| File: | tools/clang/lib/CodeGen/CGBuiltin.cpp |
| Warning: | line 11747, column 12 Value stored to 'Store' during its initialization is never read |
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 "CGCXXABI.h" |
| 14 | #include "CGObjCRuntime.h" |
| 15 | #include "CGOpenCLRuntime.h" |
| 16 | #include "CGRecordLayout.h" |
| 17 | #include "CodeGenFunction.h" |
| 18 | #include "CodeGenModule.h" |
| 19 | #include "ConstantEmitter.h" |
| 20 | #include "PatternInit.h" |
| 21 | #include "TargetInfo.h" |
| 22 | #include "clang/AST/ASTContext.h" |
| 23 | #include "clang/AST/Decl.h" |
| 24 | #include "clang/AST/OSLog.h" |
| 25 | #include "clang/Basic/TargetBuiltins.h" |
| 26 | #include "clang/Basic/TargetInfo.h" |
| 27 | #include "clang/CodeGen/CGFunctionInfo.h" |
| 28 | #include "llvm/ADT/SmallPtrSet.h" |
| 29 | #include "llvm/ADT/StringExtras.h" |
| 30 | #include "llvm/IR/DataLayout.h" |
| 31 | #include "llvm/IR/InlineAsm.h" |
| 32 | #include "llvm/IR/Intrinsics.h" |
| 33 | #include "llvm/IR/MDBuilder.h" |
| 34 | #include "llvm/Support/ConvertUTF.h" |
| 35 | #include "llvm/Support/ScopedPrinter.h" |
| 36 | #include "llvm/Support/TargetParser.h" |
| 37 | #include <sstream> |
| 38 | |
| 39 | using namespace clang; |
| 40 | using namespace CodeGen; |
| 41 | using namespace llvm; |
| 42 | |
| 43 | static |
| 44 | int64_t clamp(int64_t Value, int64_t Low, int64_t High) { |
| 45 | return std::min(High, std::max(Low, Value)); |
| 46 | } |
| 47 | |
| 48 | static void initializeAlloca(CodeGenFunction &CGF, AllocaInst *AI, Value *Size, unsigned AlignmentInBytes) { |
| 49 | ConstantInt *Byte; |
| 50 | switch (CGF.getLangOpts().getTrivialAutoVarInit()) { |
| 51 | case LangOptions::TrivialAutoVarInitKind::Uninitialized: |
| 52 | // Nothing to initialize. |
| 53 | return; |
| 54 | case LangOptions::TrivialAutoVarInitKind::Zero: |
| 55 | Byte = CGF.Builder.getInt8(0x00); |
| 56 | break; |
| 57 | case LangOptions::TrivialAutoVarInitKind::Pattern: { |
| 58 | llvm::Type *Int8 = llvm::IntegerType::getInt8Ty(CGF.CGM.getLLVMContext()); |
| 59 | Byte = llvm::dyn_cast<llvm::ConstantInt>( |
| 60 | initializationPatternFor(CGF.CGM, Int8)); |
| 61 | break; |
| 62 | } |
| 63 | } |
| 64 | CGF.Builder.CreateMemSet(AI, Byte, Size, AlignmentInBytes); |
| 65 | } |
| 66 | |
| 67 | /// getBuiltinLibFunction - Given a builtin id for a function like |
| 68 | /// "__builtin_fabsf", return a Function* for "fabsf". |
| 69 | llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD, |
| 70 | unsigned BuiltinID) { |
| 71 | assert(Context.BuiltinInfo.isLibFunction(BuiltinID))((Context.BuiltinInfo.isLibFunction(BuiltinID)) ? static_cast <void> (0) : __assert_fail ("Context.BuiltinInfo.isLibFunction(BuiltinID)" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 71, __PRETTY_FUNCTION__)); |
| 72 | |
| 73 | // Get the name, skip over the __builtin_ prefix (if necessary). |
| 74 | StringRef Name; |
| 75 | GlobalDecl D(FD); |
| 76 | |
| 77 | // If the builtin has been declared explicitly with an assembler label, |
| 78 | // use the mangled name. This differs from the plain label on platforms |
| 79 | // that prefix labels. |
| 80 | if (FD->hasAttr<AsmLabelAttr>()) |
| 81 | Name = getMangledName(D); |
| 82 | else |
| 83 | Name = Context.BuiltinInfo.getName(BuiltinID) + 10; |
| 84 | |
| 85 | llvm::FunctionType *Ty = |
| 86 | cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType())); |
| 87 | |
| 88 | return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false); |
| 89 | } |
| 90 | |
| 91 | /// Emit the conversions required to turn the given value into an |
| 92 | /// integer of the given size. |
| 93 | static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V, |
| 94 | QualType T, llvm::IntegerType *IntType) { |
| 95 | V = CGF.EmitToMemory(V, T); |
| 96 | |
| 97 | if (V->getType()->isPointerTy()) |
| 98 | return CGF.Builder.CreatePtrToInt(V, IntType); |
| 99 | |
| 100 | assert(V->getType() == IntType)((V->getType() == IntType) ? static_cast<void> (0) : __assert_fail ("V->getType() == IntType", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 100, __PRETTY_FUNCTION__)); |
| 101 | return V; |
| 102 | } |
| 103 | |
| 104 | static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V, |
| 105 | QualType T, llvm::Type *ResultType) { |
| 106 | V = CGF.EmitFromMemory(V, T); |
| 107 | |
| 108 | if (ResultType->isPointerTy()) |
| 109 | return CGF.Builder.CreateIntToPtr(V, ResultType); |
| 110 | |
| 111 | assert(V->getType() == ResultType)((V->getType() == ResultType) ? static_cast<void> (0 ) : __assert_fail ("V->getType() == ResultType", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 111, __PRETTY_FUNCTION__)); |
| 112 | return V; |
| 113 | } |
| 114 | |
| 115 | /// Utility to insert an atomic instruction based on Intrinsic::ID |
| 116 | /// and the expression node. |
| 117 | static Value *MakeBinaryAtomicValue( |
| 118 | CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E, |
| 119 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 120 | QualType T = E->getType(); |
| 121 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 121, __PRETTY_FUNCTION__)); |
| 122 | assert(CGF.getContext().hasSameUnqualifiedType(T,((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 123, __PRETTY_FUNCTION__)) |
| 123 | E->getArg(0)->getType()->getPointeeType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 123, __PRETTY_FUNCTION__)); |
| 124 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)-> getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 124, __PRETTY_FUNCTION__)); |
| 125 | |
| 126 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 127 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 128 | |
| 129 | llvm::IntegerType *IntType = |
| 130 | llvm::IntegerType::get(CGF.getLLVMContext(), |
| 131 | CGF.getContext().getTypeSize(T)); |
| 132 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 133 | |
| 134 | llvm::Value *Args[2]; |
| 135 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 136 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 137 | llvm::Type *ValueType = Args[1]->getType(); |
| 138 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 139 | |
| 140 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
| 141 | Kind, Args[0], Args[1], Ordering); |
| 142 | return EmitFromInt(CGF, Result, T, ValueType); |
| 143 | } |
| 144 | |
| 145 | static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) { |
| 146 | Value *Val = CGF.EmitScalarExpr(E->getArg(0)); |
| 147 | Value *Address = CGF.EmitScalarExpr(E->getArg(1)); |
| 148 | |
| 149 | // Convert the type of the pointer to a pointer to the stored type. |
| 150 | Val = CGF.EmitToMemory(Val, E->getArg(0)->getType()); |
| 151 | Value *BC = CGF.Builder.CreateBitCast( |
| 152 | Address, llvm::PointerType::getUnqual(Val->getType()), "cast"); |
| 153 | LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType()); |
| 154 | LV.setNontemporal(true); |
| 155 | CGF.EmitStoreOfScalar(Val, LV, false); |
| 156 | return nullptr; |
| 157 | } |
| 158 | |
| 159 | static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) { |
| 160 | Value *Address = CGF.EmitScalarExpr(E->getArg(0)); |
| 161 | |
| 162 | LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType()); |
| 163 | LV.setNontemporal(true); |
| 164 | return CGF.EmitLoadOfScalar(LV, E->getExprLoc()); |
| 165 | } |
| 166 | |
| 167 | static RValue EmitBinaryAtomic(CodeGenFunction &CGF, |
| 168 | llvm::AtomicRMWInst::BinOp Kind, |
| 169 | const CallExpr *E) { |
| 170 | return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E)); |
| 171 | } |
| 172 | |
| 173 | /// Utility to insert an atomic instruction based Intrinsic::ID and |
| 174 | /// the expression node, where the return value is the result of the |
| 175 | /// operation. |
| 176 | static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF, |
| 177 | llvm::AtomicRMWInst::BinOp Kind, |
| 178 | const CallExpr *E, |
| 179 | Instruction::BinaryOps Op, |
| 180 | bool Invert = false) { |
| 181 | QualType T = E->getType(); |
| 182 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 182, __PRETTY_FUNCTION__)); |
| 183 | assert(CGF.getContext().hasSameUnqualifiedType(T,((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 184, __PRETTY_FUNCTION__)) |
| 184 | E->getArg(0)->getType()->getPointeeType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)-> getType()->getPointeeType())) ? static_cast<void> (0 ) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 184, __PRETTY_FUNCTION__)); |
| 185 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))((CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)-> getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 185, __PRETTY_FUNCTION__)); |
| 186 | |
| 187 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 188 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 189 | |
| 190 | llvm::IntegerType *IntType = |
| 191 | llvm::IntegerType::get(CGF.getLLVMContext(), |
| 192 | CGF.getContext().getTypeSize(T)); |
| 193 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 194 | |
| 195 | llvm::Value *Args[2]; |
| 196 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 197 | llvm::Type *ValueType = Args[1]->getType(); |
| 198 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 199 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 200 | |
| 201 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
| 202 | Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent); |
| 203 | Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]); |
| 204 | if (Invert) |
| 205 | Result = CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result, |
| 206 | llvm::ConstantInt::get(IntType, -1)); |
| 207 | Result = EmitFromInt(CGF, Result, T, ValueType); |
| 208 | return RValue::get(Result); |
| 209 | } |
| 210 | |
| 211 | /// Utility to insert an atomic cmpxchg instruction. |
| 212 | /// |
| 213 | /// @param CGF The current codegen function. |
| 214 | /// @param E Builtin call expression to convert to cmpxchg. |
| 215 | /// arg0 - address to operate on |
| 216 | /// arg1 - value to compare with |
| 217 | /// arg2 - new value |
| 218 | /// @param ReturnBool Specifies whether to return success flag of |
| 219 | /// cmpxchg result or the old value. |
| 220 | /// |
| 221 | /// @returns result of cmpxchg, according to ReturnBool |
| 222 | /// |
| 223 | /// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics |
| 224 | /// invoke the function EmitAtomicCmpXchgForMSIntrin. |
| 225 | static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E, |
| 226 | bool ReturnBool) { |
| 227 | QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType(); |
| 228 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 229 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 230 | |
| 231 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
| 232 | CGF.getLLVMContext(), CGF.getContext().getTypeSize(T)); |
| 233 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 234 | |
| 235 | Value *Args[3]; |
| 236 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 237 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 238 | llvm::Type *ValueType = Args[1]->getType(); |
| 239 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 240 | Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType); |
| 241 | |
| 242 | Value *Pair = CGF.Builder.CreateAtomicCmpXchg( |
| 243 | Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent, |
| 244 | llvm::AtomicOrdering::SequentiallyConsistent); |
| 245 | if (ReturnBool) |
| 246 | // Extract boolean success flag and zext it to int. |
| 247 | return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1), |
| 248 | CGF.ConvertType(E->getType())); |
| 249 | else |
| 250 | // Extract old value and emit it using the same type as compare value. |
| 251 | return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T, |
| 252 | ValueType); |
| 253 | } |
| 254 | |
| 255 | /// This function should be invoked to emit atomic cmpxchg for Microsoft's |
| 256 | /// _InterlockedCompareExchange* intrinsics which have the following signature: |
| 257 | /// T _InterlockedCompareExchange(T volatile *Destination, |
| 258 | /// T Exchange, |
| 259 | /// T Comparand); |
| 260 | /// |
| 261 | /// Whereas the llvm 'cmpxchg' instruction has the following syntax: |
| 262 | /// cmpxchg *Destination, Comparand, Exchange. |
| 263 | /// So we need to swap Comparand and Exchange when invoking |
| 264 | /// CreateAtomicCmpXchg. That is the reason we could not use the above utility |
| 265 | /// function MakeAtomicCmpXchgValue since it expects the arguments to be |
| 266 | /// already swapped. |
| 267 | |
| 268 | static |
| 269 | Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E, |
| 270 | AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) { |
| 271 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 271, __PRETTY_FUNCTION__)); |
| 272 | assert(CGF.getContext().hasSameUnqualifiedType(((CGF.getContext().hasSameUnqualifiedType( E->getType(), E ->getArg(0)->getType()->getPointeeType())) ? static_cast <void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 273, __PRETTY_FUNCTION__)) |
| 273 | E->getType(), E->getArg(0)->getType()->getPointeeType()))((CGF.getContext().hasSameUnqualifiedType( E->getType(), E ->getArg(0)->getType()->getPointeeType())) ? static_cast <void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 273, __PRETTY_FUNCTION__)); |
| 274 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(1)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 275, __PRETTY_FUNCTION__)) |
| 275 | E->getArg(1)->getType()))((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(1)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 275, __PRETTY_FUNCTION__)); |
| 276 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(2)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 277, __PRETTY_FUNCTION__)) |
| 277 | E->getArg(2)->getType()))((CGF.getContext().hasSameUnqualifiedType(E->getType(), E-> getArg(2)->getType())) ? static_cast<void> (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 277, __PRETTY_FUNCTION__)); |
| 278 | |
| 279 | auto *Destination = CGF.EmitScalarExpr(E->getArg(0)); |
| 280 | auto *Comparand = CGF.EmitScalarExpr(E->getArg(2)); |
| 281 | auto *Exchange = CGF.EmitScalarExpr(E->getArg(1)); |
| 282 | |
| 283 | // For Release ordering, the failure ordering should be Monotonic. |
| 284 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ? |
| 285 | AtomicOrdering::Monotonic : |
| 286 | SuccessOrdering; |
| 287 | |
| 288 | auto *Result = CGF.Builder.CreateAtomicCmpXchg( |
| 289 | Destination, Comparand, Exchange, |
| 290 | SuccessOrdering, FailureOrdering); |
| 291 | Result->setVolatile(true); |
| 292 | return CGF.Builder.CreateExtractValue(Result, 0); |
| 293 | } |
| 294 | |
| 295 | static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E, |
| 296 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 297 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 297, __PRETTY_FUNCTION__)); |
| 298 | |
| 299 | auto *IntTy = CGF.ConvertType(E->getType()); |
| 300 | auto *Result = CGF.Builder.CreateAtomicRMW( |
| 301 | AtomicRMWInst::Add, |
| 302 | CGF.EmitScalarExpr(E->getArg(0)), |
| 303 | ConstantInt::get(IntTy, 1), |
| 304 | Ordering); |
| 305 | return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1)); |
| 306 | } |
| 307 | |
| 308 | static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E, |
| 309 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 310 | assert(E->getArg(0)->getType()->isPointerType())((E->getArg(0)->getType()->isPointerType()) ? static_cast <void> (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 310, __PRETTY_FUNCTION__)); |
| 311 | |
| 312 | auto *IntTy = CGF.ConvertType(E->getType()); |
| 313 | auto *Result = CGF.Builder.CreateAtomicRMW( |
| 314 | AtomicRMWInst::Sub, |
| 315 | CGF.EmitScalarExpr(E->getArg(0)), |
| 316 | ConstantInt::get(IntTy, 1), |
| 317 | Ordering); |
| 318 | return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1)); |
| 319 | } |
| 320 | |
| 321 | // Build a plain volatile load. |
| 322 | static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) { |
| 323 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
| 324 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 325 | CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy); |
| 326 | llvm::Type *ITy = |
| 327 | llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8); |
| 328 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 329 | llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(Ptr, LoadSize); |
| 330 | Load->setVolatile(true); |
| 331 | return Load; |
| 332 | } |
| 333 | |
| 334 | // Build a plain volatile store. |
| 335 | static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) { |
| 336 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
| 337 | Value *Value = CGF.EmitScalarExpr(E->getArg(1)); |
| 338 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 339 | CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy); |
| 340 | llvm::Type *ITy = |
| 341 | llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8); |
| 342 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 343 | llvm::StoreInst *Store = |
| 344 | CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize); |
| 345 | Store->setVolatile(true); |
| 346 | return Store; |
| 347 | } |
| 348 | |
| 349 | // Emit a simple mangled intrinsic that has 1 argument and a return type |
| 350 | // matching the argument type. |
| 351 | static Value *emitUnaryBuiltin(CodeGenFunction &CGF, |
| 352 | const CallExpr *E, |
| 353 | unsigned IntrinsicID) { |
| 354 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 355 | |
| 356 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 357 | return CGF.Builder.CreateCall(F, Src0); |
| 358 | } |
| 359 | |
| 360 | // Emit an intrinsic that has 2 operands of the same type as its result. |
| 361 | static Value *emitBinaryBuiltin(CodeGenFunction &CGF, |
| 362 | const CallExpr *E, |
| 363 | unsigned IntrinsicID) { |
| 364 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 365 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 366 | |
| 367 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 368 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); |
| 369 | } |
| 370 | |
| 371 | // Emit an intrinsic that has 3 operands of the same type as its result. |
| 372 | static Value *emitTernaryBuiltin(CodeGenFunction &CGF, |
| 373 | const CallExpr *E, |
| 374 | unsigned IntrinsicID) { |
| 375 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 376 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 377 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); |
| 378 | |
| 379 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 380 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); |
| 381 | } |
| 382 | |
| 383 | // Emit an intrinsic that has 1 float or double operand, and 1 integer. |
| 384 | static Value *emitFPIntBuiltin(CodeGenFunction &CGF, |
| 385 | const CallExpr *E, |
| 386 | unsigned IntrinsicID) { |
| 387 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 388 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 389 | |
| 390 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 391 | return CGF.Builder.CreateCall(F, {Src0, Src1}); |
| 392 | } |
| 393 | |
| 394 | // Emit an intrinsic that has overloaded integer result and fp operand. |
| 395 | static Value *emitFPToIntRoundBuiltin(CodeGenFunction &CGF, |
| 396 | const CallExpr *E, |
| 397 | unsigned IntrinsicID) { |
| 398 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); |
| 399 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 400 | |
| 401 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, |
| 402 | {ResultType, Src0->getType()}); |
| 403 | return CGF.Builder.CreateCall(F, Src0); |
| 404 | } |
| 405 | |
| 406 | /// EmitFAbs - Emit a call to @llvm.fabs(). |
| 407 | static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) { |
| 408 | Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType()); |
| 409 | llvm::CallInst *Call = CGF.Builder.CreateCall(F, V); |
| 410 | Call->setDoesNotAccessMemory(); |
| 411 | return Call; |
| 412 | } |
| 413 | |
| 414 | /// Emit the computation of the sign bit for a floating point value. Returns |
| 415 | /// the i1 sign bit value. |
| 416 | static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) { |
| 417 | LLVMContext &C = CGF.CGM.getLLVMContext(); |
| 418 | |
| 419 | llvm::Type *Ty = V->getType(); |
| 420 | int Width = Ty->getPrimitiveSizeInBits(); |
| 421 | llvm::Type *IntTy = llvm::IntegerType::get(C, Width); |
| 422 | V = CGF.Builder.CreateBitCast(V, IntTy); |
| 423 | if (Ty->isPPC_FP128Ty()) { |
| 424 | // We want the sign bit of the higher-order double. The bitcast we just |
| 425 | // did works as if the double-double was stored to memory and then |
| 426 | // read as an i128. The "store" will put the higher-order double in the |
| 427 | // lower address in both little- and big-Endian modes, but the "load" |
| 428 | // will treat those bits as a different part of the i128: the low bits in |
| 429 | // little-Endian, the high bits in big-Endian. Therefore, on big-Endian |
| 430 | // we need to shift the high bits down to the low before truncating. |
| 431 | Width >>= 1; |
| 432 | if (CGF.getTarget().isBigEndian()) { |
| 433 | Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width); |
| 434 | V = CGF.Builder.CreateLShr(V, ShiftCst); |
| 435 | } |
| 436 | // We are truncating value in order to extract the higher-order |
| 437 | // double, which we will be using to extract the sign from. |
| 438 | IntTy = llvm::IntegerType::get(C, Width); |
| 439 | V = CGF.Builder.CreateTrunc(V, IntTy); |
| 440 | } |
| 441 | Value *Zero = llvm::Constant::getNullValue(IntTy); |
| 442 | return CGF.Builder.CreateICmpSLT(V, Zero); |
| 443 | } |
| 444 | |
| 445 | static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD, |
| 446 | const CallExpr *E, llvm::Constant *calleeValue) { |
| 447 | CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD)); |
| 448 | return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot()); |
| 449 | } |
| 450 | |
| 451 | /// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.* |
| 452 | /// depending on IntrinsicID. |
| 453 | /// |
| 454 | /// \arg CGF The current codegen function. |
| 455 | /// \arg IntrinsicID The ID for the Intrinsic we wish to generate. |
| 456 | /// \arg X The first argument to the llvm.*.with.overflow.*. |
| 457 | /// \arg Y The second argument to the llvm.*.with.overflow.*. |
| 458 | /// \arg Carry The carry returned by the llvm.*.with.overflow.*. |
| 459 | /// \returns The result (i.e. sum/product) returned by the intrinsic. |
| 460 | static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF, |
| 461 | const llvm::Intrinsic::ID IntrinsicID, |
| 462 | llvm::Value *X, llvm::Value *Y, |
| 463 | llvm::Value *&Carry) { |
| 464 | // Make sure we have integers of the same width. |
| 465 | assert(X->getType() == Y->getType() &&((X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? static_cast< 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?)\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 467, __PRETTY_FUNCTION__)) |
| 466 | "Arguments must be the same type. (Did you forget to make sure both "((X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? static_cast< 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?)\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 467, __PRETTY_FUNCTION__)) |
| 467 | "arguments have the same integer width?)")((X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? static_cast< 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?)\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 467, __PRETTY_FUNCTION__)); |
| 468 | |
| 469 | Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType()); |
| 470 | llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y}); |
| 471 | Carry = CGF.Builder.CreateExtractValue(Tmp, 1); |
| 472 | return CGF.Builder.CreateExtractValue(Tmp, 0); |
| 473 | } |
| 474 | |
| 475 | static Value *emitRangedBuiltin(CodeGenFunction &CGF, |
| 476 | unsigned IntrinsicID, |
| 477 | int low, int high) { |
| 478 | llvm::MDBuilder MDHelper(CGF.getLLVMContext()); |
| 479 | llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high)); |
| 480 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {}); |
| 481 | llvm::Instruction *Call = CGF.Builder.CreateCall(F); |
| 482 | Call->setMetadata(llvm::LLVMContext::MD_range, RNode); |
| 483 | return Call; |
| 484 | } |
| 485 | |
| 486 | namespace { |
| 487 | struct WidthAndSignedness { |
| 488 | unsigned Width; |
| 489 | bool Signed; |
| 490 | }; |
| 491 | } |
| 492 | |
| 493 | static WidthAndSignedness |
| 494 | getIntegerWidthAndSignedness(const clang::ASTContext &context, |
| 495 | const clang::QualType Type) { |
| 496 | assert(Type->isIntegerType() && "Given type is not an integer.")((Type->isIntegerType() && "Given type is not an integer." ) ? static_cast<void> (0) : __assert_fail ("Type->isIntegerType() && \"Given type is not an integer.\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 496, __PRETTY_FUNCTION__)); |
| 497 | unsigned Width = Type->isBooleanType() ? 1 : context.getTypeInfo(Type).Width; |
| 498 | bool Signed = Type->isSignedIntegerType(); |
| 499 | return {Width, Signed}; |
| 500 | } |
| 501 | |
| 502 | // Given one or more integer types, this function produces an integer type that |
| 503 | // encompasses them: any value in one of the given types could be expressed in |
| 504 | // the encompassing type. |
| 505 | static struct WidthAndSignedness |
| 506 | EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) { |
| 507 | assert(Types.size() > 0 && "Empty list of types.")((Types.size() > 0 && "Empty list of types.") ? static_cast <void> (0) : __assert_fail ("Types.size() > 0 && \"Empty list of types.\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 507, __PRETTY_FUNCTION__)); |
| 508 | |
| 509 | // If any of the given types is signed, we must return a signed type. |
| 510 | bool Signed = false; |
| 511 | for (const auto &Type : Types) { |
| 512 | Signed |= Type.Signed; |
| 513 | } |
| 514 | |
| 515 | // The encompassing type must have a width greater than or equal to the width |
| 516 | // of the specified types. Additionally, if the encompassing type is signed, |
| 517 | // its width must be strictly greater than the width of any unsigned types |
| 518 | // given. |
| 519 | unsigned Width = 0; |
| 520 | for (const auto &Type : Types) { |
| 521 | unsigned MinWidth = Type.Width + (Signed && !Type.Signed); |
| 522 | if (Width < MinWidth) { |
| 523 | Width = MinWidth; |
| 524 | } |
| 525 | } |
| 526 | |
| 527 | return {Width, Signed}; |
| 528 | } |
| 529 | |
| 530 | Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) { |
| 531 | llvm::Type *DestType = Int8PtrTy; |
| 532 | if (ArgValue->getType() != DestType) |
| 533 | ArgValue = |
| 534 | Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data()); |
| 535 | |
| 536 | Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend; |
| 537 | return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue); |
| 538 | } |
| 539 | |
| 540 | /// Checks if using the result of __builtin_object_size(p, @p From) in place of |
| 541 | /// __builtin_object_size(p, @p To) is correct |
| 542 | static bool areBOSTypesCompatible(int From, int To) { |
| 543 | // Note: Our __builtin_object_size implementation currently treats Type=0 and |
| 544 | // Type=2 identically. Encoding this implementation detail here may make |
| 545 | // improving __builtin_object_size difficult in the future, so it's omitted. |
| 546 | return From == To || (From == 0 && To == 1) || (From == 3 && To == 2); |
| 547 | } |
| 548 | |
| 549 | static llvm::Value * |
| 550 | getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) { |
| 551 | return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true); |
| 552 | } |
| 553 | |
| 554 | llvm::Value * |
| 555 | CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type, |
| 556 | llvm::IntegerType *ResType, |
| 557 | llvm::Value *EmittedE, |
| 558 | bool IsDynamic) { |
| 559 | uint64_t ObjectSize; |
| 560 | if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type)) |
| 561 | return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic); |
| 562 | return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true); |
| 563 | } |
| 564 | |
| 565 | /// Returns a Value corresponding to the size of the given expression. |
| 566 | /// This Value may be either of the following: |
| 567 | /// - A llvm::Argument (if E is a param with the pass_object_size attribute on |
| 568 | /// it) |
| 569 | /// - A call to the @llvm.objectsize intrinsic |
| 570 | /// |
| 571 | /// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null |
| 572 | /// and we wouldn't otherwise try to reference a pass_object_size parameter, |
| 573 | /// we'll call @llvm.objectsize on EmittedE, rather than emitting E. |
| 574 | llvm::Value * |
| 575 | CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type, |
| 576 | llvm::IntegerType *ResType, |
| 577 | llvm::Value *EmittedE, bool IsDynamic) { |
| 578 | // We need to reference an argument if the pointer is a parameter with the |
| 579 | // pass_object_size attribute. |
| 580 | if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) { |
| 581 | auto *Param = dyn_cast<ParmVarDecl>(D->getDecl()); |
| 582 | auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>(); |
| 583 | if (Param != nullptr && PS != nullptr && |
| 584 | areBOSTypesCompatible(PS->getType(), Type)) { |
| 585 | auto Iter = SizeArguments.find(Param); |
| 586 | assert(Iter != SizeArguments.end())((Iter != SizeArguments.end()) ? static_cast<void> (0) : __assert_fail ("Iter != SizeArguments.end()", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 586, __PRETTY_FUNCTION__)); |
| 587 | |
| 588 | const ImplicitParamDecl *D = Iter->second; |
| 589 | auto DIter = LocalDeclMap.find(D); |
| 590 | assert(DIter != LocalDeclMap.end())((DIter != LocalDeclMap.end()) ? static_cast<void> (0) : __assert_fail ("DIter != LocalDeclMap.end()", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 590, __PRETTY_FUNCTION__)); |
| 591 | |
| 592 | return EmitLoadOfScalar(DIter->second, /*volatile=*/false, |
| 593 | getContext().getSizeType(), E->getBeginLoc()); |
| 594 | } |
| 595 | } |
| 596 | |
| 597 | // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't |
| 598 | // evaluate E for side-effects. In either case, we shouldn't lower to |
| 599 | // @llvm.objectsize. |
| 600 | if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext()))) |
| 601 | return getDefaultBuiltinObjectSizeResult(Type, ResType); |
| 602 | |
| 603 | Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E); |
| 604 | assert(Ptr->getType()->isPointerTy() &&((Ptr->getType()->isPointerTy() && "Non-pointer passed to __builtin_object_size?" ) ? static_cast<void> (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 605, __PRETTY_FUNCTION__)) |
| 605 | "Non-pointer passed to __builtin_object_size?")((Ptr->getType()->isPointerTy() && "Non-pointer passed to __builtin_object_size?" ) ? static_cast<void> (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 605, __PRETTY_FUNCTION__)); |
| 606 | |
| 607 | Function *F = |
| 608 | CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()}); |
| 609 | |
| 610 | // LLVM only supports 0 and 2, make sure that we pass along that as a boolean. |
| 611 | Value *Min = Builder.getInt1((Type & 2) != 0); |
| 612 | // For GCC compatibility, __builtin_object_size treat NULL as unknown size. |
| 613 | Value *NullIsUnknown = Builder.getTrue(); |
| 614 | Value *Dynamic = Builder.getInt1(IsDynamic); |
| 615 | return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic}); |
| 616 | } |
| 617 | |
| 618 | namespace { |
| 619 | /// A struct to generically describe a bit test intrinsic. |
| 620 | struct BitTest { |
| 621 | enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set }; |
| 622 | enum InterlockingKind : uint8_t { |
| 623 | Unlocked, |
| 624 | Sequential, |
| 625 | Acquire, |
| 626 | Release, |
| 627 | NoFence |
| 628 | }; |
| 629 | |
| 630 | ActionKind Action; |
| 631 | InterlockingKind Interlocking; |
| 632 | bool Is64Bit; |
| 633 | |
| 634 | static BitTest decodeBitTestBuiltin(unsigned BuiltinID); |
| 635 | }; |
| 636 | } // namespace |
| 637 | |
| 638 | BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) { |
| 639 | switch (BuiltinID) { |
| 640 | // Main portable variants. |
| 641 | case Builtin::BI_bittest: |
| 642 | return {TestOnly, Unlocked, false}; |
| 643 | case Builtin::BI_bittestandcomplement: |
| 644 | return {Complement, Unlocked, false}; |
| 645 | case Builtin::BI_bittestandreset: |
| 646 | return {Reset, Unlocked, false}; |
| 647 | case Builtin::BI_bittestandset: |
| 648 | return {Set, Unlocked, false}; |
| 649 | case Builtin::BI_interlockedbittestandreset: |
| 650 | return {Reset, Sequential, false}; |
| 651 | case Builtin::BI_interlockedbittestandset: |
| 652 | return {Set, Sequential, false}; |
| 653 | |
| 654 | // X86-specific 64-bit variants. |
| 655 | case Builtin::BI_bittest64: |
| 656 | return {TestOnly, Unlocked, true}; |
| 657 | case Builtin::BI_bittestandcomplement64: |
| 658 | return {Complement, Unlocked, true}; |
| 659 | case Builtin::BI_bittestandreset64: |
| 660 | return {Reset, Unlocked, true}; |
| 661 | case Builtin::BI_bittestandset64: |
| 662 | return {Set, Unlocked, true}; |
| 663 | case Builtin::BI_interlockedbittestandreset64: |
| 664 | return {Reset, Sequential, true}; |
| 665 | case Builtin::BI_interlockedbittestandset64: |
| 666 | return {Set, Sequential, true}; |
| 667 | |
| 668 | // ARM/AArch64-specific ordering variants. |
| 669 | case Builtin::BI_interlockedbittestandset_acq: |
| 670 | return {Set, Acquire, false}; |
| 671 | case Builtin::BI_interlockedbittestandset_rel: |
| 672 | return {Set, Release, false}; |
| 673 | case Builtin::BI_interlockedbittestandset_nf: |
| 674 | return {Set, NoFence, false}; |
| 675 | case Builtin::BI_interlockedbittestandreset_acq: |
| 676 | return {Reset, Acquire, false}; |
| 677 | case Builtin::BI_interlockedbittestandreset_rel: |
| 678 | return {Reset, Release, false}; |
| 679 | case Builtin::BI_interlockedbittestandreset_nf: |
| 680 | return {Reset, NoFence, false}; |
| 681 | } |
| 682 | llvm_unreachable("expected only bittest intrinsics")::llvm::llvm_unreachable_internal("expected only bittest intrinsics" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 682); |
| 683 | } |
| 684 | |
| 685 | static char bitActionToX86BTCode(BitTest::ActionKind A) { |
| 686 | switch (A) { |
| 687 | case BitTest::TestOnly: return '\0'; |
| 688 | case BitTest::Complement: return 'c'; |
| 689 | case BitTest::Reset: return 'r'; |
| 690 | case BitTest::Set: return 's'; |
| 691 | } |
| 692 | llvm_unreachable("invalid action")::llvm::llvm_unreachable_internal("invalid action", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 692); |
| 693 | } |
| 694 | |
| 695 | static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF, |
| 696 | BitTest BT, |
| 697 | const CallExpr *E, Value *BitBase, |
| 698 | Value *BitPos) { |
| 699 | char Action = bitActionToX86BTCode(BT.Action); |
| 700 | char SizeSuffix = BT.Is64Bit ? 'q' : 'l'; |
| 701 | |
| 702 | // Build the assembly. |
| 703 | SmallString<64> Asm; |
| 704 | raw_svector_ostream AsmOS(Asm); |
| 705 | if (BT.Interlocking != BitTest::Unlocked) |
| 706 | AsmOS << "lock "; |
| 707 | AsmOS << "bt"; |
| 708 | if (Action) |
| 709 | AsmOS << Action; |
| 710 | AsmOS << SizeSuffix << " $2, ($1)\n\tsetc ${0:b}"; |
| 711 | |
| 712 | // Build the constraints. FIXME: We should support immediates when possible. |
| 713 | std::string Constraints = "=r,r,r,~{cc},~{flags},~{fpsr}"; |
| 714 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
| 715 | CGF.getLLVMContext(), |
| 716 | CGF.getContext().getTypeSize(E->getArg(1)->getType())); |
| 717 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
| 718 | llvm::FunctionType *FTy = |
| 719 | llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false); |
| 720 | |
| 721 | llvm::InlineAsm *IA = |
| 722 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*SideEffects=*/true); |
| 723 | return CGF.Builder.CreateCall(IA, {BitBase, BitPos}); |
| 724 | } |
| 725 | |
| 726 | static llvm::AtomicOrdering |
| 727 | getBitTestAtomicOrdering(BitTest::InterlockingKind I) { |
| 728 | switch (I) { |
| 729 | case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic; |
| 730 | case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent; |
| 731 | case BitTest::Acquire: return llvm::AtomicOrdering::Acquire; |
| 732 | case BitTest::Release: return llvm::AtomicOrdering::Release; |
| 733 | case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic; |
| 734 | } |
| 735 | llvm_unreachable("invalid interlocking")::llvm::llvm_unreachable_internal("invalid interlocking", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 735); |
| 736 | } |
| 737 | |
| 738 | /// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of |
| 739 | /// bits and a bit position and read and optionally modify the bit at that |
| 740 | /// position. The position index can be arbitrarily large, i.e. it can be larger |
| 741 | /// than 31 or 63, so we need an indexed load in the general case. |
| 742 | static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF, |
| 743 | unsigned BuiltinID, |
| 744 | const CallExpr *E) { |
| 745 | Value *BitBase = CGF.EmitScalarExpr(E->getArg(0)); |
| 746 | Value *BitPos = CGF.EmitScalarExpr(E->getArg(1)); |
| 747 | |
| 748 | BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID); |
| 749 | |
| 750 | // X86 has special BT, BTC, BTR, and BTS instructions that handle the array |
| 751 | // indexing operation internally. Use them if possible. |
| 752 | llvm::Triple::ArchType Arch = CGF.getTarget().getTriple().getArch(); |
| 753 | if (Arch == llvm::Triple::x86 || Arch == llvm::Triple::x86_64) |
| 754 | return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos); |
| 755 | |
| 756 | // Otherwise, use generic code to load one byte and test the bit. Use all but |
| 757 | // the bottom three bits as the array index, and the bottom three bits to form |
| 758 | // a mask. |
| 759 | // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0; |
| 760 | Value *ByteIndex = CGF.Builder.CreateAShr( |
| 761 | BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx"); |
| 762 | Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy); |
| 763 | Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8, |
| 764 | ByteIndex, "bittest.byteaddr"), |
| 765 | CharUnits::One()); |
| 766 | Value *PosLow = |
| 767 | CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty), |
| 768 | llvm::ConstantInt::get(CGF.Int8Ty, 0x7)); |
| 769 | |
| 770 | // The updating instructions will need a mask. |
| 771 | Value *Mask = nullptr; |
| 772 | if (BT.Action != BitTest::TestOnly) { |
| 773 | Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow, |
| 774 | "bittest.mask"); |
| 775 | } |
| 776 | |
| 777 | // Check the action and ordering of the interlocked intrinsics. |
| 778 | llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking); |
| 779 | |
| 780 | Value *OldByte = nullptr; |
| 781 | if (Ordering != llvm::AtomicOrdering::NotAtomic) { |
| 782 | // Emit a combined atomicrmw load/store operation for the interlocked |
| 783 | // intrinsics. |
| 784 | llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or; |
| 785 | if (BT.Action == BitTest::Reset) { |
| 786 | Mask = CGF.Builder.CreateNot(Mask); |
| 787 | RMWOp = llvm::AtomicRMWInst::And; |
| 788 | } |
| 789 | OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask, |
| 790 | Ordering); |
| 791 | } else { |
| 792 | // Emit a plain load for the non-interlocked intrinsics. |
| 793 | OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte"); |
| 794 | Value *NewByte = nullptr; |
| 795 | switch (BT.Action) { |
| 796 | case BitTest::TestOnly: |
| 797 | // Don't store anything. |
| 798 | break; |
| 799 | case BitTest::Complement: |
| 800 | NewByte = CGF.Builder.CreateXor(OldByte, Mask); |
| 801 | break; |
| 802 | case BitTest::Reset: |
| 803 | NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask)); |
| 804 | break; |
| 805 | case BitTest::Set: |
| 806 | NewByte = CGF.Builder.CreateOr(OldByte, Mask); |
| 807 | break; |
| 808 | } |
| 809 | if (NewByte) |
| 810 | CGF.Builder.CreateStore(NewByte, ByteAddr); |
| 811 | } |
| 812 | |
| 813 | // However we loaded the old byte, either by plain load or atomicrmw, shift |
| 814 | // the bit into the low position and mask it to 0 or 1. |
| 815 | Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr"); |
| 816 | return CGF.Builder.CreateAnd( |
| 817 | ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res"); |
| 818 | } |
| 819 | |
| 820 | namespace { |
| 821 | enum class MSVCSetJmpKind { |
| 822 | _setjmpex, |
| 823 | _setjmp3, |
| 824 | _setjmp |
| 825 | }; |
| 826 | } |
| 827 | |
| 828 | /// MSVC handles setjmp a bit differently on different platforms. On every |
| 829 | /// architecture except 32-bit x86, the frame address is passed. On x86, extra |
| 830 | /// parameters can be passed as variadic arguments, but we always pass none. |
| 831 | static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind, |
| 832 | const CallExpr *E) { |
| 833 | llvm::Value *Arg1 = nullptr; |
| 834 | llvm::Type *Arg1Ty = nullptr; |
| 835 | StringRef Name; |
| 836 | bool IsVarArg = false; |
| 837 | if (SJKind == MSVCSetJmpKind::_setjmp3) { |
| 838 | Name = "_setjmp3"; |
| 839 | Arg1Ty = CGF.Int32Ty; |
| 840 | Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0); |
| 841 | IsVarArg = true; |
| 842 | } else { |
| 843 | Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex"; |
| 844 | Arg1Ty = CGF.Int8PtrTy; |
| 845 | if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) { |
| 846 | Arg1 = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(Intrinsic::sponentry)); |
| 847 | } else |
| 848 | Arg1 = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(Intrinsic::frameaddress), |
| 849 | llvm::ConstantInt::get(CGF.Int32Ty, 0)); |
| 850 | } |
| 851 | |
| 852 | // Mark the call site and declaration with ReturnsTwice. |
| 853 | llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty}; |
| 854 | llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get( |
| 855 | CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex, |
| 856 | llvm::Attribute::ReturnsTwice); |
| 857 | llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction( |
| 858 | llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name, |
| 859 | ReturnsTwiceAttr, /*Local=*/true); |
| 860 | |
| 861 | llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast( |
| 862 | CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy); |
| 863 | llvm::Value *Args[] = {Buf, Arg1}; |
| 864 | llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args); |
| 865 | CB->setAttributes(ReturnsTwiceAttr); |
| 866 | return RValue::get(CB); |
| 867 | } |
| 868 | |
| 869 | // Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code, |
| 870 | // we handle them here. |
| 871 | enum class CodeGenFunction::MSVCIntrin { |
| 872 | _BitScanForward, |
| 873 | _BitScanReverse, |
| 874 | _InterlockedAnd, |
| 875 | _InterlockedDecrement, |
| 876 | _InterlockedExchange, |
| 877 | _InterlockedExchangeAdd, |
| 878 | _InterlockedExchangeSub, |
| 879 | _InterlockedIncrement, |
| 880 | _InterlockedOr, |
| 881 | _InterlockedXor, |
| 882 | _InterlockedExchangeAdd_acq, |
| 883 | _InterlockedExchangeAdd_rel, |
| 884 | _InterlockedExchangeAdd_nf, |
| 885 | _InterlockedExchange_acq, |
| 886 | _InterlockedExchange_rel, |
| 887 | _InterlockedExchange_nf, |
