ID = Intrinsic::riscv_xperm_w; | |||
| 19006 | break; | |||
| 19007 | ||||
| 19008 | // Zbr | |||
| 19009 | case RISCV::BI__builtin_riscv_crc32_b: | |||
| 19010 | ID = Intrinsic::riscv_crc32_b; | |||
| 19011 | break; | |||
| 19012 | case RISCV::BI__builtin_riscv_crc32_h: | |||
| 19013 | ID = Intrinsic::riscv_crc32_h; | |||
| 19014 | break; | |||
| 19015 | case RISCV::BI__builtin_riscv_crc32_w: | |||
| 19016 | ID = Intrinsic::riscv_crc32_w; | |||
| 19017 | break; | |||
| 19018 | case RISCV::BI__builtin_riscv_crc32_d: | |||
| 19019 | ID = Intrinsic::riscv_crc32_d; | |||
| 19020 | break; | |||
| 19021 | case RISCV::BI__builtin_riscv_crc32c_b: | |||
| 19022 | ID = Intrinsic::riscv_crc32c_b; | |||
| 19023 | break; | |||
| 19024 | case RISCV::BI__builtin_riscv_crc32c_h: | |||
| 19025 | ID = Intrinsic::riscv_crc32c_h; | |||
| 19026 | break; | |||
| 19027 | case RISCV::BI__builtin_riscv_crc32c_w: | |||
| 19028 | ID = Intrinsic::riscv_crc32c_w; | |||
| 19029 | break; | |||
| 19030 | case RISCV::BI__builtin_riscv_crc32c_d: | |||
| 19031 | ID = Intrinsic::riscv_crc32c_d; | |||
| 19032 | break; | |||
| 19033 | ||||
| 19034 | // Zbt | |||
| 19035 | case RISCV::BI__builtin_riscv_fsl_32: | |||
| 19036 | case RISCV::BI__builtin_riscv_fsl_64: | |||
| 19037 | ID = Intrinsic::riscv_fsl; | |||
| 19038 | break; | |||
| 19039 | case RISCV::BI__builtin_riscv_fsr_32: | |||
| 19040 | case RISCV::BI__builtin_riscv_fsr_64: | |||
| 19041 | ID = Intrinsic::riscv_fsr; | |||
| 19042 | break; | |||
| 19043 | ||||
| 19044 | // Zbkx | |||
| 19045 | case RISCV::BI__builtin_riscv_xperm8: | |||
| 19046 | ID = Intrinsic::riscv_xperm8; | |||
| 19047 | break; | |||
| 19048 | case RISCV::BI__builtin_riscv_xperm4: | |||
| 19049 | ID = Intrinsic::riscv_xperm4; | |||
| 19050 | break; | |||
| 19051 | ||||
| 19052 | // Zbkb | |||
| 19053 | case RISCV::BI__builtin_riscv_brev8: | |||
| 19054 | ID = Intrinsic::riscv_brev8; | |||
| 19055 | break; | |||
| 19056 | case RISCV::BI__builtin_riscv_zip_32: | |||
| 19057 | ID = Intrinsic::riscv_zip; | |||
| 19058 | break; | |||
| 19059 | case RISCV::BI__builtin_riscv_unzip_32: | |||
| 19060 | ID = Intrinsic::riscv_unzip; | |||
| 19061 | break; | |||
| 19062 | } | |||
| 19063 | ||||
| 19064 | IntrinsicTypes = {ResultType}; | |||
| 19065 | break; | |||
| 19066 | } | |||
| 19067 | ||||
| 19068 | // Zk builtins | |||
| 19069 | ||||
| 19070 | // Zknd | |||
| 19071 | case RISCV::BI__builtin_riscv_aes32dsi_32: | |||
| 19072 | ID = Intrinsic::riscv_aes32dsi; | |||
| 19073 | break; | |||
| 19074 | case RISCV::BI__builtin_riscv_aes32dsmi_32: | |||
| 19075 | ID = Intrinsic::riscv_aes32dsmi; | |||
| 19076 | break; | |||
| 19077 | case RISCV::BI__builtin_riscv_aes64ds_64: | |||
| 19078 | ID = Intrinsic::riscv_aes64ds; | |||
| 19079 | break; | |||
| 19080 | case RISCV::BI__builtin_riscv_aes64dsm_64: | |||
| 19081 | ID = Intrinsic::riscv_aes64dsm; | |||
| 19082 | break; | |||
| 19083 | case RISCV::BI__builtin_riscv_aes64im_64: | |||
| 19084 | ID = Intrinsic::riscv_aes64im; | |||
| 19085 | break; | |||
| 19086 | ||||
| 19087 | // Zkne | |||
| 19088 | case RISCV::BI__builtin_riscv_aes32esi_32: | |||
| 19089 | ID = Intrinsic::riscv_aes32esi; | |||
| 19090 | break; | |||
| 19091 | case RISCV::BI__builtin_riscv_aes32esmi_32: | |||
| 19092 | ID = Intrinsic::riscv_aes32esmi; | |||
| 19093 | break; | |||
| 19094 | case RISCV::BI__builtin_riscv_aes64es_64: | |||
| 19095 | ID = Intrinsic::riscv_aes64es; | |||
| 19096 | break; | |||
| 19097 | case RISCV::BI__builtin_riscv_aes64esm_64: | |||
| 19098 | ID = Intrinsic::riscv_aes64esm; | |||
| 19099 | break; | |||
| 19100 | ||||
| 19101 | // Zknd & Zkne | |||