| 888 | _InterlockedCompareExchange_acq, |
| 889 | _InterlockedCompareExchange_rel, |
| 890 | _InterlockedCompareExchange_nf, |
| 891 | _InterlockedOr_acq, |
| 892 | _InterlockedOr_rel, |
| 893 | _InterlockedOr_nf, |
| 894 | _InterlockedXor_acq, |
| 895 | _InterlockedXor_rel, |
| 896 | _InterlockedXor_nf, |
| 897 | _InterlockedAnd_acq, |
| 898 | _InterlockedAnd_rel, |
| 899 | _InterlockedAnd_nf, |
| 900 | _InterlockedIncrement_acq, |
| 901 | _InterlockedIncrement_rel, |
| 902 | _InterlockedIncrement_nf, |
| 903 | _InterlockedDecrement_acq, |
| 904 | _InterlockedDecrement_rel, |
| 905 | _InterlockedDecrement_nf, |
| 906 | __fastfail, |
| 907 | }; |
| 908 | |
| 909 | Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID, |
| 910 | const CallExpr *E) { |
| 911 | switch (BuiltinID) { |
| 912 | case MSVCIntrin::_BitScanForward: |
| 913 | case MSVCIntrin::_BitScanReverse: { |
| 914 | Value *ArgValue = EmitScalarExpr(E->getArg(1)); |
| 915 | |
| 916 | llvm::Type *ArgType = ArgValue->getType(); |
| 917 | llvm::Type *IndexType = |
| 918 | EmitScalarExpr(E->getArg(0))->getType()->getPointerElementType(); |
| 919 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 920 | |
| 921 | Value *ArgZero = llvm::Constant::getNullValue(ArgType); |
| 922 | Value *ResZero = llvm::Constant::getNullValue(ResultType); |
| 923 | Value *ResOne = llvm::ConstantInt::get(ResultType, 1); |
| 924 | |
| 925 | BasicBlock *Begin = Builder.GetInsertBlock(); |
| 926 | BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn); |
| 927 | Builder.SetInsertPoint(End); |
| 928 | PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result"); |
| 929 | |
| 930 | Builder.SetInsertPoint(Begin); |
| 931 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero); |
| 932 | BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn); |
| 933 | Builder.CreateCondBr(IsZero, End, NotZero); |
| 934 | Result->addIncoming(ResZero, Begin); |
| 935 | |
| 936 | Builder.SetInsertPoint(NotZero); |
| 937 | Address IndexAddress = EmitPointerWithAlignment(E->getArg(0)); |
| 938 | |
| 939 | if (BuiltinID == MSVCIntrin::_BitScanForward) { |
| 940 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 941 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
| 942 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
| 943 | Builder.CreateStore(ZeroCount, IndexAddress, false); |
| 944 | } else { |
| 945 | unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth(); |
| 946 | Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1); |
| 947 | |
| 948 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 949 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
| 950 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
| 951 | Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount); |
| 952 | Builder.CreateStore(Index, IndexAddress, false); |
| 953 | } |
| 954 | Builder.CreateBr(End); |
| 955 | Result->addIncoming(ResOne, NotZero); |
| 956 | |
| 957 | Builder.SetInsertPoint(End); |
| 958 | return Result; |
| 959 | } |
| 960 | case MSVCIntrin::_InterlockedAnd: |
| 961 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E); |
| 962 | case MSVCIntrin::_InterlockedExchange: |
| 963 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E); |
| 964 | case MSVCIntrin::_InterlockedExchangeAdd: |
| 965 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E); |
| 966 | case MSVCIntrin::_InterlockedExchangeSub: |
| 967 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E); |
| 968 | case MSVCIntrin::_InterlockedOr: |
| 969 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E); |
| 970 | case MSVCIntrin::_InterlockedXor: |
| 971 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E); |
| 972 | case MSVCIntrin::_InterlockedExchangeAdd_acq: |
| 973 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 974 | AtomicOrdering::Acquire); |
| 975 | case MSVCIntrin::_InterlockedExchangeAdd_rel: |
| 976 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 977 | AtomicOrdering::Release); |
| 978 | case MSVCIntrin::_InterlockedExchangeAdd_nf: |
| 979 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 980 | AtomicOrdering::Monotonic); |
| 981 | case MSVCIntrin::_InterlockedExchange_acq: |
| 982 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 983 | AtomicOrdering::Acquire); |
| 984 | case MSVCIntrin::_InterlockedExchange_rel: |
| 985 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 986 | AtomicOrdering::Release); |
| 987 | case MSVCIntrin::_InterlockedExchange_nf: |
| 988 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 989 | AtomicOrdering::Monotonic); |
| 990 | case MSVCIntrin::_InterlockedCompareExchange_acq: |
| 991 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire); |
| 992 | case MSVCIntrin::_InterlockedCompareExchange_rel: |
| 993 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release); |
| 994 | case MSVCIntrin::_InterlockedCompareExchange_nf: |
| 995 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic); |
| 996 | case MSVCIntrin::_InterlockedOr_acq: |
| 997 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 998 | AtomicOrdering::Acquire); |
| 999 | case MSVCIntrin::_InterlockedOr_rel: |
| 1000 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1001 | AtomicOrdering::Release); |
| 1002 | case MSVCIntrin::_InterlockedOr_nf: |
| 1003 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1004 | AtomicOrdering::Monotonic); |
| 1005 | case MSVCIntrin::_InterlockedXor_acq: |
| 1006 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1007 | AtomicOrdering::Acquire); |
| 1008 | case MSVCIntrin::_InterlockedXor_rel: |
| 1009 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1010 | AtomicOrdering::Release); |
| 1011 | case MSVCIntrin::_InterlockedXor_nf: |
| 1012 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1013 | AtomicOrdering::Monotonic); |
| 1014 | case MSVCIntrin::_InterlockedAnd_acq: |
| 1015 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1016 | AtomicOrdering::Acquire); |
| 1017 | case MSVCIntrin::_InterlockedAnd_rel: |
| 1018 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1019 | AtomicOrdering::Release); |
| 1020 | case MSVCIntrin::_InterlockedAnd_nf: |
| 1021 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1022 | AtomicOrdering::Monotonic); |
| 1023 | case MSVCIntrin::_InterlockedIncrement_acq: |
| 1024 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire); |
| 1025 | case MSVCIntrin::_InterlockedIncrement_rel: |
| 1026 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release); |
| 1027 | case MSVCIntrin::_InterlockedIncrement_nf: |
| 1028 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic); |
| 1029 | case MSVCIntrin::_InterlockedDecrement_acq: |
| 1030 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire); |
| 1031 | case MSVCIntrin::_InterlockedDecrement_rel: |
| 1032 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release); |
| 1033 | case MSVCIntrin::_InterlockedDecrement_nf: |
| 1034 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic); |
| 1035 | |
| 1036 | case MSVCIntrin::_InterlockedDecrement: |
| 1037 | return EmitAtomicDecrementValue(*this, E); |
| 1038 | case MSVCIntrin::_InterlockedIncrement: |
| 1039 | return EmitAtomicIncrementValue(*this, E); |
| 1040 | |
| 1041 | case MSVCIntrin::__fastfail: { |
| 1042 | // Request immediate process termination from the kernel. The instruction |
| 1043 | // sequences to do this are documented on MSDN: |
| 1044 | // https://msdn.microsoft.com/en-us/library/dn774154.aspx |
| 1045 | llvm::Triple::ArchType ISA = getTarget().getTriple().getArch(); |
| 1046 | StringRef Asm, Constraints; |
| 1047 | switch (ISA) { |
| 1048 | default: |
| 1049 | ErrorUnsupported(E, "__fastfail call for this architecture"); |
| 1050 | break; |
| 1051 | case llvm::Triple::x86: |
| 1052 | case llvm::Triple::x86_64: |
| 1053 | Asm = "int $$0x29"; |
| 1054 | Constraints = "{cx}"; |
| 1055 | break; |
| 1056 | case llvm::Triple::thumb: |
| 1057 | Asm = "udf #251"; |
| 1058 | Constraints = "{r0}"; |
| 1059 | break; |
| 1060 | case llvm::Triple::aarch64: |
| 1061 | Asm = "brk #0xF003"; |
| 1062 | Constraints = "{w0}"; |
| 1063 | } |
| 1064 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false); |
| 1065 | llvm::InlineAsm *IA = |
| 1066 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*SideEffects=*/true); |
| 1067 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( |
| 1068 | getLLVMContext(), llvm::AttributeList::FunctionIndex, |
| 1069 | llvm::Attribute::NoReturn); |
| 1070 | llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0))); |
| 1071 | CI->setAttributes(NoReturnAttr); |
| 1072 | return CI; |
| 1073 | } |
| 1074 | } |
| 1075 | llvm_unreachable("Incorrect MSVC intrinsic!")::llvm::llvm_unreachable_internal("Incorrect MSVC intrinsic!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1075); |
| 1076 | } |
| 1077 | |
| 1078 | namespace { |
| 1079 | // ARC cleanup for __builtin_os_log_format |
| 1080 | struct CallObjCArcUse final : EHScopeStack::Cleanup { |
| 1081 | CallObjCArcUse(llvm::Value *object) : object(object) {} |
| 1082 | llvm::Value *object; |
| 1083 | |
| 1084 | void Emit(CodeGenFunction &CGF, Flags flags) override { |
| 1085 | CGF.EmitARCIntrinsicUse(object); |
| 1086 | } |
| 1087 | }; |
| 1088 | } |
| 1089 | |
| 1090 | Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E, |
| 1091 | BuiltinCheckKind Kind) { |
| 1092 | assert((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero)(((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind") ? static_cast<void> ( 0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1093, __PRETTY_FUNCTION__)) |
| 1093 | && "Unsupported builtin check kind")(((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind") ? static_cast<void> ( 0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1093, __PRETTY_FUNCTION__)); |
| 1094 | |
| 1095 | Value *ArgValue = EmitScalarExpr(E); |
| 1096 | if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef()) |
| 1097 | return ArgValue; |
| 1098 | |
| 1099 | SanitizerScope SanScope(this); |
| 1100 | Value *Cond = Builder.CreateICmpNE( |
| 1101 | ArgValue, llvm::Constant::getNullValue(ArgValue->getType())); |
| 1102 | EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin), |
| 1103 | SanitizerHandler::InvalidBuiltin, |
| 1104 | {EmitCheckSourceLocation(E->getExprLoc()), |
| 1105 | llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)}, |
| 1106 | None); |
| 1107 | return ArgValue; |
| 1108 | } |
| 1109 | |
| 1110 | /// Get the argument type for arguments to os_log_helper. |
| 1111 | static CanQualType getOSLogArgType(ASTContext &C, int Size) { |
| 1112 | QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false); |
| 1113 | return C.getCanonicalType(UnsignedTy); |
| 1114 | } |
| 1115 | |
| 1116 | llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction( |
| 1117 | const analyze_os_log::OSLogBufferLayout &Layout, |
| 1118 | CharUnits BufferAlignment) { |
| 1119 | ASTContext &Ctx = getContext(); |
| 1120 | |
| 1121 | llvm::SmallString<64> Name; |
| 1122 | { |
| 1123 | raw_svector_ostream OS(Name); |
| 1124 | OS << "__os_log_helper"; |
| 1125 | OS << "_" << BufferAlignment.getQuantity(); |
| 1126 | OS << "_" << int(Layout.getSummaryByte()); |
| 1127 | OS << "_" << int(Layout.getNumArgsByte()); |
| 1128 | for (const auto &Item : Layout.Items) |
| 1129 | OS << "_" << int(Item.getSizeByte()) << "_" |
| 1130 | << int(Item.getDescriptorByte()); |
| 1131 | } |
| 1132 | |
| 1133 | if (llvm::Function *F = CGM.getModule().getFunction(Name)) |
| 1134 | return F; |
| 1135 | |
| 1136 | llvm::SmallVector<QualType, 4> ArgTys; |
| 1137 | FunctionArgList Args; |
| 1138 | Args.push_back(ImplicitParamDecl::Create( |
| 1139 | Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"), Ctx.VoidPtrTy, |
| 1140 | ImplicitParamDecl::Other)); |
| 1141 | ArgTys.emplace_back(Ctx.VoidPtrTy); |
| 1142 | |
| 1143 | for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) { |
| 1144 | char Size = Layout.Items[I].getSizeByte(); |
| 1145 | if (!Size) |
| 1146 | continue; |
| 1147 | |
| 1148 | QualType ArgTy = getOSLogArgType(Ctx, Size); |
| 1149 | Args.push_back(ImplicitParamDecl::Create( |
| 1150 | Ctx, nullptr, SourceLocation(), |
| 1151 | &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy, |
| 1152 | ImplicitParamDecl::Other)); |
| 1153 | ArgTys.emplace_back(ArgTy); |
| 1154 | } |
| 1155 | |
| 1156 | QualType ReturnTy = Ctx.VoidTy; |
| 1157 | QualType FuncionTy = Ctx.getFunctionType(ReturnTy, ArgTys, {}); |
| 1158 | |
| 1159 | // The helper function has linkonce_odr linkage to enable the linker to merge |
| 1160 | // identical functions. To ensure the merging always happens, 'noinline' is |
| 1161 | // attached to the function when compiling with -Oz. |
| 1162 | const CGFunctionInfo &FI = |
| 1163 | CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args); |
| 1164 | llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI); |
| 1165 | llvm::Function *Fn = llvm::Function::Create( |
| 1166 | FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule()); |
| 1167 | Fn->setVisibility(llvm::GlobalValue::HiddenVisibility); |
| 1168 | CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn); |
| 1169 | CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn); |
| 1170 | Fn->setDoesNotThrow(); |
| 1171 | |
| 1172 | // Attach 'noinline' at -Oz. |
| 1173 | if (CGM.getCodeGenOpts().OptimizeSize == 2) |
| 1174 | Fn->addFnAttr(llvm::Attribute::NoInline); |
| 1175 | |
| 1176 | auto NL = ApplyDebugLocation::CreateEmpty(*this); |
| 1177 | IdentifierInfo *II = &Ctx.Idents.get(Name); |
| 1178 | FunctionDecl *FD = FunctionDecl::Create( |
| 1179 | Ctx, Ctx.getTranslationUnitDecl(), SourceLocation(), SourceLocation(), II, |
| 1180 | FuncionTy, nullptr, SC_PrivateExtern, false, false); |
| 1181 | |
| 1182 | StartFunction(FD, ReturnTy, Fn, FI, Args); |
| 1183 | |
| 1184 | // Create a scope with an artificial location for the body of this function. |
| 1185 | auto AL = ApplyDebugLocation::CreateArtificial(*this); |
| 1186 | |
| 1187 | CharUnits Offset; |
| 1188 | Address BufAddr(Builder.CreateLoad(GetAddrOfLocalVar(Args[0]), "buf"), |
| 1189 | BufferAlignment); |
| 1190 | Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()), |
| 1191 | Builder.CreateConstByteGEP(BufAddr, Offset++, "summary")); |
| 1192 | Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()), |
| 1193 | Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs")); |
| 1194 | |
| 1195 | unsigned I = 1; |
| 1196 | for (const auto &Item : Layout.Items) { |
| 1197 | Builder.CreateStore( |
| 1198 | Builder.getInt8(Item.getDescriptorByte()), |
| 1199 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor")); |
| 1200 | Builder.CreateStore( |
| 1201 | Builder.getInt8(Item.getSizeByte()), |
| 1202 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize")); |
| 1203 | |
| 1204 | CharUnits Size = Item.size(); |
| 1205 | if (!Size.getQuantity()) |
| 1206 | continue; |
| 1207 | |
| 1208 | Address Arg = GetAddrOfLocalVar(Args[I]); |
| 1209 | Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData"); |
| 1210 | Addr = Builder.CreateBitCast(Addr, Arg.getPointer()->getType(), |
| 1211 | "argDataCast"); |
| 1212 | Builder.CreateStore(Builder.CreateLoad(Arg), Addr); |
| 1213 | Offset += Size; |
| 1214 | ++I; |
| 1215 | } |
| 1216 | |
| 1217 | FinishFunction(); |
| 1218 | |
| 1219 | return Fn; |
| 1220 | } |
| 1221 | |
| 1222 | RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) { |
| 1223 | assert(E.getNumArgs() >= 2 &&((E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1224, __PRETTY_FUNCTION__)) |
| 1224 | "__builtin_os_log_format takes at least 2 arguments")((E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1224, __PRETTY_FUNCTION__)); |
| 1225 | ASTContext &Ctx = getContext(); |
| 1226 | analyze_os_log::OSLogBufferLayout Layout; |
| 1227 | analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout); |
| 1228 | Address BufAddr = EmitPointerWithAlignment(E.getArg(0)); |
| 1229 | llvm::SmallVector<llvm::Value *, 4> RetainableOperands; |
| 1230 | |
| 1231 | // Ignore argument 1, the format string. It is not currently used. |
| 1232 | CallArgList Args; |
| 1233 | Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy); |
| 1234 | |
| 1235 | for (const auto &Item : Layout.Items) { |
| 1236 | int Size = Item.getSizeByte(); |
| 1237 | if (!Size) |
| 1238 | continue; |
| 1239 | |
| 1240 | llvm::Value *ArgVal; |
| 1241 | |
| 1242 | if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) { |
| 1243 | uint64_t Val = 0; |
| 1244 | for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I) |
| 1245 | Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8; |
| 1246 | ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val)); |
| 1247 | } else if (const Expr *TheExpr = Item.getExpr()) { |
| 1248 | ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false); |
| 1249 | |
| 1250 | // Check if this is a retainable type. |
| 1251 | if (TheExpr->getType()->isObjCRetainableType()) { |
| 1252 | assert(getEvaluationKind(TheExpr->getType()) == TEK_Scalar &&((getEvaluationKind(TheExpr->getType()) == TEK_Scalar && "Only scalar can be a ObjC retainable type") ? static_cast< void> (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1253, __PRETTY_FUNCTION__)) |
| 1253 | "Only scalar can be a ObjC retainable type")((getEvaluationKind(TheExpr->getType()) == TEK_Scalar && "Only scalar can be a ObjC retainable type") ? static_cast< void> (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1253, __PRETTY_FUNCTION__)); |
| 1254 | // Check if the object is constant, if not, save it in |
| 1255 | // RetainableOperands. |
| 1256 | if (!isa<Constant>(ArgVal)) |
| 1257 | RetainableOperands.push_back(ArgVal); |
| 1258 | } |
| 1259 | } else { |
| 1260 | ArgVal = Builder.getInt32(Item.getConstValue().getQuantity()); |
| 1261 | } |
| 1262 | |
| 1263 | unsigned ArgValSize = |
| 1264 | CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType()); |
| 1265 | llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(), |
| 1266 | ArgValSize); |
| 1267 | ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy); |
| 1268 | CanQualType ArgTy = getOSLogArgType(Ctx, Size); |
| 1269 | // If ArgVal has type x86_fp80, zero-extend ArgVal. |
| 1270 | ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy)); |
| 1271 | Args.add(RValue::get(ArgVal), ArgTy); |
| 1272 | } |
| 1273 | |
| 1274 | const CGFunctionInfo &FI = |
| 1275 | CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args); |
| 1276 | llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction( |
| 1277 | Layout, BufAddr.getAlignment()); |
| 1278 | EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args); |
| 1279 | |
| 1280 | // Push a clang.arc.use cleanup for each object in RetainableOperands. The |
| 1281 | // cleanup will cause the use to appear after the final log call, keeping |
| 1282 | // the object valid while it’s held in the log buffer. Note that if there’s |
| 1283 | // a release cleanup on the object, it will already be active; since |
| 1284 | // cleanups are emitted in reverse order, the use will occur before the |
| 1285 | // object is released. |
| 1286 | if (!RetainableOperands.empty() && getLangOpts().ObjCAutoRefCount && |
| 1287 | CGM.getCodeGenOpts().OptimizationLevel != 0) |
| 1288 | for (llvm::Value *Object : RetainableOperands) |
| 1289 | pushFullExprCleanup<CallObjCArcUse>(getARCCleanupKind(), Object); |
| 1290 | |
| 1291 | return RValue::get(BufAddr.getPointer()); |
| 1292 | } |
| 1293 | |
| 1294 | /// Determine if a binop is a checked mixed-sign multiply we can specialize. |
| 1295 | static bool isSpecialMixedSignMultiply(unsigned BuiltinID, |
| 1296 | WidthAndSignedness Op1Info, |
| 1297 | WidthAndSignedness Op2Info, |
| 1298 | WidthAndSignedness ResultInfo) { |
| 1299 | return BuiltinID == Builtin::BI__builtin_mul_overflow && |
| 1300 | std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width && |
| 1301 | Op1Info.Signed != Op2Info.Signed; |
| 1302 | } |
| 1303 | |
| 1304 | /// Emit a checked mixed-sign multiply. This is a cheaper specialization of |
| 1305 | /// the generic checked-binop irgen. |
| 1306 | static RValue |
| 1307 | EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1, |
| 1308 | WidthAndSignedness Op1Info, const clang::Expr *Op2, |
| 1309 | WidthAndSignedness Op2Info, |
| 1310 | const clang::Expr *ResultArg, QualType ResultQTy, |
| 1311 | WidthAndSignedness ResultInfo) { |
| 1312 | assert(isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info,((isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize" ) ? static_cast<void> (0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1314, __PRETTY_FUNCTION__)) |
| 1313 | Op2Info, ResultInfo) &&((isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize" ) ? static_cast<void> (0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1314, __PRETTY_FUNCTION__)) |
| 1314 | "Not a mixed-sign multipliction we can specialize")((isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow , Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize" ) ? static_cast<void> (0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1314, __PRETTY_FUNCTION__)); |
| 1315 | |
| 1316 | // Emit the signed and unsigned operands. |
| 1317 | const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2; |
| 1318 | const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1; |
| 1319 | llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp); |
| 1320 | llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp); |
| 1321 | unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width; |
| 1322 | unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width; |
| 1323 | |
| 1324 | // One of the operands may be smaller than the other. If so, [s|z]ext it. |
| 1325 | if (SignedOpWidth < UnsignedOpWidth) |
| 1326 | Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext"); |
| 1327 | if (UnsignedOpWidth < SignedOpWidth) |
| 1328 | Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext"); |
| 1329 | |
| 1330 | llvm::Type *OpTy = Signed->getType(); |
| 1331 | llvm::Value *Zero = llvm::Constant::getNullValue(OpTy); |
| 1332 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); |
| 1333 | llvm::Type *ResTy = ResultPtr.getElementType(); |
| 1334 | unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width); |
| 1335 | |
| 1336 | // Take the absolute value of the signed operand. |
| 1337 | llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero); |
| 1338 | llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed); |
| 1339 | llvm::Value *AbsSigned = |
| 1340 | CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed); |
| 1341 | |
| 1342 | // Perform a checked unsigned multiplication. |
| 1343 | llvm::Value *UnsignedOverflow; |
| 1344 | llvm::Value *UnsignedResult = |
| 1345 | EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned, |
| 1346 | Unsigned, UnsignedOverflow); |
| 1347 | |
| 1348 | llvm::Value *Overflow, *Result; |
| 1349 | if (ResultInfo.Signed) { |
| 1350 | // Signed overflow occurs if the result is greater than INT_MAX or lesser |
| 1351 | // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative). |
| 1352 | auto IntMax = |
| 1353 | llvm::APInt::getSignedMaxValue(ResultInfo.Width).zextOrSelf(OpWidth); |
| 1354 | llvm::Value *MaxResult = |
| 1355 | CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax), |
| 1356 | CGF.Builder.CreateZExt(IsNegative, OpTy)); |
| 1357 | llvm::Value *SignedOverflow = |
| 1358 | CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult); |
| 1359 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow); |
| 1360 | |
| 1361 | // Prepare the signed result (possibly by negating it). |
| 1362 | llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult); |
| 1363 | llvm::Value *SignedResult = |
| 1364 | CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult); |
| 1365 | Result = CGF.Builder.CreateTrunc(SignedResult, ResTy); |
| 1366 | } else { |
| 1367 | // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX. |
| 1368 | llvm::Value *Underflow = CGF.Builder.CreateAnd( |
| 1369 | IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult)); |
| 1370 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow); |
| 1371 | if (ResultInfo.Width < OpWidth) { |
| 1372 | auto IntMax = |
| 1373 | llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth); |
| 1374 | llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT( |
| 1375 | UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax)); |
| 1376 | Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow); |
| 1377 | } |
| 1378 | |
| 1379 | // Negate the product if it would be negative in infinite precision. |
| 1380 | Result = CGF.Builder.CreateSelect( |
| 1381 | IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult); |
| 1382 | |
| 1383 | Result = CGF.Builder.CreateTrunc(Result, ResTy); |
| 1384 | } |
| 1385 | assert(Overflow && Result && "Missing overflow or result")((Overflow && Result && "Missing overflow or result" ) ? static_cast<void> (0) : __assert_fail ("Overflow && Result && \"Missing overflow or result\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1385, __PRETTY_FUNCTION__)); |
| 1386 | |
| 1387 | bool isVolatile = |
| 1388 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 1389 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, |
| 1390 | isVolatile); |
| 1391 | return RValue::get(Overflow); |
| 1392 | } |
| 1393 | |
| 1394 | static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType, |
| 1395 | Value *&RecordPtr, CharUnits Align, |
| 1396 | llvm::FunctionCallee Func, int Lvl) { |
| 1397 | const auto *RT = RType->getAs<RecordType>(); |
| 1398 | ASTContext &Context = CGF.getContext(); |
| 1399 | RecordDecl *RD = RT->getDecl()->getDefinition(); |
| 1400 | std::string Pad = std::string(Lvl * 4, ' '); |
| 1401 | |
| 1402 | Value *GString = |
| 1403 | CGF.Builder.CreateGlobalStringPtr(RType.getAsString() + " {\n"); |
| 1404 | Value *Res = CGF.Builder.CreateCall(Func, {GString}); |
| 1405 | |
| 1406 | static llvm::DenseMap<QualType, const char *> Types; |
| 1407 | if (Types.empty()) { |
| 1408 | Types[Context.CharTy] = "%c"; |
| 1409 | Types[Context.BoolTy] = "%d"; |
| 1410 | Types[Context.SignedCharTy] = "%hhd"; |
| 1411 | Types[Context.UnsignedCharTy] = "%hhu"; |
| 1412 | Types[Context.IntTy] = "%d"; |
| 1413 | Types[Context.UnsignedIntTy] = "%u"; |
| 1414 | Types[Context.LongTy] = "%ld"; |
| 1415 | Types[Context.UnsignedLongTy] = "%lu"; |
| 1416 | Types[Context.LongLongTy] = "%lld"; |
| 1417 | Types[Context.UnsignedLongLongTy] = "%llu"; |
| 1418 | Types[Context.ShortTy] = "%hd"; |
| 1419 | Types[Context.UnsignedShortTy] = "%hu"; |
| 1420 | Types[Context.VoidPtrTy] = "%p"; |
| 1421 | Types[Context.FloatTy] = "%f"; |
| 1422 | Types[Context.DoubleTy] = "%f"; |
| 1423 | Types[Context.LongDoubleTy] = "%Lf"; |
| 1424 | Types[Context.getPointerType(Context.CharTy)] = "%s"; |
| 1425 | Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s"; |
| 1426 | } |
| 1427 | |
| 1428 | for (const auto *FD : RD->fields()) { |
| 1429 | Value *FieldPtr = RecordPtr; |
| 1430 | if (RD->isUnion()) |
| 1431 | FieldPtr = CGF.Builder.CreatePointerCast( |
| 1432 | FieldPtr, CGF.ConvertType(Context.getPointerType(FD->getType()))); |
| 1433 | else |
| 1434 | FieldPtr = CGF.Builder.CreateStructGEP(CGF.ConvertType(RType), FieldPtr, |
| 1435 | FD->getFieldIndex()); |
| 1436 | |
| 1437 | GString = CGF.Builder.CreateGlobalStringPtr( |
| 1438 | llvm::Twine(Pad) |
| 1439 | .concat(FD->getType().getAsString()) |
| 1440 | .concat(llvm::Twine(' ')) |
| 1441 | .concat(FD->getNameAsString()) |
| 1442 | .concat(" : ") |
| 1443 | .str()); |
| 1444 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); |
| 1445 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
| 1446 | |
| 1447 | QualType CanonicalType = |
| 1448 | FD->getType().getUnqualifiedType().getCanonicalType(); |
| 1449 | |
| 1450 | // We check whether we are in a recursive type |
| 1451 | if (CanonicalType->isRecordType()) { |
| 1452 | Value *TmpRes = |
| 1453 | dumpRecord(CGF, CanonicalType, FieldPtr, Align, Func, Lvl + 1); |
| 1454 | Res = CGF.Builder.CreateAdd(TmpRes, Res); |
| 1455 | continue; |
| 1456 | } |
| 1457 | |
| 1458 | // We try to determine the best format to print the current field |
| 1459 | llvm::Twine Format = Types.find(CanonicalType) == Types.end() |
| 1460 | ? Types[Context.VoidPtrTy] |
| 1461 | : Types[CanonicalType]; |
| 1462 | |
| 1463 | Address FieldAddress = Address(FieldPtr, Align); |
| 1464 | FieldPtr = CGF.Builder.CreateLoad(FieldAddress); |
| 1465 | |
| 1466 | // FIXME Need to handle bitfield here |
| 1467 | GString = CGF.Builder.CreateGlobalStringPtr( |
| 1468 | Format.concat(llvm::Twine('\n')).str()); |
| 1469 | TmpRes = CGF.Builder.CreateCall(Func, {GString, FieldPtr}); |
| 1470 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
| 1471 | } |
| 1472 | |
| 1473 | GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n"); |
| 1474 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); |
| 1475 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
| 1476 | return Res; |
| 1477 | } |
| 1478 | |
| 1479 | static bool |
| 1480 | TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty, |
| 1481 | llvm::SmallPtrSetImpl<const Decl *> &Seen) { |
| 1482 | if (const auto *Arr = Ctx.getAsArrayType(Ty)) |
| 1483 | Ty = Ctx.getBaseElementType(Arr); |
| 1484 | |
| 1485 | const auto *Record = Ty->getAsCXXRecordDecl(); |
| 1486 | if (!Record) |
| 1487 | return false; |
| 1488 | |
| 1489 | // We've already checked this type, or are in the process of checking it. |
| 1490 | if (!Seen.insert(Record).second) |
| 1491 | return false; |
| 1492 | |
| 1493 | assert(Record->hasDefinition() &&((Record->hasDefinition() && "Incomplete types should already be diagnosed" ) ? static_cast<void> (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1494, __PRETTY_FUNCTION__)) |
| 1494 | "Incomplete types should already be diagnosed")((Record->hasDefinition() && "Incomplete types should already be diagnosed" ) ? static_cast<void> (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 1494, __PRETTY_FUNCTION__)); |
| 1495 | |
| 1496 | if (Record->isDynamicClass()) |
| 1497 | return true; |
| 1498 | |
| 1499 | for (FieldDecl *F : Record->fields()) { |
| 1500 | if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen)) |
| 1501 | return true; |
| 1502 | } |
| 1503 | return false; |
| 1504 | } |
| 1505 | |
| 1506 | /// Determine if the specified type requires laundering by checking if it is a |
| 1507 | /// dynamic class type or contains a subobject which is a dynamic class type. |
| 1508 | static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) { |
| 1509 | if (!CGM.getCodeGenOpts().StrictVTablePointers) |
| 1510 | return false; |
| 1511 | llvm::SmallPtrSet<const Decl *, 16> Seen; |
| 1512 | return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen); |
| 1513 | } |
| 1514 | |
| 1515 | RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) { |
| 1516 | llvm::Value *Src = EmitScalarExpr(E->getArg(0)); |
| 1517 | llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1)); |
| 1518 | |
| 1519 | // The builtin's shift arg may have a different type than the source arg and |
| 1520 | // result, but the LLVM intrinsic uses the same type for all values. |
| 1521 | llvm::Type *Ty = Src->getType(); |
| 1522 | ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false); |
| 1523 | |
| 1524 | // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same. |
| 1525 | unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl; |
| 1526 | Function *F = CGM.getIntrinsic(IID, Ty); |
| 1527 | return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt })); |
| 1528 | } |
| 1529 | |
| 1530 | RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, |
| 1531 | const CallExpr *E, |
| 1532 | ReturnValueSlot ReturnValue) { |
| 1533 | const FunctionDecl *FD = GD.getDecl()->getAsFunction(); |
| 1534 | // See if we can constant fold this builtin. If so, don't emit it at all. |
| 1535 | Expr::EvalResult Result; |
| 1536 | if (E->EvaluateAsRValue(Result, CGM.getContext()) && |
| 1537 | !Result.hasSideEffects()) { |
| 1538 | if (Result.Val.isInt()) |
| 1539 | return RValue::get(llvm::ConstantInt::get(getLLVMContext(), |
| 1540 | Result.Val.getInt())); |
| 1541 | if (Result.Val.isFloat()) |
| 1542 | return RValue::get(llvm::ConstantFP::get(getLLVMContext(), |
| 1543 | Result.Val.getFloat())); |
| 1544 | } |
| 1545 | |
| 1546 | // There are LLVM math intrinsics/instructions corresponding to math library |
| 1547 | // functions except the LLVM op will never set errno while the math library |
| 1548 | // might. Also, math builtins have the same semantics as their math library |
| 1549 | // twins. Thus, we can transform math library and builtin calls to their |
| 1550 | // LLVM counterparts if the call is marked 'const' (known to never set errno). |
| 1551 | if (FD->hasAttr<ConstAttr>()) { |
| 1552 | switch (BuiltinID) { |
| 1553 | case Builtin::BIceil: |
| 1554 | case Builtin::BIceilf: |
| 1555 | case Builtin::BIceill: |
| 1556 | case Builtin::BI__builtin_ceil: |
| 1557 | case Builtin::BI__builtin_ceilf: |
| 1558 | case Builtin::BI__builtin_ceill: |
| 1559 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::ceil)); |
| 1560 | |
| 1561 | case Builtin::BIcopysign: |
| 1562 | case Builtin::BIcopysignf: |
| 1563 | case Builtin::BIcopysignl: |
| 1564 | case Builtin::BI__builtin_copysign: |
| 1565 | case Builtin::BI__builtin_copysignf: |
| 1566 | case Builtin::BI__builtin_copysignl: |
| 1567 | case Builtin::BI__builtin_copysignf128: |
| 1568 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign)); |
| 1569 | |
| 1570 | case Builtin::BIcos: |
| 1571 | case Builtin::BIcosf: |
| 1572 | case Builtin::BIcosl: |
| 1573 | case Builtin::BI__builtin_cos: |
| 1574 | case Builtin::BI__builtin_cosf: |
| 1575 | case Builtin::BI__builtin_cosl: |
| 1576 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::cos)); |
| 1577 | |
| 1578 | case Builtin::BIexp: |
| 1579 | case Builtin::BIexpf: |
| 1580 | case Builtin::BIexpl: |
| 1581 | case Builtin::BI__builtin_exp: |
| 1582 | case Builtin::BI__builtin_expf: |
| 1583 | case Builtin::BI__builtin_expl: |
| 1584 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp)); |
| 1585 | |
| 1586 | case Builtin::BIexp2: |
| 1587 | case Builtin::BIexp2f: |
| 1588 | case Builtin::BIexp2l: |
| 1589 | case Builtin::BI__builtin_exp2: |
| 1590 | case Builtin::BI__builtin_exp2f: |
| 1591 | case Builtin::BI__builtin_exp2l: |
| 1592 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp2)); |
| 1593 | |
| 1594 | case Builtin::BIfabs: |
| 1595 | case Builtin::BIfabsf: |
| 1596 | case Builtin::BIfabsl: |
| 1597 | case Builtin::BI__builtin_fabs: |
| 1598 | case Builtin::BI__builtin_fabsf: |
| 1599 | case Builtin::BI__builtin_fabsl: |
| 1600 | case Builtin::BI__builtin_fabsf128: |
| 1601 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs)); |
| 1602 | |
| 1603 | case Builtin::BIfloor: |
| 1604 | case Builtin::BIfloorf: |
| 1605 | case Builtin::BIfloorl: |
| 1606 | case Builtin::BI__builtin_floor: |
| 1607 | case Builtin::BI__builtin_floorf: |
| 1608 | case Builtin::BI__builtin_floorl: |
| 1609 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::floor)); |
| 1610 | |
| 1611 | case Builtin::BIfma: |
| 1612 | case Builtin::BIfmaf: |
| 1613 | case Builtin::BIfmal: |
| 1614 | case Builtin::BI__builtin_fma: |
| 1615 | case Builtin::BI__builtin_fmaf: |
| 1616 | case Builtin::BI__builtin_fmal: |
| 1617 | return RValue::get(emitTernaryBuiltin(*this, E, Intrinsic::fma)); |
| 1618 | |
| 1619 | case Builtin::BIfmax: |
| 1620 | case Builtin::BIfmaxf: |
| 1621 | case Builtin::BIfmaxl: |
| 1622 | case Builtin::BI__builtin_fmax: |
| 1623 | case Builtin::BI__builtin_fmaxf: |
| 1624 | case Builtin::BI__builtin_fmaxl: |
| 1625 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::maxnum)); |
| 1626 | |
| 1627 | case Builtin::BIfmin: |
| 1628 | case Builtin::BIfminf: |
| 1629 | case Builtin::BIfminl: |
| 1630 | case Builtin::BI__builtin_fmin: |
| 1631 | case Builtin::BI__builtin_fminf: |
| 1632 | case Builtin::BI__builtin_fminl: |
| 1633 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::minnum)); |
| 1634 | |
| 1635 | // fmod() is a special-case. It maps to the frem instruction rather than an |
| 1636 | // LLVM intrinsic. |
| 1637 | case Builtin::BIfmod: |
| 1638 | case Builtin::BIfmodf: |
| 1639 | case Builtin::BIfmodl: |
| 1640 | case Builtin::BI__builtin_fmod: |
| 1641 | case Builtin::BI__builtin_fmodf: |
| 1642 | case Builtin::BI__builtin_fmodl: { |
| 1643 | Value *Arg1 = EmitScalarExpr(E->getArg(0)); |
| 1644 | Value *Arg2 = EmitScalarExpr(E->getArg(1)); |
| 1645 | return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod")); |
| 1646 | } |
| 1647 | |
| 1648 | case Builtin::BIlog: |
| 1649 | case Builtin::BIlogf: |
| 1650 | case Builtin::BIlogl: |
| 1651 | case Builtin::BI__builtin_log: |
| 1652 | case Builtin::BI__builtin_logf: |
| 1653 | case Builtin::BI__builtin_logl: |
| 1654 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log)); |
| 1655 | |
| 1656 | case Builtin::BIlog10: |
| 1657 | case Builtin::BIlog10f: |
| 1658 | case Builtin::BIlog10l: |
| 1659 | case Builtin::BI__builtin_log10: |
| 1660 | case Builtin::BI__builtin_log10f: |
| 1661 | case Builtin::BI__builtin_log10l: |
| 1662 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log10)); |
| 1663 | |
| 1664 | case Builtin::BIlog2: |
| 1665 | case Builtin::BIlog2f: |
| 1666 | case Builtin::BIlog2l: |
| 1667 | case Builtin::BI__builtin_log2: |
| 1668 | case Builtin::BI__builtin_log2f: |
| 1669 | case Builtin::BI__builtin_log2l: |
| 1670 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log2)); |
| 1671 | |
| 1672 | case Builtin::BInearbyint: |
| 1673 | case Builtin::BInearbyintf: |
| 1674 | case Builtin::BInearbyintl: |
| 1675 | case Builtin::BI__builtin_nearbyint: |
| 1676 | case Builtin::BI__builtin_nearbyintf: |
| 1677 | case Builtin::BI__builtin_nearbyintl: |
| 1678 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::nearbyint)); |
| 1679 | |
| 1680 | case Builtin::BIpow: |
| 1681 | case Builtin::BIpowf: |
| 1682 | case Builtin::BIpowl: |
| 1683 | case Builtin::BI__builtin_pow: |
| 1684 | case Builtin::BI__builtin_powf: |
| 1685 | case Builtin::BI__builtin_powl: |
| 1686 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::pow)); |
| 1687 | |
| 1688 | case Builtin::BIrint: |
| 1689 | case Builtin::BIrintf: |
| 1690 | case Builtin::BIrintl: |
| 1691 | case Builtin::BI__builtin_rint: |
| 1692 | case Builtin::BI__builtin_rintf: |
| 1693 | case Builtin::BI__builtin_rintl: |
| 1694 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::rint)); |
| 1695 | |
| 1696 | case Builtin::BIround: |
| 1697 | case Builtin::BIroundf: |
| 1698 | case Builtin::BIroundl: |
| 1699 | case Builtin::BI__builtin_round: |
| 1700 | case Builtin::BI__builtin_roundf: |
| 1701 | case Builtin::BI__builtin_roundl: |
| 1702 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::round)); |
| 1703 | |
| 1704 | case Builtin::BIsin: |
| 1705 | case Builtin::BIsinf: |
| 1706 | case Builtin::BIsinl: |
| 1707 | case Builtin::BI__builtin_sin: |
| 1708 | case Builtin::BI__builtin_sinf: |
| 1709 | case Builtin::BI__builtin_sinl: |
| 1710 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sin)); |
| 1711 | |
| 1712 | case Builtin::BIsqrt: |
| 1713 | case Builtin::BIsqrtf: |
| 1714 | case Builtin::BIsqrtl: |
| 1715 | case Builtin::BI__builtin_sqrt: |
| 1716 | case Builtin::BI__builtin_sqrtf: |
| 1717 | case Builtin::BI__builtin_sqrtl: |
| 1718 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sqrt)); |
| 1719 | |
| 1720 | case Builtin::BItrunc: |
| 1721 | case Builtin::BItruncf: |
| 1722 | case Builtin::BItruncl: |
| 1723 | case Builtin::BI__builtin_trunc: |
| 1724 | case Builtin::BI__builtin_truncf: |
| 1725 | case Builtin::BI__builtin_truncl: |
| 1726 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::trunc)); |
| 1727 | |
| 1728 | case Builtin::BIlround: |
| 1729 | case Builtin::BIlroundf: |
| 1730 | case Builtin::BIlroundl: |
| 1731 | case Builtin::BI__builtin_lround: |
| 1732 | case Builtin::BI__builtin_lroundf: |
| 1733 | case Builtin::BI__builtin_lroundl: |
| 1734 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::lround)); |
| 1735 | |
| 1736 | case Builtin::BIllround: |
| 1737 | case Builtin::BIllroundf: |
| 1738 | case Builtin::BIllroundl: |
| 1739 | case Builtin::BI__builtin_llround: |
| 1740 | case Builtin::BI__builtin_llroundf: |
| 1741 | case Builtin::BI__builtin_llroundl: |
| 1742 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::llround)); |
| 1743 | |
| 1744 | case Builtin::BIlrint: |
| 1745 | case Builtin::BIlrintf: |
| 1746 | case Builtin::BIlrintl: |
| 1747 | case Builtin::BI__builtin_lrint: |
| 1748 | case Builtin::BI__builtin_lrintf: |
| 1749 | case Builtin::BI__builtin_lrintl: |
| 1750 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::lrint)); |
| 1751 | |
| 1752 | case Builtin::BIllrint: |
| 1753 | case Builtin::BIllrintf: |
| 1754 | case Builtin::BIllrintl: |
| 1755 | case Builtin::BI__builtin_llrint: |
| 1756 | case Builtin::BI__builtin_llrintf: |
| 1757 | case Builtin::BI__builtin_llrintl: |
| 1758 | return RValue::get(emitFPToIntRoundBuiltin(*this, E, Intrinsic::llrint)); |
| 1759 | |
| 1760 | default: |
| 1761 | break; |
| 1762 | } |
| 1763 | } |
| 1764 | |
| 1765 | switch (BuiltinID) { |
| 1766 | default: break; |
| 1767 | case Builtin::BI__builtin___CFStringMakeConstantString: |
| 1768 | case Builtin::BI__builtin___NSStringMakeConstantString: |
| 1769 | return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType())); |
| 1770 | case Builtin::BI__builtin_stdarg_start: |
| 1771 | case Builtin::BI__builtin_va_start: |
| 1772 | case Builtin::BI__va_start: |
| 1773 | case Builtin::BI__builtin_va_end: |
| 1774 | return RValue::get( |
| 1775 | EmitVAStartEnd(BuiltinID == Builtin::BI__va_start |
| 1776 | ? EmitScalarExpr(E->getArg(0)) |
| 1777 | : EmitVAListRef(E->getArg(0)).getPointer(), |
| 1778 | BuiltinID != Builtin::BI__builtin_va_end)); |
| 1779 | case Builtin::BI__builtin_va_copy: { |
| 1780 | Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer(); |
| 1781 | Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer(); |
| 1782 | |
| 1783 | llvm::Type *Type = Int8PtrTy; |
| 1784 | |
| 1785 | DstPtr = Builder.CreateBitCast(DstPtr, Type); |
| 1786 | SrcPtr = Builder.CreateBitCast(SrcPtr, Type); |
| 1787 | return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy), |
| 1788 | {DstPtr, SrcPtr})); |
| 1789 | } |
| 1790 | case Builtin::BI__builtin_abs: |
| 1791 | case Builtin::BI__builtin_labs: |
| 1792 | case Builtin::BI__builtin_llabs: { |
| 1793 | // X < 0 ? -X : X |
| 1794 | // The negation has 'nsw' because abs of INT_MIN is undefined. |
| 1795 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 1796 | Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg"); |
| 1797 | Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType()); |
| 1798 | Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond"); |
| 1799 | Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs"); |
| 1800 | return RValue::get(Result); |
| 1801 | } |
| 1802 | case Builtin::BI__builtin_conj: |
| 1803 | case Builtin::BI__builtin_conjf: |
| 1804 | case Builtin::BI__builtin_conjl: { |
| 1805 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 1806 | Value *Real = ComplexVal.first; |
| 1807 | Value *Imag = ComplexVal.second; |
| 1808 | Value *Zero = |
| 1809 | Imag->getType()->isFPOrFPVectorTy() |
| 1810 | ? llvm::ConstantFP::getZeroValueForNegation(Imag->getType()) |
| 1811 | : llvm::Constant::getNullValue(Imag->getType()); |
| 1812 | |
| 1813 | Imag = Builder.CreateFSub(Zero, Imag, "sub"); |
| 1814 | return RValue::getComplex(std::make_pair(Real, Imag)); |
| 1815 | } |
| 1816 | case Builtin::BI__builtin_creal: |
| 1817 | case Builtin::BI__builtin_crealf: |
| 1818 | case Builtin::BI__builtin_creall: |
| 1819 | case Builtin::BIcreal: |
| 1820 | case Builtin::BIcrealf: |
| 1821 | case Builtin::BIcreall: { |
| 1822 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 1823 | return RValue::get(ComplexVal.first); |
| 1824 | } |
| 1825 | |
| 1826 | case Builtin::BI__builtin_dump_struct: { |
| 1827 | llvm::Type *LLVMIntTy = getTypes().ConvertType(getContext().IntTy); |
| 1828 | llvm::FunctionType *LLVMFuncType = llvm::FunctionType::get( |
| 1829 | LLVMIntTy, {llvm::Type::getInt8PtrTy(getLLVMContext())}, true); |
| 1830 | |
| 1831 | Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts()); |
| 1832 | CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment(); |
| 1833 | |
| 1834 | const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts(); |
| 1835 | QualType Arg0Type = Arg0->getType()->getPointeeType(); |
| 1836 | |
| 1837 | Value *RecordPtr = EmitScalarExpr(Arg0); |
| 1838 | Value *Res = dumpRecord(*this, Arg0Type, RecordPtr, Arg0Align, |
| 1839 | {LLVMFuncType, Func}, 0); |
| 1840 | return RValue::get(Res); |
| 1841 | } |
| 1842 | |
| 1843 | case Builtin::BI__builtin_cimag: |
| 1844 | case Builtin::BI__builtin_cimagf: |
| 1845 | case Builtin::BI__builtin_cimagl: |
| 1846 | case Builtin::BIcimag: |
| 1847 | case Builtin::BIcimagf: |
| 1848 | case Builtin::BIcimagl: { |
| 1849 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 1850 | return RValue::get(ComplexVal.second); |
| 1851 | } |
| 1852 | |
| 1853 | case Builtin::BI__builtin_clrsb: |
| 1854 | case Builtin::BI__builtin_clrsbl: |
| 1855 | case Builtin::BI__builtin_clrsbll: { |
| 1856 | // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or |
| 1857 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 1858 | |
| 1859 | llvm::Type *ArgType = ArgValue->getType(); |
| 1860 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 1861 | |
| 1862 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1863 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
| 1864 | Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg"); |
| 1865 | Value *Inverse = Builder.CreateNot(ArgValue, "not"); |
| 1866 | Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue); |
| 1867 | Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()}); |
| 1868 | Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1)); |
| 1869 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 1870 | "cast"); |
| 1871 | return RValue::get(Result); |
| 1872 | } |
| 1873 | case Builtin::BI__builtin_ctzs: |
| 1874 | case Builtin::BI__builtin_ctz: |
| 1875 | case Builtin::BI__builtin_ctzl: |
| 1876 | case Builtin::BI__builtin_ctzll: { |
| 1877 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero); |
| 1878 | |
| 1879 | llvm::Type *ArgType = ArgValue->getType(); |
| 1880 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 1881 | |
| 1882 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1883 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
| 1884 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
| 1885 | if (Result->getType() != ResultType) |
| 1886 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 1887 | "cast"); |
| 1888 | return RValue::get(Result); |
| 1889 | } |
| 1890 | case Builtin::BI__builtin_clzs: |
| 1891 | case Builtin::BI__builtin_clz: |
| 1892 | case Builtin::BI__builtin_clzl: |
| 1893 | case Builtin::BI__builtin_clzll: { |
| 1894 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero); |
| 1895 | |
| 1896 | llvm::Type *ArgType = ArgValue->getType(); |
| 1897 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 1898 | |
| 1899 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1900 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
| 1901 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
| 1902 | if (Result->getType() != ResultType) |
| 1903 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 1904 | "cast"); |
| 1905 | return RValue::get(Result); |
| 1906 | } |
| 1907 | case Builtin::BI__builtin_ffs: |
| 1908 | case Builtin::BI__builtin_ffsl: |
| 1909 | case Builtin::BI__builtin_ffsll: { |
| 1910 | // ffs(x) -> x ? cttz(x) + 1 : 0 |
| 1911 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 1912 | |
| 1913 | llvm::Type *ArgType = ArgValue->getType(); |
| 1914 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 1915 | |
| 1916 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1917 | Value *Tmp = |
| 1918 | Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}), |
| 1919 | llvm::ConstantInt::get(ArgType, 1)); |
| 1920 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
| 1921 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero"); |
| 1922 | Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs"); |
| 1923 | if (Result->getType() != ResultType) |
| 1924 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 1925 | "cast"); |
| 1926 | return RValue::get(Result); |
| 1927 | } |
| 1928 | case Builtin::BI__builtin_parity: |
| 1929 | case Builtin::BI__builtin_parityl: |
| 1930 | case Builtin::BI__builtin_parityll: { |
| 1931 | // parity(x) -> ctpop(x) & 1 |
| 1932 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 1933 | |
| 1934 | llvm::Type *ArgType = ArgValue->getType(); |
| 1935 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
| 1936 | |
| 1937 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1938 | Value *Tmp = Builder.CreateCall(F, ArgValue); |
| 1939 | Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1)); |
| 1940 | if (Result->getType() != ResultType) |
| 1941 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 1942 | "cast"); |
| 1943 | return RValue::get(Result); |
| 1944 | } |
| 1945 | case Builtin::BI__lzcnt16: |
| 1946 | case Builtin::BI__lzcnt: |
| 1947 | case Builtin::BI__lzcnt64: { |
| 1948 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 1949 | |
| 1950 | llvm::Type *ArgType = ArgValue->getType(); |
| 1951 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 1952 | |
| 1953 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1954 | Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()}); |
| 1955 | if (Result->getType() != ResultType) |
| 1956 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 1957 | "cast"); |
| 1958 | return RValue::get(Result); |
| 1959 | } |
| 1960 | case Builtin::BI__popcnt16: |
| 1961 | case Builtin::BI__popcnt: |
| 1962 | case Builtin::BI__popcnt64: |
| 1963 | case Builtin::BI__builtin_popcount: |
| 1964 | case Builtin::BI__builtin_popcountl: |
| 1965 | case Builtin::BI__builtin_popcountll: { |
| 1966 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 1967 | |
| 1968 | llvm::Type *ArgType = ArgValue->getType(); |
| 1969 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
| 1970 | |
| 1971 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1972 | Value *Result = Builder.CreateCall(F, ArgValue); |
| 1973 | if (Result->getType() != ResultType) |
| 1974 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 1975 | "cast"); |
| 1976 | return RValue::get(Result); |
| 1977 | } |
| 1978 | case Builtin::BI__builtin_unpredictable: { |
| 1979 | // Always return the argument of __builtin_unpredictable. LLVM does not |
| 1980 | // handle this builtin. Metadata for this builtin should be added directly |
| 1981 | // to instructions such as branches or switches that use it. |
| 1982 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
| 1983 | } |
| 1984 | case Builtin::BI__builtin_expect: { |
| 1985 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 1986 | llvm::Type *ArgType = ArgValue->getType(); |
| 1987 | |
| 1988 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); |
| 1989 | // Don't generate llvm.expect on -O0 as the backend won't use it for |
| 1990 | // anything. |
| 1991 | // Note, we still IRGen ExpectedValue because it could have side-effects. |
| 1992 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
| 1993 | return RValue::get(ArgValue); |
| 1994 | |
| 1995 | Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType); |
| 1996 | Value *Result = |
| 1997 | Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval"); |
| 1998 | return RValue::get(Result); |
| 1999 | } |
| 2000 | case Builtin::BI__builtin_assume_aligned: { |
| 2001 | const Expr *Ptr = E->getArg(0); |
| 2002 | Value *PtrValue = EmitScalarExpr(Ptr); |
| 2003 | Value *OffsetValue = |
| 2004 | (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr; |
| 2005 | |
| 2006 | Value *AlignmentValue = EmitScalarExpr(E->getArg(1)); |
| 2007 | ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue); |
| 2008 | unsigned Alignment = (unsigned)AlignmentCI->getZExtValue(); |
| 2009 | |
| 2010 | EmitAlignmentAssumption(PtrValue, Ptr, |
| 2011 | /*The expr loc is sufficient.*/ SourceLocation(), |
| 2012 | Alignment, OffsetValue); |
| 2013 | return RValue::get(PtrValue); |
| 2014 | } |
| 2015 | case Builtin::BI__assume: |
| 2016 | case Builtin::BI__builtin_assume: { |
| 2017 | if (E->getArg(0)->HasSideEffects(getContext())) |
| 2018 | return RValue::get(nullptr); |
| 2019 | |
| 2020 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2021 | Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume); |
| 2022 | return RValue::get(Builder.CreateCall(FnAssume, ArgValue)); |
| 2023 | } |
| 2024 | case Builtin::BI__builtin_bswap16: |
| 2025 | case Builtin::BI__builtin_bswap32: |
| 2026 | case Builtin::BI__builtin_bswap64: { |
| 2027 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap)); |
| 2028 | } |
| 2029 | case Builtin::BI__builtin_bitreverse8: |
| 2030 | case Builtin::BI__builtin_bitreverse16: |
| 2031 | case Builtin::BI__builtin_bitreverse32: |
| 2032 | case Builtin::BI__builtin_bitreverse64: { |
| 2033 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse)); |
| 2034 | } |
| 2035 | case Builtin::BI__builtin_rotateleft8: |
| 2036 | case Builtin::BI__builtin_rotateleft16: |
| 2037 | case Builtin::BI__builtin_rotateleft32: |
| 2038 | case Builtin::BI__builtin_rotateleft64: |
| 2039 | case Builtin::BI_rotl8: // Microsoft variants of rotate left |
| 2040 | case Builtin::BI_rotl16: |
| 2041 | case Builtin::BI_rotl: |
| 2042 | case Builtin::BI_lrotl: |
| 2043 | case Builtin::BI_rotl64: |
| 2044 | return emitRotate(E, false); |
| 2045 | |
| 2046 | case Builtin::BI__builtin_rotateright8: |
| 2047 | case Builtin::BI__builtin_rotateright16: |
| 2048 | case Builtin::BI__builtin_rotateright32: |
| 2049 | case Builtin::BI__builtin_rotateright64: |
| 2050 | case Builtin::BI_rotr8: // Microsoft variants of rotate right |
| 2051 | case Builtin::BI_rotr16: |
| 2052 | case Builtin::BI_rotr: |
| 2053 | case Builtin::BI_lrotr: |
| 2054 | case Builtin::BI_rotr64: |
| 2055 | return emitRotate(E, true); |
| 2056 | |
| 2057 | case Builtin::BI__builtin_constant_p: { |
| 2058 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2059 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
| 2060 | // At -O0, we don't perform inlining, so we don't need to delay the |
| 2061 | // processing. |
| 2062 | return RValue::get(ConstantInt::get(ResultType, 0)); |
| 2063 | |
| 2064 | const Expr *Arg = E->getArg(0); |
| 2065 | QualType ArgType = Arg->getType(); |
| 2066 | // FIXME: The allowance for Obj-C pointers and block pointers is historical |
| 2067 | // and likely a mistake. |
| 2068 | if (!ArgType->isIntegralOrEnumerationType() && !ArgType->isFloatingType() && |
| 2069 | !ArgType->isObjCObjectPointerType() && !ArgType->isBlockPointerType()) |
| 2070 | // Per the GCC documentation, only numeric constants are recognized after |
| 2071 | // inlining. |
| 2072 | return RValue::get(ConstantInt::get(ResultType, 0)); |
| 2073 | |
| 2074 | if (Arg->HasSideEffects(getContext())) |
| 2075 | // The argument is unevaluated, so be conservative if it might have |
| 2076 | // side-effects. |
| 2077 | return RValue::get(ConstantInt::get(ResultType, 0)); |
| 2078 | |
| 2079 | Value *ArgValue = EmitScalarExpr(Arg); |
| 2080 | if (ArgType->isObjCObjectPointerType()) { |
| 2081 | // Convert Objective-C objects to id because we cannot distinguish between |
| 2082 | // LLVM types for Obj-C classes as they are opaque. |
| 2083 | ArgType = CGM.getContext().getObjCIdType(); |
| 2084 | ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType)); |
| 2085 | } |
| 2086 | Function *F = |
| 2087 | CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType)); |
| 2088 | Value *Result = Builder.CreateCall(F, ArgValue); |
| 2089 | if (Result->getType() != ResultType) |
| 2090 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false); |
| 2091 | return RValue::get(Result); |
| 2092 | } |
| 2093 | case Builtin::BI__builtin_dynamic_object_size: |
| 2094 | case Builtin::BI__builtin_object_size: { |
| 2095 | unsigned Type = |
| 2096 | E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
| 2097 | auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType())); |
| 2098 | |
| 2099 | // We pass this builtin onto the optimizer so that it can figure out the |
| 2100 | // object size in more complex cases. |
| 2101 | bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size; |
| 2102 | return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType, |
| 2103 | /*EmittedE=*/nullptr, IsDynamic)); |
| 2104 | } |
| 2105 | case Builtin::BI__builtin_prefetch: { |
| 2106 | Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0)); |
| 2107 | // FIXME: Technically these constants should of type 'int', yes? |
| 2108 | RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) : |
| 2109 | llvm::ConstantInt::get(Int32Ty, 0); |
| 2110 | Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : |
| 2111 | llvm::ConstantInt::get(Int32Ty, 3); |
| 2112 | Value *Data = llvm::ConstantInt::get(Int32Ty, 1); |
| 2113 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch); |
| 2114 | return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data})); |
| 2115 | } |
| 2116 | case Builtin::BI__builtin_readcyclecounter: { |
| 2117 | Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter); |
| 2118 | return RValue::get(Builder.CreateCall(F)); |
| 2119 | } |
| 2120 | case Builtin::BI__builtin___clear_cache: { |
| 2121 | Value *Begin = EmitScalarExpr(E->getArg(0)); |
| 2122 | Value *End = EmitScalarExpr(E->getArg(1)); |
| 2123 | Function *F = CGM.getIntrinsic(Intrinsic::clear_cache); |
| 2124 | return RValue::get(Builder.CreateCall(F, {Begin, End})); |
| 2125 | } |
| 2126 | case Builtin::BI__builtin_trap: |
| 2127 | return RValue::get(EmitTrapCall(Intrinsic::trap)); |
| 2128 | case Builtin::BI__debugbreak: |
| 2129 | return RValue::get(EmitTrapCall(Intrinsic::debugtrap)); |
| 2130 | case Builtin::BI__builtin_unreachable: { |
| 2131 | EmitUnreachable(E->getExprLoc()); |
| 2132 | |
| 2133 | // We do need to preserve an insertion point. |
| 2134 | EmitBlock(createBasicBlock("unreachable.cont")); |
| 2135 | |
| 2136 | return RValue::get(nullptr); |
| 2137 | } |
| 2138 | |
| 2139 | case Builtin::BI__builtin_powi: |
| 2140 | case Builtin::BI__builtin_powif: |
| 2141 | case Builtin::BI__builtin_powil: { |
| 2142 | Value *Base = EmitScalarExpr(E->getArg(0)); |
| 2143 | Value *Exponent = EmitScalarExpr(E->getArg(1)); |
| 2144 | llvm::Type *ArgType = Base->getType(); |
| 2145 | Function *F = CGM.getIntrinsic(Intrinsic::powi, ArgType); |
| 2146 | return RValue::get(Builder.CreateCall(F, {Base, Exponent})); |
| 2147 | } |
| 2148 | |
| 2149 | case Builtin::BI__builtin_isgreater: |
| 2150 | case Builtin::BI__builtin_isgreaterequal: |
| 2151 | case Builtin::BI__builtin_isless: |
| 2152 | case Builtin::BI__builtin_islessequal: |
| 2153 | case Builtin::BI__builtin_islessgreater: |
| 2154 | case Builtin::BI__builtin_isunordered: { |
| 2155 | // Ordered comparisons: we know the arguments to these are matching scalar |
| 2156 | // floating point values. |
| 2157 | Value *LHS = EmitScalarExpr(E->getArg(0)); |
| 2158 | Value *RHS = EmitScalarExpr(E->getArg(1)); |
| 2159 | |
| 2160 | switch (BuiltinID) { |
| 2161 | default: llvm_unreachable("Unknown ordered comparison")::llvm::llvm_unreachable_internal("Unknown ordered comparison" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2161); |
| 2162 | case Builtin::BI__builtin_isgreater: |
| 2163 | LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp"); |
| 2164 | break; |
| 2165 | case Builtin::BI__builtin_isgreaterequal: |
| 2166 | LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp"); |
| 2167 | break; |
| 2168 | case Builtin::BI__builtin_isless: |
| 2169 | LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp"); |
| 2170 | break; |
| 2171 | case Builtin::BI__builtin_islessequal: |
| 2172 | LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp"); |
| 2173 | break; |
| 2174 | case Builtin::BI__builtin_islessgreater: |
| 2175 | LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp"); |
| 2176 | break; |
| 2177 | case Builtin::BI__builtin_isunordered: |
| 2178 | LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp"); |
| 2179 | break; |
| 2180 | } |
| 2181 | // ZExt bool to int type. |
| 2182 | return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType()))); |
| 2183 | } |
| 2184 | case Builtin::BI__builtin_isnan: { |
| 2185 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 2186 | V = Builder.CreateFCmpUNO(V, V, "cmp"); |
| 2187 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
| 2188 | } |
| 2189 | |
| 2190 | case Builtin::BIfinite: |
| 2191 | case Builtin::BI__finite: |
| 2192 | case Builtin::BIfinitef: |
| 2193 | case Builtin::BI__finitef: |
| 2194 | case Builtin::BIfinitel: |
| 2195 | case Builtin::BI__finitel: |
| 2196 | case Builtin::BI__builtin_isinf: |
| 2197 | case Builtin::BI__builtin_isfinite: { |
| 2198 | // isinf(x) --> fabs(x) == infinity |
| 2199 | // isfinite(x) --> fabs(x) != infinity |
| 2200 | // x != NaN via the ordered compare in either case. |
| 2201 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 2202 | Value *Fabs = EmitFAbs(*this, V); |
| 2203 | Constant *Infinity = ConstantFP::getInfinity(V->getType()); |
| 2204 | CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf) |
| 2205 | ? CmpInst::FCMP_OEQ |
| 2206 | : CmpInst::FCMP_ONE; |
| 2207 | Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf"); |
| 2208 | return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType()))); |
| 2209 | } |
| 2210 | |
| 2211 | case Builtin::BI__builtin_isinf_sign: { |
| 2212 | // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0 |
| 2213 | Value *Arg = EmitScalarExpr(E->getArg(0)); |
| 2214 | Value *AbsArg = EmitFAbs(*this, Arg); |
| 2215 | Value *IsInf = Builder.CreateFCmpOEQ( |
| 2216 | AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf"); |
| 2217 | Value *IsNeg = EmitSignBit(*this, Arg); |
| 2218 | |
| 2219 | llvm::Type *IntTy = ConvertType(E->getType()); |
| 2220 | Value *Zero = Constant::getNullValue(IntTy); |
| 2221 | Value *One = ConstantInt::get(IntTy, 1); |
| 2222 | Value *NegativeOne = ConstantInt::get(IntTy, -1); |
| 2223 | Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One); |
| 2224 | Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero); |
| 2225 | return RValue::get(Result); |
| 2226 | } |
| 2227 | |
| 2228 | case Builtin::BI__builtin_isnormal: { |
| 2229 | // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min |
| 2230 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 2231 | Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq"); |
| 2232 | |
| 2233 | Value *Abs = EmitFAbs(*this, V); |
| 2234 | Value *IsLessThanInf = |
| 2235 | Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf"); |
| 2236 | APFloat Smallest = APFloat::getSmallestNormalized( |
| 2237 | getContext().getFloatTypeSemantics(E->getArg(0)->getType())); |
| 2238 | Value *IsNormal = |
| 2239 | Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest), |
| 2240 | "isnormal"); |
| 2241 | V = Builder.CreateAnd(Eq, IsLessThanInf, "and"); |
| 2242 | V = Builder.CreateAnd(V, IsNormal, "and"); |
| 2243 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
| 2244 | } |
| 2245 | |
| 2246 | case Builtin::BI__builtin_flt_rounds: { |
| 2247 | Function *F = CGM.getIntrinsic(Intrinsic::flt_rounds); |
| 2248 | |
| 2249 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2250 | Value *Result = Builder.CreateCall(F); |
| 2251 | if (Result->getType() != ResultType) |
| 2252 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2253 | "cast"); |
| 2254 | return RValue::get(Result); |
| 2255 | } |
| 2256 | |
| 2257 | case Builtin::BI__builtin_fpclassify: { |
| 2258 | Value *V = EmitScalarExpr(E->getArg(5)); |
| 2259 | llvm::Type *Ty = ConvertType(E->getArg(5)->getType()); |
| 2260 | |
| 2261 | // Create Result |
| 2262 | BasicBlock *Begin = Builder.GetInsertBlock(); |
| 2263 | BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn); |
| 2264 | Builder.SetInsertPoint(End); |
| 2265 | PHINode *Result = |
| 2266 | Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4, |
| 2267 | "fpclassify_result"); |
| 2268 | |
| 2269 | // if (V==0) return FP_ZERO |
| 2270 | Builder.SetInsertPoint(Begin); |
| 2271 | Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty), |
| 2272 | "iszero"); |
| 2273 | Value *ZeroLiteral = EmitScalarExpr(E->getArg(4)); |
| 2274 | BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn); |
| 2275 | Builder.CreateCondBr(IsZero, End, NotZero); |
| 2276 | Result->addIncoming(ZeroLiteral, Begin); |
| 2277 | |
| 2278 | // if (V != V) return FP_NAN |
| 2279 | Builder.SetInsertPoint(NotZero); |
| 2280 | Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp"); |
| 2281 | Value *NanLiteral = EmitScalarExpr(E->getArg(0)); |
| 2282 | BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn); |
| 2283 | Builder.CreateCondBr(IsNan, End, NotNan); |
| 2284 | Result->addIncoming(NanLiteral, NotZero); |
| 2285 | |
| 2286 | // if (fabs(V) == infinity) return FP_INFINITY |
| 2287 | Builder.SetInsertPoint(NotNan); |
| 2288 | Value *VAbs = EmitFAbs(*this, V); |
| 2289 | Value *IsInf = |
| 2290 | Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()), |
| 2291 | "isinf"); |
| 2292 | Value *InfLiteral = EmitScalarExpr(E->getArg(1)); |
| 2293 | BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn); |
| 2294 | Builder.CreateCondBr(IsInf, End, NotInf); |
| 2295 | Result->addIncoming(InfLiteral, NotNan); |
| 2296 | |
| 2297 | // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL |
| 2298 | Builder.SetInsertPoint(NotInf); |
| 2299 | APFloat Smallest = APFloat::getSmallestNormalized( |
| 2300 | getContext().getFloatTypeSemantics(E->getArg(5)->getType())); |
| 2301 | Value *IsNormal = |
| 2302 | Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest), |
| 2303 | "isnormal"); |
| 2304 | Value *NormalResult = |
| 2305 | Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)), |
| 2306 | EmitScalarExpr(E->getArg(3))); |
| 2307 | Builder.CreateBr(End); |
| 2308 | Result->addIncoming(NormalResult, NotInf); |
| 2309 | |
| 2310 | // return Result |
| 2311 | Builder.SetInsertPoint(End); |
| 2312 | return RValue::get(Result); |
| 2313 | } |
| 2314 | |
| 2315 | case Builtin::BIalloca: |
| 2316 | case Builtin::BI_alloca: |
| 2317 | case Builtin::BI__builtin_alloca: { |
| 2318 | Value *Size = EmitScalarExpr(E->getArg(0)); |
| 2319 | const TargetInfo &TI = getContext().getTargetInfo(); |
| 2320 | // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__. |
| 2321 | unsigned SuitableAlignmentInBytes = |
| 2322 | CGM.getContext() |
| 2323 | .toCharUnitsFromBits(TI.getSuitableAlign()) |
| 2324 | .getQuantity(); |
| 2325 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
| 2326 | AI->setAlignment(SuitableAlignmentInBytes); |
| 2327 | initializeAlloca(*this, AI, Size, SuitableAlignmentInBytes); |
| 2328 | return RValue::get(AI); |
| 2329 | } |
| 2330 | |
| 2331 | case Builtin::BI__builtin_alloca_with_align: { |
| 2332 | Value *Size = EmitScalarExpr(E->getArg(0)); |
| 2333 | Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1)); |
| 2334 | auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue); |
| 2335 | unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue(); |
| 2336 | unsigned AlignmentInBytes = |
| 2337 | CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getQuantity(); |
| 2338 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
| 2339 | AI->setAlignment(AlignmentInBytes); |
| 2340 | initializeAlloca(*this, AI, Size, AlignmentInBytes); |
| 2341 | return RValue::get(AI); |
| 2342 | } |
| 2343 | |
| 2344 | case Builtin::BIbzero: |
| 2345 | case Builtin::BI__builtin_bzero: { |
| 2346 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 2347 | Value *SizeVal = EmitScalarExpr(E->getArg(1)); |
| 2348 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 2349 | E->getArg(0)->getExprLoc(), FD, 0); |
| 2350 | Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false); |
| 2351 | return RValue::get(nullptr); |
| 2352 | } |
| 2353 | case Builtin::BImemcpy: |
| 2354 | case Builtin::BI__builtin_memcpy: { |
| 2355 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 2356 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 2357 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 2358 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 2359 | E->getArg(0)->getExprLoc(), FD, 0); |
| 2360 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 2361 | E->getArg(1)->getExprLoc(), FD, 1); |
| 2362 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
| 2363 | return RValue::get(Dest.getPointer()); |
| 2364 | } |
| 2365 | |
| 2366 | case Builtin::BI__builtin_char_memchr: |
| 2367 | BuiltinID = Builtin::BI__builtin_memchr; |
| 2368 | break; |
| 2369 | |
| 2370 | case Builtin::BI__builtin___memcpy_chk: { |
| 2371 | // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2. |
| 2372 | Expr::EvalResult SizeResult, DstSizeResult; |
| 2373 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 2374 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 2375 | break; |
| 2376 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 2377 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 2378 | if (Size.ugt(DstSize)) |
| 2379 | break; |
| 2380 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 2381 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 2382 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 2383 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
| 2384 | return RValue::get(Dest.getPointer()); |
| 2385 | } |
| 2386 | |
| 2387 | case Builtin::BI__builtin_objc_memmove_collectable: { |
| 2388 | Address DestAddr = EmitPointerWithAlignment(E->getArg(0)); |
| 2389 | Address SrcAddr = EmitPointerWithAlignment(E->getArg(1)); |
| 2390 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 2391 | CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, |
| 2392 | DestAddr, SrcAddr, SizeVal); |
| 2393 | return RValue::get(DestAddr.getPointer()); |
| 2394 | } |
| 2395 | |
| 2396 | case Builtin::BI__builtin___memmove_chk: { |
| 2397 | // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2. |
| 2398 | Expr::EvalResult SizeResult, DstSizeResult; |
| 2399 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 2400 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 2401 | break; |
| 2402 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 2403 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 2404 | if (Size.ugt(DstSize)) |
| 2405 | break; |
| 2406 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 2407 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 2408 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 2409 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
| 2410 | return RValue::get(Dest.getPointer()); |
| 2411 | } |
| 2412 | |
| 2413 | case Builtin::BImemmove: |
| 2414 | case Builtin::BI__builtin_memmove: { |
| 2415 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 2416 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 2417 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 2418 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 2419 | E->getArg(0)->getExprLoc(), FD, 0); |
| 2420 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 2421 | E->getArg(1)->getExprLoc(), FD, 1); |
| 2422 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
| 2423 | return RValue::get(Dest.getPointer()); |
| 2424 | } |
| 2425 | case Builtin::BImemset: |
| 2426 | case Builtin::BI__builtin_memset: { |
| 2427 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 2428 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
| 2429 | Builder.getInt8Ty()); |
| 2430 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 2431 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 2432 | E->getArg(0)->getExprLoc(), FD, 0); |
| 2433 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
| 2434 | return RValue::get(Dest.getPointer()); |
| 2435 | } |
| 2436 | case Builtin::BI__builtin___memset_chk: { |
| 2437 | // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2. |
| 2438 | Expr::EvalResult SizeResult, DstSizeResult; |
| 2439 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 2440 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 2441 | break; |
| 2442 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 2443 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 2444 | if (Size.ugt(DstSize)) |
| 2445 | break; |
| 2446 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 2447 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
| 2448 | Builder.getInt8Ty()); |
| 2449 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 2450 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
| 2451 | return RValue::get(Dest.getPointer()); |
| 2452 | } |
| 2453 | case Builtin::BI__builtin_wmemcmp: { |
| 2454 | // The MSVC runtime library does not provide a definition of wmemcmp, so we |
| 2455 | // need an inline implementation. |
| 2456 | if (!getTarget().getTriple().isOSMSVCRT()) |
| 2457 | break; |
| 2458 | |
| 2459 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); |
| 2460 | |
| 2461 | Value *Dst = EmitScalarExpr(E->getArg(0)); |
| 2462 | Value *Src = EmitScalarExpr(E->getArg(1)); |
| 2463 | Value *Size = EmitScalarExpr(E->getArg(2)); |
| 2464 | |
| 2465 | BasicBlock *Entry = Builder.GetInsertBlock(); |
| 2466 | BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt"); |
| 2467 | BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt"); |
| 2468 | BasicBlock *Next = createBasicBlock("wmemcmp.next"); |
| 2469 | BasicBlock *Exit = createBasicBlock("wmemcmp.exit"); |
| 2470 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); |
| 2471 | Builder.CreateCondBr(SizeEq0, Exit, CmpGT); |
| 2472 | |
| 2473 | EmitBlock(CmpGT); |
| 2474 | PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2); |
| 2475 | DstPhi->addIncoming(Dst, Entry); |
| 2476 | PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2); |
| 2477 | SrcPhi->addIncoming(Src, Entry); |
| 2478 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); |
| 2479 | SizePhi->addIncoming(Size, Entry); |
| 2480 | CharUnits WCharAlign = |
| 2481 | getContext().getTypeAlignInChars(getContext().WCharTy); |
| 2482 | Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign); |
| 2483 | Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign); |
| 2484 | Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh); |
| 2485 | Builder.CreateCondBr(DstGtSrc, Exit, CmpLT); |
| 2486 | |
| 2487 | EmitBlock(CmpLT); |
| 2488 | Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh); |
| 2489 | Builder.CreateCondBr(DstLtSrc, Exit, Next); |
| 2490 | |
| 2491 | EmitBlock(Next); |
| 2492 | Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1); |
| 2493 | Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1); |
| 2494 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); |
| 2495 | Value *NextSizeEq0 = |
| 2496 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); |
| 2497 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT); |
| 2498 | DstPhi->addIncoming(NextDst, Next); |
| 2499 | SrcPhi->addIncoming(NextSrc, Next); |
| 2500 | SizePhi->addIncoming(NextSize, Next); |
| 2501 | |
| 2502 | EmitBlock(Exit); |
| 2503 | PHINode *Ret = Builder.CreatePHI(IntTy, 4); |
| 2504 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry); |
| 2505 | Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT); |
| 2506 | Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT); |
| 2507 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Next); |
| 2508 | return RValue::get(Ret); |
| 2509 | } |
| 2510 | case Builtin::BI__builtin_dwarf_cfa: { |
| 2511 | // The offset in bytes from the first argument to the CFA. |
| 2512 | // |
| 2513 | // Why on earth is this in the frontend? Is there any reason at |
| 2514 | // all that the backend can't reasonably determine this while |
| 2515 | // lowering llvm.eh.dwarf.cfa()? |
| 2516 | // |
| 2517 | // TODO: If there's a satisfactory reason, add a target hook for |
| 2518 | // this instead of hard-coding 0, which is correct for most targets. |
| 2519 | int32_t Offset = 0; |
| 2520 | |
| 2521 | Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa); |
| 2522 | return RValue::get(Builder.CreateCall(F, |
| 2523 | llvm::ConstantInt::get(Int32Ty, Offset))); |
| 2524 | } |
| 2525 | case Builtin::BI__builtin_return_address: { |
| 2526 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
| 2527 | getContext().UnsignedIntTy); |
| 2528 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
| 2529 | return RValue::get(Builder.CreateCall(F, Depth)); |
| 2530 | } |
| 2531 | case Builtin::BI_ReturnAddress: { |
| 2532 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
| 2533 | return RValue::get(Builder.CreateCall(F, Builder.getInt32(0))); |
| 2534 | } |
| 2535 | case Builtin::BI__builtin_frame_address: { |
| 2536 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
| 2537 | getContext().UnsignedIntTy); |
| 2538 | Function *F = CGM.getIntrinsic(Intrinsic::frameaddress); |
| 2539 | return RValue::get(Builder.CreateCall(F, Depth)); |
| 2540 | } |
| 2541 | case Builtin::BI__builtin_extract_return_addr: { |
| 2542 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 2543 | Value *Result = getTargetHooks().decodeReturnAddress(*this, Address); |
| 2544 | return RValue::get(Result); |
| 2545 | } |
| 2546 | case Builtin::BI__builtin_frob_return_addr: { |
| 2547 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 2548 | Value *Result = getTargetHooks().encodeReturnAddress(*this, Address); |
| 2549 | return RValue::get(Result); |
| 2550 | } |
| 2551 | case Builtin::BI__builtin_dwarf_sp_column: { |
| 2552 | llvm::IntegerType *Ty |
| 2553 | = cast<llvm::IntegerType>(ConvertType(E->getType())); |
| 2554 | int Column = getTargetHooks().getDwarfEHStackPointer(CGM); |
| 2555 | if (Column == -1) { |
| 2556 | CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column"); |
| 2557 | return RValue::get(llvm::UndefValue::get(Ty)); |
| 2558 | } |
| 2559 | return RValue::get(llvm::ConstantInt::get(Ty, Column, true)); |
| 2560 | } |
| 2561 | case Builtin::BI__builtin_init_dwarf_reg_size_table: { |
| 2562 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 2563 | if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address)) |
| 2564 | CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table"); |
| 2565 | return RValue::get(llvm::UndefValue::get(ConvertType(E->getType()))); |
| 2566 | } |
| 2567 | case Builtin::BI__builtin_eh_return: { |
| 2568 | Value *Int = EmitScalarExpr(E->getArg(0)); |
| 2569 | Value *Ptr = EmitScalarExpr(E->getArg(1)); |
| 2570 | |
| 2571 | llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType()); |
| 2572 | assert((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) &&(((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? static_cast<void> (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2573, __PRETTY_FUNCTION__)) |
| 2573 | "LLVM's __builtin_eh_return only supports 32- and 64-bit variants")(((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? static_cast<void> (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2573, __PRETTY_FUNCTION__)); |
| 2574 | Function *F = |
| 2575 | CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32 |
| 2576 | : Intrinsic::eh_return_i64); |
| 2577 | Builder.CreateCall(F, {Int, Ptr}); |
| 2578 | Builder.CreateUnreachable(); |
| 2579 | |
| 2580 | // We do need to preserve an insertion point. |
| 2581 | EmitBlock(createBasicBlock("builtin_eh_return.cont")); |
| 2582 | |
| 2583 | return RValue::get(nullptr); |
| 2584 | } |
| 2585 | case Builtin::BI__builtin_unwind_init: { |
| 2586 | Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init); |
| 2587 | return RValue::get(Builder.CreateCall(F)); |
| 2588 | } |
| 2589 | case Builtin::BI__builtin_extend_pointer: { |
| 2590 | // Extends a pointer to the size of an _Unwind_Word, which is |
| 2591 | // uint64_t on all platforms. Generally this gets poked into a |
| 2592 | // register and eventually used as an address, so if the |
| 2593 | // addressing registers are wider than pointers and the platform |
| 2594 | // doesn't implicitly ignore high-order bits when doing |
| 2595 | // addressing, we need to make sure we zext / sext based on |
| 2596 | // the platform's expectations. |
| 2597 | // |
| 2598 | // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html |
| 2599 | |
| 2600 | // Cast the pointer to intptr_t. |
| 2601 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 2602 | Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast"); |
| 2603 | |
| 2604 | // If that's 64 bits, we're done. |
| 2605 | if (IntPtrTy->getBitWidth() == 64) |
| 2606 | return RValue::get(Result); |
| 2607 | |
| 2608 | // Otherwise, ask the codegen data what to do. |
| 2609 | if (getTargetHooks().extendPointerWithSExt()) |
| 2610 | return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext")); |
| 2611 | else |
| 2612 | return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext")); |
| 2613 | } |
| 2614 | case Builtin::BI__builtin_setjmp: { |
| 2615 | // Buffer is a void**. |
| 2616 | Address Buf = EmitPointerWithAlignment(E->getArg(0)); |
| 2617 | |
| 2618 | // Store the frame pointer to the setjmp buffer. |
| 2619 | Value *FrameAddr = |
| 2620 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::frameaddress), |
| 2621 | ConstantInt::get(Int32Ty, 0)); |
| 2622 | Builder.CreateStore(FrameAddr, Buf); |
| 2623 | |
| 2624 | // Store the stack pointer to the setjmp buffer. |
| 2625 | Value *StackAddr = |
| 2626 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave)); |
| 2627 | Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2); |
| 2628 | Builder.CreateStore(StackAddr, StackSaveSlot); |
| 2629 | |
| 2630 | // Call LLVM's EH setjmp, which is lightweight. |
| 2631 | Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp); |
| 2632 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); |
| 2633 | return RValue::get(Builder.CreateCall(F, Buf.getPointer())); |
| 2634 | } |
| 2635 | case Builtin::BI__builtin_longjmp: { |
| 2636 | Value *Buf = EmitScalarExpr(E->getArg(0)); |
| 2637 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); |
| 2638 | |
| 2639 | // Call LLVM's EH longjmp, which is lightweight. |
| 2640 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf); |
| 2641 | |
| 2642 | // longjmp doesn't return; mark this as unreachable. |
| 2643 | Builder.CreateUnreachable(); |
| 2644 | |
| 2645 | // We do need to preserve an insertion point. |
| 2646 | EmitBlock(createBasicBlock("longjmp.cont")); |
| 2647 | |
| 2648 | return RValue::get(nullptr); |
| 2649 | } |
| 2650 | case Builtin::BI__builtin_launder: { |
| 2651 | const Expr *Arg = E->getArg(0); |
| 2652 | QualType ArgTy = Arg->getType()->getPointeeType(); |
| 2653 | Value *Ptr = EmitScalarExpr(Arg); |
| 2654 | if (TypeRequiresBuiltinLaunder(CGM, ArgTy)) |
| 2655 | Ptr = Builder.CreateLaunderInvariantGroup(Ptr); |
| 2656 | |
| 2657 | return RValue::get(Ptr); |
| 2658 | } |
| 2659 | case Builtin::BI__sync_fetch_and_add: |
| 2660 | case Builtin::BI__sync_fetch_and_sub: |
| 2661 | case Builtin::BI__sync_fetch_and_or: |
| 2662 | case Builtin::BI__sync_fetch_and_and: |
| 2663 | case Builtin::BI__sync_fetch_and_xor: |
| 2664 | case Builtin::BI__sync_fetch_and_nand: |
| 2665 | case Builtin::BI__sync_add_and_fetch: |
| 2666 | case Builtin::BI__sync_sub_and_fetch: |
| 2667 | case Builtin::BI__sync_and_and_fetch: |
| 2668 | case Builtin::BI__sync_or_and_fetch: |
| 2669 | case Builtin::BI__sync_xor_and_fetch: |
| 2670 | case Builtin::BI__sync_nand_and_fetch: |
| 2671 | case Builtin::BI__sync_val_compare_and_swap: |
| 2672 | case Builtin::BI__sync_bool_compare_and_swap: |
| 2673 | case Builtin::BI__sync_lock_test_and_set: |
| 2674 | case Builtin::BI__sync_lock_release: |
| 2675 | case Builtin::BI__sync_swap: |
| 2676 | llvm_unreachable("Shouldn't make it through sema")::llvm::llvm_unreachable_internal("Shouldn't make it through sema" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 2676); |
| 2677 | case Builtin::BI__sync_fetch_and_add_1: |
| 2678 | case Builtin::BI__sync_fetch_and_add_2: |
| 2679 | case Builtin::BI__sync_fetch_and_add_4: |
| 2680 | case Builtin::BI__sync_fetch_and_add_8: |
| 2681 | case Builtin::BI__sync_fetch_and_add_16: |
| 2682 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E); |
| 2683 | case Builtin::BI__sync_fetch_and_sub_1: |
| 2684 | case Builtin::BI__sync_fetch_and_sub_2: |
| 2685 | case Builtin::BI__sync_fetch_and_sub_4: |
| 2686 | case Builtin::BI__sync_fetch_and_sub_8: |
| 2687 | case Builtin::BI__sync_fetch_and_sub_16: |
| 2688 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E); |
| 2689 | case Builtin::BI__sync_fetch_and_or_1: |
| 2690 | case Builtin::BI__sync_fetch_and_or_2: |
| 2691 | case Builtin::BI__sync_fetch_and_or_4: |
| 2692 | case Builtin::BI__sync_fetch_and_or_8: |
| 2693 | case Builtin::BI__sync_fetch_and_or_16: |
| 2694 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E); |
| 2695 | case Builtin::BI__sync_fetch_and_and_1: |
| 2696 | case Builtin::BI__sync_fetch_and_and_2: |
| 2697 | case Builtin::BI__sync_fetch_and_and_4: |
| 2698 | case Builtin::BI__sync_fetch_and_and_8: |
| 2699 | case Builtin::BI__sync_fetch_and_and_16: |
| 2700 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E); |
| 2701 | case Builtin::BI__sync_fetch_and_xor_1: |
| 2702 | case Builtin::BI__sync_fetch_and_xor_2: |
| 2703 | case Builtin::BI__sync_fetch_and_xor_4: |
| 2704 | case Builtin::BI__sync_fetch_and_xor_8: |
| 2705 | case Builtin::BI__sync_fetch_and_xor_16: |
| 2706 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E); |
| 2707 | case Builtin::BI__sync_fetch_and_nand_1: |
| 2708 | case Builtin::BI__sync_fetch_and_nand_2: |
| 2709 | case Builtin::BI__sync_fetch_and_nand_4: |
| 2710 | case Builtin::BI__sync_fetch_and_nand_8: |
| 2711 | case Builtin::BI__sync_fetch_and_nand_16: |
| 2712 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E); |
| 2713 | |
| 2714 | // Clang extensions: not overloaded yet. |
| 2715 | case Builtin::BI__sync_fetch_and_min: |
| 2716 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E); |
| 2717 | case Builtin::BI__sync_fetch_and_max: |
| 2718 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E); |
| 2719 | case Builtin::BI__sync_fetch_and_umin: |
| 2720 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E); |
| 2721 | case Builtin::BI__sync_fetch_and_umax: |
| 2722 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E); |
| 2723 | |
| 2724 | case Builtin::BI__sync_add_and_fetch_1: |
| 2725 | case Builtin::BI__sync_add_and_fetch_2: |
| 2726 | case Builtin::BI__sync_add_and_fetch_4: |
| 2727 | case Builtin::BI__sync_add_and_fetch_8: |
| 2728 | case Builtin::BI__sync_add_and_fetch_16: |
| 2729 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E, |