| 19102 | case RISCV::BI__builtin_riscv_aes64ks1i_64: | |||
| 19103 | ID = Intrinsic::riscv_aes64ks1i; | |||
| 19104 | break; | |||
| 19105 | case RISCV::BI__builtin_riscv_aes64ks2_64: | |||
| 19106 | ID = Intrinsic::riscv_aes64ks2; | |||
| 19107 | break; | |||
| 19108 | ||||
| 19109 | // Zknh | |||
| 19110 | case RISCV::BI__builtin_riscv_sha256sig0: | |||
| 19111 | ID = Intrinsic::riscv_sha256sig0; | |||
| 19112 | IntrinsicTypes = {ResultType}; | |||
| 19113 | break; | |||
| 19114 | case RISCV::BI__builtin_riscv_sha256sig1: | |||
| 19115 | ID = Intrinsic::riscv_sha256sig1; | |||
| 19116 | IntrinsicTypes = {ResultType}; | |||
| 19117 | break; | |||
| 19118 | case RISCV::BI__builtin_riscv_sha256sum0: | |||
| 19119 | ID = Intrinsic::riscv_sha256sum0; | |||
| 19120 | IntrinsicTypes = {ResultType}; | |||
| 19121 | break; | |||
| 19122 | case RISCV::BI__builtin_riscv_sha256sum1: | |||
| 19123 | ID = Intrinsic::riscv_sha256sum1; | |||
| 19124 | IntrinsicTypes = {ResultType}; | |||
| 19125 | break; | |||
| 19126 | case RISCV::BI__builtin_riscv_sha512sig0_64: | |||
| 19127 | ID = Intrinsic::riscv_sha512sig0; | |||
| 19128 | break; | |||
| 19129 | case RISCV::BI__builtin_riscv_sha512sig0h_32: | |||
| 19130 | ID = Intrinsic::riscv_sha512sig0h; | |||
| 19131 | break; | |||
| 19132 | case RISCV::BI__builtin_riscv_sha512sig0l_32: | |||
| 19133 | ID = Intrinsic::riscv_sha512sig0l; | |||
| 19134 | break; | |||
| 19135 | case RISCV::BI__builtin_riscv_sha512sig1_64: | |||
| 19136 | ID = Intrinsic::riscv_sha512sig1; | |||
| 19137 | break; | |||
| 19138 | case RISCV::BI__builtin_riscv_sha512sig1h_32: | |||
| 19139 | ID = Intrinsic::riscv_sha512sig1h; | |||
| 19140 | break; | |||
| 19141 | case RISCV::BI__builtin_riscv_sha512sig1l_32: | |||
| 19142 | ID = Intrinsic::riscv_sha512sig1l; | |||
| 19143 | break; | |||
| 19144 | case RISCV::BI__builtin_riscv_sha512sum0_64: | |||
| 19145 | ID = Intrinsic::riscv_sha512sum0; | |||
| 19146 | break; | |||
| 19147 | case RISCV::BI__builtin_riscv_sha512sum0r_32: | |||
| 19148 | ID = Intrinsic::riscv_sha512sum0r; | |||
| 19149 | break; | |||
| 19150 | case RISCV::BI__builtin_riscv_sha512sum1_64: | |||
| 19151 | ID = Intrinsic::riscv_sha512sum1; | |||
| 19152 | break; | |||
| 19153 | case RISCV::BI__builtin_riscv_sha512sum1r_32: | |||
| 19154 | ID = Intrinsic::riscv_sha512sum1r; | |||
| 19155 | break; | |||
| 19156 | ||||
| 19157 | // Zksed | |||
| 19158 | case RISCV::BI__builtin_riscv_sm4ks: | |||
| 19159 | ID = Intrinsic::riscv_sm4ks; | |||
| 19160 | IntrinsicTypes = {ResultType}; | |||
| 19161 | break; | |||
| 19162 | case RISCV::BI__builtin_riscv_sm4ed: | |||
| 19163 | ID = Intrinsic::riscv_sm4ed; | |||
| 19164 | IntrinsicTypes = {ResultType}; | |||
| 19165 | break; | |||
| 19166 | ||||
| 19167 | // Zksh | |||
| 19168 | case RISCV::BI__builtin_riscv_sm3p0: | |||
| 19169 | ID = Intrinsic::riscv_sm3p0; | |||
| 19170 | IntrinsicTypes = {ResultType}; | |||
| 19171 | break; | |||
| 19172 | case RISCV::BI__builtin_riscv_sm3p1: | |||
| 19173 | ID = Intrinsic::riscv_sm3p1; | |||
| 19174 | IntrinsicTypes = {ResultType}; | |||
| 19175 | break; | |||
| 19176 | ||||
| 19177 | // Vector builtins are handled from here. | |||
| 19178 | #include "clang/Basic/riscv_vector_builtin_cg.inc" | |||
| 19179 | } | |||
| 19180 | ||||
| 19181 | assert(ID != Intrinsic::not_intrinsic)(static_cast <bool> (ID != Intrinsic::not_intrinsic) ? void (0) : __assert_fail ("ID != Intrinsic::not_intrinsic", "clang/lib/CodeGen/CGBuiltin.cpp" , 19181, __extension__ __PRETTY_FUNCTION__)); | |||
| 19182 | ||||
| 19183 | llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes); | |||
| 19184 | return Builder.CreateCall(F, Ops, ""); | |||
| 19185 | } |