| 2730 | llvm::Instruction::Add); |
| 2731 | case Builtin::BI__sync_sub_and_fetch_1: |
| 2732 | case Builtin::BI__sync_sub_and_fetch_2: |
| 2733 | case Builtin::BI__sync_sub_and_fetch_4: |
| 2734 | case Builtin::BI__sync_sub_and_fetch_8: |
| 2735 | case Builtin::BI__sync_sub_and_fetch_16: |
| 2736 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E, |
| 2737 | llvm::Instruction::Sub); |
| 2738 | case Builtin::BI__sync_and_and_fetch_1: |
| 2739 | case Builtin::BI__sync_and_and_fetch_2: |
| 2740 | case Builtin::BI__sync_and_and_fetch_4: |
| 2741 | case Builtin::BI__sync_and_and_fetch_8: |
| 2742 | case Builtin::BI__sync_and_and_fetch_16: |
| 2743 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E, |
| 2744 | llvm::Instruction::And); |
| 2745 | case Builtin::BI__sync_or_and_fetch_1: |
| 2746 | case Builtin::BI__sync_or_and_fetch_2: |
| 2747 | case Builtin::BI__sync_or_and_fetch_4: |
| 2748 | case Builtin::BI__sync_or_and_fetch_8: |
| 2749 | case Builtin::BI__sync_or_and_fetch_16: |
| 2750 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E, |
| 2751 | llvm::Instruction::Or); |
| 2752 | case Builtin::BI__sync_xor_and_fetch_1: |
| 2753 | case Builtin::BI__sync_xor_and_fetch_2: |
| 2754 | case Builtin::BI__sync_xor_and_fetch_4: |
| 2755 | case Builtin::BI__sync_xor_and_fetch_8: |
| 2756 | case Builtin::BI__sync_xor_and_fetch_16: |
| 2757 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E, |
| 2758 | llvm::Instruction::Xor); |
| 2759 | case Builtin::BI__sync_nand_and_fetch_1: |
| 2760 | case Builtin::BI__sync_nand_and_fetch_2: |
| 2761 | case Builtin::BI__sync_nand_and_fetch_4: |
| 2762 | case Builtin::BI__sync_nand_and_fetch_8: |
| 2763 | case Builtin::BI__sync_nand_and_fetch_16: |
| 2764 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E, |
| 2765 | llvm::Instruction::And, true); |
| 2766 | |
| 2767 | case Builtin::BI__sync_val_compare_and_swap_1: |
| 2768 | case Builtin::BI__sync_val_compare_and_swap_2: |
| 2769 | case Builtin::BI__sync_val_compare_and_swap_4: |
| 2770 | case Builtin::BI__sync_val_compare_and_swap_8: |
| 2771 | case Builtin::BI__sync_val_compare_and_swap_16: |
| 2772 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, false)); |
| 2773 | |
| 2774 | case Builtin::BI__sync_bool_compare_and_swap_1: |
| 2775 | case Builtin::BI__sync_bool_compare_and_swap_2: |
| 2776 | case Builtin::BI__sync_bool_compare_and_swap_4: |
| 2777 | case Builtin::BI__sync_bool_compare_and_swap_8: |
| 2778 | case Builtin::BI__sync_bool_compare_and_swap_16: |
| 2779 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, true)); |
| 2780 | |
| 2781 | case Builtin::BI__sync_swap_1: |
| 2782 | case Builtin::BI__sync_swap_2: |
| 2783 | case Builtin::BI__sync_swap_4: |
| 2784 | case Builtin::BI__sync_swap_8: |
| 2785 | case Builtin::BI__sync_swap_16: |
| 2786 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
| 2787 | |
| 2788 | case Builtin::BI__sync_lock_test_and_set_1: |
| 2789 | case Builtin::BI__sync_lock_test_and_set_2: |
| 2790 | case Builtin::BI__sync_lock_test_and_set_4: |
| 2791 | case Builtin::BI__sync_lock_test_and_set_8: |
| 2792 | case Builtin::BI__sync_lock_test_and_set_16: |
| 2793 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
| 2794 | |
| 2795 | case Builtin::BI__sync_lock_release_1: |
| 2796 | case Builtin::BI__sync_lock_release_2: |
| 2797 | case Builtin::BI__sync_lock_release_4: |
| 2798 | case Builtin::BI__sync_lock_release_8: |
| 2799 | case Builtin::BI__sync_lock_release_16: { |
| 2800 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 2801 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 2802 | CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy); |
| 2803 | llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(), |
| 2804 | StoreSize.getQuantity() * 8); |
| 2805 | Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 2806 | llvm::StoreInst *Store = |
| 2807 | Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr, |
| 2808 | StoreSize); |
| 2809 | Store->setAtomic(llvm::AtomicOrdering::Release); |
| 2810 | return RValue::get(nullptr); |
| 2811 | } |
| 2812 | |
| 2813 | case Builtin::BI__sync_synchronize: { |
| 2814 | // We assume this is supposed to correspond to a C++0x-style |
| 2815 | // sequentially-consistent fence (i.e. this is only usable for |
| 2816 | // synchronization, not device I/O or anything like that). This intrinsic |
| 2817 | // is really badly designed in the sense that in theory, there isn't |
| 2818 | // any way to safely use it... but in practice, it mostly works |
| 2819 | // to use it with non-atomic loads and stores to get acquire/release |
| 2820 | // semantics. |
| 2821 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent); |
| 2822 | return RValue::get(nullptr); |
| 2823 | } |
| 2824 | |
| 2825 | case Builtin::BI__builtin_nontemporal_load: |
| 2826 | return RValue::get(EmitNontemporalLoad(*this, E)); |
| 2827 | case Builtin::BI__builtin_nontemporal_store: |
| 2828 | return RValue::get(EmitNontemporalStore(*this, E)); |
| 2829 | case Builtin::BI__c11_atomic_is_lock_free: |
| 2830 | case Builtin::BI__atomic_is_lock_free: { |
| 2831 | // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the |
| 2832 | // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since |
| 2833 | // _Atomic(T) is always properly-aligned. |
| 2834 | const char *LibCallName = "__atomic_is_lock_free"; |
| 2835 | CallArgList Args; |
| 2836 | Args.add(RValue::get(EmitScalarExpr(E->getArg(0))), |
| 2837 | getContext().getSizeType()); |
| 2838 | if (BuiltinID == Builtin::BI__atomic_is_lock_free) |
| 2839 | Args.add(RValue::get(EmitScalarExpr(E->getArg(1))), |
| 2840 | getContext().VoidPtrTy); |
| 2841 | else |
| 2842 | Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)), |
| 2843 | getContext().VoidPtrTy); |
| 2844 | const CGFunctionInfo &FuncInfo = |
| 2845 | CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args); |
| 2846 | llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo); |
| 2847 | llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName); |
| 2848 | return EmitCall(FuncInfo, CGCallee::forDirect(Func), |
| 2849 | ReturnValueSlot(), Args); |
| 2850 | } |
| 2851 | |
| 2852 | case Builtin::BI__atomic_test_and_set: { |
| 2853 | // Look at the argument type to determine whether this is a volatile |
| 2854 | // operation. The parameter type is always volatile. |
| 2855 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
| 2856 | bool Volatile = |
| 2857 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
| 2858 | |
| 2859 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 2860 | unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace(); |
| 2861 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); |
| 2862 | Value *NewVal = Builder.getInt8(1); |
| 2863 | Value *Order = EmitScalarExpr(E->getArg(1)); |
| 2864 | if (isa<llvm::ConstantInt>(Order)) { |
| 2865 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 2866 | AtomicRMWInst *Result = nullptr; |
| 2867 | switch (ord) { |
| 2868 | case 0: // memory_order_relaxed |
| 2869 | default: // invalid order |
| 2870 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 2871 | llvm::AtomicOrdering::Monotonic); |
| 2872 | break; |
| 2873 | case 1: // memory_order_consume |
| 2874 | case 2: // memory_order_acquire |
| 2875 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 2876 | llvm::AtomicOrdering::Acquire); |
| 2877 | break; |
| 2878 | case 3: // memory_order_release |
| 2879 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 2880 | llvm::AtomicOrdering::Release); |
| 2881 | break; |
| 2882 | case 4: // memory_order_acq_rel |
| 2883 | |
| 2884 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 2885 | llvm::AtomicOrdering::AcquireRelease); |
| 2886 | break; |
| 2887 | case 5: // memory_order_seq_cst |
| 2888 | Result = Builder.CreateAtomicRMW( |
| 2889 | llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 2890 | llvm::AtomicOrdering::SequentiallyConsistent); |
| 2891 | break; |
| 2892 | } |
| 2893 | Result->setVolatile(Volatile); |
| 2894 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
| 2895 | } |
| 2896 | |
| 2897 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 2898 | |
| 2899 | llvm::BasicBlock *BBs[5] = { |
| 2900 | createBasicBlock("monotonic", CurFn), |
| 2901 | createBasicBlock("acquire", CurFn), |
| 2902 | createBasicBlock("release", CurFn), |
| 2903 | createBasicBlock("acqrel", CurFn), |
| 2904 | createBasicBlock("seqcst", CurFn) |
| 2905 | }; |
| 2906 | llvm::AtomicOrdering Orders[5] = { |
| 2907 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire, |
| 2908 | llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease, |
| 2909 | llvm::AtomicOrdering::SequentiallyConsistent}; |
| 2910 | |
| 2911 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 2912 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
| 2913 | |
| 2914 | Builder.SetInsertPoint(ContBB); |
| 2915 | PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set"); |
| 2916 | |
| 2917 | for (unsigned i = 0; i < 5; ++i) { |
| 2918 | Builder.SetInsertPoint(BBs[i]); |
| 2919 | AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, |
| 2920 | Ptr, NewVal, Orders[i]); |
| 2921 | RMW->setVolatile(Volatile); |
| 2922 | Result->addIncoming(RMW, BBs[i]); |
| 2923 | Builder.CreateBr(ContBB); |
| 2924 | } |
| 2925 | |
| 2926 | SI->addCase(Builder.getInt32(0), BBs[0]); |
| 2927 | SI->addCase(Builder.getInt32(1), BBs[1]); |
| 2928 | SI->addCase(Builder.getInt32(2), BBs[1]); |
| 2929 | SI->addCase(Builder.getInt32(3), BBs[2]); |
| 2930 | SI->addCase(Builder.getInt32(4), BBs[3]); |
| 2931 | SI->addCase(Builder.getInt32(5), BBs[4]); |
| 2932 | |
| 2933 | Builder.SetInsertPoint(ContBB); |
| 2934 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
| 2935 | } |
| 2936 | |
| 2937 | case Builtin::BI__atomic_clear: { |
| 2938 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
| 2939 | bool Volatile = |
| 2940 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
| 2941 | |
| 2942 | Address Ptr = EmitPointerWithAlignment(E->getArg(0)); |
| 2943 | unsigned AddrSpace = Ptr.getPointer()->getType()->getPointerAddressSpace(); |
| 2944 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); |
| 2945 | Value *NewVal = Builder.getInt8(0); |
| 2946 | Value *Order = EmitScalarExpr(E->getArg(1)); |
| 2947 | if (isa<llvm::ConstantInt>(Order)) { |
| 2948 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 2949 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
| 2950 | switch (ord) { |
| 2951 | case 0: // memory_order_relaxed |
| 2952 | default: // invalid order |
| 2953 | Store->setOrdering(llvm::AtomicOrdering::Monotonic); |
| 2954 | break; |
| 2955 | case 3: // memory_order_release |
| 2956 | Store->setOrdering(llvm::AtomicOrdering::Release); |
| 2957 | break; |
| 2958 | case 5: // memory_order_seq_cst |
| 2959 | Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent); |
| 2960 | break; |
| 2961 | } |
| 2962 | return RValue::get(nullptr); |
| 2963 | } |
| 2964 | |
| 2965 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 2966 | |
| 2967 | llvm::BasicBlock *BBs[3] = { |
| 2968 | createBasicBlock("monotonic", CurFn), |
| 2969 | createBasicBlock("release", CurFn), |
| 2970 | createBasicBlock("seqcst", CurFn) |
| 2971 | }; |
| 2972 | llvm::AtomicOrdering Orders[3] = { |
| 2973 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release, |
| 2974 | llvm::AtomicOrdering::SequentiallyConsistent}; |
| 2975 | |
| 2976 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 2977 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
| 2978 | |
| 2979 | for (unsigned i = 0; i < 3; ++i) { |
| 2980 | Builder.SetInsertPoint(BBs[i]); |
| 2981 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
| 2982 | Store->setOrdering(Orders[i]); |
| 2983 | Builder.CreateBr(ContBB); |
| 2984 | } |
| 2985 | |
| 2986 | SI->addCase(Builder.getInt32(0), BBs[0]); |
| 2987 | SI->addCase(Builder.getInt32(3), BBs[1]); |
| 2988 | SI->addCase(Builder.getInt32(5), BBs[2]); |
| 2989 | |
| 2990 | Builder.SetInsertPoint(ContBB); |
| 2991 | return RValue::get(nullptr); |
| 2992 | } |
| 2993 | |
| 2994 | case Builtin::BI__atomic_thread_fence: |
| 2995 | case Builtin::BI__atomic_signal_fence: |
| 2996 | case Builtin::BI__c11_atomic_thread_fence: |
| 2997 | case Builtin::BI__c11_atomic_signal_fence: { |
| 2998 | llvm::SyncScope::ID SSID; |
| 2999 | if (BuiltinID == Builtin::BI__atomic_signal_fence || |
| 3000 | BuiltinID == Builtin::BI__c11_atomic_signal_fence) |
| 3001 | SSID = llvm::SyncScope::SingleThread; |
| 3002 | else |
| 3003 | SSID = llvm::SyncScope::System; |
| 3004 | Value *Order = EmitScalarExpr(E->getArg(0)); |
| 3005 | if (isa<llvm::ConstantInt>(Order)) { |
| 3006 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 3007 | switch (ord) { |
| 3008 | case 0: // memory_order_relaxed |
| 3009 | default: // invalid order |
| 3010 | break; |
| 3011 | case 1: // memory_order_consume |
| 3012 | case 2: // memory_order_acquire |
| 3013 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
| 3014 | break; |
| 3015 | case 3: // memory_order_release |
| 3016 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
| 3017 | break; |
| 3018 | case 4: // memory_order_acq_rel |
| 3019 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
| 3020 | break; |
| 3021 | case 5: // memory_order_seq_cst |
| 3022 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
| 3023 | break; |
| 3024 | } |
| 3025 | return RValue::get(nullptr); |
| 3026 | } |
| 3027 | |
| 3028 | llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB; |
| 3029 | AcquireBB = createBasicBlock("acquire", CurFn); |
| 3030 | ReleaseBB = createBasicBlock("release", CurFn); |
| 3031 | AcqRelBB = createBasicBlock("acqrel", CurFn); |
| 3032 | SeqCstBB = createBasicBlock("seqcst", CurFn); |
| 3033 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 3034 | |
| 3035 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 3036 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB); |
| 3037 | |
| 3038 | Builder.SetInsertPoint(AcquireBB); |
| 3039 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
| 3040 | Builder.CreateBr(ContBB); |
| 3041 | SI->addCase(Builder.getInt32(1), AcquireBB); |
| 3042 | SI->addCase(Builder.getInt32(2), AcquireBB); |
| 3043 | |
| 3044 | Builder.SetInsertPoint(ReleaseBB); |
| 3045 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
| 3046 | Builder.CreateBr(ContBB); |
| 3047 | SI->addCase(Builder.getInt32(3), ReleaseBB); |
| 3048 | |
| 3049 | Builder.SetInsertPoint(AcqRelBB); |
| 3050 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
| 3051 | Builder.CreateBr(ContBB); |
| 3052 | SI->addCase(Builder.getInt32(4), AcqRelBB); |
| 3053 | |
| 3054 | Builder.SetInsertPoint(SeqCstBB); |
| 3055 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
| 3056 | Builder.CreateBr(ContBB); |
| 3057 | SI->addCase(Builder.getInt32(5), SeqCstBB); |
| 3058 | |
| 3059 | Builder.SetInsertPoint(ContBB); |
| 3060 | return RValue::get(nullptr); |
| 3061 | } |
| 3062 | |
| 3063 | case Builtin::BI__builtin_signbit: |
| 3064 | case Builtin::BI__builtin_signbitf: |
| 3065 | case Builtin::BI__builtin_signbitl: { |
| 3066 | return RValue::get( |
| 3067 | Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))), |
| 3068 | ConvertType(E->getType()))); |
| 3069 | } |
| 3070 | case Builtin::BI__annotation: { |
| 3071 | // Re-encode each wide string to UTF8 and make an MDString. |
| 3072 | SmallVector<Metadata *, 1> Strings; |
| 3073 | for (const Expr *Arg : E->arguments()) { |
| 3074 | const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts()); |
| 3075 | assert(Str->getCharByteWidth() == 2)((Str->getCharByteWidth() == 2) ? static_cast<void> ( 0) : __assert_fail ("Str->getCharByteWidth() == 2", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3075, __PRETTY_FUNCTION__)); |
| 3076 | StringRef WideBytes = Str->getBytes(); |
| 3077 | std::string StrUtf8; |
| 3078 | if (!convertUTF16ToUTF8String( |
| 3079 | makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) { |
| 3080 | CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument"); |
| 3081 | continue; |
| 3082 | } |
| 3083 | Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8)); |
| 3084 | } |
| 3085 | |
| 3086 | // Build and MDTuple of MDStrings and emit the intrinsic call. |
| 3087 | llvm::Function *F = |
| 3088 | CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {}); |
| 3089 | MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings); |
| 3090 | Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple)); |
| 3091 | return RValue::getIgnored(); |
| 3092 | } |
| 3093 | case Builtin::BI__builtin_annotation: { |
| 3094 | llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0)); |
| 3095 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::annotation, |
| 3096 | AnnVal->getType()); |
| 3097 | |
| 3098 | // Get the annotation string, go through casts. Sema requires this to be a |
| 3099 | // non-wide string literal, potentially casted, so the cast<> is safe. |
| 3100 | const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts(); |
| 3101 | StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString(); |
| 3102 | return RValue::get(EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc())); |
| 3103 | } |
| 3104 | case Builtin::BI__builtin_addcb: |
| 3105 | case Builtin::BI__builtin_addcs: |
| 3106 | case Builtin::BI__builtin_addc: |
| 3107 | case Builtin::BI__builtin_addcl: |
| 3108 | case Builtin::BI__builtin_addcll: |
| 3109 | case Builtin::BI__builtin_subcb: |
| 3110 | case Builtin::BI__builtin_subcs: |
| 3111 | case Builtin::BI__builtin_subc: |
| 3112 | case Builtin::BI__builtin_subcl: |
| 3113 | case Builtin::BI__builtin_subcll: { |
| 3114 | |
| 3115 | // We translate all of these builtins from expressions of the form: |
| 3116 | // int x = ..., y = ..., carryin = ..., carryout, result; |
| 3117 | // result = __builtin_addc(x, y, carryin, &carryout); |
| 3118 | // |
| 3119 | // to LLVM IR of the form: |
| 3120 | // |
| 3121 | // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y) |
| 3122 | // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0 |
| 3123 | // %carry1 = extractvalue {i32, i1} %tmp1, 1 |
| 3124 | // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1, |
| 3125 | // i32 %carryin) |
| 3126 | // %result = extractvalue {i32, i1} %tmp2, 0 |
| 3127 | // %carry2 = extractvalue {i32, i1} %tmp2, 1 |
| 3128 | // %tmp3 = or i1 %carry1, %carry2 |
| 3129 | // %tmp4 = zext i1 %tmp3 to i32 |
| 3130 | // store i32 %tmp4, i32* %carryout |
| 3131 | |
| 3132 | // Scalarize our inputs. |
| 3133 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
| 3134 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
| 3135 | llvm::Value *Carryin = EmitScalarExpr(E->getArg(2)); |
| 3136 | Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3)); |
| 3137 | |
| 3138 | // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow. |
| 3139 | llvm::Intrinsic::ID IntrinsicId; |
| 3140 | switch (BuiltinID) { |
| 3141 | default: llvm_unreachable("Unknown multiprecision builtin id.")::llvm::llvm_unreachable_internal("Unknown multiprecision builtin id." , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3141); |
| 3142 | case Builtin::BI__builtin_addcb: |
| 3143 | case Builtin::BI__builtin_addcs: |
| 3144 | case Builtin::BI__builtin_addc: |
| 3145 | case Builtin::BI__builtin_addcl: |
| 3146 | case Builtin::BI__builtin_addcll: |
| 3147 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
| 3148 | break; |
| 3149 | case Builtin::BI__builtin_subcb: |
| 3150 | case Builtin::BI__builtin_subcs: |
| 3151 | case Builtin::BI__builtin_subc: |
| 3152 | case Builtin::BI__builtin_subcl: |
| 3153 | case Builtin::BI__builtin_subcll: |
| 3154 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
| 3155 | break; |
| 3156 | } |
| 3157 | |
| 3158 | // Construct our resulting LLVM IR expression. |
| 3159 | llvm::Value *Carry1; |
| 3160 | llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId, |
| 3161 | X, Y, Carry1); |
| 3162 | llvm::Value *Carry2; |
| 3163 | llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId, |
| 3164 | Sum1, Carryin, Carry2); |
| 3165 | llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2), |
| 3166 | X->getType()); |
| 3167 | Builder.CreateStore(CarryOut, CarryOutPtr); |
| 3168 | return RValue::get(Sum2); |
| 3169 | } |
| 3170 | |
| 3171 | case Builtin::BI__builtin_add_overflow: |
| 3172 | case Builtin::BI__builtin_sub_overflow: |
| 3173 | case Builtin::BI__builtin_mul_overflow: { |
| 3174 | const clang::Expr *LeftArg = E->getArg(0); |
| 3175 | const clang::Expr *RightArg = E->getArg(1); |
| 3176 | const clang::Expr *ResultArg = E->getArg(2); |
| 3177 | |
| 3178 | clang::QualType ResultQTy = |
| 3179 | ResultArg->getType()->castAs<PointerType>()->getPointeeType(); |
| 3180 | |
| 3181 | WidthAndSignedness LeftInfo = |
| 3182 | getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType()); |
| 3183 | WidthAndSignedness RightInfo = |
| 3184 | getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType()); |
| 3185 | WidthAndSignedness ResultInfo = |
| 3186 | getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy); |
| 3187 | |
| 3188 | // Handle mixed-sign multiplication as a special case, because adding |
| 3189 | // runtime or backend support for our generic irgen would be too expensive. |
| 3190 | if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo)) |
| 3191 | return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg, |
| 3192 | RightInfo, ResultArg, ResultQTy, |
| 3193 | ResultInfo); |
| 3194 | |
| 3195 | WidthAndSignedness EncompassingInfo = |
| 3196 | EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo}); |
| 3197 | |
| 3198 | llvm::Type *EncompassingLLVMTy = |
| 3199 | llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width); |
| 3200 | |
| 3201 | llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy); |
| 3202 | |
| 3203 | llvm::Intrinsic::ID IntrinsicId; |
| 3204 | switch (BuiltinID) { |
| 3205 | default: |
| 3206 | llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3206); |
| 3207 | case Builtin::BI__builtin_add_overflow: |
| 3208 | IntrinsicId = EncompassingInfo.Signed |
| 3209 | ? llvm::Intrinsic::sadd_with_overflow |
| 3210 | : llvm::Intrinsic::uadd_with_overflow; |
| 3211 | break; |
| 3212 | case Builtin::BI__builtin_sub_overflow: |
| 3213 | IntrinsicId = EncompassingInfo.Signed |
| 3214 | ? llvm::Intrinsic::ssub_with_overflow |
| 3215 | : llvm::Intrinsic::usub_with_overflow; |
| 3216 | break; |
| 3217 | case Builtin::BI__builtin_mul_overflow: |
| 3218 | IntrinsicId = EncompassingInfo.Signed |
| 3219 | ? llvm::Intrinsic::smul_with_overflow |
| 3220 | : llvm::Intrinsic::umul_with_overflow; |
| 3221 | break; |
| 3222 | } |
| 3223 | |
| 3224 | llvm::Value *Left = EmitScalarExpr(LeftArg); |
| 3225 | llvm::Value *Right = EmitScalarExpr(RightArg); |
| 3226 | Address ResultPtr = EmitPointerWithAlignment(ResultArg); |
| 3227 | |
| 3228 | // Extend each operand to the encompassing type. |
| 3229 | Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed); |
| 3230 | Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed); |
| 3231 | |
| 3232 | // Perform the operation on the extended values. |
| 3233 | llvm::Value *Overflow, *Result; |
| 3234 | Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow); |
| 3235 | |
| 3236 | if (EncompassingInfo.Width > ResultInfo.Width) { |
| 3237 | // The encompassing type is wider than the result type, so we need to |
| 3238 | // truncate it. |
| 3239 | llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy); |
| 3240 | |
| 3241 | // To see if the truncation caused an overflow, we will extend |
| 3242 | // the result and then compare it to the original result. |
| 3243 | llvm::Value *ResultTruncExt = Builder.CreateIntCast( |
| 3244 | ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed); |
| 3245 | llvm::Value *TruncationOverflow = |
| 3246 | Builder.CreateICmpNE(Result, ResultTruncExt); |
| 3247 | |
| 3248 | Overflow = Builder.CreateOr(Overflow, TruncationOverflow); |
| 3249 | Result = ResultTrunc; |
| 3250 | } |
| 3251 | |
| 3252 | // Finally, store the result using the pointer. |
| 3253 | bool isVolatile = |
| 3254 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 3255 | Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile); |
| 3256 | |
| 3257 | return RValue::get(Overflow); |
| 3258 | } |
| 3259 | |
| 3260 | case Builtin::BI__builtin_uadd_overflow: |
| 3261 | case Builtin::BI__builtin_uaddl_overflow: |
| 3262 | case Builtin::BI__builtin_uaddll_overflow: |
| 3263 | case Builtin::BI__builtin_usub_overflow: |
| 3264 | case Builtin::BI__builtin_usubl_overflow: |
| 3265 | case Builtin::BI__builtin_usubll_overflow: |
| 3266 | case Builtin::BI__builtin_umul_overflow: |
| 3267 | case Builtin::BI__builtin_umull_overflow: |
| 3268 | case Builtin::BI__builtin_umulll_overflow: |
| 3269 | case Builtin::BI__builtin_sadd_overflow: |
| 3270 | case Builtin::BI__builtin_saddl_overflow: |
| 3271 | case Builtin::BI__builtin_saddll_overflow: |
| 3272 | case Builtin::BI__builtin_ssub_overflow: |
| 3273 | case Builtin::BI__builtin_ssubl_overflow: |
| 3274 | case Builtin::BI__builtin_ssubll_overflow: |
| 3275 | case Builtin::BI__builtin_smul_overflow: |
| 3276 | case Builtin::BI__builtin_smull_overflow: |
| 3277 | case Builtin::BI__builtin_smulll_overflow: { |
| 3278 | |
| 3279 | // We translate all of these builtins directly to the relevant llvm IR node. |
| 3280 | |
| 3281 | // Scalarize our inputs. |
| 3282 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
| 3283 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
| 3284 | Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2)); |
| 3285 | |
| 3286 | // Decide which of the overflow intrinsics we are lowering to: |
| 3287 | llvm::Intrinsic::ID IntrinsicId; |
| 3288 | switch (BuiltinID) { |
| 3289 | default: llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3289); |
| 3290 | case Builtin::BI__builtin_uadd_overflow: |
| 3291 | case Builtin::BI__builtin_uaddl_overflow: |
| 3292 | case Builtin::BI__builtin_uaddll_overflow: |
| 3293 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
| 3294 | break; |
| 3295 | case Builtin::BI__builtin_usub_overflow: |
| 3296 | case Builtin::BI__builtin_usubl_overflow: |
| 3297 | case Builtin::BI__builtin_usubll_overflow: |
| 3298 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
| 3299 | break; |
| 3300 | case Builtin::BI__builtin_umul_overflow: |
| 3301 | case Builtin::BI__builtin_umull_overflow: |
| 3302 | case Builtin::BI__builtin_umulll_overflow: |
| 3303 | IntrinsicId = llvm::Intrinsic::umul_with_overflow; |
| 3304 | break; |
| 3305 | case Builtin::BI__builtin_sadd_overflow: |
| 3306 | case Builtin::BI__builtin_saddl_overflow: |
| 3307 | case Builtin::BI__builtin_saddll_overflow: |
| 3308 | IntrinsicId = llvm::Intrinsic::sadd_with_overflow; |
| 3309 | break; |
| 3310 | case Builtin::BI__builtin_ssub_overflow: |
| 3311 | case Builtin::BI__builtin_ssubl_overflow: |
| 3312 | case Builtin::BI__builtin_ssubll_overflow: |
| 3313 | IntrinsicId = llvm::Intrinsic::ssub_with_overflow; |
| 3314 | break; |
| 3315 | case Builtin::BI__builtin_smul_overflow: |
| 3316 | case Builtin::BI__builtin_smull_overflow: |
| 3317 | case Builtin::BI__builtin_smulll_overflow: |
| 3318 | IntrinsicId = llvm::Intrinsic::smul_with_overflow; |
| 3319 | break; |
| 3320 | } |
| 3321 | |
| 3322 | |
| 3323 | llvm::Value *Carry; |
| 3324 | llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry); |
| 3325 | Builder.CreateStore(Sum, SumOutPtr); |
| 3326 | |
| 3327 | return RValue::get(Carry); |
| 3328 | } |
| 3329 | case Builtin::BI__builtin_addressof: |
| 3330 | return RValue::get(EmitLValue(E->getArg(0)).getPointer()); |
| 3331 | case Builtin::BI__builtin_operator_new: |
| 3332 | return EmitBuiltinNewDeleteCall( |
| 3333 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false); |
| 3334 | case Builtin::BI__builtin_operator_delete: |
| 3335 | return EmitBuiltinNewDeleteCall( |
| 3336 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true); |
| 3337 | |
| 3338 | case Builtin::BI__noop: |
| 3339 | // __noop always evaluates to an integer literal zero. |
| 3340 | return RValue::get(ConstantInt::get(IntTy, 0)); |
| 3341 | case Builtin::BI__builtin_call_with_static_chain: { |
| 3342 | const CallExpr *Call = cast<CallExpr>(E->getArg(0)); |
| 3343 | const Expr *Chain = E->getArg(1); |
| 3344 | return EmitCall(Call->getCallee()->getType(), |
| 3345 | EmitCallee(Call->getCallee()), Call, ReturnValue, |
| 3346 | EmitScalarExpr(Chain)); |
| 3347 | } |
| 3348 | case Builtin::BI_InterlockedExchange8: |
| 3349 | case Builtin::BI_InterlockedExchange16: |
| 3350 | case Builtin::BI_InterlockedExchange: |
| 3351 | case Builtin::BI_InterlockedExchangePointer: |
| 3352 | return RValue::get( |
| 3353 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E)); |
| 3354 | case Builtin::BI_InterlockedCompareExchangePointer: |
| 3355 | case Builtin::BI_InterlockedCompareExchangePointer_nf: { |
| 3356 | llvm::Type *RTy; |
| 3357 | llvm::IntegerType *IntType = |
| 3358 | IntegerType::get(getLLVMContext(), |
| 3359 | getContext().getTypeSize(E->getType())); |
| 3360 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
| 3361 | |
| 3362 | llvm::Value *Destination = |
| 3363 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType); |
| 3364 | |
| 3365 | llvm::Value *Exchange = EmitScalarExpr(E->getArg(1)); |
| 3366 | RTy = Exchange->getType(); |
| 3367 | Exchange = Builder.CreatePtrToInt(Exchange, IntType); |
| 3368 | |
| 3369 | llvm::Value *Comparand = |
| 3370 | Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType); |
| 3371 | |
| 3372 | auto Ordering = |
| 3373 | BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ? |
| 3374 | AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent; |
| 3375 | |
| 3376 | auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
| 3377 | Ordering, Ordering); |
| 3378 | Result->setVolatile(true); |
| 3379 | |
| 3380 | return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result, |
| 3381 | 0), |
| 3382 | RTy)); |
| 3383 | } |
| 3384 | case Builtin::BI_InterlockedCompareExchange8: |
| 3385 | case Builtin::BI_InterlockedCompareExchange16: |
| 3386 | case Builtin::BI_InterlockedCompareExchange: |
| 3387 | case Builtin::BI_InterlockedCompareExchange64: |
| 3388 | return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E)); |
| 3389 | case Builtin::BI_InterlockedIncrement16: |
| 3390 | case Builtin::BI_InterlockedIncrement: |
| 3391 | return RValue::get( |
| 3392 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E)); |
| 3393 | case Builtin::BI_InterlockedDecrement16: |
| 3394 | case Builtin::BI_InterlockedDecrement: |
| 3395 | return RValue::get( |
| 3396 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E)); |
| 3397 | case Builtin::BI_InterlockedAnd8: |
| 3398 | case Builtin::BI_InterlockedAnd16: |
| 3399 | case Builtin::BI_InterlockedAnd: |
| 3400 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E)); |
| 3401 | case Builtin::BI_InterlockedExchangeAdd8: |
| 3402 | case Builtin::BI_InterlockedExchangeAdd16: |
| 3403 | case Builtin::BI_InterlockedExchangeAdd: |
| 3404 | return RValue::get( |
| 3405 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E)); |
| 3406 | case Builtin::BI_InterlockedExchangeSub8: |
| 3407 | case Builtin::BI_InterlockedExchangeSub16: |
| 3408 | case Builtin::BI_InterlockedExchangeSub: |
| 3409 | return RValue::get( |
| 3410 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E)); |
| 3411 | case Builtin::BI_InterlockedOr8: |
| 3412 | case Builtin::BI_InterlockedOr16: |
| 3413 | case Builtin::BI_InterlockedOr: |
| 3414 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E)); |
| 3415 | case Builtin::BI_InterlockedXor8: |
| 3416 | case Builtin::BI_InterlockedXor16: |
| 3417 | case Builtin::BI_InterlockedXor: |
| 3418 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E)); |
| 3419 | |
| 3420 | case Builtin::BI_bittest64: |
| 3421 | case Builtin::BI_bittest: |
| 3422 | case Builtin::BI_bittestandcomplement64: |
| 3423 | case Builtin::BI_bittestandcomplement: |
| 3424 | case Builtin::BI_bittestandreset64: |
| 3425 | case Builtin::BI_bittestandreset: |
| 3426 | case Builtin::BI_bittestandset64: |
| 3427 | case Builtin::BI_bittestandset: |
| 3428 | case Builtin::BI_interlockedbittestandreset: |
| 3429 | case Builtin::BI_interlockedbittestandreset64: |
| 3430 | case Builtin::BI_interlockedbittestandset64: |
| 3431 | case Builtin::BI_interlockedbittestandset: |
| 3432 | case Builtin::BI_interlockedbittestandset_acq: |
| 3433 | case Builtin::BI_interlockedbittestandset_rel: |
| 3434 | case Builtin::BI_interlockedbittestandset_nf: |
| 3435 | case Builtin::BI_interlockedbittestandreset_acq: |
| 3436 | case Builtin::BI_interlockedbittestandreset_rel: |
| 3437 | case Builtin::BI_interlockedbittestandreset_nf: |
| 3438 | return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E)); |
| 3439 | |
| 3440 | // These builtins exist to emit regular volatile loads and stores not |
| 3441 | // affected by the -fms-volatile setting. |
| 3442 | case Builtin::BI__iso_volatile_load8: |
| 3443 | case Builtin::BI__iso_volatile_load16: |
| 3444 | case Builtin::BI__iso_volatile_load32: |
| 3445 | case Builtin::BI__iso_volatile_load64: |
| 3446 | return RValue::get(EmitISOVolatileLoad(*this, E)); |
| 3447 | case Builtin::BI__iso_volatile_store8: |
| 3448 | case Builtin::BI__iso_volatile_store16: |
| 3449 | case Builtin::BI__iso_volatile_store32: |
| 3450 | case Builtin::BI__iso_volatile_store64: |
| 3451 | return RValue::get(EmitISOVolatileStore(*this, E)); |
| 3452 | |
| 3453 | case Builtin::BI__exception_code: |
| 3454 | case Builtin::BI_exception_code: |
| 3455 | return RValue::get(EmitSEHExceptionCode()); |
| 3456 | case Builtin::BI__exception_info: |
| 3457 | case Builtin::BI_exception_info: |
| 3458 | return RValue::get(EmitSEHExceptionInfo()); |
| 3459 | case Builtin::BI__abnormal_termination: |
| 3460 | case Builtin::BI_abnormal_termination: |
| 3461 | return RValue::get(EmitSEHAbnormalTermination()); |
| 3462 | case Builtin::BI_setjmpex: |
| 3463 | if (getTarget().getTriple().isOSMSVCRT()) |
| 3464 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
| 3465 | break; |
| 3466 | case Builtin::BI_setjmp: |
| 3467 | if (getTarget().getTriple().isOSMSVCRT()) { |
| 3468 | if (getTarget().getTriple().getArch() == llvm::Triple::x86) |
| 3469 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E); |
| 3470 | else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64) |
| 3471 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
| 3472 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E); |
| 3473 | } |
| 3474 | break; |
| 3475 | |
| 3476 | case Builtin::BI__GetExceptionInfo: { |
| 3477 | if (llvm::GlobalVariable *GV = |
| 3478 | CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType())) |
| 3479 | return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy)); |
| 3480 | break; |
| 3481 | } |
| 3482 | |
| 3483 | case Builtin::BI__fastfail: |
| 3484 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E)); |
| 3485 | |
| 3486 | case Builtin::BI__builtin_coro_size: { |
| 3487 | auto & Context = getContext(); |
| 3488 | auto SizeTy = Context.getSizeType(); |
| 3489 | auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy)); |
| 3490 | Function *F = CGM.getIntrinsic(Intrinsic::coro_size, T); |
| 3491 | return RValue::get(Builder.CreateCall(F)); |
| 3492 | } |
| 3493 | |
| 3494 | case Builtin::BI__builtin_coro_id: |
| 3495 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_id); |
| 3496 | case Builtin::BI__builtin_coro_promise: |
| 3497 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise); |
| 3498 | case Builtin::BI__builtin_coro_resume: |
| 3499 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume); |
| 3500 | case Builtin::BI__builtin_coro_frame: |
| 3501 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame); |
| 3502 | case Builtin::BI__builtin_coro_noop: |
| 3503 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop); |
| 3504 | case Builtin::BI__builtin_coro_free: |
| 3505 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_free); |
| 3506 | case Builtin::BI__builtin_coro_destroy: |
| 3507 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy); |
| 3508 | case Builtin::BI__builtin_coro_done: |
| 3509 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_done); |
| 3510 | case Builtin::BI__builtin_coro_alloc: |
| 3511 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc); |
| 3512 | case Builtin::BI__builtin_coro_begin: |
| 3513 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin); |
| 3514 | case Builtin::BI__builtin_coro_end: |
| 3515 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_end); |
| 3516 | case Builtin::BI__builtin_coro_suspend: |
| 3517 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend); |
| 3518 | case Builtin::BI__builtin_coro_param: |
| 3519 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_param); |
| 3520 | |
| 3521 | // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions |
| 3522 | case Builtin::BIread_pipe: |
| 3523 | case Builtin::BIwrite_pipe: { |
| 3524 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 3525 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 3526 | CGOpenCLRuntime OpenCLRT(CGM); |
| 3527 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 3528 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 3529 | |
| 3530 | // Type of the generic packet parameter. |
| 3531 | unsigned GenericAS = |
| 3532 | getContext().getTargetAddressSpace(LangAS::opencl_generic); |
| 3533 | llvm::Type *I8PTy = llvm::PointerType::get( |
| 3534 | llvm::Type::getInt8Ty(getLLVMContext()), GenericAS); |
| 3535 | |
| 3536 | // Testing which overloaded version we should generate the call for. |
| 3537 | if (2U == E->getNumArgs()) { |
| 3538 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2" |
| 3539 | : "__write_pipe_2"; |
| 3540 | // Creating a generic function type to be able to call with any builtin or |
| 3541 | // user defined type. |
| 3542 | llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty}; |
| 3543 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3544 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3545 | Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy); |
| 3546 | return RValue::get( |
| 3547 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 3548 | {Arg0, BCast, PacketSize, PacketAlign})); |
| 3549 | } else { |
| 3550 | assert(4 == E->getNumArgs() &&((4 == E->getNumArgs() && "Illegal number of parameters to pipe function" ) ? static_cast<void> (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3551, __PRETTY_FUNCTION__)) |
| 3551 | "Illegal number of parameters to pipe function")((4 == E->getNumArgs() && "Illegal number of parameters to pipe function" ) ? static_cast<void> (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3551, __PRETTY_FUNCTION__)); |
| 3552 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4" |
| 3553 | : "__write_pipe_4"; |
| 3554 | |
| 3555 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy, |
| 3556 | Int32Ty, Int32Ty}; |
| 3557 | Value *Arg2 = EmitScalarExpr(E->getArg(2)), |
| 3558 | *Arg3 = EmitScalarExpr(E->getArg(3)); |
| 3559 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3560 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3561 | Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy); |
| 3562 | // We know the third argument is an integer type, but we may need to cast |
| 3563 | // it to i32. |
| 3564 | if (Arg2->getType() != Int32Ty) |
| 3565 | Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty); |
| 3566 | return RValue::get(Builder.CreateCall( |
| 3567 | CGM.CreateRuntimeFunction(FTy, Name), |
| 3568 | {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign})); |
| 3569 | } |
| 3570 | } |
| 3571 | // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write |
| 3572 | // functions |
| 3573 | case Builtin::BIreserve_read_pipe: |
| 3574 | case Builtin::BIreserve_write_pipe: |
| 3575 | case Builtin::BIwork_group_reserve_read_pipe: |
| 3576 | case Builtin::BIwork_group_reserve_write_pipe: |
| 3577 | case Builtin::BIsub_group_reserve_read_pipe: |
| 3578 | case Builtin::BIsub_group_reserve_write_pipe: { |
| 3579 | // Composing the mangled name for the function. |
| 3580 | const char *Name; |
| 3581 | if (BuiltinID == Builtin::BIreserve_read_pipe) |
| 3582 | Name = "__reserve_read_pipe"; |
| 3583 | else if (BuiltinID == Builtin::BIreserve_write_pipe) |
| 3584 | Name = "__reserve_write_pipe"; |
| 3585 | else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe) |
| 3586 | Name = "__work_group_reserve_read_pipe"; |
| 3587 | else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe) |
| 3588 | Name = "__work_group_reserve_write_pipe"; |
| 3589 | else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe) |
| 3590 | Name = "__sub_group_reserve_read_pipe"; |
| 3591 | else |
| 3592 | Name = "__sub_group_reserve_write_pipe"; |
| 3593 | |
| 3594 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 3595 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 3596 | llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy); |
| 3597 | CGOpenCLRuntime OpenCLRT(CGM); |
| 3598 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 3599 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 3600 | |
| 3601 | // Building the generic function prototype. |
| 3602 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty}; |
| 3603 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3604 | ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3605 | // We know the second argument is an integer type, but we may need to cast |
| 3606 | // it to i32. |
| 3607 | if (Arg1->getType() != Int32Ty) |
| 3608 | Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty); |
| 3609 | return RValue::get( |
| 3610 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 3611 | {Arg0, Arg1, PacketSize, PacketAlign})); |
| 3612 | } |
| 3613 | // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write |
| 3614 | // functions |
| 3615 | case Builtin::BIcommit_read_pipe: |
| 3616 | case Builtin::BIcommit_write_pipe: |
| 3617 | case Builtin::BIwork_group_commit_read_pipe: |
| 3618 | case Builtin::BIwork_group_commit_write_pipe: |
| 3619 | case Builtin::BIsub_group_commit_read_pipe: |
| 3620 | case Builtin::BIsub_group_commit_write_pipe: { |
| 3621 | const char *Name; |
| 3622 | if (BuiltinID == Builtin::BIcommit_read_pipe) |
| 3623 | Name = "__commit_read_pipe"; |
| 3624 | else if (BuiltinID == Builtin::BIcommit_write_pipe) |
| 3625 | Name = "__commit_write_pipe"; |
| 3626 | else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe) |
| 3627 | Name = "__work_group_commit_read_pipe"; |
| 3628 | else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe) |
| 3629 | Name = "__work_group_commit_write_pipe"; |
| 3630 | else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe) |
| 3631 | Name = "__sub_group_commit_read_pipe"; |
| 3632 | else |
| 3633 | Name = "__sub_group_commit_write_pipe"; |
| 3634 | |
| 3635 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 3636 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 3637 | CGOpenCLRuntime OpenCLRT(CGM); |
| 3638 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 3639 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 3640 | |
| 3641 | // Building the generic function prototype. |
| 3642 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty}; |
| 3643 | llvm::FunctionType *FTy = |
| 3644 | llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()), |
| 3645 | llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3646 | |
| 3647 | return RValue::get( |
| 3648 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 3649 | {Arg0, Arg1, PacketSize, PacketAlign})); |
| 3650 | } |
| 3651 | // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions |
| 3652 | case Builtin::BIget_pipe_num_packets: |
| 3653 | case Builtin::BIget_pipe_max_packets: { |
| 3654 | const char *BaseName; |
| 3655 | const PipeType *PipeTy = E->getArg(0)->getType()->getAs<PipeType>(); |
| 3656 | if (BuiltinID == Builtin::BIget_pipe_num_packets) |
| 3657 | BaseName = "__get_pipe_num_packets"; |
| 3658 | else |
| 3659 | BaseName = "__get_pipe_max_packets"; |
| 3660 | auto Name = std::string(BaseName) + |
| 3661 | std::string(PipeTy->isReadOnly() ? "_ro" : "_wo"); |
| 3662 | |
| 3663 | // Building the generic function prototype. |
| 3664 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); |
| 3665 | CGOpenCLRuntime OpenCLRT(CGM); |
| 3666 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 3667 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 3668 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty}; |
| 3669 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3670 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3671 | |
| 3672 | return RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 3673 | {Arg0, PacketSize, PacketAlign})); |
| 3674 | } |
| 3675 | |
| 3676 | // OpenCL v2.0 s6.13.9 - Address space qualifier functions. |
| 3677 | case Builtin::BIto_global: |
| 3678 | case Builtin::BIto_local: |
| 3679 | case Builtin::BIto_private: { |
| 3680 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
| 3681 | auto NewArgT = llvm::PointerType::get(Int8Ty, |
| 3682 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 3683 | auto NewRetT = llvm::PointerType::get(Int8Ty, |
| 3684 | CGM.getContext().getTargetAddressSpace( |
| 3685 | E->getType()->getPointeeType().getAddressSpace())); |
| 3686 | auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false); |
| 3687 | llvm::Value *NewArg; |
| 3688 | if (Arg0->getType()->getPointerAddressSpace() != |
| 3689 | NewArgT->getPointerAddressSpace()) |
| 3690 | NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT); |
| 3691 | else |
| 3692 | NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT); |
| 3693 | auto NewName = std::string("__") + E->getDirectCallee()->getName().str(); |
| 3694 | auto NewCall = |
| 3695 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg}); |
| 3696 | return RValue::get(Builder.CreateBitOrPointerCast(NewCall, |
| 3697 | ConvertType(E->getType()))); |
| 3698 | } |
| 3699 | |
| 3700 | // OpenCL v2.0, s6.13.17 - Enqueue kernel function. |
| 3701 | // It contains four different overload formats specified in Table 6.13.17.1. |
| 3702 | case Builtin::BIenqueue_kernel: { |
| 3703 | StringRef Name; // Generated function call name |
| 3704 | unsigned NumArgs = E->getNumArgs(); |
| 3705 | |
| 3706 | llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy); |
| 3707 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 3708 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 3709 | |
| 3710 | llvm::Value *Queue = EmitScalarExpr(E->getArg(0)); |
| 3711 | llvm::Value *Flags = EmitScalarExpr(E->getArg(1)); |
| 3712 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(2)); |
| 3713 | llvm::Value *Range = NDRangeL.getAddress().getPointer(); |
| 3714 | llvm::Type *RangeTy = NDRangeL.getAddress().getType(); |
| 3715 | |
| 3716 | if (NumArgs == 4) { |
| 3717 | // The most basic form of the call with parameters: |
| 3718 | // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void) |
| 3719 | Name = "__enqueue_kernel_basic"; |
| 3720 | llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy, |
| 3721 | GenericVoidPtrTy}; |
| 3722 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3723 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3724 | |
| 3725 | auto Info = |
| 3726 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
| 3727 | llvm::Value *Kernel = |
| 3728 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 3729 | llvm::Value *Block = |
| 3730 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 3731 | |
| 3732 | AttrBuilder B; |
| 3733 | B.addAttribute(Attribute::ByVal); |
| 3734 | llvm::AttributeList ByValAttrSet = |
| 3735 | llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B); |
| 3736 | |
| 3737 | auto RTCall = |
| 3738 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet), |
| 3739 | {Queue, Flags, Range, Kernel, Block}); |
| 3740 | RTCall->setAttributes(ByValAttrSet); |
| 3741 | return RValue::get(RTCall); |
| 3742 | } |
| 3743 | assert(NumArgs >= 5 && "Invalid enqueue_kernel signature")((NumArgs >= 5 && "Invalid enqueue_kernel signature" ) ? static_cast<void> (0) : __assert_fail ("NumArgs >= 5 && \"Invalid enqueue_kernel signature\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 3743, __PRETTY_FUNCTION__)); |
| 3744 | |
| 3745 | // Create a temporary array to hold the sizes of local pointer arguments |
| 3746 | // for the block. \p First is the position of the first size argument. |
| 3747 | auto CreateArrayForSizeVar = [=](unsigned First) |
| 3748 | -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> { |
| 3749 | llvm::APInt ArraySize(32, NumArgs - First); |
| 3750 | QualType SizeArrayTy = getContext().getConstantArrayType( |
| 3751 | getContext().getSizeType(), ArraySize, ArrayType::Normal, |
| 3752 | /*IndexTypeQuals=*/0); |
| 3753 | auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes"); |
| 3754 | llvm::Value *TmpPtr = Tmp.getPointer(); |
| 3755 | llvm::Value *TmpSize = EmitLifetimeStart( |
| 3756 | CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr); |
| 3757 | llvm::Value *ElemPtr; |
| 3758 | // Each of the following arguments specifies the size of the corresponding |
| 3759 | // argument passed to the enqueued block. |
| 3760 | auto *Zero = llvm::ConstantInt::get(IntTy, 0); |
| 3761 | for (unsigned I = First; I < NumArgs; ++I) { |
| 3762 | auto *Index = llvm::ConstantInt::get(IntTy, I - First); |
| 3763 | auto *GEP = Builder.CreateGEP(TmpPtr, {Zero, Index}); |
| 3764 | if (I == First) |
| 3765 | ElemPtr = GEP; |
| 3766 | auto *V = |
| 3767 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy); |
| 3768 | Builder.CreateAlignedStore( |
| 3769 | V, GEP, CGM.getDataLayout().getPrefTypeAlignment(SizeTy)); |
| 3770 | } |
| 3771 | return std::tie(ElemPtr, TmpSize, TmpPtr); |
| 3772 | }; |
| 3773 | |
| 3774 | // Could have events and/or varargs. |
| 3775 | if (E->getArg(3)->getType()->isBlockPointerType()) { |
| 3776 | // No events passed, but has variadic arguments. |
| 3777 | Name = "__enqueue_kernel_varargs"; |
| 3778 | auto Info = |
| 3779 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
| 3780 | llvm::Value *Kernel = |
| 3781 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 3782 | auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 3783 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
| 3784 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4); |
| 3785 | |
| 3786 | // Create a vector of the arguments, as well as a constant value to |
| 3787 | // express to the runtime the number of variadic arguments. |
| 3788 | std::vector<llvm::Value *> Args = { |
| 3789 | Queue, Flags, Range, |
| 3790 | Kernel, Block, ConstantInt::get(IntTy, NumArgs - 4), |
| 3791 | ElemPtr}; |
| 3792 | std::vector<llvm::Type *> ArgTys = { |
| 3793 | QueueTy, IntTy, RangeTy, GenericVoidPtrTy, |
| 3794 | GenericVoidPtrTy, IntTy, ElemPtr->getType()}; |
| 3795 | |
| 3796 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3797 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3798 | auto Call = |
| 3799 | RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 3800 | llvm::ArrayRef<llvm::Value *>(Args))); |
| 3801 | if (TmpSize) |
| 3802 | EmitLifetimeEnd(TmpSize, TmpPtr); |
| 3803 | return Call; |
| 3804 | } |
| 3805 | // Any calls now have event arguments passed. |
| 3806 | if (NumArgs >= 7) { |
| 3807 | llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy); |
| 3808 | llvm::PointerType *EventPtrTy = EventTy->getPointerTo( |
| 3809 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 3810 | |
| 3811 | llvm::Value *NumEvents = |
| 3812 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty); |
| 3813 | |
| 3814 | // Since SemaOpenCLBuiltinEnqueueKernel allows fifth and sixth arguments |
| 3815 | // to be a null pointer constant (including `0` literal), we can take it |
| 3816 | // into account and emit null pointer directly. |
| 3817 | llvm::Value *EventWaitList = nullptr; |
| 3818 | if (E->getArg(4)->isNullPointerConstant( |
| 3819 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
| 3820 | EventWaitList = llvm::ConstantPointerNull::get(EventPtrTy); |
| 3821 | } else { |
| 3822 | EventWaitList = E->getArg(4)->getType()->isArrayType() |
| 3823 | ? EmitArrayToPointerDecay(E->getArg(4)).getPointer() |
| 3824 | : EmitScalarExpr(E->getArg(4)); |
| 3825 | // Convert to generic address space. |
| 3826 | EventWaitList = Builder.CreatePointerCast(EventWaitList, EventPtrTy); |
| 3827 | } |
| 3828 | llvm::Value *EventRet = nullptr; |
| 3829 | if (E->getArg(5)->isNullPointerConstant( |
| 3830 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
| 3831 | EventRet = llvm::ConstantPointerNull::get(EventPtrTy); |
| 3832 | } else { |
| 3833 | EventRet = |
| 3834 | Builder.CreatePointerCast(EmitScalarExpr(E->getArg(5)), EventPtrTy); |
| 3835 | } |
| 3836 | |
| 3837 | auto Info = |
| 3838 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6)); |
| 3839 | llvm::Value *Kernel = |
| 3840 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 3841 | llvm::Value *Block = |
| 3842 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 3843 | |
| 3844 | std::vector<llvm::Type *> ArgTys = { |
| 3845 | QueueTy, Int32Ty, RangeTy, Int32Ty, |
| 3846 | EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy}; |
| 3847 | |
| 3848 | std::vector<llvm::Value *> Args = {Queue, Flags, Range, |
| 3849 | NumEvents, EventWaitList, EventRet, |
| 3850 | Kernel, Block}; |
| 3851 | |
| 3852 | if (NumArgs == 7) { |
| 3853 | // Has events but no variadics. |
| 3854 | Name = "__enqueue_kernel_basic_events"; |
| 3855 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3856 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3857 | return RValue::get( |
| 3858 | Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 3859 | llvm::ArrayRef<llvm::Value *>(Args))); |
| 3860 | } |
| 3861 | // Has event info and variadics |
| 3862 | // Pass the number of variadics to the runtime function too. |
| 3863 | Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7)); |
| 3864 | ArgTys.push_back(Int32Ty); |
| 3865 | Name = "__enqueue_kernel_events_varargs"; |
| 3866 | |
| 3867 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
| 3868 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7); |
| 3869 | Args.push_back(ElemPtr); |
| 3870 | ArgTys.push_back(ElemPtr->getType()); |
| 3871 | |
| 3872 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 3873 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 3874 | auto Call = |
| 3875 | RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 3876 | llvm::ArrayRef<llvm::Value *>(Args))); |
| 3877 | if (TmpSize) |
| 3878 | EmitLifetimeEnd(TmpSize, TmpPtr); |
| 3879 | return Call; |
| 3880 | } |
| 3881 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 3882 | } |
| 3883 | // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block |
| 3884 | // parameter. |
| 3885 | case Builtin::BIget_kernel_work_group_size: { |
| 3886 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 3887 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 3888 | auto Info = |
| 3889 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
| 3890 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 3891 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 3892 | return RValue::get(Builder.CreateCall( |
| 3893 | CGM.CreateRuntimeFunction( |
| 3894 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
| 3895 | false), |
| 3896 | "__get_kernel_work_group_size_impl"), |
| 3897 | {Kernel, Arg})); |
| 3898 | } |
| 3899 | case Builtin::BIget_kernel_preferred_work_group_size_multiple: { |
| 3900 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 3901 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 3902 | auto Info = |
| 3903 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
| 3904 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 3905 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 3906 | return RValue::get(Builder.CreateCall( |
| 3907 | CGM.CreateRuntimeFunction( |
| 3908 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
| 3909 | false), |
| 3910 | "__get_kernel_preferred_work_group_size_multiple_impl"), |
| 3911 | {Kernel, Arg})); |
| 3912 | } |
| 3913 | case Builtin::BIget_kernel_max_sub_group_size_for_ndrange: |
| 3914 | case Builtin::BIget_kernel_sub_group_count_for_ndrange: { |
| 3915 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 3916 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 3917 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(0)); |
| 3918 | llvm::Value *NDRange = NDRangeL.getAddress().getPointer(); |
| 3919 | auto Info = |
| 3920 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1)); |
| 3921 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 3922 | Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 3923 | const char *Name = |
| 3924 | BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange |
| 3925 | ? "__get_kernel_max_sub_group_size_for_ndrange_impl" |
| 3926 | : "__get_kernel_sub_group_count_for_ndrange_impl"; |
| 3927 | return RValue::get(Builder.CreateCall( |
| 3928 | CGM.CreateRuntimeFunction( |
| 3929 | llvm::FunctionType::get( |
| 3930 | IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy}, |
| 3931 | false), |
| 3932 | Name), |
| 3933 | {NDRange, Kernel, Block})); |
| 3934 | } |
| 3935 | |
| 3936 | case Builtin::BI__builtin_store_half: |
| 3937 | case Builtin::BI__builtin_store_halff: { |
| 3938 | Value *Val = EmitScalarExpr(E->getArg(0)); |
| 3939 | Address Address = EmitPointerWithAlignment(E->getArg(1)); |
| 3940 | Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy()); |
| 3941 | return RValue::get(Builder.CreateStore(HalfVal, Address)); |
| 3942 | } |
| 3943 | case Builtin::BI__builtin_load_half: { |
| 3944 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
| 3945 | Value *HalfVal = Builder.CreateLoad(Address); |
| 3946 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy())); |
| 3947 | } |
| 3948 | case Builtin::BI__builtin_load_halff: { |
| 3949 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
| 3950 | Value *HalfVal = Builder.CreateLoad(Address); |
| 3951 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy())); |
| 3952 | } |
| 3953 | case Builtin::BIprintf: |
| 3954 | if (getTarget().getTriple().isNVPTX()) |
| 3955 | return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue); |
| 3956 | break; |
| 3957 | case Builtin::BI__builtin_canonicalize: |
| 3958 | case Builtin::BI__builtin_canonicalizef: |
| 3959 | case Builtin::BI__builtin_canonicalizel: |
| 3960 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize)); |
| 3961 | |
| 3962 | case Builtin::BI__builtin_thread_pointer: { |
| 3963 | if (!getContext().getTargetInfo().isTLSSupported()) |
| 3964 | CGM.ErrorUnsupported(E, "__builtin_thread_pointer"); |
| 3965 | // Fall through - it's already mapped to the intrinsic by GCCBuiltin. |
| 3966 | break; |
| 3967 | } |
| 3968 | case Builtin::BI__builtin_os_log_format: |
| 3969 | return emitBuiltinOSLogFormat(*E); |
| 3970 | |
| 3971 | case Builtin::BI__xray_customevent: { |
| 3972 | if (!ShouldXRayInstrumentFunction()) |
| 3973 | return RValue::getIgnored(); |
| 3974 | |
| 3975 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
| 3976 | XRayInstrKind::Custom)) |
| 3977 | return RValue::getIgnored(); |
| 3978 | |
| 3979 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
| 3980 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents()) |
| 3981 | return RValue::getIgnored(); |
| 3982 | |
| 3983 | Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent); |
| 3984 | auto FTy = F->getFunctionType(); |
| 3985 | auto Arg0 = E->getArg(0); |
| 3986 | auto Arg0Val = EmitScalarExpr(Arg0); |
| 3987 | auto Arg0Ty = Arg0->getType(); |
| 3988 | auto PTy0 = FTy->getParamType(0); |
| 3989 | if (PTy0 != Arg0Val->getType()) { |
| 3990 | if (Arg0Ty->isArrayType()) |
| 3991 | Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer(); |
| 3992 | else |
| 3993 | Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0); |
| 3994 | } |
| 3995 | auto Arg1 = EmitScalarExpr(E->getArg(1)); |
| 3996 | auto PTy1 = FTy->getParamType(1); |
| 3997 | if (PTy1 != Arg1->getType()) |
| 3998 | Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1); |
| 3999 | return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1})); |
| 4000 | } |
| 4001 | |
| 4002 | case Builtin::BI__xray_typedevent: { |
| 4003 | // TODO: There should be a way to always emit events even if the current |
| 4004 | // function is not instrumented. Losing events in a stream can cripple |
| 4005 | // a trace. |
| 4006 | if (!ShouldXRayInstrumentFunction()) |
| 4007 | return RValue::getIgnored(); |
| 4008 | |
| 4009 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
| 4010 | XRayInstrKind::Typed)) |
| 4011 | return RValue::getIgnored(); |
| 4012 | |
| 4013 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
| 4014 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents()) |
| 4015 | return RValue::getIgnored(); |
| 4016 | |
| 4017 | Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent); |
| 4018 | auto FTy = F->getFunctionType(); |
| 4019 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
| 4020 | auto PTy0 = FTy->getParamType(0); |
| 4021 | if (PTy0 != Arg0->getType()) |
| 4022 | Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0); |
| 4023 | auto Arg1 = E->getArg(1); |
| 4024 | auto Arg1Val = EmitScalarExpr(Arg1); |
| 4025 | auto Arg1Ty = Arg1->getType(); |
| 4026 | auto PTy1 = FTy->getParamType(1); |
| 4027 | if (PTy1 != Arg1Val->getType()) { |
| 4028 | if (Arg1Ty->isArrayType()) |
| 4029 | Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer(); |
| 4030 | else |
| 4031 | Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1); |
| 4032 | } |
| 4033 | auto Arg2 = EmitScalarExpr(E->getArg(2)); |
| 4034 | auto PTy2 = FTy->getParamType(2); |
| 4035 | if (PTy2 != Arg2->getType()) |
| 4036 | Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2); |
| 4037 | return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2})); |
| 4038 | } |
| 4039 | |
| 4040 | case Builtin::BI__builtin_ms_va_start: |
| 4041 | case Builtin::BI__builtin_ms_va_end: |
| 4042 | return RValue::get( |
| 4043 | EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(), |
| 4044 | BuiltinID == Builtin::BI__builtin_ms_va_start)); |
| 4045 | |
| 4046 | case Builtin::BI__builtin_ms_va_copy: { |
| 4047 | // Lower this manually. We can't reliably determine whether or not any |
| 4048 | // given va_copy() is for a Win64 va_list from the calling convention |
| 4049 | // alone, because it's legal to do this from a System V ABI function. |
| 4050 | // With opaque pointer types, we won't have enough information in LLVM |
| 4051 | // IR to determine this from the argument types, either. Best to do it |
| 4052 | // now, while we have enough information. |
| 4053 | Address DestAddr = EmitMSVAListRef(E->getArg(0)); |
| 4054 | Address SrcAddr = EmitMSVAListRef(E->getArg(1)); |
| 4055 | |
| 4056 | llvm::Type *BPP = Int8PtrPtrTy; |
| 4057 | |
| 4058 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"), |
| 4059 | DestAddr.getAlignment()); |
| 4060 | SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"), |
| 4061 | SrcAddr.getAlignment()); |
| 4062 | |
| 4063 | Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val"); |
| 4064 | return RValue::get(Builder.CreateStore(ArgPtr, DestAddr)); |
| 4065 | } |
| 4066 | } |
| 4067 | |
| 4068 | // If this is an alias for a lib function (e.g. __builtin_sin), emit |
| 4069 | // the call using the normal call path, but using the unmangled |
| 4070 | // version of the function name. |
| 4071 | if (getContext().BuiltinInfo.isLibFunction(BuiltinID)) |
| 4072 | return emitLibraryCall(*this, FD, E, |
| 4073 | CGM.getBuiltinLibFunction(FD, BuiltinID)); |
| 4074 | |
| 4075 | // If this is a predefined lib function (e.g. malloc), emit the call |
| 4076 | // using exactly the normal call path. |
| 4077 | if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID)) |
| 4078 | return emitLibraryCall(*this, FD, E, |
| 4079 | cast<llvm::Constant>(EmitScalarExpr(E->getCallee()))); |
| 4080 | |
| 4081 | // Check that a call to a target specific builtin has the correct target |
| 4082 | // features. |
| 4083 | // This is down here to avoid non-target specific builtins, however, if |
| 4084 | // generic builtins start to require generic target features then we |
| 4085 | // can move this up to the beginning of the function. |
| 4086 | checkTargetFeatures(E, FD); |
| 4087 | |
| 4088 | if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID)) |
| 4089 | LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth); |
| 4090 | |
| 4091 | // See if we have a target specific intrinsic. |
| 4092 | const char *Name = getContext().BuiltinInfo.getName(BuiltinID); |
| 4093 | Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic; |
| 4094 | StringRef Prefix = |
| 4095 | llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch()); |
| 4096 | if (!Prefix.empty()) { |
| 4097 | IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name); |
| 4098 | // NOTE we don't need to perform a compatibility flag check here since the |
| 4099 | // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the |
| 4100 | // MS builtins via ALL_MS_LANGUAGES and are filtered earlier. |
| 4101 | if (IntrinsicID == Intrinsic::not_intrinsic) |
| 4102 | IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name); |
| 4103 | } |
| 4104 | |
| 4105 | if (IntrinsicID != Intrinsic::not_intrinsic) { |
| 4106 | SmallVector<Value*, 16> Args; |
| 4107 | |
| 4108 | // Find out if any arguments are required to be integer constant |
| 4109 | // expressions. |
| 4110 | unsigned ICEArguments = 0; |
| 4111 | ASTContext::GetBuiltinTypeError Error; |
| 4112 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); |
| 4113 | assert(Error == ASTContext::GE_None && "Should not codegen an error")((Error == ASTContext::GE_None && "Should not codegen an error" ) ? static_cast<void> (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4113, __PRETTY_FUNCTION__)); |
| 4114 | |
| 4115 | Function *F = CGM.getIntrinsic(IntrinsicID); |
| 4116 | llvm::FunctionType *FTy = F->getFunctionType(); |
| 4117 | |
| 4118 | for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) { |
| 4119 | Value *ArgValue; |
| 4120 | // If this is a normal argument, just emit it as a scalar. |
| 4121 | if ((ICEArguments & (1 << i)) == 0) { |
| 4122 | ArgValue = EmitScalarExpr(E->getArg(i)); |
| 4123 | } else { |
| 4124 | // If this is required to be a constant, constant fold it so that we |
| 4125 | // know that the generated intrinsic gets a ConstantInt. |
| 4126 | llvm::APSInt Result; |
| 4127 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result,getContext()); |
| 4128 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4128, __PRETTY_FUNCTION__)); |
| 4129 | (void)IsConst; |
| 4130 | ArgValue = llvm::ConstantInt::get(getLLVMContext(), Result); |
| 4131 | } |
| 4132 | |
| 4133 | // If the intrinsic arg type is different from the builtin arg type |
| 4134 | // we need to do a bit cast. |
| 4135 | llvm::Type *PTy = FTy->getParamType(i); |
| 4136 | if (PTy != ArgValue->getType()) { |
| 4137 | // XXX - vector of pointers? |
| 4138 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) { |
| 4139 | if (PtrTy->getAddressSpace() != |
| 4140 | ArgValue->getType()->getPointerAddressSpace()) { |
| 4141 | ArgValue = Builder.CreateAddrSpaceCast( |
| 4142 | ArgValue, |
| 4143 | ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace())); |
| 4144 | } |
| 4145 | } |
| 4146 | |
| 4147 | assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&((PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && "Must be able to losslessly bit cast to param") ? static_cast <void> (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4148, __PRETTY_FUNCTION__)) |
| 4148 | "Must be able to losslessly bit cast to param")((PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && "Must be able to losslessly bit cast to param") ? static_cast <void> (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4148, __PRETTY_FUNCTION__)); |
| 4149 | ArgValue = Builder.CreateBitCast(ArgValue, PTy); |
| 4150 | } |
| 4151 | |
| 4152 | Args.push_back(ArgValue); |
| 4153 | } |
| 4154 | |
| 4155 | Value *V = Builder.CreateCall(F, Args); |
| 4156 | QualType BuiltinRetType = E->getType(); |
| 4157 | |
| 4158 | llvm::Type *RetTy = VoidTy; |
| 4159 | if (!BuiltinRetType->isVoidType()) |
| 4160 | RetTy = ConvertType(BuiltinRetType); |
| 4161 | |
| 4162 | if (RetTy != V->getType()) { |
| 4163 | // XXX - vector of pointers? |
| 4164 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) { |
| 4165 | if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) { |
| 4166 | V = Builder.CreateAddrSpaceCast( |
| 4167 | V, V->getType()->getPointerTo(PtrTy->getAddressSpace())); |
| 4168 | } |
| 4169 | } |
| 4170 | |
| 4171 | assert(V->getType()->canLosslesslyBitCastTo(RetTy) &&((V->getType()->canLosslesslyBitCastTo(RetTy) && "Must be able to losslessly bit cast result type") ? static_cast <void> (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4172, __PRETTY_FUNCTION__)) |
| 4172 | "Must be able to losslessly bit cast result type")((V->getType()->canLosslesslyBitCastTo(RetTy) && "Must be able to losslessly bit cast result type") ? static_cast <void> (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4172, __PRETTY_FUNCTION__)); |
| 4173 | V = Builder.CreateBitCast(V, RetTy); |
| 4174 | } |
| 4175 | |
| 4176 | return RValue::get(V); |
| 4177 | } |
| 4178 | |
| 4179 | // See if we have a target specific builtin that needs to be lowered. |
| 4180 | if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E)) |
| 4181 | return RValue::get(V); |
| 4182 | |
| 4183 | ErrorUnsupported(E, "builtin function"); |
| 4184 | |
| 4185 | // Unknown builtin, for now just dump it out and return undef. |
| 4186 | return GetUndefRValue(E->getType()); |
| 4187 | } |
| 4188 | |
| 4189 | static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF, |
| 4190 | unsigned BuiltinID, const CallExpr *E, |
| 4191 | llvm::Triple::ArchType Arch) { |
| 4192 | switch (Arch) { |
| 4193 | case llvm::Triple::arm: |
| 4194 | case llvm::Triple::armeb: |
| 4195 | case llvm::Triple::thumb: |
| 4196 | case llvm::Triple::thumbeb: |
| 4197 | return CGF->EmitARMBuiltinExpr(BuiltinID, E, Arch); |
| 4198 | case llvm::Triple::aarch64: |
| 4199 | case llvm::Triple::aarch64_be: |
| 4200 | return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch); |
| 4201 | case llvm::Triple::x86: |
| 4202 | case llvm::Triple::x86_64: |
| 4203 | return CGF->EmitX86BuiltinExpr(BuiltinID, E); |
| 4204 | case llvm::Triple::ppc: |
| 4205 | case llvm::Triple::ppc64: |
| 4206 | case llvm::Triple::ppc64le: |
| 4207 | return CGF->EmitPPCBuiltinExpr(BuiltinID, E); |
| 4208 | case llvm::Triple::r600: |
| 4209 | case llvm::Triple::amdgcn: |
| 4210 | return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E); |
| 4211 | case llvm::Triple::systemz: |
| 4212 | return CGF->EmitSystemZBuiltinExpr(BuiltinID, E); |
| 4213 | case llvm::Triple::nvptx: |
| 4214 | case llvm::Triple::nvptx64: |
| 4215 | return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E); |
| 4216 | case llvm::Triple::wasm32: |
| 4217 | case llvm::Triple::wasm64: |
| 4218 | return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E); |
| 4219 | case llvm::Triple::hexagon: |
| 4220 | return CGF->EmitHexagonBuiltinExpr(BuiltinID, E); |
| 4221 | default: |
| 4222 | return nullptr; |
| 4223 | } |
| 4224 | } |
| 4225 | |
| 4226 | Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID, |
| 4227 | const CallExpr *E) { |
| 4228 | if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) { |
| 4229 | assert(getContext().getAuxTargetInfo() && "Missing aux target info")((getContext().getAuxTargetInfo() && "Missing aux target info" ) ? static_cast<void> (0) : __assert_fail ("getContext().getAuxTargetInfo() && \"Missing aux target info\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4229, __PRETTY_FUNCTION__)); |
| 4230 | return EmitTargetArchBuiltinExpr( |
| 4231 | this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E, |
| 4232 | getContext().getAuxTargetInfo()->getTriple().getArch()); |
| 4233 | } |
| 4234 | |
| 4235 | return EmitTargetArchBuiltinExpr(this, BuiltinID, E, |
| 4236 | getTarget().getTriple().getArch()); |
| 4237 | } |
| 4238 | |
| 4239 | static llvm::VectorType *GetNeonType(CodeGenFunction *CGF, |
| 4240 | NeonTypeFlags TypeFlags, |
| 4241 | bool HasLegalHalfType=true, |
| 4242 | bool V1Ty=false) { |
| 4243 | int IsQuad = TypeFlags.isQuad(); |
| 4244 | switch (TypeFlags.getEltType()) { |
| 4245 | case NeonTypeFlags::Int8: |
| 4246 | case NeonTypeFlags::Poly8: |
| 4247 | return llvm::VectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad)); |
| 4248 | case NeonTypeFlags::Int16: |
| 4249 | case NeonTypeFlags::Poly16: |
| 4250 | return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 4251 | case NeonTypeFlags::Float16: |
| 4252 | if (HasLegalHalfType) |
| 4253 | return llvm::VectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad)); |
| 4254 | else |
| 4255 | return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 4256 | case NeonTypeFlags::Int32: |
| 4257 | return llvm::VectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad)); |
| 4258 | case NeonTypeFlags::Int64: |
| 4259 | case NeonTypeFlags::Poly64: |
| 4260 | return llvm::VectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad)); |
| 4261 | case NeonTypeFlags::Poly128: |
| 4262 | // FIXME: i128 and f128 doesn't get fully support in Clang and llvm. |
| 4263 | // There is a lot of i128 and f128 API missing. |
| 4264 | // so we use v16i8 to represent poly128 and get pattern matched. |
| 4265 | return llvm::VectorType::get(CGF->Int8Ty, 16); |
| 4266 | case NeonTypeFlags::Float32: |
| 4267 | return llvm::VectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad)); |
| 4268 | case NeonTypeFlags::Float64: |
| 4269 | return llvm::VectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad)); |
| 4270 | } |
| 4271 | llvm_unreachable("Unknown vector element type!")::llvm::llvm_unreachable_internal("Unknown vector element type!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4271); |
| 4272 | } |
| 4273 | |
| 4274 | static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF, |
| 4275 | NeonTypeFlags IntTypeFlags) { |
| 4276 | int IsQuad = IntTypeFlags.isQuad(); |
| 4277 | switch (IntTypeFlags.getEltType()) { |
| 4278 | case NeonTypeFlags::Int16: |
| 4279 | return llvm::VectorType::get(CGF->HalfTy, (4 << IsQuad)); |
| 4280 | case NeonTypeFlags::Int32: |
| 4281 | return llvm::VectorType::get(CGF->FloatTy, (2 << IsQuad)); |
| 4282 | case NeonTypeFlags::Int64: |
| 4283 | return llvm::VectorType::get(CGF->DoubleTy, (1 << IsQuad)); |
| 4284 | default: |
| 4285 | llvm_unreachable("Type can't be converted to floating-point!")::llvm::llvm_unreachable_internal("Type can't be converted to floating-point!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 4285); |
| 4286 | } |
| 4287 | } |
| 4288 | |
| 4289 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) { |
| 4290 | unsigned nElts = V->getType()->getVectorNumElements(); |
| 4291 | Value* SV = llvm::ConstantVector::getSplat(nElts, C); |
| 4292 | return Builder.CreateShuffleVector(V, V, SV, "lane"); |
| 4293 | } |
| 4294 | |
| 4295 | Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops, |
| 4296 | const char *name, |
| 4297 | unsigned shift, bool rightshift) { |
| 4298 | unsigned j = 0; |
| 4299 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); |
| 4300 | ai != ae; ++ai, ++j) |
| 4301 | if (shift > 0 && shift == j) |
| 4302 | Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift); |
| 4303 | else |
| 4304 | Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name); |
| 4305 | |
| 4306 | return Builder.CreateCall(F, Ops, name); |
| 4307 | } |
| 4308 | |
| 4309 | Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty, |
| 4310 | bool neg) { |
| 4311 | int SV = cast<ConstantInt>(V)->getSExtValue(); |
| 4312 | return ConstantInt::get(Ty, neg ? -SV : SV); |
| 4313 | } |
| 4314 | |
| 4315 | // Right-shift a vector by a constant. |
| 4316 | Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift, |
| 4317 | llvm::Type *Ty, bool usgn, |
| 4318 | const char *name) { |
| 4319 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); |
| 4320 | |
| 4321 | int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue(); |
| 4322 | int EltSize = VTy->getScalarSizeInBits(); |
| 4323 | |
| 4324 | Vec = Builder.CreateBitCast(Vec, Ty); |
| 4325 | |
| 4326 | // lshr/ashr are undefined when the shift amount is equal to the vector |
| 4327 | // element size. |
| 4328 | if (ShiftAmt == EltSize) { |
| 4329 | if (usgn) { |
| 4330 | // Right-shifting an unsigned value by its size yields 0. |
| 4331 | return llvm::ConstantAggregateZero::get(VTy); |
| 4332 | } else { |
| 4333 | // Right-shifting a signed value by its size is equivalent |
| 4334 | // to a shift of size-1. |
| 4335 | --ShiftAmt; |
| 4336 | Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt); |
| 4337 | } |
| 4338 | } |
| 4339 | |
| 4340 | Shift = EmitNeonShiftVector(Shift, Ty, false); |
| 4341 | if (usgn) |
| 4342 | return Builder.CreateLShr(Vec, Shift, name); |
| 4343 | else |
| 4344 | return Builder.CreateAShr(Vec, Shift, name); |
| 4345 | } |
| 4346 | |
| 4347 | enum { |
| 4348 | AddRetType = (1 << 0), |
| 4349 | Add1ArgType = (1 << 1), |
| 4350 | Add2ArgTypes = (1 << 2), |
| 4351 | |
| 4352 | VectorizeRetType = (1 << 3), |
| 4353 | VectorizeArgTypes = (1 << 4), |
| 4354 | |
| 4355 | InventFloatType = (1 << 5), |
| 4356 | UnsignedAlts = (1 << 6), |
| 4357 | |
| 4358 | Use64BitVectors = (1 << 7), |
| 4359 | Use128BitVectors = (1 << 8), |
| 4360 | |
| 4361 | Vectorize1ArgType = Add1ArgType | VectorizeArgTypes, |
| 4362 | VectorRet = AddRetType | VectorizeRetType, |
| 4363 | VectorRetGetArgs01 = |
| 4364 | AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes, |
| 4365 | FpCmpzModifiers = |
| 4366 | AddRetType | VectorizeRetType | Add1ArgType | InventFloatType |
| 4367 | }; |
| 4368 | |
| 4369 | namespace { |
| 4370 | struct NeonIntrinsicInfo { |
| 4371 | const char *NameHint; |
| 4372 | unsigned BuiltinID; |
| 4373 | unsigned LLVMIntrinsic; |
| 4374 | unsigned AltLLVMIntrinsic; |
| 4375 | unsigned TypeModifier; |
| 4376 | |
| 4377 | bool operator<(unsigned RHSBuiltinID) const { |
| 4378 | return BuiltinID < RHSBuiltinID; |
| 4379 | } |
| 4380 | bool operator<(const NeonIntrinsicInfo &TE) const { |
| 4381 | return BuiltinID < TE.BuiltinID; |
| 4382 | } |
| 4383 | }; |
| 4384 | } // end anonymous namespace |
| 4385 | |
| 4386 | #define NEONMAP0(NameBase) \ |
| 4387 | { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 } |
| 4388 | |
| 4389 | #define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \ |
| 4390 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
| 4391 | Intrinsic::LLVMIntrinsic, 0, TypeModifier } |
| 4392 | |
| 4393 | #define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \ |
| 4394 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
| 4395 | Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \ |
| 4396 | TypeModifier } |
| 4397 | |
| 4398 | static const NeonIntrinsicInfo ARMSIMDIntrinsicMap [] = { |
| 4399 | NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
| 4400 | NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
| 4401 | NEONMAP1(vabs_v, arm_neon_vabs, 0), |
| 4402 | NEONMAP1(vabsq_v, arm_neon_vabs, 0), |
| 4403 | NEONMAP0(vaddhn_v), |
| 4404 | NEONMAP1(vaesdq_v, arm_neon_aesd, 0), |
| 4405 | NEONMAP1(vaeseq_v, arm_neon_aese, 0), |
| 4406 | NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0), |
| 4407 | NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0), |
| 4408 | NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType), |
| 4409 | NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType), |
| 4410 | NEONMAP1(vcage_v, arm_neon_vacge, 0), |
| 4411 | NEONMAP1(vcageq_v, arm_neon_vacge, 0), |
| 4412 | NEONMAP1(vcagt_v, arm_neon_vacgt, 0), |
| 4413 | NEONMAP1(vcagtq_v, arm_neon_vacgt, 0), |
| 4414 | NEONMAP1(vcale_v, arm_neon_vacge, 0), |
| 4415 | NEONMAP1(vcaleq_v, arm_neon_vacge, 0), |
| 4416 | NEONMAP1(vcalt_v, arm_neon_vacgt, 0), |
| 4417 | NEONMAP1(vcaltq_v, arm_neon_vacgt, 0), |
| 4418 | NEONMAP0(vceqz_v), |
| 4419 | NEONMAP0(vceqzq_v), |
| 4420 | NEONMAP0(vcgez_v), |
| 4421 | NEONMAP0(vcgezq_v), |
| 4422 | NEONMAP0(vcgtz_v), |
| 4423 | NEONMAP0(vcgtzq_v), |
| 4424 | NEONMAP0(vclez_v), |
| 4425 | NEONMAP0(vclezq_v), |
| 4426 | NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType), |
| 4427 | NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType), |
| 4428 | NEONMAP0(vcltz_v), |
| 4429 | NEONMAP0(vcltzq_v), |
| 4430 | NEONMAP1(vclz_v, ctlz, Add1ArgType), |
| 4431 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), |
| 4432 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), |
| 4433 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), |
| 4434 | NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0), |
| 4435 | NEONMAP0(vcvt_f16_v), |
| 4436 | NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0), |
| 4437 | NEONMAP0(vcvt_f32_v), |
| 4438 | NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 4439 | NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 4440 | NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0), |
| 4441 | NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
| 4442 | NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
| 4443 | NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0), |
| 4444 | NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
| 4445 | NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
| 4446 | NEONMAP0(vcvt_s16_v), |
| 4447 | NEONMAP0(vcvt_s32_v), |
| 4448 | NEONMAP0(vcvt_s64_v), |
| 4449 | NEONMAP0(vcvt_u16_v), |
| 4450 | NEONMAP0(vcvt_u32_v), |
| 4451 | NEONMAP0(vcvt_u64_v), |
| 4452 | NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0), |
| 4453 | NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0), |
| 4454 | NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0), |
| 4455 | NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0), |
| 4456 | NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0), |
| 4457 | NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0), |
| 4458 | NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0), |
| 4459 | NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0), |
| 4460 | NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0), |
| 4461 | NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0), |
| 4462 | NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0), |
| 4463 | NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0), |
| 4464 | NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0), |
| 4465 | NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0), |
| 4466 | NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0), |
| 4467 | NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0), |
| 4468 | NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0), |
| 4469 | NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0), |
| 4470 | NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0), |
| 4471 | NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0), |
| 4472 | NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0), |
| 4473 | NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0), |
| 4474 | NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0), |
| 4475 | NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0), |
| 4476 | NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0), |
| 4477 | NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0), |
| 4478 | NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0), |
| 4479 | NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0), |
| 4480 | NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0), |
| 4481 | NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0), |
| 4482 | NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0), |
| 4483 | NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0), |
| 4484 | NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0), |
| 4485 | NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0), |
| 4486 | NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0), |
| 4487 | NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0), |
| 4488 | NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0), |
| 4489 | NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0), |
| 4490 | NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0), |
| 4491 | NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0), |
| 4492 | NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0), |
| 4493 | NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0), |
| 4494 | NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0), |
| 4495 | NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0), |
| 4496 | NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0), |
| 4497 | NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0), |
| 4498 | NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0), |
| 4499 | NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0), |
| 4500 | NEONMAP0(vcvtq_f16_v), |
| 4501 | NEONMAP0(vcvtq_f32_v), |
| 4502 | NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 4503 | NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 4504 | NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0), |
| 4505 | NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
| 4506 | NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
| 4507 | NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0), |
| 4508 | NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
| 4509 | NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
| 4510 | NEONMAP0(vcvtq_s16_v), |
| 4511 | NEONMAP0(vcvtq_s32_v), |
| 4512 | NEONMAP0(vcvtq_s64_v), |
| 4513 | NEONMAP0(vcvtq_u16_v), |
| 4514 | NEONMAP0(vcvtq_u32_v), |
| 4515 | NEONMAP0(vcvtq_u64_v), |
| 4516 | NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0), |
| 4517 | NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0), |
| 4518 | NEONMAP0(vext_v), |
| 4519 | NEONMAP0(vextq_v), |
| 4520 | NEONMAP0(vfma_v), |
| 4521 | NEONMAP0(vfmaq_v), |
| 4522 | NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), |
| 4523 | NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), |
| 4524 | NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), |
| 4525 | NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), |
| 4526 | NEONMAP0(vld1_dup_v), |
| 4527 | NEONMAP1(vld1_v, arm_neon_vld1, 0), |
| 4528 | NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0), |
| 4529 | NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0), |
| 4530 | NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0), |
| 4531 | NEONMAP0(vld1q_dup_v), |
| 4532 | NEONMAP1(vld1q_v, arm_neon_vld1, 0), |
| 4533 | NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0), |
| 4534 | NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0), |
| 4535 | NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0), |
| 4536 | NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0), |
| 4537 | NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0), |
| 4538 | NEONMAP1(vld2_v, arm_neon_vld2, 0), |
| 4539 | NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0), |
| 4540 | NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0), |
| 4541 | NEONMAP1(vld2q_v, arm_neon_vld2, 0), |
| 4542 | NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0), |
| 4543 | NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0), |
| 4544 | NEONMAP1(vld3_v, arm_neon_vld3, 0), |
| 4545 | NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0), |
| 4546 | NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0), |
| 4547 | NEONMAP1(vld3q_v, arm_neon_vld3, 0), |
| 4548 | NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0), |
| 4549 | NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0), |
| 4550 | NEONMAP1(vld4_v, arm_neon_vld4, 0), |
| 4551 | NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0), |
| 4552 | NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0), |
| 4553 | NEONMAP1(vld4q_v, arm_neon_vld4, 0), |
| 4554 | NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), |
| 4555 | NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType), |
| 4556 | NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType), |
| 4557 | NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), |
| 4558 | NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), |
| 4559 | NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType), |
| 4560 | NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType), |
| 4561 | NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), |
| 4562 | NEONMAP0(vmovl_v), |
| 4563 | NEONMAP0(vmovn_v), |
| 4564 | NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType), |
| 4565 | NEONMAP0(vmull_v), |
| 4566 | NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType), |
| 4567 | NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), |
| 4568 | NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), |
| 4569 | NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType), |
| 4570 | NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), |
| 4571 | NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), |
| 4572 | NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType), |
| 4573 | NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts), |
| 4574 | NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts), |
| 4575 | NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType), |
| 4576 | NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType), |
| 4577 | NEONMAP2(vqadd_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts), |
| 4578 | NEONMAP2(vqaddq_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts), |
| 4579 | NEONMAP2(vqdmlal_v, arm_neon_vqdmull, arm_neon_vqadds, 0), |
| 4580 | NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, arm_neon_vqsubs, 0), |
| 4581 | NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType), |
| 4582 | NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType), |
| 4583 | NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType), |
| 4584 | NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts), |
| 4585 | NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType), |
| 4586 | NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType), |
| 4587 | NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType), |
| 4588 | NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType), |
| 4589 | NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType), |
| 4590 | NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), |
| 4591 | NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), |
| 4592 | NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), |
| 4593 | NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), |
| 4594 | NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), |
| 4595 | NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), |
| 4596 | NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0), |
| 4597 | NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0), |
| 4598 | NEONMAP2(vqsub_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts), |
| 4599 | NEONMAP2(vqsubq_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts), |
| 4600 | NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType), |
| 4601 | NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), |
| 4602 | NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), |
| 4603 | NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType), |
| 4604 | NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType), |
| 4605 | NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), |
| 4606 | NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), |
| 4607 | NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType), |
| 4608 | NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType), |
| 4609 | NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType), |
| 4610 | NEONMAP0(vrndi_v), |
| 4611 | NEONMAP0(vrndiq_v), |
| 4612 | NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType), |
| 4613 | NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType), |
| 4614 | NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType), |
| 4615 | NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType), |
| 4616 | NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType), |
| 4617 | NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType), |
| 4618 | NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType), |
| 4619 | NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType), |
| 4620 | NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType), |
| 4621 | NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), |
| 4622 | NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), |
| 4623 | NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), |
| 4624 | NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), |
| 4625 | NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), |
| 4626 | NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), |
| 4627 | NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType), |
| 4628 | NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType), |
| 4629 | NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType), |
| 4630 | NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0), |
| 4631 | NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0), |
| 4632 | NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0), |
| 4633 | NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0), |
| 4634 | NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0), |
| 4635 | NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0), |
| 4636 | NEONMAP0(vshl_n_v), |
| 4637 | NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), |
| 4638 | NEONMAP0(vshll_n_v), |
| 4639 | NEONMAP0(vshlq_n_v), |
| 4640 | NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), |
| 4641 | NEONMAP0(vshr_n_v), |
| 4642 | NEONMAP0(vshrn_n_v), |
| 4643 | NEONMAP0(vshrq_n_v), |
| 4644 | NEONMAP1(vst1_v, arm_neon_vst1, 0), |
| 4645 | NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0), |
| 4646 | NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0), |
| 4647 | NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0), |
| 4648 | NEONMAP1(vst1q_v, arm_neon_vst1, 0), |
| 4649 | NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0), |
| 4650 | NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0), |
| 4651 | NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0), |
| 4652 | NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0), |
| 4653 | NEONMAP1(vst2_v, arm_neon_vst2, 0), |
| 4654 | NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0), |
| 4655 | NEONMAP1(vst2q_v, arm_neon_vst2, 0), |
| 4656 | NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0), |
| 4657 | NEONMAP1(vst3_v, arm_neon_vst3, 0), |
| 4658 | NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0), |
| 4659 | NEONMAP1(vst3q_v, arm_neon_vst3, 0), |
| 4660 | NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0), |
| 4661 | NEONMAP1(vst4_v, arm_neon_vst4, 0), |
| 4662 | NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0), |
| 4663 | NEONMAP1(vst4q_v, arm_neon_vst4, 0), |
| 4664 | NEONMAP0(vsubhn_v), |
| 4665 | NEONMAP0(vtrn_v), |
| 4666 | NEONMAP0(vtrnq_v), |
| 4667 | NEONMAP0(vtst_v), |
| 4668 | NEONMAP0(vtstq_v), |
| 4669 | NEONMAP0(vuzp_v), |
| 4670 | NEONMAP0(vuzpq_v), |
| 4671 | NEONMAP0(vzip_v), |
| 4672 | NEONMAP0(vzipq_v) |
| 4673 | }; |
| 4674 | |
| 4675 | static const NeonIntrinsicInfo AArch64SIMDIntrinsicMap[] = { |
| 4676 | NEONMAP1(vabs_v, aarch64_neon_abs, 0), |
| 4677 | NEONMAP1(vabsq_v, aarch64_neon_abs, 0), |
| 4678 | NEONMAP0(vaddhn_v), |
| 4679 | NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0), |
| 4680 | NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0), |
| 4681 | NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0), |
| 4682 | NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0), |
| 4683 | NEONMAP1(vcage_v, aarch64_neon_facge, 0), |
| 4684 | NEONMAP1(vcageq_v, aarch64_neon_facge, 0), |
| 4685 | NEONMAP1(vcagt_v, aarch64_neon_facgt, 0), |
| 4686 | NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0), |
| 4687 | NEONMAP1(vcale_v, aarch64_neon_facge, 0), |
| 4688 | NEONMAP1(vcaleq_v, aarch64_neon_facge, 0), |
| 4689 | NEONMAP1(vcalt_v, aarch64_neon_facgt, 0), |
| 4690 | NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0), |
| 4691 | NEONMAP0(vceqz_v), |
| 4692 | NEONMAP0(vceqzq_v), |
| 4693 | NEONMAP0(vcgez_v), |
| 4694 | NEONMAP0(vcgezq_v), |
| 4695 | NEONMAP0(vcgtz_v), |
| 4696 | NEONMAP0(vcgtzq_v), |
| 4697 | NEONMAP0(vclez_v), |
| 4698 | NEONMAP0(vclezq_v), |
| 4699 | NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType), |
| 4700 | NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType), |
| 4701 | NEONMAP0(vcltz_v), |
| 4702 | NEONMAP0(vcltzq_v), |
| 4703 | NEONMAP1(vclz_v, ctlz, Add1ArgType), |
| 4704 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), |
| 4705 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), |
| 4706 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), |
| 4707 | NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0), |
| 4708 | NEONMAP0(vcvt_f16_v), |
| 4709 | NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0), |
| 4710 | NEONMAP0(vcvt_f32_v), |
| 4711 | NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 4712 | NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 4713 | NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 4714 | NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), |
| 4715 | NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), |
| 4716 | NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), |
| 4717 | NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), |
| 4718 | NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), |
| 4719 | NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), |
| 4720 | NEONMAP0(vcvtq_f16_v), |
| 4721 | NEONMAP0(vcvtq_f32_v), |
| 4722 | NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 4723 | NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 4724 | NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 4725 | NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), |
| 4726 | NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), |
| 4727 | NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), |
| 4728 | NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), |
| 4729 | NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), |
| 4730 | NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), |
| 4731 | NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType), |
| 4732 | NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0), |
| 4733 | NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0), |
| 4734 | NEONMAP0(vext_v), |
| 4735 | NEONMAP0(vextq_v), |
| 4736 | NEONMAP0(vfma_v), |
| 4737 | NEONMAP0(vfmaq_v), |
| 4738 | NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0), |
| 4739 | NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0), |
| 4740 | NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0), |
| 4741 | NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0), |
| 4742 | NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0), |
| 4743 | NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0), |
| 4744 | NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0), |
| 4745 | NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0), |
| 4746 | NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), |
| 4747 | NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), |
| 4748 | NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), |
| 4749 | NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), |
| 4750 | NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0), |
| 4751 | NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0), |
| 4752 | NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0), |
| 4753 | NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0), |
| 4754 | NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0), |
| 4755 | NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0), |
| 4756 | NEONMAP0(vmovl_v), |
| 4757 | NEONMAP0(vmovn_v), |
| 4758 | NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType), |
| 4759 | NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType), |
| 4760 | NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType), |
| 4761 | NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), |
| 4762 | NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), |
| 4763 | NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType), |
| 4764 | NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType), |
| 4765 | NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType), |
| 4766 | NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), |
| 4767 | NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), |
| 4768 | NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0), |
| 4769 | NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0), |
| 4770 | NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType), |
| 4771 | NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType), |
| 4772 | NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType), |
| 4773 | NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts), |
| 4774 | NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType), |
| 4775 | NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType), |
| 4776 | NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType), |
| 4777 | NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType), |
| 4778 | NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType), |
| 4779 | NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), |
| 4780 | NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), |
| 4781 | NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts), |
| 4782 | NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), |
| 4783 | NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts), |
| 4784 | NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), |
| 4785 | NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0), |
| 4786 | NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0), |
| 4787 | NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), |
| 4788 | NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), |
| 4789 | NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType), |
| 4790 | NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), |
| 4791 | NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), |
| 4792 | NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType), |
| 4793 | NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType), |
| 4794 | NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), |
| 4795 | NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), |
| 4796 | NEONMAP0(vrndi_v), |
| 4797 | NEONMAP0(vrndiq_v), |
| 4798 | NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), |
| 4799 | NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), |
| 4800 | NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), |
| 4801 | NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), |
| 4802 | NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), |
| 4803 | NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), |
| 4804 | NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType), |
| 4805 | NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType), |
| 4806 | NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType), |
| 4807 | NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0), |
| 4808 | NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0), |
| 4809 | NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0), |
| 4810 | NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0), |
| 4811 | NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0), |
| 4812 | NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0), |
| 4813 | NEONMAP0(vshl_n_v), |
| 4814 | NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), |
| 4815 | NEONMAP0(vshll_n_v), |
| 4816 | NEONMAP0(vshlq_n_v), |
| 4817 | NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), |
| 4818 | NEONMAP0(vshr_n_v), |
| 4819 | NEONMAP0(vshrn_n_v), |
| 4820 | NEONMAP0(vshrq_n_v), |
| 4821 | NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0), |
| 4822 | NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0), |
| 4823 | NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0), |
| 4824 | NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0), |
| 4825 | NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0), |
| 4826 | NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0), |
| 4827 | NEONMAP0(vsubhn_v), |
| 4828 | NEONMAP0(vtst_v), |
| 4829 | NEONMAP0(vtstq_v), |
| 4830 | }; |
| 4831 | |
| 4832 | static const NeonIntrinsicInfo AArch64SISDIntrinsicMap[] = { |
| 4833 | NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType), |
| 4834 | NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType), |
| 4835 | NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType), |
| 4836 | NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), |
| 4837 | NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), |
| 4838 | NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), |
| 4839 | NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), |
| 4840 | NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), |
| 4841 | NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), |
| 4842 | NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 4843 | NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), |
| 4844 | NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType), |
| 4845 | NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), |
| 4846 | NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType), |
| 4847 | NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 4848 | NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 4849 | NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 4850 | NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 4851 | NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 4852 | NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 4853 | NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 4854 | NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 4855 | NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 4856 | NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 4857 | NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
| 4858 | NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
| 4859 | NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
| 4860 | NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
| 4861 | NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
| 4862 | NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
| 4863 | NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
| 4864 | NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
| 4865 | NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
| 4866 | NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
| 4867 | NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
| 4868 | NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
| 4869 | NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
| 4870 | NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
| 4871 | NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
| 4872 | NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
| 4873 | NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
| 4874 | NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
| 4875 | NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
| 4876 | NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
| 4877 | NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
| 4878 | NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
| 4879 | NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
| 4880 | NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
| 4881 | NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0), |
| 4882 | NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 4883 | NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 4884 | NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 4885 | NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 4886 | NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), |
| 4887 | NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), |
| 4888 | NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 4889 | NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 4890 | NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), |
| 4891 | NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), |
| 4892 | NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 4893 | NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 4894 | NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 4895 | NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 4896 | NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), |
| 4897 | NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), |
| 4898 | NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 4899 | NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 4900 | NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), |
| 4901 | NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), |
| 4902 | NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0), |
| 4903 | NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType), |
| 4904 | NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType), |
| 4905 | NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 4906 | NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 4907 | NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 4908 | NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 4909 | NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 4910 | NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 4911 | NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 4912 | NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 4913 | NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 4914 | NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 4915 | NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), |
| 4916 | NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType), |
| 4917 | NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), |
| 4918 | NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType), |
| 4919 | NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), |
| 4920 | NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), |
| 4921 | NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType), |
| 4922 | NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType), |
| 4923 | NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), |
| 4924 | NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), |
| 4925 | NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType), |
| 4926 | NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType), |
| 4927 | NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors), |
| 4928 | NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType), |
| 4929 | NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors), |
| 4930 | NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0), |
| 4931 | NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType), |
| 4932 | NEONMAP1(vqmovnd_u64, aarch64_neon_scalar_uqxtn, AddRetType | Add1ArgType), |
| 4933 | NEONMAP1(vqmovnh_s16, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), |
| 4934 | NEONMAP1(vqmovnh_u16, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), |
| 4935 | NEONMAP1(vqmovns_s32, aarch64_neon_sqxtn, VectorRet | Use64BitVectors), |
| 4936 | NEONMAP1(vqmovns_u32, aarch64_neon_uqxtn, VectorRet | Use64BitVectors), |
| 4937 | NEONMAP1(vqmovund_s64, aarch64_neon_scalar_sqxtun, AddRetType | Add1ArgType), |
| 4938 | NEONMAP1(vqmovunh_s16, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), |
| 4939 | NEONMAP1(vqmovuns_s32, aarch64_neon_sqxtun, VectorRet | Use64BitVectors), |
| 4940 | NEONMAP1(vqnegb_s8, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), |
| 4941 | NEONMAP1(vqnegd_s64, aarch64_neon_sqneg, Add1ArgType), |
| 4942 | NEONMAP1(vqnegh_s16, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors), |
| 4943 | NEONMAP1(vqnegs_s32, aarch64_neon_sqneg, Add1ArgType), |
| 4944 | NEONMAP1(vqrdmulhh_s16, aarch64_neon_sqrdmulh, Vectorize1ArgType | Use64BitVectors), |
| 4945 | NEONMAP1(vqrdmulhs_s32, aarch64_neon_sqrdmulh, Add1ArgType), |
| 4946 | NEONMAP1(vqrshlb_s8, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), |
| 4947 | NEONMAP1(vqrshlb_u8, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), |
| 4948 | NEONMAP1(vqrshld_s64, aarch64_neon_sqrshl, Add1ArgType), |
| 4949 | NEONMAP1(vqrshld_u64, aarch64_neon_uqrshl, Add1ArgType), |
| 4950 | NEONMAP1(vqrshlh_s16, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors), |
| 4951 | NEONMAP1(vqrshlh_u16, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors), |
| 4952 | NEONMAP1(vqrshls_s32, aarch64_neon_sqrshl, Add1ArgType), |
| 4953 | NEONMAP1(vqrshls_u32, aarch64_neon_uqrshl, Add1ArgType), |
| 4954 | NEONMAP1(vqrshrnd_n_s64, aarch64_neon_sqrshrn, AddRetType), |
| 4955 | NEONMAP1(vqrshrnd_n_u64, aarch64_neon_uqrshrn, AddRetType), |
| 4956 | NEONMAP1(vqrshrnh_n_s16, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), |
| 4957 | NEONMAP1(vqrshrnh_n_u16, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), |
| 4958 | NEONMAP1(vqrshrns_n_s32, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors), |
| 4959 | NEONMAP1(vqrshrns_n_u32, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors), |
| 4960 | NEONMAP1(vqrshrund_n_s64, aarch64_neon_sqrshrun, AddRetType), |
| 4961 | NEONMAP1(vqrshrunh_n_s16, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), |
| 4962 | NEONMAP1(vqrshruns_n_s32, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors), |
| 4963 | NEONMAP1(vqshlb_n_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
| 4964 | NEONMAP1(vqshlb_n_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
| 4965 | NEONMAP1(vqshlb_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
| 4966 | NEONMAP1(vqshlb_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
| 4967 | NEONMAP1(vqshld_s64, aarch64_neon_sqshl, Add1ArgType), |
| 4968 | NEONMAP1(vqshld_u64, aarch64_neon_uqshl, Add1ArgType), |
| 4969 | NEONMAP1(vqshlh_n_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
| 4970 | NEONMAP1(vqshlh_n_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
| 4971 | NEONMAP1(vqshlh_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors), |
| 4972 | NEONMAP1(vqshlh_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors), |
| 4973 | NEONMAP1(vqshls_n_s32, aarch64_neon_sqshl, Add1ArgType), |
| 4974 | NEONMAP1(vqshls_n_u32, aarch64_neon_uqshl, Add1ArgType), |
| 4975 | NEONMAP1(vqshls_s32, aarch64_neon_sqshl, Add1ArgType), |
| 4976 | NEONMAP1(vqshls_u32, aarch64_neon_uqshl, Add1ArgType), |
| 4977 | NEONMAP1(vqshlub_n_s8, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), |
| 4978 | NEONMAP1(vqshluh_n_s16, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors), |
| 4979 | NEONMAP1(vqshlus_n_s32, aarch64_neon_sqshlu, Add1ArgType), |
| 4980 | NEONMAP1(vqshrnd_n_s64, aarch64_neon_sqshrn, AddRetType), |
| 4981 | NEONMAP1(vqshrnd_n_u64, aarch64_neon_uqshrn, AddRetType), |
| 4982 | NEONMAP1(vqshrnh_n_s16, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), |
| 4983 | NEONMAP1(vqshrnh_n_u16, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), |
| 4984 | NEONMAP1(vqshrns_n_s32, aarch64_neon_sqshrn, VectorRet | Use64BitVectors), |
| 4985 | NEONMAP1(vqshrns_n_u32, aarch64_neon_uqshrn, VectorRet | Use64BitVectors), |
| 4986 | NEONMAP1(vqshrund_n_s64, aarch64_neon_sqshrun, AddRetType), |
| 4987 | NEONMAP1(vqshrunh_n_s16, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), |
| 4988 | NEONMAP1(vqshruns_n_s32, aarch64_neon_sqshrun, VectorRet | Use64BitVectors), |
| 4989 | NEONMAP1(vqsubb_s8, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors), |
| 4990 | NEONMAP1(vqsubb_u8, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors), |
| 4991 | NEONMAP1(vqsubd_s64, aarch64_neon_sqsub, Add1ArgType), |
| 4992 | NEONMAP1(vqsubd_u64, aarch64_neon_uqsub, Add1ArgType), |
| 4993 | NEONMAP1(vqsubh_s16, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors), |
| 4994 | NEONMAP1(vqsubh_u16, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors), |
| 4995 | NEONMAP1(vqsubs_s32, aarch64_neon_sqsub, Add1ArgType), |
| 4996 | NEONMAP1(vqsubs_u32, aarch64_neon_uqsub, Add1ArgType), |
| 4997 | NEONMAP1(vrecped_f64, aarch64_neon_frecpe, Add1ArgType), |
| 4998 | NEONMAP1(vrecpes_f32, aarch64_neon_frecpe, Add1ArgType), |
| 4999 | NEONMAP1(vrecpxd_f64, aarch64_neon_frecpx, Add1ArgType), |
| 5000 | NEONMAP1(vrecpxs_f32, aarch64_neon_frecpx, Add1ArgType), |
| 5001 | NEONMAP1(vrshld_s64, aarch64_neon_srshl, Add1ArgType), |
| 5002 | NEONMAP1(vrshld_u64, aarch64_neon_urshl, Add1ArgType), |
| 5003 | NEONMAP1(vrsqrted_f64, aarch64_neon_frsqrte, Add1ArgType), |
| 5004 | NEONMAP1(vrsqrtes_f32, aarch64_neon_frsqrte, Add1ArgType), |
| 5005 | NEONMAP1(vrsqrtsd_f64, aarch64_neon_frsqrts, Add1ArgType), |
| 5006 | NEONMAP1(vrsqrtss_f32, aarch64_neon_frsqrts, Add1ArgType), |
| 5007 | NEONMAP1(vsha1cq_u32, aarch64_crypto_sha1c, 0), |
| 5008 | NEONMAP1(vsha1h_u32, aarch64_crypto_sha1h, 0), |
| 5009 | NEONMAP1(vsha1mq_u32, aarch64_crypto_sha1m, 0), |
| 5010 | NEONMAP1(vsha1pq_u32, aarch64_crypto_sha1p, 0), |
| 5011 | NEONMAP1(vshld_s64, aarch64_neon_sshl, Add1ArgType), |
| 5012 | NEONMAP1(vshld_u64, aarch64_neon_ushl, Add1ArgType), |
| 5013 | NEONMAP1(vslid_n_s64, aarch64_neon_vsli, Vectorize1ArgType), |
| 5014 | NEONMAP1(vslid_n_u64, aarch64_neon_vsli, Vectorize1ArgType), |
| 5015 | NEONMAP1(vsqaddb_u8, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors), |
| 5016 | NEONMAP1(vsqaddd_u64, aarch64_neon_usqadd, Add1ArgType), |
| 5017 | NEONMAP1(vsqaddh_u16, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors), |
| 5018 | NEONMAP1(vsqadds_u32, aarch64_neon_usqadd, Add1ArgType), |
| 5019 | NEONMAP1(vsrid_n_s64, aarch64_neon_vsri, Vectorize1ArgType), |
| 5020 | NEONMAP1(vsrid_n_u64, aarch64_neon_vsri, Vectorize1ArgType), |
| 5021 | NEONMAP1(vuqaddb_s8, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors), |
| 5022 | NEONMAP1(vuqaddd_s64, aarch64_neon_suqadd, Add1ArgType), |
| 5023 | NEONMAP1(vuqaddh_s16, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors), |
| 5024 | NEONMAP1(vuqadds_s32, aarch64_neon_suqadd, Add1ArgType), |
| 5025 | // FP16 scalar intrinisics go here. |
| 5026 | NEONMAP1(vabdh_f16, aarch64_sisd_fabd, Add1ArgType), |
| 5027 | NEONMAP1(vcvtah_s32_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
| 5028 | NEONMAP1(vcvtah_s64_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
| 5029 | NEONMAP1(vcvtah_u32_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
| 5030 | NEONMAP1(vcvtah_u64_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
| 5031 | NEONMAP1(vcvth_n_f16_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
| 5032 | NEONMAP1(vcvth_n_f16_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
| 5033 | NEONMAP1(vcvth_n_f16_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
| 5034 | NEONMAP1(vcvth_n_f16_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
| 5035 | NEONMAP1(vcvth_n_s32_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
| 5036 | NEONMAP1(vcvth_n_s64_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
| 5037 | NEONMAP1(vcvth_n_u32_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
| 5038 | NEONMAP1(vcvth_n_u64_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
| 5039 | NEONMAP1(vcvtmh_s32_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
| 5040 | NEONMAP1(vcvtmh_s64_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
| 5041 | NEONMAP1(vcvtmh_u32_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
| 5042 | NEONMAP1(vcvtmh_u64_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
| 5043 | NEONMAP1(vcvtnh_s32_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
| 5044 | NEONMAP1(vcvtnh_s64_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
| 5045 | NEONMAP1(vcvtnh_u32_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
| 5046 | NEONMAP1(vcvtnh_u64_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
| 5047 | NEONMAP1(vcvtph_s32_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
| 5048 | NEONMAP1(vcvtph_s64_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
| 5049 | NEONMAP1(vcvtph_u32_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
| 5050 | NEONMAP1(vcvtph_u64_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
| 5051 | NEONMAP1(vmulxh_f16, aarch64_neon_fmulx, Add1ArgType), |
| 5052 | NEONMAP1(vrecpeh_f16, aarch64_neon_frecpe, Add1ArgType), |
| 5053 | NEONMAP1(vrecpxh_f16, aarch64_neon_frecpx, Add1ArgType), |
| 5054 | NEONMAP1(vrsqrteh_f16, aarch64_neon_frsqrte, Add1ArgType), |
| 5055 | NEONMAP1(vrsqrtsh_f16, aarch64_neon_frsqrts, Add1ArgType), |
| 5056 | }; |
| 5057 | |
| 5058 | #undef NEONMAP0 |
| 5059 | #undef NEONMAP1 |
| 5060 | #undef NEONMAP2 |
| 5061 | |
| 5062 | static bool NEONSIMDIntrinsicsProvenSorted = false; |
| 5063 | |
| 5064 | static bool AArch64SIMDIntrinsicsProvenSorted = false; |
| 5065 | static bool AArch64SISDIntrinsicsProvenSorted = false; |
| 5066 | |
| 5067 | |
| 5068 | static const NeonIntrinsicInfo * |
| 5069 | findNeonIntrinsicInMap(ArrayRef<NeonIntrinsicInfo> IntrinsicMap, |
| 5070 | unsigned BuiltinID, bool &MapProvenSorted) { |
| 5071 | |
| 5072 | #ifndef NDEBUG |
| 5073 | if (!MapProvenSorted) { |
| 5074 | assert(std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap)))((std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap ))) ? static_cast<void> (0) : __assert_fail ("std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap))" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5074, __PRETTY_FUNCTION__)); |
| 5075 | MapProvenSorted = true; |
| 5076 | } |
| 5077 | #endif |
| 5078 | |
| 5079 | const NeonIntrinsicInfo *Builtin = |
| 5080 | std::lower_bound(IntrinsicMap.begin(), IntrinsicMap.end(), BuiltinID); |
| 5081 | |
| 5082 | if (Builtin != IntrinsicMap.end() && Builtin->BuiltinID == BuiltinID) |
| 5083 | return Builtin; |
| 5084 | |
| 5085 | return nullptr; |
| 5086 | } |
| 5087 | |
| 5088 | Function *CodeGenFunction::LookupNeonLLVMIntrinsic(unsigned IntrinsicID, |
| 5089 | unsigned Modifier, |
| 5090 | llvm::Type *ArgType, |
| 5091 | const CallExpr *E) { |
| 5092 | int VectorSize = 0; |
| 5093 | if (Modifier & Use64BitVectors) |
| 5094 | VectorSize = 64; |
| 5095 | else if (Modifier & Use128BitVectors) |
| 5096 | VectorSize = 128; |
| 5097 | |
| 5098 | // Return type. |
| 5099 | SmallVector<llvm::Type *, 3> Tys; |
| 5100 | if (Modifier & AddRetType) { |
| 5101 | llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext())); |
| 5102 | if (Modifier & VectorizeRetType) |
| 5103 | Ty = llvm::VectorType::get( |
| 5104 | Ty, VectorSize ? VectorSize / Ty->getPrimitiveSizeInBits() : 1); |
| 5105 | |
| 5106 | Tys.push_back(Ty); |
| 5107 | } |
| 5108 | |
| 5109 | // Arguments. |
| 5110 | if (Modifier & VectorizeArgTypes) { |
| 5111 | int Elts = VectorSize ? VectorSize / ArgType->getPrimitiveSizeInBits() : 1; |
| 5112 | ArgType = llvm::VectorType::get(ArgType, Elts); |
| 5113 | } |
| 5114 | |
| 5115 | if (Modifier & (Add1ArgType | Add2ArgTypes)) |
| 5116 | Tys.push_back(ArgType); |
| 5117 | |
| 5118 | if (Modifier & Add2ArgTypes) |
| 5119 | Tys.push_back(ArgType); |
| 5120 | |
| 5121 | if (Modifier & InventFloatType) |
| 5122 | Tys.push_back(FloatTy); |
| 5123 | |
| 5124 | return CGM.getIntrinsic(IntrinsicID, Tys); |
| 5125 | } |
| 5126 | |
| 5127 | static Value *EmitCommonNeonSISDBuiltinExpr(CodeGenFunction &CGF, |
| 5128 | const NeonIntrinsicInfo &SISDInfo, |
| 5129 | SmallVectorImpl<Value *> &Ops, |
| 5130 | const CallExpr *E) { |
| 5131 | unsigned BuiltinID = SISDInfo.BuiltinID; |
| 5132 | unsigned int Int = SISDInfo.LLVMIntrinsic; |
| 5133 | unsigned Modifier = SISDInfo.TypeModifier; |
| 5134 | const char *s = SISDInfo.NameHint; |
| 5135 | |
| 5136 | switch (BuiltinID) { |
| 5137 | case NEON::BI__builtin_neon_vcled_s64: |
| 5138 | case NEON::BI__builtin_neon_vcled_u64: |
| 5139 | case NEON::BI__builtin_neon_vcles_f32: |
| 5140 | case NEON::BI__builtin_neon_vcled_f64: |
| 5141 | case NEON::BI__builtin_neon_vcltd_s64: |
| 5142 | case NEON::BI__builtin_neon_vcltd_u64: |
| 5143 | case NEON::BI__builtin_neon_vclts_f32: |
| 5144 | case NEON::BI__builtin_neon_vcltd_f64: |
| 5145 | case NEON::BI__builtin_neon_vcales_f32: |
| 5146 | case NEON::BI__builtin_neon_vcaled_f64: |
| 5147 | case NEON::BI__builtin_neon_vcalts_f32: |
| 5148 | case NEON::BI__builtin_neon_vcaltd_f64: |
| 5149 | // Only one direction of comparisons actually exist, cmle is actually a cmge |
| 5150 | // with swapped operands. The table gives us the right intrinsic but we |
| 5151 | // still need to do the swap. |
| 5152 | std::swap(Ops[0], Ops[1]); |
| 5153 | break; |
| 5154 | } |
| 5155 | |
| 5156 | assert(Int && "Generic code assumes a valid intrinsic")((Int && "Generic code assumes a valid intrinsic") ? static_cast <void> (0) : __assert_fail ("Int && \"Generic code assumes a valid intrinsic\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5156, __PRETTY_FUNCTION__)); |
| 5157 | |
| 5158 | // Determine the type(s) of this overloaded AArch64 intrinsic. |
| 5159 | const Expr *Arg = E->getArg(0); |
| 5160 | llvm::Type *ArgTy = CGF.ConvertType(Arg->getType()); |
| 5161 | Function *F = CGF.LookupNeonLLVMIntrinsic(Int, Modifier, ArgTy, E); |
| 5162 | |
| 5163 | int j = 0; |
| 5164 | ConstantInt *C0 = ConstantInt::get(CGF.SizeTy, 0); |
| 5165 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); |
| 5166 | ai != ae; ++ai, ++j) { |
| 5167 | llvm::Type *ArgTy = ai->getType(); |
| 5168 | if (Ops[j]->getType()->getPrimitiveSizeInBits() == |
| 5169 | ArgTy->getPrimitiveSizeInBits()) |
| 5170 | continue; |
| 5171 | |
| 5172 | assert(ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy())((ArgTy->isVectorTy() && !Ops[j]->getType()-> isVectorTy()) ? static_cast<void> (0) : __assert_fail ( "ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy()" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5172, __PRETTY_FUNCTION__)); |
| 5173 | // The constant argument to an _n_ intrinsic always has Int32Ty, so truncate |
| 5174 | // it before inserting. |
| 5175 | Ops[j] = |
| 5176 | CGF.Builder.CreateTruncOrBitCast(Ops[j], ArgTy->getVectorElementType()); |
| 5177 | Ops[j] = |
| 5178 | CGF.Builder.CreateInsertElement(UndefValue::get(ArgTy), Ops[j], C0); |
| 5179 | } |
| 5180 | |
| 5181 | Value *Result = CGF.EmitNeonCall(F, Ops, s); |
| 5182 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); |
| 5183 | if (ResultType->getPrimitiveSizeInBits() < |
| 5184 | Result->getType()->getPrimitiveSizeInBits()) |
| 5185 | return CGF.Builder.CreateExtractElement(Result, C0); |
| 5186 | |
| 5187 | return CGF.Builder.CreateBitCast(Result, ResultType, s); |
| 5188 | } |
| 5189 | |
| 5190 | Value *CodeGenFunction::EmitCommonNeonBuiltinExpr( |
| 5191 | unsigned BuiltinID, unsigned LLVMIntrinsic, unsigned AltLLVMIntrinsic, |
| 5192 | const char *NameHint, unsigned Modifier, const CallExpr *E, |
| 5193 | SmallVectorImpl<llvm::Value *> &Ops, Address PtrOp0, Address PtrOp1, |
| 5194 | llvm::Triple::ArchType Arch) { |
| 5195 | // Get the last argument, which specifies the vector type. |
| 5196 | llvm::APSInt NeonTypeConst; |
| 5197 | const Expr *Arg = E->getArg(E->getNumArgs() - 1); |
| 5198 | if (!Arg->isIntegerConstantExpr(NeonTypeConst, getContext())) |
| 5199 | return nullptr; |
| 5200 | |
| 5201 | // Determine the type of this overloaded NEON intrinsic. |
| 5202 | NeonTypeFlags Type(NeonTypeConst.getZExtValue()); |
| 5203 | bool Usgn = Type.isUnsigned(); |
| 5204 | bool Quad = Type.isQuad(); |
| 5205 | const bool HasLegalHalfType = getTarget().hasLegalHalfType(); |
| 5206 | |
| 5207 | llvm::VectorType *VTy = GetNeonType(this, Type, HasLegalHalfType); |
| 5208 | llvm::Type *Ty = VTy; |
| 5209 | if (!Ty) |
| 5210 | return nullptr; |
| 5211 | |
| 5212 | auto getAlignmentValue32 = [&](Address addr) -> Value* { |
| 5213 | return Builder.getInt32(addr.getAlignment().getQuantity()); |
| 5214 | }; |
| 5215 | |
| 5216 | unsigned Int = LLVMIntrinsic; |
| 5217 | if ((Modifier & UnsignedAlts) && !Usgn) |
| 5218 | Int = AltLLVMIntrinsic; |
| 5219 | |
| 5220 | switch (BuiltinID) { |
| 5221 | default: break; |
| 5222 | case NEON::BI__builtin_neon_vpadd_v: |
| 5223 | case NEON::BI__builtin_neon_vpaddq_v: |
| 5224 | // We don't allow fp/int overloading of intrinsics. |
| 5225 | if (VTy->getElementType()->isFloatingPointTy() && |
| 5226 | Int == Intrinsic::aarch64_neon_addp) |
| 5227 | Int = Intrinsic::aarch64_neon_faddp; |
| 5228 | break; |
| 5229 | case NEON::BI__builtin_neon_vabs_v: |
| 5230 | case NEON::BI__builtin_neon_vabsq_v: |
| 5231 | if (VTy->getElementType()->isFloatingPointTy()) |
| 5232 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, Ty), Ops, "vabs"); |
| 5233 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), Ops, "vabs"); |
| 5234 | case NEON::BI__builtin_neon_vaddhn_v: { |
| 5235 | llvm::VectorType *SrcTy = |
| 5236 | llvm::VectorType::getExtendedElementVectorType(VTy); |
| 5237 | |
| 5238 | // %sum = add <4 x i32> %lhs, %rhs |
| 5239 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); |
| 5240 | Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy); |
| 5241 | Ops[0] = Builder.CreateAdd(Ops[0], Ops[1], "vaddhn"); |
| 5242 | |
| 5243 | // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16> |
| 5244 | Constant *ShiftAmt = |
| 5245 | ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2); |
| 5246 | Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vaddhn"); |
| 5247 | |
| 5248 | // %res = trunc <4 x i32> %high to <4 x i16> |
| 5249 | return Builder.CreateTrunc(Ops[0], VTy, "vaddhn"); |
| 5250 | } |
| 5251 | case NEON::BI__builtin_neon_vcale_v: |
| 5252 | case NEON::BI__builtin_neon_vcaleq_v: |
| 5253 | case NEON::BI__builtin_neon_vcalt_v: |
| 5254 | case NEON::BI__builtin_neon_vcaltq_v: |
| 5255 | std::swap(Ops[0], Ops[1]); |
| 5256 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 5257 | case NEON::BI__builtin_neon_vcage_v: |
| 5258 | case NEON::BI__builtin_neon_vcageq_v: |
| 5259 | case NEON::BI__builtin_neon_vcagt_v: |
| 5260 | case NEON::BI__builtin_neon_vcagtq_v: { |
| 5261 | llvm::Type *Ty; |
| 5262 | switch (VTy->getScalarSizeInBits()) { |
| 5263 | default: llvm_unreachable("unexpected type")::llvm::llvm_unreachable_internal("unexpected type", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5263); |
| 5264 | case 32: |
| 5265 | Ty = FloatTy; |
| 5266 | break; |
| 5267 | case 64: |
| 5268 | Ty = DoubleTy; |
| 5269 | break; |
| 5270 | case 16: |
| 5271 | Ty = HalfTy; |
| 5272 | break; |
| 5273 | } |
| 5274 | llvm::Type *VecFlt = llvm::VectorType::get(Ty, VTy->getNumElements()); |
| 5275 | llvm::Type *Tys[] = { VTy, VecFlt }; |
| 5276 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); |
| 5277 | return EmitNeonCall(F, Ops, NameHint); |
| 5278 | } |
| 5279 | case NEON::BI__builtin_neon_vceqz_v: |
| 5280 | case NEON::BI__builtin_neon_vceqzq_v: |
| 5281 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OEQ, |
| 5282 | ICmpInst::ICMP_EQ, "vceqz"); |
| 5283 | case NEON::BI__builtin_neon_vcgez_v: |
| 5284 | case NEON::BI__builtin_neon_vcgezq_v: |
| 5285 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGE, |
| 5286 | ICmpInst::ICMP_SGE, "vcgez"); |
| 5287 | case NEON::BI__builtin_neon_vclez_v: |
| 5288 | case NEON::BI__builtin_neon_vclezq_v: |
| 5289 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLE, |
| 5290 | ICmpInst::ICMP_SLE, "vclez"); |
| 5291 | case NEON::BI__builtin_neon_vcgtz_v: |
| 5292 | case NEON::BI__builtin_neon_vcgtzq_v: |
| 5293 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGT, |
| 5294 | ICmpInst::ICMP_SGT, "vcgtz"); |
| 5295 | case NEON::BI__builtin_neon_vcltz_v: |
| 5296 | case NEON::BI__builtin_neon_vcltzq_v: |
| 5297 | return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLT, |
| 5298 | ICmpInst::ICMP_SLT, "vcltz"); |
| 5299 | case NEON::BI__builtin_neon_vclz_v: |
| 5300 | case NEON::BI__builtin_neon_vclzq_v: |
| 5301 | // We generate target-independent intrinsic, which needs a second argument |
| 5302 | // for whether or not clz of zero is undefined; on ARM it isn't. |
| 5303 | Ops.push_back(Builder.getInt1(getTarget().isCLZForZeroUndef())); |
| 5304 | break; |
| 5305 | case NEON::BI__builtin_neon_vcvt_f32_v: |
| 5306 | case NEON::BI__builtin_neon_vcvtq_f32_v: |
| 5307 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5308 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float32, false, Quad), |
| 5309 | HasLegalHalfType); |
| 5310 | return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") |
| 5311 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); |
| 5312 | case NEON::BI__builtin_neon_vcvt_f16_v: |
| 5313 | case NEON::BI__builtin_neon_vcvtq_f16_v: |
| 5314 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5315 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float16, false, Quad), |
| 5316 | HasLegalHalfType); |
| 5317 | return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") |
| 5318 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); |
| 5319 | case NEON::BI__builtin_neon_vcvt_n_f16_v: |
| 5320 | case NEON::BI__builtin_neon_vcvt_n_f32_v: |
| 5321 | case NEON::BI__builtin_neon_vcvt_n_f64_v: |
| 5322 | case NEON::BI__builtin_neon_vcvtq_n_f16_v: |
| 5323 | case NEON::BI__builtin_neon_vcvtq_n_f32_v: |
| 5324 | case NEON::BI__builtin_neon_vcvtq_n_f64_v: { |
| 5325 | llvm::Type *Tys[2] = { GetFloatNeonType(this, Type), Ty }; |
| 5326 | Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic; |
| 5327 | Function *F = CGM.getIntrinsic(Int, Tys); |
| 5328 | return EmitNeonCall(F, Ops, "vcvt_n"); |
| 5329 | } |
| 5330 | case NEON::BI__builtin_neon_vcvt_n_s16_v: |
| 5331 | case NEON::BI__builtin_neon_vcvt_n_s32_v: |
| 5332 | case NEON::BI__builtin_neon_vcvt_n_u16_v: |
| 5333 | case NEON::BI__builtin_neon_vcvt_n_u32_v: |
| 5334 | case NEON::BI__builtin_neon_vcvt_n_s64_v: |
| 5335 | case NEON::BI__builtin_neon_vcvt_n_u64_v: |
| 5336 | case NEON::BI__builtin_neon_vcvtq_n_s16_v: |
| 5337 | case NEON::BI__builtin_neon_vcvtq_n_s32_v: |
| 5338 | case NEON::BI__builtin_neon_vcvtq_n_u16_v: |
| 5339 | case NEON::BI__builtin_neon_vcvtq_n_u32_v: |
| 5340 | case NEON::BI__builtin_neon_vcvtq_n_s64_v: |
| 5341 | case NEON::BI__builtin_neon_vcvtq_n_u64_v: { |
| 5342 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; |
| 5343 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); |
| 5344 | return EmitNeonCall(F, Ops, "vcvt_n"); |
| 5345 | } |
| 5346 | case NEON::BI__builtin_neon_vcvt_s32_v: |
| 5347 | case NEON::BI__builtin_neon_vcvt_u32_v: |
| 5348 | case NEON::BI__builtin_neon_vcvt_s64_v: |
| 5349 | case NEON::BI__builtin_neon_vcvt_u64_v: |
| 5350 | case NEON::BI__builtin_neon_vcvt_s16_v: |
| 5351 | case NEON::BI__builtin_neon_vcvt_u16_v: |
| 5352 | case NEON::BI__builtin_neon_vcvtq_s32_v: |
| 5353 | case NEON::BI__builtin_neon_vcvtq_u32_v: |
| 5354 | case NEON::BI__builtin_neon_vcvtq_s64_v: |
| 5355 | case NEON::BI__builtin_neon_vcvtq_u64_v: |
| 5356 | case NEON::BI__builtin_neon_vcvtq_s16_v: |
| 5357 | case NEON::BI__builtin_neon_vcvtq_u16_v: { |
| 5358 | Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type)); |
| 5359 | return Usgn ? Builder.CreateFPToUI(Ops[0], Ty, "vcvt") |
| 5360 | : Builder.CreateFPToSI(Ops[0], Ty, "vcvt"); |
| 5361 | } |
| 5362 | case NEON::BI__builtin_neon_vcvta_s16_v: |
| 5363 | case NEON::BI__builtin_neon_vcvta_s32_v: |
| 5364 | case NEON::BI__builtin_neon_vcvta_s64_v: |
| 5365 | case NEON::BI__builtin_neon_vcvta_u16_v: |
| 5366 | case NEON::BI__builtin_neon_vcvta_u32_v: |
| 5367 | case NEON::BI__builtin_neon_vcvta_u64_v: |
| 5368 | case NEON::BI__builtin_neon_vcvtaq_s16_v: |
| 5369 | case NEON::BI__builtin_neon_vcvtaq_s32_v: |
| 5370 | case NEON::BI__builtin_neon_vcvtaq_s64_v: |
| 5371 | case NEON::BI__builtin_neon_vcvtaq_u16_v: |
| 5372 | case NEON::BI__builtin_neon_vcvtaq_u32_v: |
| 5373 | case NEON::BI__builtin_neon_vcvtaq_u64_v: |
| 5374 | case NEON::BI__builtin_neon_vcvtn_s16_v: |
| 5375 | case NEON::BI__builtin_neon_vcvtn_s32_v: |
| 5376 | case NEON::BI__builtin_neon_vcvtn_s64_v: |
| 5377 | case NEON::BI__builtin_neon_vcvtn_u16_v: |
| 5378 | case NEON::BI__builtin_neon_vcvtn_u32_v: |
| 5379 | case NEON::BI__builtin_neon_vcvtn_u64_v: |
| 5380 | case NEON::BI__builtin_neon_vcvtnq_s16_v: |
| 5381 | case NEON::BI__builtin_neon_vcvtnq_s32_v: |
| 5382 | case NEON::BI__builtin_neon_vcvtnq_s64_v: |
| 5383 | case NEON::BI__builtin_neon_vcvtnq_u16_v: |
| 5384 | case NEON::BI__builtin_neon_vcvtnq_u32_v: |
| 5385 | case NEON::BI__builtin_neon_vcvtnq_u64_v: |
| 5386 | case NEON::BI__builtin_neon_vcvtp_s16_v: |
| 5387 | case NEON::BI__builtin_neon_vcvtp_s32_v: |
| 5388 | case NEON::BI__builtin_neon_vcvtp_s64_v: |
| 5389 | case NEON::BI__builtin_neon_vcvtp_u16_v: |
| 5390 | case NEON::BI__builtin_neon_vcvtp_u32_v: |
| 5391 | case NEON::BI__builtin_neon_vcvtp_u64_v: |
| 5392 | case NEON::BI__builtin_neon_vcvtpq_s16_v: |
| 5393 | case NEON::BI__builtin_neon_vcvtpq_s32_v: |
| 5394 | case NEON::BI__builtin_neon_vcvtpq_s64_v: |
| 5395 | case NEON::BI__builtin_neon_vcvtpq_u16_v: |
| 5396 | case NEON::BI__builtin_neon_vcvtpq_u32_v: |
| 5397 | case NEON::BI__builtin_neon_vcvtpq_u64_v: |
| 5398 | case NEON::BI__builtin_neon_vcvtm_s16_v: |
| 5399 | case NEON::BI__builtin_neon_vcvtm_s32_v: |
| 5400 | case NEON::BI__builtin_neon_vcvtm_s64_v: |
| 5401 | case NEON::BI__builtin_neon_vcvtm_u16_v: |
| 5402 | case NEON::BI__builtin_neon_vcvtm_u32_v: |
| 5403 | case NEON::BI__builtin_neon_vcvtm_u64_v: |
| 5404 | case NEON::BI__builtin_neon_vcvtmq_s16_v: |
| 5405 | case NEON::BI__builtin_neon_vcvtmq_s32_v: |
| 5406 | case NEON::BI__builtin_neon_vcvtmq_s64_v: |
| 5407 | case NEON::BI__builtin_neon_vcvtmq_u16_v: |
| 5408 | case NEON::BI__builtin_neon_vcvtmq_u32_v: |
| 5409 | case NEON::BI__builtin_neon_vcvtmq_u64_v: { |
| 5410 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; |
| 5411 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint); |
| 5412 | } |
| 5413 | case NEON::BI__builtin_neon_vext_v: |
| 5414 | case NEON::BI__builtin_neon_vextq_v: { |
| 5415 | int CV = cast<ConstantInt>(Ops[2])->getSExtValue(); |
| 5416 | SmallVector<uint32_t, 16> Indices; |
| 5417 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) |
| 5418 | Indices.push_back(i+CV); |
| 5419 | |
| 5420 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5421 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 5422 | return Builder.CreateShuffleVector(Ops[0], Ops[1], Indices, "vext"); |
| 5423 | } |
| 5424 | case NEON::BI__builtin_neon_vfma_v: |
| 5425 | case NEON::BI__builtin_neon_vfmaq_v: { |
| 5426 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); |
| 5427 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5428 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 5429 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 5430 | |
| 5431 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. |
| 5432 | return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]}); |
| 5433 | } |
| 5434 | case NEON::BI__builtin_neon_vld1_v: |
| 5435 | case NEON::BI__builtin_neon_vld1q_v: { |
| 5436 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; |
| 5437 | Ops.push_back(getAlignmentValue32(PtrOp0)); |
| 5438 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "vld1"); |
| 5439 | } |
| 5440 | case NEON::BI__builtin_neon_vld1_x2_v: |
| 5441 | case NEON::BI__builtin_neon_vld1q_x2_v: |
| 5442 | case NEON::BI__builtin_neon_vld1_x3_v: |
| 5443 | case NEON::BI__builtin_neon_vld1q_x3_v: |
| 5444 | case NEON::BI__builtin_neon_vld1_x4_v: |
| 5445 | case NEON::BI__builtin_neon_vld1q_x4_v: { |
| 5446 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType()); |
| 5447 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); |
| 5448 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 5449 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); |
| 5450 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld1xN"); |
| 5451 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 5452 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5453 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 5454 | } |
| 5455 | case NEON::BI__builtin_neon_vld2_v: |
| 5456 | case NEON::BI__builtin_neon_vld2q_v: |
| 5457 | case NEON::BI__builtin_neon_vld3_v: |
| 5458 | case NEON::BI__builtin_neon_vld3q_v: |
| 5459 | case NEON::BI__builtin_neon_vld4_v: |
| 5460 | case NEON::BI__builtin_neon_vld4q_v: |
| 5461 | case NEON::BI__builtin_neon_vld2_dup_v: |
| 5462 | case NEON::BI__builtin_neon_vld2q_dup_v: |
| 5463 | case NEON::BI__builtin_neon_vld3_dup_v: |
| 5464 | case NEON::BI__builtin_neon_vld3q_dup_v: |
| 5465 | case NEON::BI__builtin_neon_vld4_dup_v: |
| 5466 | case NEON::BI__builtin_neon_vld4q_dup_v: { |
| 5467 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; |
| 5468 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); |
| 5469 | Value *Align = getAlignmentValue32(PtrOp1); |
| 5470 | Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint); |
| 5471 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 5472 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5473 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 5474 | } |
| 5475 | case NEON::BI__builtin_neon_vld1_dup_v: |
| 5476 | case NEON::BI__builtin_neon_vld1q_dup_v: { |
| 5477 | Value *V = UndefValue::get(Ty); |
| 5478 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); |
| 5479 | PtrOp0 = Builder.CreateBitCast(PtrOp0, Ty); |
| 5480 | LoadInst *Ld = Builder.CreateLoad(PtrOp0); |
| 5481 | llvm::Constant *CI = ConstantInt::get(SizeTy, 0); |
| 5482 | Ops[0] = Builder.CreateInsertElement(V, Ld, CI); |
| 5483 | return EmitNeonSplat(Ops[0], CI); |
| 5484 | } |
| 5485 | case NEON::BI__builtin_neon_vld2_lane_v: |
| 5486 | case NEON::BI__builtin_neon_vld2q_lane_v: |
| 5487 | case NEON::BI__builtin_neon_vld3_lane_v: |
| 5488 | case NEON::BI__builtin_neon_vld3q_lane_v: |
| 5489 | case NEON::BI__builtin_neon_vld4_lane_v: |
| 5490 | case NEON::BI__builtin_neon_vld4q_lane_v: { |
| 5491 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; |
| 5492 | Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys); |
| 5493 | for (unsigned I = 2; I < Ops.size() - 1; ++I) |
| 5494 | Ops[I] = Builder.CreateBitCast(Ops[I], Ty); |
| 5495 | Ops.push_back(getAlignmentValue32(PtrOp1)); |
| 5496 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), NameHint); |
| 5497 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 5498 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5499 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 5500 | } |
| 5501 | case NEON::BI__builtin_neon_vmovl_v: { |
| 5502 | llvm::Type *DTy =llvm::VectorType::getTruncatedElementVectorType(VTy); |
| 5503 | Ops[0] = Builder.CreateBitCast(Ops[0], DTy); |
| 5504 | if (Usgn) |
| 5505 | return Builder.CreateZExt(Ops[0], Ty, "vmovl"); |
| 5506 | return Builder.CreateSExt(Ops[0], Ty, "vmovl"); |
| 5507 | } |
| 5508 | case NEON::BI__builtin_neon_vmovn_v: { |
| 5509 | llvm::Type *QTy = llvm::VectorType::getExtendedElementVectorType(VTy); |
| 5510 | Ops[0] = Builder.CreateBitCast(Ops[0], QTy); |
| 5511 | return Builder.CreateTrunc(Ops[0], Ty, "vmovn"); |
| 5512 | } |
| 5513 | case NEON::BI__builtin_neon_vmull_v: |
| 5514 | // FIXME: the integer vmull operations could be emitted in terms of pure |
| 5515 | // LLVM IR (2 exts followed by a mul). Unfortunately LLVM has a habit of |
| 5516 | // hoisting the exts outside loops. Until global ISel comes along that can |
| 5517 | // see through such movement this leads to bad CodeGen. So we need an |
| 5518 | // intrinsic for now. |
| 5519 | Int = Usgn ? Intrinsic::arm_neon_vmullu : Intrinsic::arm_neon_vmulls; |
| 5520 | Int = Type.isPoly() ? (unsigned)Intrinsic::arm_neon_vmullp : Int; |
| 5521 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull"); |
| 5522 | case NEON::BI__builtin_neon_vpadal_v: |
| 5523 | case NEON::BI__builtin_neon_vpadalq_v: { |
| 5524 | // The source operand type has twice as many elements of half the size. |
| 5525 | unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits(); |
| 5526 | llvm::Type *EltTy = |
| 5527 | llvm::IntegerType::get(getLLVMContext(), EltBits / 2); |
| 5528 | llvm::Type *NarrowTy = |
| 5529 | llvm::VectorType::get(EltTy, VTy->getNumElements() * 2); |
| 5530 | llvm::Type *Tys[2] = { Ty, NarrowTy }; |
| 5531 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint); |
| 5532 | } |
| 5533 | case NEON::BI__builtin_neon_vpaddl_v: |
| 5534 | case NEON::BI__builtin_neon_vpaddlq_v: { |
| 5535 | // The source operand type has twice as many elements of half the size. |
| 5536 | unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits(); |
| 5537 | llvm::Type *EltTy = llvm::IntegerType::get(getLLVMContext(), EltBits / 2); |
| 5538 | llvm::Type *NarrowTy = |
| 5539 | llvm::VectorType::get(EltTy, VTy->getNumElements() * 2); |
| 5540 | llvm::Type *Tys[2] = { Ty, NarrowTy }; |
| 5541 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vpaddl"); |
| 5542 | } |
| 5543 | case NEON::BI__builtin_neon_vqdmlal_v: |
| 5544 | case NEON::BI__builtin_neon_vqdmlsl_v: { |
| 5545 | SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end()); |
| 5546 | Ops[1] = |
| 5547 | EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), MulOps, "vqdmlal"); |
| 5548 | Ops.resize(2); |
| 5549 | return EmitNeonCall(CGM.getIntrinsic(AltLLVMIntrinsic, Ty), Ops, NameHint); |
| 5550 | } |
| 5551 | case NEON::BI__builtin_neon_vqshl_n_v: |
| 5552 | case NEON::BI__builtin_neon_vqshlq_n_v: |
| 5553 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshl_n", |
| 5554 | 1, false); |
| 5555 | case NEON::BI__builtin_neon_vqshlu_n_v: |
| 5556 | case NEON::BI__builtin_neon_vqshluq_n_v: |
| 5557 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshlu_n", |
| 5558 | 1, false); |
| 5559 | case NEON::BI__builtin_neon_vrecpe_v: |
| 5560 | case NEON::BI__builtin_neon_vrecpeq_v: |
| 5561 | case NEON::BI__builtin_neon_vrsqrte_v: |
| 5562 | case NEON::BI__builtin_neon_vrsqrteq_v: |
| 5563 | Int = Ty->isFPOrFPVectorTy() ? LLVMIntrinsic : AltLLVMIntrinsic; |
| 5564 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint); |
| 5565 | case NEON::BI__builtin_neon_vrndi_v: |
| 5566 | case NEON::BI__builtin_neon_vrndiq_v: |
| 5567 | Int = Intrinsic::nearbyint; |
| 5568 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint); |
| 5569 | case NEON::BI__builtin_neon_vrshr_n_v: |
| 5570 | case NEON::BI__builtin_neon_vrshrq_n_v: |
| 5571 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshr_n", |
| 5572 | 1, true); |
| 5573 | case NEON::BI__builtin_neon_vshl_n_v: |
| 5574 | case NEON::BI__builtin_neon_vshlq_n_v: |
| 5575 | Ops[1] = EmitNeonShiftVector(Ops[1], Ty, false); |
| 5576 | return Builder.CreateShl(Builder.CreateBitCast(Ops[0],Ty), Ops[1], |
| 5577 | "vshl_n"); |
| 5578 | case NEON::BI__builtin_neon_vshll_n_v: { |
| 5579 | llvm::Type *SrcTy = llvm::VectorType::getTruncatedElementVectorType(VTy); |
| 5580 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); |
| 5581 | if (Usgn) |
| 5582 | Ops[0] = Builder.CreateZExt(Ops[0], VTy); |
| 5583 | else |
| 5584 | Ops[0] = Builder.CreateSExt(Ops[0], VTy); |
| 5585 | Ops[1] = EmitNeonShiftVector(Ops[1], VTy, false); |
| 5586 | return Builder.CreateShl(Ops[0], Ops[1], "vshll_n"); |
| 5587 | } |
| 5588 | case NEON::BI__builtin_neon_vshrn_n_v: { |
| 5589 | llvm::Type *SrcTy = llvm::VectorType::getExtendedElementVectorType(VTy); |
| 5590 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); |
| 5591 | Ops[1] = EmitNeonShiftVector(Ops[1], SrcTy, false); |
| 5592 | if (Usgn) |
| 5593 | Ops[0] = Builder.CreateLShr(Ops[0], Ops[1]); |
| 5594 | else |
| 5595 | Ops[0] = Builder.CreateAShr(Ops[0], Ops[1]); |
| 5596 | return Builder.CreateTrunc(Ops[0], Ty, "vshrn_n"); |
| 5597 | } |
| 5598 | case NEON::BI__builtin_neon_vshr_n_v: |
| 5599 | case NEON::BI__builtin_neon_vshrq_n_v: |
| 5600 | return EmitNeonRShiftImm(Ops[0], Ops[1], Ty, Usgn, "vshr_n"); |
| 5601 | case NEON::BI__builtin_neon_vst1_v: |
| 5602 | case NEON::BI__builtin_neon_vst1q_v: |
| 5603 | case NEON::BI__builtin_neon_vst2_v: |
| 5604 | case NEON::BI__builtin_neon_vst2q_v: |
| 5605 | case NEON::BI__builtin_neon_vst3_v: |
| 5606 | case NEON::BI__builtin_neon_vst3q_v: |
| 5607 | case NEON::BI__builtin_neon_vst4_v: |
| 5608 | case NEON::BI__builtin_neon_vst4q_v: |
| 5609 | case NEON::BI__builtin_neon_vst2_lane_v: |
| 5610 | case NEON::BI__builtin_neon_vst2q_lane_v: |
| 5611 | case NEON::BI__builtin_neon_vst3_lane_v: |
| 5612 | case NEON::BI__builtin_neon_vst3q_lane_v: |
| 5613 | case NEON::BI__builtin_neon_vst4_lane_v: |
| 5614 | case NEON::BI__builtin_neon_vst4q_lane_v: { |
| 5615 | llvm::Type *Tys[] = {Int8PtrTy, Ty}; |
| 5616 | Ops.push_back(getAlignmentValue32(PtrOp0)); |
| 5617 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, ""); |
| 5618 | } |
| 5619 | case NEON::BI__builtin_neon_vst1_x2_v: |
| 5620 | case NEON::BI__builtin_neon_vst1q_x2_v: |
| 5621 | case NEON::BI__builtin_neon_vst1_x3_v: |
| 5622 | case NEON::BI__builtin_neon_vst1q_x3_v: |
| 5623 | case NEON::BI__builtin_neon_vst1_x4_v: |
| 5624 | case NEON::BI__builtin_neon_vst1q_x4_v: { |
| 5625 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType()); |
| 5626 | // TODO: Currently in AArch32 mode the pointer operand comes first, whereas |
| 5627 | // in AArch64 it comes last. We may want to stick to one or another. |
| 5628 | if (Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_be) { |
| 5629 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 5630 | std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end()); |
| 5631 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, ""); |
| 5632 | } |
| 5633 | llvm::Type *Tys[2] = { PTy, VTy }; |
| 5634 | return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, ""); |
| 5635 | } |
| 5636 | case NEON::BI__builtin_neon_vsubhn_v: { |
| 5637 | llvm::VectorType *SrcTy = |
| 5638 | llvm::VectorType::getExtendedElementVectorType(VTy); |
| 5639 | |
| 5640 | // %sum = add <4 x i32> %lhs, %rhs |
| 5641 | Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy); |
| 5642 | Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy); |
| 5643 | Ops[0] = Builder.CreateSub(Ops[0], Ops[1], "vsubhn"); |
| 5644 | |
| 5645 | // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16> |
| 5646 | Constant *ShiftAmt = |
| 5647 | ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2); |
| 5648 | Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vsubhn"); |
| 5649 | |
| 5650 | // %res = trunc <4 x i32> %high to <4 x i16> |
| 5651 | return Builder.CreateTrunc(Ops[0], VTy, "vsubhn"); |
| 5652 | } |
| 5653 | case NEON::BI__builtin_neon_vtrn_v: |
| 5654 | case NEON::BI__builtin_neon_vtrnq_v: { |
| 5655 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); |
| 5656 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 5657 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 5658 | Value *SV = nullptr; |
| 5659 | |
| 5660 | for (unsigned vi = 0; vi != 2; ++vi) { |
| 5661 | SmallVector<uint32_t, 16> Indices; |
| 5662 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { |
| 5663 | Indices.push_back(i+vi); |
| 5664 | Indices.push_back(i+e+vi); |
| 5665 | } |
| 5666 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); |
| 5667 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn"); |
| 5668 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); |
| 5669 | } |
| 5670 | return SV; |
| 5671 | } |
| 5672 | case NEON::BI__builtin_neon_vtst_v: |
| 5673 | case NEON::BI__builtin_neon_vtstq_v: { |
| 5674 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 5675 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 5676 | Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]); |
| 5677 | Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0], |
| 5678 | ConstantAggregateZero::get(Ty)); |
| 5679 | return Builder.CreateSExt(Ops[0], Ty, "vtst"); |
| 5680 | } |
| 5681 | case NEON::BI__builtin_neon_vuzp_v: |
| 5682 | case NEON::BI__builtin_neon_vuzpq_v: { |
| 5683 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); |
| 5684 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 5685 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 5686 | Value *SV = nullptr; |
| 5687 | |
| 5688 | for (unsigned vi = 0; vi != 2; ++vi) { |
| 5689 | SmallVector<uint32_t, 16> Indices; |
| 5690 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) |
| 5691 | Indices.push_back(2*i+vi); |
| 5692 | |
| 5693 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); |
| 5694 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp"); |
| 5695 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); |
| 5696 | } |
| 5697 | return SV; |
| 5698 | } |
| 5699 | case NEON::BI__builtin_neon_vzip_v: |
| 5700 | case NEON::BI__builtin_neon_vzipq_v: { |
| 5701 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); |
| 5702 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 5703 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 5704 | Value *SV = nullptr; |
| 5705 | |
| 5706 | for (unsigned vi = 0; vi != 2; ++vi) { |
| 5707 | SmallVector<uint32_t, 16> Indices; |
| 5708 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { |
| 5709 | Indices.push_back((i + vi*e) >> 1); |
| 5710 | Indices.push_back(((i + vi*e) >> 1)+e); |
| 5711 | } |
| 5712 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); |
| 5713 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip"); |
| 5714 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); |
| 5715 | } |
| 5716 | return SV; |
| 5717 | } |
| 5718 | case NEON::BI__builtin_neon_vdot_v: |
| 5719 | case NEON::BI__builtin_neon_vdotq_v: { |
| 5720 | llvm::Type *InputTy = |
| 5721 | llvm::VectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8); |
| 5722 | llvm::Type *Tys[2] = { Ty, InputTy }; |
| 5723 | Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic; |
| 5724 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vdot"); |
| 5725 | } |
| 5726 | case NEON::BI__builtin_neon_vfmlal_low_v: |
| 5727 | case NEON::BI__builtin_neon_vfmlalq_low_v: { |
| 5728 | llvm::Type *InputTy = |
| 5729 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); |
| 5730 | llvm::Type *Tys[2] = { Ty, InputTy }; |
| 5731 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_low"); |
| 5732 | } |
| 5733 | case NEON::BI__builtin_neon_vfmlsl_low_v: |
| 5734 | case NEON::BI__builtin_neon_vfmlslq_low_v: { |
| 5735 | llvm::Type *InputTy = |
| 5736 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); |
| 5737 | llvm::Type *Tys[2] = { Ty, InputTy }; |
| 5738 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_low"); |
| 5739 | } |
| 5740 | case NEON::BI__builtin_neon_vfmlal_high_v: |
| 5741 | case NEON::BI__builtin_neon_vfmlalq_high_v: { |
| 5742 | llvm::Type *InputTy = |
| 5743 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); |
| 5744 | llvm::Type *Tys[2] = { Ty, InputTy }; |
| 5745 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_high"); |
| 5746 | } |
| 5747 | case NEON::BI__builtin_neon_vfmlsl_high_v: |
| 5748 | case NEON::BI__builtin_neon_vfmlslq_high_v: { |
| 5749 | llvm::Type *InputTy = |
| 5750 | llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16); |
| 5751 | llvm::Type *Tys[2] = { Ty, InputTy }; |
| 5752 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_high"); |
| 5753 | } |
| 5754 | } |
| 5755 | |
| 5756 | assert(Int && "Expected valid intrinsic number")((Int && "Expected valid intrinsic number") ? static_cast <void> (0) : __assert_fail ("Int && \"Expected valid intrinsic number\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5756, __PRETTY_FUNCTION__)); |
| 5757 | |
| 5758 | // Determine the type(s) of this overloaded AArch64 intrinsic. |
| 5759 | Function *F = LookupNeonLLVMIntrinsic(Int, Modifier, Ty, E); |
| 5760 | |
| 5761 | Value *Result = EmitNeonCall(F, Ops, NameHint); |
| 5762 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 5763 | // AArch64 intrinsic one-element vector type cast to |
| 5764 | // scalar type expected by the builtin |
| 5765 | return Builder.CreateBitCast(Result, ResultType, NameHint); |
| 5766 | } |
| 5767 | |
| 5768 | Value *CodeGenFunction::EmitAArch64CompareBuiltinExpr( |
| 5769 | Value *Op, llvm::Type *Ty, const CmpInst::Predicate Fp, |
| 5770 | const CmpInst::Predicate Ip, const Twine &Name) { |
| 5771 | llvm::Type *OTy = Op->getType(); |
| 5772 | |
| 5773 | // FIXME: this is utterly horrific. We should not be looking at previous |
| 5774 | // codegen context to find out what needs doing. Unfortunately TableGen |
| 5775 | // currently gives us exactly the same calls for vceqz_f32 and vceqz_s32 |
| 5776 | // (etc). |
| 5777 | if (BitCastInst *BI = dyn_cast<BitCastInst>(Op)) |
| 5778 | OTy = BI->getOperand(0)->getType(); |
| 5779 | |
| 5780 | Op = Builder.CreateBitCast(Op, OTy); |
| 5781 | if (OTy->getScalarType()->isFloatingPointTy()) { |
| 5782 | Op = Builder.CreateFCmp(Fp, Op, Constant::getNullValue(OTy)); |
| 5783 | } else { |
| 5784 | Op = Builder.CreateICmp(Ip, Op, Constant::getNullValue(OTy)); |
| 5785 | } |
| 5786 | return Builder.CreateSExt(Op, Ty, Name); |
| 5787 | } |
| 5788 | |
| 5789 | static Value *packTBLDVectorList(CodeGenFunction &CGF, ArrayRef<Value *> Ops, |
| 5790 | Value *ExtOp, Value *IndexOp, |
| 5791 | llvm::Type *ResTy, unsigned IntID, |
| 5792 | const char *Name) { |
| 5793 | SmallVector<Value *, 2> TblOps; |
| 5794 | if (ExtOp) |
| 5795 | TblOps.push_back(ExtOp); |
| 5796 | |
| 5797 | // Build a vector containing sequential number like (0, 1, 2, ..., 15) |
| 5798 | SmallVector<uint32_t, 16> Indices; |
| 5799 | llvm::VectorType *TblTy = cast<llvm::VectorType>(Ops[0]->getType()); |
| 5800 | for (unsigned i = 0, e = TblTy->getNumElements(); i != e; ++i) { |
| 5801 | Indices.push_back(2*i); |
| 5802 | Indices.push_back(2*i+1); |
| 5803 | } |
| 5804 | |
| 5805 | int PairPos = 0, End = Ops.size() - 1; |
| 5806 | while (PairPos < End) { |
| 5807 | TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos], |
| 5808 | Ops[PairPos+1], Indices, |
| 5809 | Name)); |
| 5810 | PairPos += 2; |
| 5811 | } |
| 5812 | |
| 5813 | // If there's an odd number of 64-bit lookup table, fill the high 64-bit |
| 5814 | // of the 128-bit lookup table with zero. |
| 5815 | if (PairPos == End) { |
| 5816 | Value *ZeroTbl = ConstantAggregateZero::get(TblTy); |
| 5817 | TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos], |
| 5818 | ZeroTbl, Indices, Name)); |
| 5819 | } |
| 5820 | |
| 5821 | Function *TblF; |
| 5822 | TblOps.push_back(IndexOp); |
| 5823 | TblF = CGF.CGM.getIntrinsic(IntID, ResTy); |
| 5824 | |
| 5825 | return CGF.EmitNeonCall(TblF, TblOps, Name); |
| 5826 | } |
| 5827 | |
| 5828 | Value *CodeGenFunction::GetValueForARMHint(unsigned BuiltinID) { |
| 5829 | unsigned Value; |
| 5830 | switch (BuiltinID) { |
| 5831 | default: |
| 5832 | return nullptr; |
| 5833 | case ARM::BI__builtin_arm_nop: |
| 5834 | Value = 0; |
| 5835 | break; |
| 5836 | case ARM::BI__builtin_arm_yield: |
| 5837 | case ARM::BI__yield: |
| 5838 | Value = 1; |
| 5839 | break; |
| 5840 | case ARM::BI__builtin_arm_wfe: |
| 5841 | case ARM::BI__wfe: |
| 5842 | Value = 2; |
| 5843 | break; |
| 5844 | case ARM::BI__builtin_arm_wfi: |
| 5845 | case ARM::BI__wfi: |
| 5846 | Value = 3; |
| 5847 | break; |
| 5848 | case ARM::BI__builtin_arm_sev: |
| 5849 | case ARM::BI__sev: |
| 5850 | Value = 4; |
| 5851 | break; |
| 5852 | case ARM::BI__builtin_arm_sevl: |
| 5853 | case ARM::BI__sevl: |
| 5854 | Value = 5; |
| 5855 | break; |
| 5856 | } |
| 5857 | |
| 5858 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_hint), |
| 5859 | llvm::ConstantInt::get(Int32Ty, Value)); |
| 5860 | } |
| 5861 | |
| 5862 | // Generates the IR for the read/write special register builtin, |
| 5863 | // ValueType is the type of the value that is to be written or read, |
| 5864 | // RegisterType is the type of the register being written to or read from. |
| 5865 | static Value *EmitSpecialRegisterBuiltin(CodeGenFunction &CGF, |
| 5866 | const CallExpr *E, |
| 5867 | llvm::Type *RegisterType, |
| 5868 | llvm::Type *ValueType, |
| 5869 | bool IsRead, |
| 5870 | StringRef SysReg = "") { |
| 5871 | // write and register intrinsics only support 32 and 64 bit operations. |
| 5872 | assert((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64))(((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy (64)) && "Unsupported size for register.") ? static_cast <void> (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5873, __PRETTY_FUNCTION__)) |
| 5873 | && "Unsupported size for register.")(((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy (64)) && "Unsupported size for register.") ? static_cast <void> (0) : __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5873, __PRETTY_FUNCTION__)); |
| 5874 | |
| 5875 | CodeGen::CGBuilderTy &Builder = CGF.Builder; |
| 5876 | CodeGen::CodeGenModule &CGM = CGF.CGM; |
| 5877 | LLVMContext &Context = CGM.getLLVMContext(); |
| 5878 | |
| 5879 | if (SysReg.empty()) { |
| 5880 | const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts(); |
| 5881 | SysReg = cast<clang::StringLiteral>(SysRegStrExpr)->getString(); |
| 5882 | } |
| 5883 | |
| 5884 | llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysReg) }; |
| 5885 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); |
| 5886 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); |
| 5887 | |
| 5888 | llvm::Type *Types[] = { RegisterType }; |
| 5889 | |
| 5890 | bool MixedTypes = RegisterType->isIntegerTy(64) && ValueType->isIntegerTy(32); |
| 5891 | assert(!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64))((!(RegisterType->isIntegerTy(32) && ValueType-> isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register" ) ? static_cast<void> (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5892, __PRETTY_FUNCTION__)) |
| 5892 | && "Can't fit 64-bit value in 32-bit register")((!(RegisterType->isIntegerTy(32) && ValueType-> isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register" ) ? static_cast<void> (0) : __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5892, __PRETTY_FUNCTION__)); |
| 5893 | |
| 5894 | if (IsRead) { |
| 5895 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types); |
| 5896 | llvm::Value *Call = Builder.CreateCall(F, Metadata); |
| 5897 | |
| 5898 | if (MixedTypes) |
| 5899 | // Read into 64 bit register and then truncate result to 32 bit. |
| 5900 | return Builder.CreateTrunc(Call, ValueType); |
| 5901 | |
| 5902 | if (ValueType->isPointerTy()) |
| 5903 | // Have i32/i64 result (Call) but want to return a VoidPtrTy (i8*). |
| 5904 | return Builder.CreateIntToPtr(Call, ValueType); |
| 5905 | |
| 5906 | return Call; |
| 5907 | } |
| 5908 | |
| 5909 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types); |
| 5910 | llvm::Value *ArgValue = CGF.EmitScalarExpr(E->getArg(1)); |
| 5911 | if (MixedTypes) { |
| 5912 | // Extend 32 bit write value to 64 bit to pass to write. |
| 5913 | ArgValue = Builder.CreateZExt(ArgValue, RegisterType); |
| 5914 | return Builder.CreateCall(F, { Metadata, ArgValue }); |
| 5915 | } |
| 5916 | |
| 5917 | if (ValueType->isPointerTy()) { |
| 5918 | // Have VoidPtrTy ArgValue but want to return an i32/i64. |
| 5919 | ArgValue = Builder.CreatePtrToInt(ArgValue, RegisterType); |
| 5920 | return Builder.CreateCall(F, { Metadata, ArgValue }); |
| 5921 | } |
| 5922 | |
| 5923 | return Builder.CreateCall(F, { Metadata, ArgValue }); |
| 5924 | } |
| 5925 | |
| 5926 | /// Return true if BuiltinID is an overloaded Neon intrinsic with an extra |
| 5927 | /// argument that specifies the vector type. |
| 5928 | static bool HasExtraNeonArgument(unsigned BuiltinID) { |
| 5929 | switch (BuiltinID) { |
| 5930 | default: break; |
| 5931 | case NEON::BI__builtin_neon_vget_lane_i8: |
| 5932 | case NEON::BI__builtin_neon_vget_lane_i16: |
| 5933 | case NEON::BI__builtin_neon_vget_lane_i32: |
| 5934 | case NEON::BI__builtin_neon_vget_lane_i64: |
| 5935 | case NEON::BI__builtin_neon_vget_lane_f32: |
| 5936 | case NEON::BI__builtin_neon_vgetq_lane_i8: |
| 5937 | case NEON::BI__builtin_neon_vgetq_lane_i16: |
| 5938 | case NEON::BI__builtin_neon_vgetq_lane_i32: |
| 5939 | case NEON::BI__builtin_neon_vgetq_lane_i64: |
| 5940 | case NEON::BI__builtin_neon_vgetq_lane_f32: |
| 5941 | case NEON::BI__builtin_neon_vset_lane_i8: |
| 5942 | case NEON::BI__builtin_neon_vset_lane_i16: |
| 5943 | case NEON::BI__builtin_neon_vset_lane_i32: |
| 5944 | case NEON::BI__builtin_neon_vset_lane_i64: |
| 5945 | case NEON::BI__builtin_neon_vset_lane_f32: |
| 5946 | case NEON::BI__builtin_neon_vsetq_lane_i8: |
| 5947 | case NEON::BI__builtin_neon_vsetq_lane_i16: |
| 5948 | case NEON::BI__builtin_neon_vsetq_lane_i32: |
| 5949 | case NEON::BI__builtin_neon_vsetq_lane_i64: |
| 5950 | case NEON::BI__builtin_neon_vsetq_lane_f32: |
| 5951 | case NEON::BI__builtin_neon_vsha1h_u32: |
| 5952 | case NEON::BI__builtin_neon_vsha1cq_u32: |
| 5953 | case NEON::BI__builtin_neon_vsha1pq_u32: |
| 5954 | case NEON::BI__builtin_neon_vsha1mq_u32: |
| 5955 | case clang::ARM::BI_MoveToCoprocessor: |
| 5956 | case clang::ARM::BI_MoveToCoprocessor2: |
| 5957 | return false; |
| 5958 | } |
| 5959 | return true; |
| 5960 | } |
| 5961 | |
| 5962 | Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID, |
| 5963 | const CallExpr *E, |
| 5964 | llvm::Triple::ArchType Arch) { |
| 5965 | if (auto Hint = GetValueForARMHint(BuiltinID)) |
| 5966 | return Hint; |
| 5967 | |
| 5968 | if (BuiltinID == ARM::BI__emit) { |
| 5969 | bool IsThumb = getTarget().getTriple().getArch() == llvm::Triple::thumb; |
| 5970 | llvm::FunctionType *FTy = |
| 5971 | llvm::FunctionType::get(VoidTy, /*Variadic=*/false); |
| 5972 | |
| 5973 | Expr::EvalResult Result; |
| 5974 | if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext())) |
| 5975 | llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 5975); |
| 5976 | |
| 5977 | llvm::APSInt Value = Result.Val.getInt(); |
| 5978 | uint64_t ZExtValue = Value.zextOrTrunc(IsThumb ? 16 : 32).getZExtValue(); |
| 5979 | |
| 5980 | llvm::InlineAsm *Emit = |
| 5981 | IsThumb ? InlineAsm::get(FTy, ".inst.n 0x" + utohexstr(ZExtValue), "", |
| 5982 | /*SideEffects=*/true) |
| 5983 | : InlineAsm::get(FTy, ".inst 0x" + utohexstr(ZExtValue), "", |
| 5984 | /*SideEffects=*/true); |
| 5985 | |
| 5986 | return Builder.CreateCall(Emit); |
| 5987 | } |
| 5988 | |
| 5989 | if (BuiltinID == ARM::BI__builtin_arm_dbg) { |
| 5990 | Value *Option = EmitScalarExpr(E->getArg(0)); |
| 5991 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_dbg), Option); |
| 5992 | } |
| 5993 | |
| 5994 | if (BuiltinID == ARM::BI__builtin_arm_prefetch) { |
| 5995 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 5996 | Value *RW = EmitScalarExpr(E->getArg(1)); |
| 5997 | Value *IsData = EmitScalarExpr(E->getArg(2)); |
| 5998 | |
| 5999 | // Locality is not supported on ARM target |
| 6000 | Value *Locality = llvm::ConstantInt::get(Int32Ty, 3); |
| 6001 | |
| 6002 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch); |
| 6003 | return Builder.CreateCall(F, {Address, RW, Locality, IsData}); |
| 6004 | } |
| 6005 | |
| 6006 | if (BuiltinID == ARM::BI__builtin_arm_rbit) { |
| 6007 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); |
| 6008 | return Builder.CreateCall( |
| 6009 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); |
| 6010 | } |
| 6011 | |
| 6012 | if (BuiltinID == ARM::BI__clear_cache) { |
| 6013 | assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")((E->getNumArgs() == 2 && "__clear_cache takes 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6013, __PRETTY_FUNCTION__)); |
| 6014 | const FunctionDecl *FD = E->getDirectCallee(); |
| 6015 | Value *Ops[2]; |
| 6016 | for (unsigned i = 0; i < 2; i++) |
| 6017 | Ops[i] = EmitScalarExpr(E->getArg(i)); |
| 6018 | llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType()); |
| 6019 | llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty); |
| 6020 | StringRef Name = FD->getName(); |
| 6021 | return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops); |
| 6022 | } |
| 6023 | |
| 6024 | if (BuiltinID == ARM::BI__builtin_arm_mcrr || |
| 6025 | BuiltinID == ARM::BI__builtin_arm_mcrr2) { |
| 6026 | Function *F; |
| 6027 | |
| 6028 | switch (BuiltinID) { |
| 6029 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6029); |
| 6030 | case ARM::BI__builtin_arm_mcrr: |
| 6031 | F = CGM.getIntrinsic(Intrinsic::arm_mcrr); |
| 6032 | break; |
| 6033 | case ARM::BI__builtin_arm_mcrr2: |
| 6034 | F = CGM.getIntrinsic(Intrinsic::arm_mcrr2); |
| 6035 | break; |
| 6036 | } |
| 6037 | |
| 6038 | // MCRR{2} instruction has 5 operands but |
| 6039 | // the intrinsic has 4 because Rt and Rt2 |
| 6040 | // are represented as a single unsigned 64 |
| 6041 | // bit integer in the intrinsic definition |
| 6042 | // but internally it's represented as 2 32 |
| 6043 | // bit integers. |
| 6044 | |
| 6045 | Value *Coproc = EmitScalarExpr(E->getArg(0)); |
| 6046 | Value *Opc1 = EmitScalarExpr(E->getArg(1)); |
| 6047 | Value *RtAndRt2 = EmitScalarExpr(E->getArg(2)); |
| 6048 | Value *CRm = EmitScalarExpr(E->getArg(3)); |
| 6049 | |
| 6050 | Value *C1 = llvm::ConstantInt::get(Int64Ty, 32); |
| 6051 | Value *Rt = Builder.CreateTruncOrBitCast(RtAndRt2, Int32Ty); |
| 6052 | Value *Rt2 = Builder.CreateLShr(RtAndRt2, C1); |
| 6053 | Rt2 = Builder.CreateTruncOrBitCast(Rt2, Int32Ty); |
| 6054 | |
| 6055 | return Builder.CreateCall(F, {Coproc, Opc1, Rt, Rt2, CRm}); |
| 6056 | } |
| 6057 | |
| 6058 | if (BuiltinID == ARM::BI__builtin_arm_mrrc || |
| 6059 | BuiltinID == ARM::BI__builtin_arm_mrrc2) { |
| 6060 | Function *F; |
| 6061 | |
| 6062 | switch (BuiltinID) { |
| 6063 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6063); |
| 6064 | case ARM::BI__builtin_arm_mrrc: |
| 6065 | F = CGM.getIntrinsic(Intrinsic::arm_mrrc); |
| 6066 | break; |
| 6067 | case ARM::BI__builtin_arm_mrrc2: |
| 6068 | F = CGM.getIntrinsic(Intrinsic::arm_mrrc2); |
| 6069 | break; |
| 6070 | } |
| 6071 | |
| 6072 | Value *Coproc = EmitScalarExpr(E->getArg(0)); |
| 6073 | Value *Opc1 = EmitScalarExpr(E->getArg(1)); |
| 6074 | Value *CRm = EmitScalarExpr(E->getArg(2)); |
| 6075 | Value *RtAndRt2 = Builder.CreateCall(F, {Coproc, Opc1, CRm}); |
| 6076 | |
| 6077 | // Returns an unsigned 64 bit integer, represented |
| 6078 | // as two 32 bit integers. |
| 6079 | |
| 6080 | Value *Rt = Builder.CreateExtractValue(RtAndRt2, 1); |
| 6081 | Value *Rt1 = Builder.CreateExtractValue(RtAndRt2, 0); |
| 6082 | Rt = Builder.CreateZExt(Rt, Int64Ty); |
| 6083 | Rt1 = Builder.CreateZExt(Rt1, Int64Ty); |
| 6084 | |
| 6085 | Value *ShiftCast = llvm::ConstantInt::get(Int64Ty, 32); |
| 6086 | RtAndRt2 = Builder.CreateShl(Rt, ShiftCast, "shl", true); |
| 6087 | RtAndRt2 = Builder.CreateOr(RtAndRt2, Rt1); |
| 6088 | |
| 6089 | return Builder.CreateBitCast(RtAndRt2, ConvertType(E->getType())); |
| 6090 | } |
| 6091 | |
| 6092 | if (BuiltinID == ARM::BI__builtin_arm_ldrexd || |
| 6093 | ((BuiltinID == ARM::BI__builtin_arm_ldrex || |
| 6094 | BuiltinID == ARM::BI__builtin_arm_ldaex) && |
| 6095 | getContext().getTypeSize(E->getType()) == 64) || |
| 6096 | BuiltinID == ARM::BI__ldrexd) { |
| 6097 | Function *F; |
| 6098 | |
| 6099 | switch (BuiltinID) { |
| 6100 | default: llvm_unreachable("unexpected builtin")::llvm::llvm_unreachable_internal("unexpected builtin", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6100); |
| 6101 | case ARM::BI__builtin_arm_ldaex: |
| 6102 | F = CGM.getIntrinsic(Intrinsic::arm_ldaexd); |
| 6103 | break; |
| 6104 | case ARM::BI__builtin_arm_ldrexd: |
| 6105 | case ARM::BI__builtin_arm_ldrex: |
| 6106 | case ARM::BI__ldrexd: |
| 6107 | F = CGM.getIntrinsic(Intrinsic::arm_ldrexd); |
| 6108 | break; |
| 6109 | } |
| 6110 | |
| 6111 | Value *LdPtr = EmitScalarExpr(E->getArg(0)); |
| 6112 | Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy), |
| 6113 | "ldrexd"); |
| 6114 | |
| 6115 | Value *Val0 = Builder.CreateExtractValue(Val, 1); |
| 6116 | Value *Val1 = Builder.CreateExtractValue(Val, 0); |
| 6117 | Val0 = Builder.CreateZExt(Val0, Int64Ty); |
| 6118 | Val1 = Builder.CreateZExt(Val1, Int64Ty); |
| 6119 | |
| 6120 | Value *ShiftCst = llvm::ConstantInt::get(Int64Ty, 32); |
| 6121 | Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */); |
| 6122 | Val = Builder.CreateOr(Val, Val1); |
| 6123 | return Builder.CreateBitCast(Val, ConvertType(E->getType())); |
| 6124 | } |
| 6125 | |
| 6126 | if (BuiltinID == ARM::BI__builtin_arm_ldrex || |
| 6127 | BuiltinID == ARM::BI__builtin_arm_ldaex) { |
| 6128 | Value *LoadAddr = EmitScalarExpr(E->getArg(0)); |
| 6129 | |
| 6130 | QualType Ty = E->getType(); |
| 6131 | llvm::Type *RealResTy = ConvertType(Ty); |
| 6132 | llvm::Type *PtrTy = llvm::IntegerType::get( |
| 6133 | getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo(); |
| 6134 | LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy); |
| 6135 | |
| 6136 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_ldaex |
| 6137 | ? Intrinsic::arm_ldaex |
| 6138 | : Intrinsic::arm_ldrex, |
| 6139 | PtrTy); |
| 6140 | Value *Val = Builder.CreateCall(F, LoadAddr, "ldrex"); |
| 6141 | |
| 6142 | if (RealResTy->isPointerTy()) |
| 6143 | return Builder.CreateIntToPtr(Val, RealResTy); |
| 6144 | else { |
| 6145 | llvm::Type *IntResTy = llvm::IntegerType::get( |
| 6146 | getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy)); |
| 6147 | Val = Builder.CreateTruncOrBitCast(Val, IntResTy); |
| 6148 | return Builder.CreateBitCast(Val, RealResTy); |
| 6149 | } |
| 6150 | } |
| 6151 | |
| 6152 | if (BuiltinID == ARM::BI__builtin_arm_strexd || |
| 6153 | ((BuiltinID == ARM::BI__builtin_arm_stlex || |
| 6154 | BuiltinID == ARM::BI__builtin_arm_strex) && |
| 6155 | getContext().getTypeSize(E->getArg(0)->getType()) == 64)) { |
| 6156 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex |
| 6157 | ? Intrinsic::arm_stlexd |
| 6158 | : Intrinsic::arm_strexd); |
| 6159 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty); |
| 6160 | |
| 6161 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); |
| 6162 | Value *Val = EmitScalarExpr(E->getArg(0)); |
| 6163 | Builder.CreateStore(Val, Tmp); |
| 6164 | |
| 6165 | Address LdPtr = Builder.CreateBitCast(Tmp,llvm::PointerType::getUnqual(STy)); |
| 6166 | Val = Builder.CreateLoad(LdPtr); |
| 6167 | |
| 6168 | Value *Arg0 = Builder.CreateExtractValue(Val, 0); |
| 6169 | Value *Arg1 = Builder.CreateExtractValue(Val, 1); |
| 6170 | Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), Int8PtrTy); |
| 6171 | return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "strexd"); |
| 6172 | } |
| 6173 | |
| 6174 | if (BuiltinID == ARM::BI__builtin_arm_strex || |
| 6175 | BuiltinID == ARM::BI__builtin_arm_stlex) { |
| 6176 | Value *StoreVal = EmitScalarExpr(E->getArg(0)); |
| 6177 | Value *StoreAddr = EmitScalarExpr(E->getArg(1)); |
| 6178 | |
| 6179 | QualType Ty = E->getArg(0)->getType(); |
| 6180 | llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(), |
| 6181 | getContext().getTypeSize(Ty)); |
| 6182 | StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo()); |
| 6183 | |
| 6184 | if (StoreVal->getType()->isPointerTy()) |
| 6185 | StoreVal = Builder.CreatePtrToInt(StoreVal, Int32Ty); |
| 6186 | else { |
| 6187 | llvm::Type *IntTy = llvm::IntegerType::get( |
| 6188 | getLLVMContext(), |
| 6189 | CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType())); |
| 6190 | StoreVal = Builder.CreateBitCast(StoreVal, IntTy); |
| 6191 | StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int32Ty); |
| 6192 | } |
| 6193 | |
| 6194 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex |
| 6195 | ? Intrinsic::arm_stlex |
| 6196 | : Intrinsic::arm_strex, |
| 6197 | StoreAddr->getType()); |
| 6198 | return Builder.CreateCall(F, {StoreVal, StoreAddr}, "strex"); |
| 6199 | } |
| 6200 | |
| 6201 | if (BuiltinID == ARM::BI__builtin_arm_clrex) { |
| 6202 | Function *F = CGM.getIntrinsic(Intrinsic::arm_clrex); |
| 6203 | return Builder.CreateCall(F); |
| 6204 | } |
| 6205 | |
| 6206 | // CRC32 |
| 6207 | Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic; |
| 6208 | switch (BuiltinID) { |
| 6209 | case ARM::BI__builtin_arm_crc32b: |
| 6210 | CRCIntrinsicID = Intrinsic::arm_crc32b; break; |
| 6211 | case ARM::BI__builtin_arm_crc32cb: |
| 6212 | CRCIntrinsicID = Intrinsic::arm_crc32cb; break; |
| 6213 | case ARM::BI__builtin_arm_crc32h: |
| 6214 | CRCIntrinsicID = Intrinsic::arm_crc32h; break; |
| 6215 | case ARM::BI__builtin_arm_crc32ch: |
| 6216 | CRCIntrinsicID = Intrinsic::arm_crc32ch; break; |
| 6217 | case ARM::BI__builtin_arm_crc32w: |
| 6218 | case ARM::BI__builtin_arm_crc32d: |
| 6219 | CRCIntrinsicID = Intrinsic::arm_crc32w; break; |
| 6220 | case ARM::BI__builtin_arm_crc32cw: |
| 6221 | case ARM::BI__builtin_arm_crc32cd: |
| 6222 | CRCIntrinsicID = Intrinsic::arm_crc32cw; break; |
| 6223 | } |
| 6224 | |
| 6225 | if (CRCIntrinsicID != Intrinsic::not_intrinsic) { |
| 6226 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); |
| 6227 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 6228 | |
| 6229 | // crc32{c,}d intrinsics are implemnted as two calls to crc32{c,}w |
| 6230 | // intrinsics, hence we need different codegen for these cases. |
| 6231 | if (BuiltinID == ARM::BI__builtin_arm_crc32d || |
| 6232 | BuiltinID == ARM::BI__builtin_arm_crc32cd) { |
| 6233 | Value *C1 = llvm::ConstantInt::get(Int64Ty, 32); |
| 6234 | Value *Arg1a = Builder.CreateTruncOrBitCast(Arg1, Int32Ty); |
| 6235 | Value *Arg1b = Builder.CreateLShr(Arg1, C1); |
| 6236 | Arg1b = Builder.CreateTruncOrBitCast(Arg1b, Int32Ty); |
| 6237 | |
| 6238 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); |
| 6239 | Value *Res = Builder.CreateCall(F, {Arg0, Arg1a}); |
| 6240 | return Builder.CreateCall(F, {Res, Arg1b}); |
| 6241 | } else { |
| 6242 | Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty); |
| 6243 | |
| 6244 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); |
| 6245 | return Builder.CreateCall(F, {Arg0, Arg1}); |
| 6246 | } |
| 6247 | } |
| 6248 | |
| 6249 | if (BuiltinID == ARM::BI__builtin_arm_rsr || |
| 6250 | BuiltinID == ARM::BI__builtin_arm_rsr64 || |
| 6251 | BuiltinID == ARM::BI__builtin_arm_rsrp || |
| 6252 | BuiltinID == ARM::BI__builtin_arm_wsr || |
| 6253 | BuiltinID == ARM::BI__builtin_arm_wsr64 || |
| 6254 | BuiltinID == ARM::BI__builtin_arm_wsrp) { |
| 6255 | |
| 6256 | bool IsRead = BuiltinID == ARM::BI__builtin_arm_rsr || |
| 6257 | BuiltinID == ARM::BI__builtin_arm_rsr64 || |
| 6258 | BuiltinID == ARM::BI__builtin_arm_rsrp; |
| 6259 | |
| 6260 | bool IsPointerBuiltin = BuiltinID == ARM::BI__builtin_arm_rsrp || |
| 6261 | BuiltinID == ARM::BI__builtin_arm_wsrp; |
| 6262 | |
| 6263 | bool Is64Bit = BuiltinID == ARM::BI__builtin_arm_rsr64 || |
| 6264 | BuiltinID == ARM::BI__builtin_arm_wsr64; |
| 6265 | |
| 6266 | llvm::Type *ValueType; |
| 6267 | llvm::Type *RegisterType; |
| 6268 | if (IsPointerBuiltin) { |
| 6269 | ValueType = VoidPtrTy; |
| 6270 | RegisterType = Int32Ty; |
| 6271 | } else if (Is64Bit) { |
| 6272 | ValueType = RegisterType = Int64Ty; |
| 6273 | } else { |
| 6274 | ValueType = RegisterType = Int32Ty; |
| 6275 | } |
| 6276 | |
| 6277 | return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead); |
| 6278 | } |
| 6279 | |
| 6280 | // Find out if any arguments are required to be integer constant |
| 6281 | // expressions. |
| 6282 | unsigned ICEArguments = 0; |
| 6283 | ASTContext::GetBuiltinTypeError Error; |
| 6284 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); |
| 6285 | assert(Error == ASTContext::GE_None && "Should not codegen an error")((Error == ASTContext::GE_None && "Should not codegen an error" ) ? static_cast<void> (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6285, __PRETTY_FUNCTION__)); |
| 6286 | |
| 6287 | auto getAlignmentValue32 = [&](Address addr) -> Value* { |
| 6288 | return Builder.getInt32(addr.getAlignment().getQuantity()); |
| 6289 | }; |
| 6290 | |
| 6291 | Address PtrOp0 = Address::invalid(); |
| 6292 | Address PtrOp1 = Address::invalid(); |
| 6293 | SmallVector<Value*, 4> Ops; |
| 6294 | bool HasExtraArg = HasExtraNeonArgument(BuiltinID); |
| 6295 | unsigned NumArgs = E->getNumArgs() - (HasExtraArg ? 1 : 0); |
| 6296 | for (unsigned i = 0, e = NumArgs; i != e; i++) { |
| 6297 | if (i == 0) { |
| 6298 | switch (BuiltinID) { |
| 6299 | case NEON::BI__builtin_neon_vld1_v: |
| 6300 | case NEON::BI__builtin_neon_vld1q_v: |
| 6301 | case NEON::BI__builtin_neon_vld1q_lane_v: |
| 6302 | case NEON::BI__builtin_neon_vld1_lane_v: |
| 6303 | case NEON::BI__builtin_neon_vld1_dup_v: |
| 6304 | case NEON::BI__builtin_neon_vld1q_dup_v: |
| 6305 | case NEON::BI__builtin_neon_vst1_v: |
| 6306 | case NEON::BI__builtin_neon_vst1q_v: |
| 6307 | case NEON::BI__builtin_neon_vst1q_lane_v: |
| 6308 | case NEON::BI__builtin_neon_vst1_lane_v: |
| 6309 | case NEON::BI__builtin_neon_vst2_v: |
| 6310 | case NEON::BI__builtin_neon_vst2q_v: |
| 6311 | case NEON::BI__builtin_neon_vst2_lane_v: |
| 6312 | case NEON::BI__builtin_neon_vst2q_lane_v: |
| 6313 | case NEON::BI__builtin_neon_vst3_v: |
| 6314 | case NEON::BI__builtin_neon_vst3q_v: |
| 6315 | case NEON::BI__builtin_neon_vst3_lane_v: |
| 6316 | case NEON::BI__builtin_neon_vst3q_lane_v: |
| 6317 | case NEON::BI__builtin_neon_vst4_v: |
| 6318 | case NEON::BI__builtin_neon_vst4q_v: |
| 6319 | case NEON::BI__builtin_neon_vst4_lane_v: |
| 6320 | case NEON::BI__builtin_neon_vst4q_lane_v: |
| 6321 | // Get the alignment for the argument in addition to the value; |
| 6322 | // we'll use it later. |
| 6323 | PtrOp0 = EmitPointerWithAlignment(E->getArg(0)); |
| 6324 | Ops.push_back(PtrOp0.getPointer()); |
| 6325 | continue; |
| 6326 | } |
| 6327 | } |
| 6328 | if (i == 1) { |
| 6329 | switch (BuiltinID) { |
| 6330 | case NEON::BI__builtin_neon_vld2_v: |
| 6331 | case NEON::BI__builtin_neon_vld2q_v: |
| 6332 | case NEON::BI__builtin_neon_vld3_v: |
| 6333 | case NEON::BI__builtin_neon_vld3q_v: |
| 6334 | case NEON::BI__builtin_neon_vld4_v: |
| 6335 | case NEON::BI__builtin_neon_vld4q_v: |
| 6336 | case NEON::BI__builtin_neon_vld2_lane_v: |
| 6337 | case NEON::BI__builtin_neon_vld2q_lane_v: |
| 6338 | case NEON::BI__builtin_neon_vld3_lane_v: |
| 6339 | case NEON::BI__builtin_neon_vld3q_lane_v: |
| 6340 | case NEON::BI__builtin_neon_vld4_lane_v: |
| 6341 | case NEON::BI__builtin_neon_vld4q_lane_v: |
| 6342 | case NEON::BI__builtin_neon_vld2_dup_v: |
| 6343 | case NEON::BI__builtin_neon_vld2q_dup_v: |
| 6344 | case NEON::BI__builtin_neon_vld3_dup_v: |
| 6345 | case NEON::BI__builtin_neon_vld3q_dup_v: |
| 6346 | case NEON::BI__builtin_neon_vld4_dup_v: |
| 6347 | case NEON::BI__builtin_neon_vld4q_dup_v: |
| 6348 | // Get the alignment for the argument in addition to the value; |
| 6349 | // we'll use it later. |
| 6350 | PtrOp1 = EmitPointerWithAlignment(E->getArg(1)); |
| 6351 | Ops.push_back(PtrOp1.getPointer()); |
| 6352 | continue; |
| 6353 | } |
| 6354 | } |
| 6355 | |
| 6356 | if ((ICEArguments & (1 << i)) == 0) { |
| 6357 | Ops.push_back(EmitScalarExpr(E->getArg(i))); |
| 6358 | } else { |
| 6359 | // If this is required to be a constant, constant fold it so that we know |
| 6360 | // that the generated intrinsic gets a ConstantInt. |
| 6361 | llvm::APSInt Result; |
| 6362 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext()); |
| 6363 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6363, __PRETTY_FUNCTION__)); (void)IsConst; |
| 6364 | Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result)); |
| 6365 | } |
| 6366 | } |
| 6367 | |
| 6368 | switch (BuiltinID) { |
| 6369 | default: break; |
| 6370 | |
| 6371 | case NEON::BI__builtin_neon_vget_lane_i8: |
| 6372 | case NEON::BI__builtin_neon_vget_lane_i16: |
| 6373 | case NEON::BI__builtin_neon_vget_lane_i32: |
| 6374 | case NEON::BI__builtin_neon_vget_lane_i64: |
| 6375 | case NEON::BI__builtin_neon_vget_lane_f32: |
| 6376 | case NEON::BI__builtin_neon_vgetq_lane_i8: |
| 6377 | case NEON::BI__builtin_neon_vgetq_lane_i16: |
| 6378 | case NEON::BI__builtin_neon_vgetq_lane_i32: |
| 6379 | case NEON::BI__builtin_neon_vgetq_lane_i64: |
| 6380 | case NEON::BI__builtin_neon_vgetq_lane_f32: |
| 6381 | return Builder.CreateExtractElement(Ops[0], Ops[1], "vget_lane"); |
| 6382 | |
| 6383 | case NEON::BI__builtin_neon_vrndns_f32: { |
| 6384 | Value *Arg = EmitScalarExpr(E->getArg(0)); |
| 6385 | llvm::Type *Tys[] = {Arg->getType()}; |
| 6386 | Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vrintn, Tys); |
| 6387 | return Builder.CreateCall(F, {Arg}, "vrndn"); } |
| 6388 | |
| 6389 | case NEON::BI__builtin_neon_vset_lane_i8: |
| 6390 | case NEON::BI__builtin_neon_vset_lane_i16: |
| 6391 | case NEON::BI__builtin_neon_vset_lane_i32: |
| 6392 | case NEON::BI__builtin_neon_vset_lane_i64: |
| 6393 | case NEON::BI__builtin_neon_vset_lane_f32: |
| 6394 | case NEON::BI__builtin_neon_vsetq_lane_i8: |
| 6395 | case NEON::BI__builtin_neon_vsetq_lane_i16: |
| 6396 | case NEON::BI__builtin_neon_vsetq_lane_i32: |
| 6397 | case NEON::BI__builtin_neon_vsetq_lane_i64: |
| 6398 | case NEON::BI__builtin_neon_vsetq_lane_f32: |
| 6399 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); |
| 6400 | |
| 6401 | case NEON::BI__builtin_neon_vsha1h_u32: |
| 6402 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1h), Ops, |
| 6403 | "vsha1h"); |
| 6404 | case NEON::BI__builtin_neon_vsha1cq_u32: |
| 6405 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1c), Ops, |
| 6406 | "vsha1h"); |
| 6407 | case NEON::BI__builtin_neon_vsha1pq_u32: |
| 6408 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1p), Ops, |
| 6409 | "vsha1h"); |
| 6410 | case NEON::BI__builtin_neon_vsha1mq_u32: |
| 6411 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1m), Ops, |
| 6412 | "vsha1h"); |
| 6413 | |
| 6414 | // The ARM _MoveToCoprocessor builtins put the input register value as |
| 6415 | // the first argument, but the LLVM intrinsic expects it as the third one. |
| 6416 | case ARM::BI_MoveToCoprocessor: |
| 6417 | case ARM::BI_MoveToCoprocessor2: { |
| 6418 | Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI_MoveToCoprocessor ? |
| 6419 | Intrinsic::arm_mcr : Intrinsic::arm_mcr2); |
| 6420 | return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0], |
| 6421 | Ops[3], Ops[4], Ops[5]}); |
| 6422 | } |
| 6423 | case ARM::BI_BitScanForward: |
| 6424 | case ARM::BI_BitScanForward64: |
| 6425 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E); |
| 6426 | case ARM::BI_BitScanReverse: |
| 6427 | case ARM::BI_BitScanReverse64: |
| 6428 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E); |
| 6429 | |
| 6430 | case ARM::BI_InterlockedAnd64: |
| 6431 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E); |
| 6432 | case ARM::BI_InterlockedExchange64: |
| 6433 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E); |
| 6434 | case ARM::BI_InterlockedExchangeAdd64: |
| 6435 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E); |
| 6436 | case ARM::BI_InterlockedExchangeSub64: |
| 6437 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E); |
| 6438 | case ARM::BI_InterlockedOr64: |
| 6439 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E); |
| 6440 | case ARM::BI_InterlockedXor64: |
| 6441 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E); |
| 6442 | case ARM::BI_InterlockedDecrement64: |
| 6443 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E); |
| 6444 | case ARM::BI_InterlockedIncrement64: |
| 6445 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E); |
| 6446 | case ARM::BI_InterlockedExchangeAdd8_acq: |
| 6447 | case ARM::BI_InterlockedExchangeAdd16_acq: |
| 6448 | case ARM::BI_InterlockedExchangeAdd_acq: |
| 6449 | case ARM::BI_InterlockedExchangeAdd64_acq: |
| 6450 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_acq, E); |
| 6451 | case ARM::BI_InterlockedExchangeAdd8_rel: |
| 6452 | case ARM::BI_InterlockedExchangeAdd16_rel: |
| 6453 | case ARM::BI_InterlockedExchangeAdd_rel: |
| 6454 | case ARM::BI_InterlockedExchangeAdd64_rel: |
| 6455 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_rel, E); |
| 6456 | case ARM::BI_InterlockedExchangeAdd8_nf: |
| 6457 | case ARM::BI_InterlockedExchangeAdd16_nf: |
| 6458 | case ARM::BI_InterlockedExchangeAdd_nf: |
| 6459 | case ARM::BI_InterlockedExchangeAdd64_nf: |
| 6460 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_nf, E); |
| 6461 | case ARM::BI_InterlockedExchange8_acq: |
| 6462 | case ARM::BI_InterlockedExchange16_acq: |
| 6463 | case ARM::BI_InterlockedExchange_acq: |
| 6464 | case ARM::BI_InterlockedExchange64_acq: |
| 6465 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_acq, E); |
| 6466 | case ARM::BI_InterlockedExchange8_rel: |
| 6467 | case ARM::BI_InterlockedExchange16_rel: |
| 6468 | case ARM::BI_InterlockedExchange_rel: |
| 6469 | case ARM::BI_InterlockedExchange64_rel: |
| 6470 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_rel, E); |
| 6471 | case ARM::BI_InterlockedExchange8_nf: |
| 6472 | case ARM::BI_InterlockedExchange16_nf: |
| 6473 | case ARM::BI_InterlockedExchange_nf: |
| 6474 | case ARM::BI_InterlockedExchange64_nf: |
| 6475 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_nf, E); |
| 6476 | case ARM::BI_InterlockedCompareExchange8_acq: |
| 6477 | case ARM::BI_InterlockedCompareExchange16_acq: |
| 6478 | case ARM::BI_InterlockedCompareExchange_acq: |
| 6479 | case ARM::BI_InterlockedCompareExchange64_acq: |
| 6480 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_acq, E); |
| 6481 | case ARM::BI_InterlockedCompareExchange8_rel: |
| 6482 | case ARM::BI_InterlockedCompareExchange16_rel: |
| 6483 | case ARM::BI_InterlockedCompareExchange_rel: |
| 6484 | case ARM::BI_InterlockedCompareExchange64_rel: |
| 6485 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_rel, E); |
| 6486 | case ARM::BI_InterlockedCompareExchange8_nf: |
| 6487 | case ARM::BI_InterlockedCompareExchange16_nf: |
| 6488 | case ARM::BI_InterlockedCompareExchange_nf: |
| 6489 | case ARM::BI_InterlockedCompareExchange64_nf: |
| 6490 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_nf, E); |
| 6491 | case ARM::BI_InterlockedOr8_acq: |
| 6492 | case ARM::BI_InterlockedOr16_acq: |
| 6493 | case ARM::BI_InterlockedOr_acq: |
| 6494 | case ARM::BI_InterlockedOr64_acq: |
| 6495 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_acq, E); |
| 6496 | case ARM::BI_InterlockedOr8_rel: |
| 6497 | case ARM::BI_InterlockedOr16_rel: |
| 6498 | case ARM::BI_InterlockedOr_rel: |
| 6499 | case ARM::BI_InterlockedOr64_rel: |
| 6500 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_rel, E); |
| 6501 | case ARM::BI_InterlockedOr8_nf: |
| 6502 | case ARM::BI_InterlockedOr16_nf: |
| 6503 | case ARM::BI_InterlockedOr_nf: |
| 6504 | case ARM::BI_InterlockedOr64_nf: |
| 6505 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_nf, E); |
| 6506 | case ARM::BI_InterlockedXor8_acq: |
| 6507 | case ARM::BI_InterlockedXor16_acq: |
| 6508 | case ARM::BI_InterlockedXor_acq: |
| 6509 | case ARM::BI_InterlockedXor64_acq: |
| 6510 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_acq, E); |
| 6511 | case ARM::BI_InterlockedXor8_rel: |
| 6512 | case ARM::BI_InterlockedXor16_rel: |
| 6513 | case ARM::BI_InterlockedXor_rel: |
| 6514 | case ARM::BI_InterlockedXor64_rel: |
| 6515 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_rel, E); |
| 6516 | case ARM::BI_InterlockedXor8_nf: |
| 6517 | case ARM::BI_InterlockedXor16_nf: |
| 6518 | case ARM::BI_InterlockedXor_nf: |
| 6519 | case ARM::BI_InterlockedXor64_nf: |
| 6520 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_nf, E); |
| 6521 | case ARM::BI_InterlockedAnd8_acq: |
| 6522 | case ARM::BI_InterlockedAnd16_acq: |
| 6523 | case ARM::BI_InterlockedAnd_acq: |
| 6524 | case ARM::BI_InterlockedAnd64_acq: |
| 6525 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_acq, E); |
| 6526 | case ARM::BI_InterlockedAnd8_rel: |
| 6527 | case ARM::BI_InterlockedAnd16_rel: |
| 6528 | case ARM::BI_InterlockedAnd_rel: |
| 6529 | case ARM::BI_InterlockedAnd64_rel: |
| 6530 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_rel, E); |
| 6531 | case ARM::BI_InterlockedAnd8_nf: |
| 6532 | case ARM::BI_InterlockedAnd16_nf: |
| 6533 | case ARM::BI_InterlockedAnd_nf: |
| 6534 | case ARM::BI_InterlockedAnd64_nf: |
| 6535 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E); |
| 6536 | case ARM::BI_InterlockedIncrement16_acq: |
| 6537 | case ARM::BI_InterlockedIncrement_acq: |
| 6538 | case ARM::BI_InterlockedIncrement64_acq: |
| 6539 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E); |
| 6540 | case ARM::BI_InterlockedIncrement16_rel: |
| 6541 | case ARM::BI_InterlockedIncrement_rel: |
| 6542 | case ARM::BI_InterlockedIncrement64_rel: |
| 6543 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E); |
| 6544 | case ARM::BI_InterlockedIncrement16_nf: |
| 6545 | case ARM::BI_InterlockedIncrement_nf: |
| 6546 | case ARM::BI_InterlockedIncrement64_nf: |
| 6547 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E); |
| 6548 | case ARM::BI_InterlockedDecrement16_acq: |
| 6549 | case ARM::BI_InterlockedDecrement_acq: |
| 6550 | case ARM::BI_InterlockedDecrement64_acq: |
| 6551 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_acq, E); |
| 6552 | case ARM::BI_InterlockedDecrement16_rel: |
| 6553 | case ARM::BI_InterlockedDecrement_rel: |
| 6554 | case ARM::BI_InterlockedDecrement64_rel: |
| 6555 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_rel, E); |
| 6556 | case ARM::BI_InterlockedDecrement16_nf: |
| 6557 | case ARM::BI_InterlockedDecrement_nf: |
| 6558 | case ARM::BI_InterlockedDecrement64_nf: |
| 6559 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_nf, E); |
| 6560 | } |
| 6561 | |
| 6562 | // Get the last argument, which specifies the vector type. |
| 6563 | assert(HasExtraArg)((HasExtraArg) ? static_cast<void> (0) : __assert_fail ( "HasExtraArg", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6563, __PRETTY_FUNCTION__)); |
| 6564 | llvm::APSInt Result; |
| 6565 | const Expr *Arg = E->getArg(E->getNumArgs()-1); |
| 6566 | if (!Arg->isIntegerConstantExpr(Result, getContext())) |
| 6567 | return nullptr; |
| 6568 | |
| 6569 | if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f || |
| 6570 | BuiltinID == ARM::BI__builtin_arm_vcvtr_d) { |
| 6571 | // Determine the overloaded type of this builtin. |
| 6572 | llvm::Type *Ty; |
| 6573 | if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f) |
| 6574 | Ty = FloatTy; |
| 6575 | else |
| 6576 | Ty = DoubleTy; |
| 6577 | |
| 6578 | // Determine whether this is an unsigned conversion or not. |
| 6579 | bool usgn = Result.getZExtValue() == 1; |
| 6580 | unsigned Int = usgn ? Intrinsic::arm_vcvtru : Intrinsic::arm_vcvtr; |
| 6581 | |
| 6582 | // Call the appropriate intrinsic. |
| 6583 | Function *F = CGM.getIntrinsic(Int, Ty); |
| 6584 | return Builder.CreateCall(F, Ops, "vcvtr"); |
| 6585 | } |
| 6586 | |
| 6587 | // Determine the type of this overloaded NEON intrinsic. |
| 6588 | NeonTypeFlags Type(Result.getZExtValue()); |
| 6589 | bool usgn = Type.isUnsigned(); |
| 6590 | bool rightShift = false; |
| 6591 | |
| 6592 | llvm::VectorType *VTy = GetNeonType(this, Type, |
| 6593 | getTarget().hasLegalHalfType()); |
| 6594 | llvm::Type *Ty = VTy; |
| 6595 | if (!Ty) |
| 6596 | return nullptr; |
| 6597 | |
| 6598 | // Many NEON builtins have identical semantics and uses in ARM and |
| 6599 | // AArch64. Emit these in a single function. |
| 6600 | auto IntrinsicMap = makeArrayRef(ARMSIMDIntrinsicMap); |
| 6601 | const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap( |
| 6602 | IntrinsicMap, BuiltinID, NEONSIMDIntrinsicsProvenSorted); |
| 6603 | if (Builtin) |
| 6604 | return EmitCommonNeonBuiltinExpr( |
| 6605 | Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic, |
| 6606 | Builtin->NameHint, Builtin->TypeModifier, E, Ops, PtrOp0, PtrOp1, Arch); |
| 6607 | |
| 6608 | unsigned Int; |
| 6609 | switch (BuiltinID) { |
| 6610 | default: return nullptr; |
| 6611 | case NEON::BI__builtin_neon_vld1q_lane_v: |
| 6612 | // Handle 64-bit integer elements as a special case. Use shuffles of |
| 6613 | // one-element vectors to avoid poor code for i64 in the backend. |
| 6614 | if (VTy->getElementType()->isIntegerTy(64)) { |
| 6615 | // Extract the other lane. |
| 6616 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 6617 | uint32_t Lane = cast<ConstantInt>(Ops[2])->getZExtValue(); |
| 6618 | Value *SV = llvm::ConstantVector::get(ConstantInt::get(Int32Ty, 1-Lane)); |
| 6619 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV); |
| 6620 | // Load the value as a one-element vector. |
| 6621 | Ty = llvm::VectorType::get(VTy->getElementType(), 1); |
| 6622 | llvm::Type *Tys[] = {Ty, Int8PtrTy}; |
| 6623 | Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vld1, Tys); |
| 6624 | Value *Align = getAlignmentValue32(PtrOp0); |
| 6625 | Value *Ld = Builder.CreateCall(F, {Ops[0], Align}); |
| 6626 | // Combine them. |
| 6627 | uint32_t Indices[] = {1 - Lane, Lane}; |
| 6628 | SV = llvm::ConstantDataVector::get(getLLVMContext(), Indices); |
| 6629 | return Builder.CreateShuffleVector(Ops[1], Ld, SV, "vld1q_lane"); |
| 6630 | } |
| 6631 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 6632 | case NEON::BI__builtin_neon_vld1_lane_v: { |
| 6633 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 6634 | PtrOp0 = Builder.CreateElementBitCast(PtrOp0, VTy->getElementType()); |
| 6635 | Value *Ld = Builder.CreateLoad(PtrOp0); |
| 6636 | return Builder.CreateInsertElement(Ops[1], Ld, Ops[2], "vld1_lane"); |
| 6637 | } |
| 6638 | case NEON::BI__builtin_neon_vqrshrn_n_v: |
| 6639 | Int = |
| 6640 | usgn ? Intrinsic::arm_neon_vqrshiftnu : Intrinsic::arm_neon_vqrshiftns; |
| 6641 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n", |
| 6642 | 1, true); |
| 6643 | case NEON::BI__builtin_neon_vqrshrun_n_v: |
| 6644 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqrshiftnsu, Ty), |
| 6645 | Ops, "vqrshrun_n", 1, true); |
| 6646 | case NEON::BI__builtin_neon_vqshrn_n_v: |
| 6647 | Int = usgn ? Intrinsic::arm_neon_vqshiftnu : Intrinsic::arm_neon_vqshiftns; |
| 6648 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n", |
| 6649 | 1, true); |
| 6650 | case NEON::BI__builtin_neon_vqshrun_n_v: |
| 6651 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqshiftnsu, Ty), |
| 6652 | Ops, "vqshrun_n", 1, true); |
| 6653 | case NEON::BI__builtin_neon_vrecpe_v: |
| 6654 | case NEON::BI__builtin_neon_vrecpeq_v: |
| 6655 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrecpe, Ty), |
| 6656 | Ops, "vrecpe"); |
| 6657 | case NEON::BI__builtin_neon_vrshrn_n_v: |
| 6658 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrshiftn, Ty), |
| 6659 | Ops, "vrshrn_n", 1, true); |
| 6660 | case NEON::BI__builtin_neon_vrsra_n_v: |
| 6661 | case NEON::BI__builtin_neon_vrsraq_n_v: |
| 6662 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 6663 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 6664 | Ops[2] = EmitNeonShiftVector(Ops[2], Ty, true); |
| 6665 | Int = usgn ? Intrinsic::arm_neon_vrshiftu : Intrinsic::arm_neon_vrshifts; |
| 6666 | Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Ty), {Ops[1], Ops[2]}); |
| 6667 | return Builder.CreateAdd(Ops[0], Ops[1], "vrsra_n"); |
| 6668 | case NEON::BI__builtin_neon_vsri_n_v: |
| 6669 | case NEON::BI__builtin_neon_vsriq_n_v: |
| 6670 | rightShift = true; |
| 6671 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 6672 | case NEON::BI__builtin_neon_vsli_n_v: |
| 6673 | case NEON::BI__builtin_neon_vsliq_n_v: |
| 6674 | Ops[2] = EmitNeonShiftVector(Ops[2], Ty, rightShift); |
| 6675 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vshiftins, Ty), |
| 6676 | Ops, "vsli_n"); |
| 6677 | case NEON::BI__builtin_neon_vsra_n_v: |
| 6678 | case NEON::BI__builtin_neon_vsraq_n_v: |
| 6679 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 6680 | Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n"); |
| 6681 | return Builder.CreateAdd(Ops[0], Ops[1]); |
| 6682 | case NEON::BI__builtin_neon_vst1q_lane_v: |
| 6683 | // Handle 64-bit integer elements as a special case. Use a shuffle to get |
| 6684 | // a one-element vector and avoid poor code for i64 in the backend. |
| 6685 | if (VTy->getElementType()->isIntegerTy(64)) { |
| 6686 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 6687 | Value *SV = llvm::ConstantVector::get(cast<llvm::Constant>(Ops[2])); |
| 6688 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV); |
| 6689 | Ops[2] = getAlignmentValue32(PtrOp0); |
| 6690 | llvm::Type *Tys[] = {Int8PtrTy, Ops[1]->getType()}; |
| 6691 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_neon_vst1, |
| 6692 | Tys), Ops); |
| 6693 | } |
| 6694 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 6695 | case NEON::BI__builtin_neon_vst1_lane_v: { |
| 6696 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 6697 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]); |
| 6698 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 6699 | auto St = Builder.CreateStore(Ops[1], Builder.CreateBitCast(PtrOp0, Ty)); |
| 6700 | return St; |
| 6701 | } |
| 6702 | case NEON::BI__builtin_neon_vtbl1_v: |
| 6703 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl1), |
| 6704 | Ops, "vtbl1"); |
| 6705 | case NEON::BI__builtin_neon_vtbl2_v: |
| 6706 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl2), |
| 6707 | Ops, "vtbl2"); |
| 6708 | case NEON::BI__builtin_neon_vtbl3_v: |
| 6709 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl3), |
| 6710 | Ops, "vtbl3"); |
| 6711 | case NEON::BI__builtin_neon_vtbl4_v: |
| 6712 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl4), |
| 6713 | Ops, "vtbl4"); |
| 6714 | case NEON::BI__builtin_neon_vtbx1_v: |
| 6715 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx1), |
| 6716 | Ops, "vtbx1"); |
| 6717 | case NEON::BI__builtin_neon_vtbx2_v: |
| 6718 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx2), |
| 6719 | Ops, "vtbx2"); |
| 6720 | case NEON::BI__builtin_neon_vtbx3_v: |
| 6721 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx3), |
| 6722 | Ops, "vtbx3"); |
| 6723 | case NEON::BI__builtin_neon_vtbx4_v: |
| 6724 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx4), |
| 6725 | Ops, "vtbx4"); |
| 6726 | } |
| 6727 | } |
| 6728 | |
| 6729 | static Value *EmitAArch64TblBuiltinExpr(CodeGenFunction &CGF, unsigned BuiltinID, |
| 6730 | const CallExpr *E, |
| 6731 | SmallVectorImpl<Value *> &Ops, |
| 6732 | llvm::Triple::ArchType Arch) { |
| 6733 | unsigned int Int = 0; |
| 6734 | const char *s = nullptr; |
| 6735 | |
| 6736 | switch (BuiltinID) { |
| 6737 | default: |
| 6738 | return nullptr; |
| 6739 | case NEON::BI__builtin_neon_vtbl1_v: |
| 6740 | case NEON::BI__builtin_neon_vqtbl1_v: |
| 6741 | case NEON::BI__builtin_neon_vqtbl1q_v: |
| 6742 | case NEON::BI__builtin_neon_vtbl2_v: |
| 6743 | case NEON::BI__builtin_neon_vqtbl2_v: |
| 6744 | case NEON::BI__builtin_neon_vqtbl2q_v: |
| 6745 | case NEON::BI__builtin_neon_vtbl3_v: |
| 6746 | case NEON::BI__builtin_neon_vqtbl3_v: |
| 6747 | case NEON::BI__builtin_neon_vqtbl3q_v: |
| 6748 | case NEON::BI__builtin_neon_vtbl4_v: |
| 6749 | case NEON::BI__builtin_neon_vqtbl4_v: |
| 6750 | case NEON::BI__builtin_neon_vqtbl4q_v: |
| 6751 | break; |
| 6752 | case NEON::BI__builtin_neon_vtbx1_v: |
| 6753 | case NEON::BI__builtin_neon_vqtbx1_v: |
| 6754 | case NEON::BI__builtin_neon_vqtbx1q_v: |
| 6755 | case NEON::BI__builtin_neon_vtbx2_v: |
| 6756 | case NEON::BI__builtin_neon_vqtbx2_v: |
| 6757 | case NEON::BI__builtin_neon_vqtbx2q_v: |
| 6758 | case NEON::BI__builtin_neon_vtbx3_v: |
| 6759 | case NEON::BI__builtin_neon_vqtbx3_v: |
| 6760 | case NEON::BI__builtin_neon_vqtbx3q_v: |
| 6761 | case NEON::BI__builtin_neon_vtbx4_v: |
| 6762 | case NEON::BI__builtin_neon_vqtbx4_v: |
| 6763 | case NEON::BI__builtin_neon_vqtbx4q_v: |
| 6764 | break; |
| 6765 | } |
| 6766 | |
| 6767 | assert(E->getNumArgs() >= 3)((E->getNumArgs() >= 3) ? static_cast<void> (0) : __assert_fail ("E->getNumArgs() >= 3", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6767, __PRETTY_FUNCTION__)); |
| 6768 | |
| 6769 | // Get the last argument, which specifies the vector type. |
| 6770 | llvm::APSInt Result; |
| 6771 | const Expr *Arg = E->getArg(E->getNumArgs() - 1); |
| 6772 | if (!Arg->isIntegerConstantExpr(Result, CGF.getContext())) |
| 6773 | return nullptr; |
| 6774 | |
| 6775 | // Determine the type of this overloaded NEON intrinsic. |
| 6776 | NeonTypeFlags Type(Result.getZExtValue()); |
| 6777 | llvm::VectorType *Ty = GetNeonType(&CGF, Type); |
| 6778 | if (!Ty) |
| 6779 | return nullptr; |
| 6780 | |
| 6781 | CodeGen::CGBuilderTy &Builder = CGF.Builder; |
| 6782 | |
| 6783 | // AArch64 scalar builtins are not overloaded, they do not have an extra |
| 6784 | // argument that specifies the vector type, need to handle each case. |
| 6785 | switch (BuiltinID) { |
| 6786 | case NEON::BI__builtin_neon_vtbl1_v: { |
| 6787 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 1), nullptr, |
| 6788 | Ops[1], Ty, Intrinsic::aarch64_neon_tbl1, |
| 6789 | "vtbl1"); |
| 6790 | } |
| 6791 | case NEON::BI__builtin_neon_vtbl2_v: { |
| 6792 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 2), nullptr, |
| 6793 | Ops[2], Ty, Intrinsic::aarch64_neon_tbl1, |
| 6794 | "vtbl1"); |
| 6795 | } |
| 6796 | case NEON::BI__builtin_neon_vtbl3_v: { |
| 6797 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 3), nullptr, |
| 6798 | Ops[3], Ty, Intrinsic::aarch64_neon_tbl2, |
| 6799 | "vtbl2"); |
| 6800 | } |
| 6801 | case NEON::BI__builtin_neon_vtbl4_v: { |
| 6802 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 4), nullptr, |
| 6803 | Ops[4], Ty, Intrinsic::aarch64_neon_tbl2, |
| 6804 | "vtbl2"); |
| 6805 | } |
| 6806 | case NEON::BI__builtin_neon_vtbx1_v: { |
| 6807 | Value *TblRes = |
| 6808 | packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 1), nullptr, Ops[2], |
| 6809 | Ty, Intrinsic::aarch64_neon_tbl1, "vtbl1"); |
| 6810 | |
| 6811 | llvm::Constant *EightV = ConstantInt::get(Ty, 8); |
| 6812 | Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[2], EightV); |
| 6813 | CmpRes = Builder.CreateSExt(CmpRes, Ty); |
| 6814 | |
| 6815 | Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]); |
| 6816 | Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes); |
| 6817 | return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx"); |
| 6818 | } |
| 6819 | case NEON::BI__builtin_neon_vtbx2_v: { |
| 6820 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 2), Ops[0], |
| 6821 | Ops[3], Ty, Intrinsic::aarch64_neon_tbx1, |
| 6822 | "vtbx1"); |
| 6823 | } |
| 6824 | case NEON::BI__builtin_neon_vtbx3_v: { |
| 6825 | Value *TblRes = |
| 6826 | packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 3), nullptr, Ops[4], |
| 6827 | Ty, Intrinsic::aarch64_neon_tbl2, "vtbl2"); |
| 6828 | |
| 6829 | llvm::Constant *TwentyFourV = ConstantInt::get(Ty, 24); |
| 6830 | Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[4], |
| 6831 | TwentyFourV); |
| 6832 | CmpRes = Builder.CreateSExt(CmpRes, Ty); |
| 6833 | |
| 6834 | Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]); |
| 6835 | Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes); |
| 6836 | return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx"); |
| 6837 | } |
| 6838 | case NEON::BI__builtin_neon_vtbx4_v: { |
| 6839 | return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 4), Ops[0], |
| 6840 | Ops[5], Ty, Intrinsic::aarch64_neon_tbx2, |
| 6841 | "vtbx2"); |
| 6842 | } |
| 6843 | case NEON::BI__builtin_neon_vqtbl1_v: |
| 6844 | case NEON::BI__builtin_neon_vqtbl1q_v: |
| 6845 | Int = Intrinsic::aarch64_neon_tbl1; s = "vtbl1"; break; |
| 6846 | case NEON::BI__builtin_neon_vqtbl2_v: |
| 6847 | case NEON::BI__builtin_neon_vqtbl2q_v: { |
| 6848 | Int = Intrinsic::aarch64_neon_tbl2; s = "vtbl2"; break; |
| 6849 | case NEON::BI__builtin_neon_vqtbl3_v: |
| 6850 | case NEON::BI__builtin_neon_vqtbl3q_v: |
| 6851 | Int = Intrinsic::aarch64_neon_tbl3; s = "vtbl3"; break; |
| 6852 | case NEON::BI__builtin_neon_vqtbl4_v: |
| 6853 | case NEON::BI__builtin_neon_vqtbl4q_v: |
| 6854 | Int = Intrinsic::aarch64_neon_tbl4; s = "vtbl4"; break; |
| 6855 | case NEON::BI__builtin_neon_vqtbx1_v: |
| 6856 | case NEON::BI__builtin_neon_vqtbx1q_v: |
| 6857 | Int = Intrinsic::aarch64_neon_tbx1; s = "vtbx1"; break; |
| 6858 | case NEON::BI__builtin_neon_vqtbx2_v: |
| 6859 | case NEON::BI__builtin_neon_vqtbx2q_v: |
| 6860 | Int = Intrinsic::aarch64_neon_tbx2; s = "vtbx2"; break; |
| 6861 | case NEON::BI__builtin_neon_vqtbx3_v: |
| 6862 | case NEON::BI__builtin_neon_vqtbx3q_v: |
| 6863 | Int = Intrinsic::aarch64_neon_tbx3; s = "vtbx3"; break; |
| 6864 | case NEON::BI__builtin_neon_vqtbx4_v: |
| 6865 | case NEON::BI__builtin_neon_vqtbx4q_v: |
| 6866 | Int = Intrinsic::aarch64_neon_tbx4; s = "vtbx4"; break; |
| 6867 | } |
| 6868 | } |
| 6869 | |
| 6870 | if (!Int) |
| 6871 | return nullptr; |
| 6872 | |
| 6873 | Function *F = CGF.CGM.getIntrinsic(Int, Ty); |
| 6874 | return CGF.EmitNeonCall(F, Ops, s); |
| 6875 | } |
| 6876 | |
| 6877 | Value *CodeGenFunction::vectorWrapScalar16(Value *Op) { |
| 6878 | llvm::Type *VTy = llvm::VectorType::get(Int16Ty, 4); |
| 6879 | Op = Builder.CreateBitCast(Op, Int16Ty); |
| 6880 | Value *V = UndefValue::get(VTy); |
| 6881 | llvm::Constant *CI = ConstantInt::get(SizeTy, 0); |
| 6882 | Op = Builder.CreateInsertElement(V, Op, CI); |
| 6883 | return Op; |
| 6884 | } |
| 6885 | |
| 6886 | Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID, |
| 6887 | const CallExpr *E, |
| 6888 | llvm::Triple::ArchType Arch) { |
| 6889 | unsigned HintID = static_cast<unsigned>(-1); |
| 6890 | switch (BuiltinID) { |
| 6891 | default: break; |
| 6892 | case AArch64::BI__builtin_arm_nop: |
| 6893 | HintID = 0; |
| 6894 | break; |
| 6895 | case AArch64::BI__builtin_arm_yield: |
| 6896 | case AArch64::BI__yield: |
| 6897 | HintID = 1; |
| 6898 | break; |
| 6899 | case AArch64::BI__builtin_arm_wfe: |
| 6900 | case AArch64::BI__wfe: |
| 6901 | HintID = 2; |
| 6902 | break; |
| 6903 | case AArch64::BI__builtin_arm_wfi: |
| 6904 | case AArch64::BI__wfi: |
| 6905 | HintID = 3; |
| 6906 | break; |
| 6907 | case AArch64::BI__builtin_arm_sev: |
| 6908 | case AArch64::BI__sev: |
| 6909 | HintID = 4; |
| 6910 | break; |
| 6911 | case AArch64::BI__builtin_arm_sevl: |
| 6912 | case AArch64::BI__sevl: |
| 6913 | HintID = 5; |
| 6914 | break; |
| 6915 | } |
| 6916 | |
| 6917 | if (HintID != static_cast<unsigned>(-1)) { |
| 6918 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_hint); |
| 6919 | return Builder.CreateCall(F, llvm::ConstantInt::get(Int32Ty, HintID)); |
| 6920 | } |
| 6921 | |
| 6922 | if (BuiltinID == AArch64::BI__builtin_arm_prefetch) { |
| 6923 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 6924 | Value *RW = EmitScalarExpr(E->getArg(1)); |
| 6925 | Value *CacheLevel = EmitScalarExpr(E->getArg(2)); |
| 6926 | Value *RetentionPolicy = EmitScalarExpr(E->getArg(3)); |
| 6927 | Value *IsData = EmitScalarExpr(E->getArg(4)); |
| 6928 | |
| 6929 | Value *Locality = nullptr; |
| 6930 | if (cast<llvm::ConstantInt>(RetentionPolicy)->isZero()) { |
| 6931 | // Temporal fetch, needs to convert cache level to locality. |
| 6932 | Locality = llvm::ConstantInt::get(Int32Ty, |
| 6933 | -cast<llvm::ConstantInt>(CacheLevel)->getValue() + 3); |
| 6934 | } else { |
| 6935 | // Streaming fetch. |
| 6936 | Locality = llvm::ConstantInt::get(Int32Ty, 0); |
| 6937 | } |
| 6938 | |
| 6939 | // FIXME: We need AArch64 specific LLVM intrinsic if we want to specify |
| 6940 | // PLDL3STRM or PLDL2STRM. |
| 6941 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch); |
| 6942 | return Builder.CreateCall(F, {Address, RW, Locality, IsData}); |
| 6943 | } |
| 6944 | |
| 6945 | if (BuiltinID == AArch64::BI__builtin_arm_rbit) { |
| 6946 | assert((getContext().getTypeSize(E->getType()) == 32) &&(((getContext().getTypeSize(E->getType()) == 32) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6947, __PRETTY_FUNCTION__)) |
| 6947 | "rbit of unusual size!")(((getContext().getTypeSize(E->getType()) == 32) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6947, __PRETTY_FUNCTION__)); |
| 6948 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); |
| 6949 | return Builder.CreateCall( |
| 6950 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); |
| 6951 | } |
| 6952 | if (BuiltinID == AArch64::BI__builtin_arm_rbit64) { |
| 6953 | assert((getContext().getTypeSize(E->getType()) == 64) &&(((getContext().getTypeSize(E->getType()) == 64) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6954, __PRETTY_FUNCTION__)) |
| 6954 | "rbit of unusual size!")(((getContext().getTypeSize(E->getType()) == 64) && "rbit of unusual size!") ? static_cast<void> (0) : __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6954, __PRETTY_FUNCTION__)); |
| 6955 | llvm::Value *Arg = EmitScalarExpr(E->getArg(0)); |
| 6956 | return Builder.CreateCall( |
| 6957 | CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit"); |
| 6958 | } |
| 6959 | |
| 6960 | if (BuiltinID == AArch64::BI__clear_cache) { |
| 6961 | assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments")((E->getNumArgs() == 2 && "__clear_cache takes 2 arguments" ) ? static_cast<void> (0) : __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 6961, __PRETTY_FUNCTION__)); |
| 6962 | const FunctionDecl *FD = E->getDirectCallee(); |
| 6963 | Value *Ops[2]; |
| 6964 | for (unsigned i = 0; i < 2; i++) |
| 6965 | Ops[i] = EmitScalarExpr(E->getArg(i)); |
| 6966 | llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType()); |
| 6967 | llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty); |
| 6968 | StringRef Name = FD->getName(); |
| 6969 | return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops); |
| 6970 | } |
| 6971 | |
| 6972 | if ((BuiltinID == AArch64::BI__builtin_arm_ldrex || |
| 6973 | BuiltinID == AArch64::BI__builtin_arm_ldaex) && |
| 6974 | getContext().getTypeSize(E->getType()) == 128) { |
| 6975 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex |
| 6976 | ? Intrinsic::aarch64_ldaxp |
| 6977 | : Intrinsic::aarch64_ldxp); |
| 6978 | |
| 6979 | Value *LdPtr = EmitScalarExpr(E->getArg(0)); |
| 6980 | Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy), |
| 6981 | "ldxp"); |
| 6982 | |
| 6983 | Value *Val0 = Builder.CreateExtractValue(Val, 1); |
| 6984 | Value *Val1 = Builder.CreateExtractValue(Val, 0); |
| 6985 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); |
| 6986 | Val0 = Builder.CreateZExt(Val0, Int128Ty); |
| 6987 | Val1 = Builder.CreateZExt(Val1, Int128Ty); |
| 6988 | |
| 6989 | Value *ShiftCst = llvm::ConstantInt::get(Int128Ty, 64); |
| 6990 | Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */); |
| 6991 | Val = Builder.CreateOr(Val, Val1); |
| 6992 | return Builder.CreateBitCast(Val, ConvertType(E->getType())); |
| 6993 | } else if (BuiltinID == AArch64::BI__builtin_arm_ldrex || |
| 6994 | BuiltinID == AArch64::BI__builtin_arm_ldaex) { |
| 6995 | Value *LoadAddr = EmitScalarExpr(E->getArg(0)); |
| 6996 | |
| 6997 | QualType Ty = E->getType(); |
| 6998 | llvm::Type *RealResTy = ConvertType(Ty); |
| 6999 | llvm::Type *PtrTy = llvm::IntegerType::get( |
| 7000 | getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo(); |
| 7001 | LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy); |
| 7002 | |
| 7003 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex |
| 7004 | ? Intrinsic::aarch64_ldaxr |
| 7005 | : Intrinsic::aarch64_ldxr, |
| 7006 | PtrTy); |
| 7007 | Value *Val = Builder.CreateCall(F, LoadAddr, "ldxr"); |
| 7008 | |
| 7009 | if (RealResTy->isPointerTy()) |
| 7010 | return Builder.CreateIntToPtr(Val, RealResTy); |
| 7011 | |
| 7012 | llvm::Type *IntResTy = llvm::IntegerType::get( |
| 7013 | getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy)); |
| 7014 | Val = Builder.CreateTruncOrBitCast(Val, IntResTy); |
| 7015 | return Builder.CreateBitCast(Val, RealResTy); |
| 7016 | } |
| 7017 | |
| 7018 | if ((BuiltinID == AArch64::BI__builtin_arm_strex || |
| 7019 | BuiltinID == AArch64::BI__builtin_arm_stlex) && |
| 7020 | getContext().getTypeSize(E->getArg(0)->getType()) == 128) { |
| 7021 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex |
| 7022 | ? Intrinsic::aarch64_stlxp |
| 7023 | : Intrinsic::aarch64_stxp); |
| 7024 | llvm::Type *STy = llvm::StructType::get(Int64Ty, Int64Ty); |
| 7025 | |
| 7026 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); |
| 7027 | EmitAnyExprToMem(E->getArg(0), Tmp, Qualifiers(), /*init*/ true); |
| 7028 | |
| 7029 | Tmp = Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(STy)); |
| 7030 | llvm::Value *Val = Builder.CreateLoad(Tmp); |
| 7031 | |
| 7032 | Value *Arg0 = Builder.CreateExtractValue(Val, 0); |
| 7033 | Value *Arg1 = Builder.CreateExtractValue(Val, 1); |
| 7034 | Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), |
| 7035 | Int8PtrTy); |
| 7036 | return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "stxp"); |
| 7037 | } |
| 7038 | |
| 7039 | if (BuiltinID == AArch64::BI__builtin_arm_strex || |
| 7040 | BuiltinID == AArch64::BI__builtin_arm_stlex) { |
| 7041 | Value *StoreVal = EmitScalarExpr(E->getArg(0)); |
| 7042 | Value *StoreAddr = EmitScalarExpr(E->getArg(1)); |
| 7043 | |
| 7044 | QualType Ty = E->getArg(0)->getType(); |
| 7045 | llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(), |
| 7046 | getContext().getTypeSize(Ty)); |
| 7047 | StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo()); |
| 7048 | |
| 7049 | if (StoreVal->getType()->isPointerTy()) |
| 7050 | StoreVal = Builder.CreatePtrToInt(StoreVal, Int64Ty); |
| 7051 | else { |
| 7052 | llvm::Type *IntTy = llvm::IntegerType::get( |
| 7053 | getLLVMContext(), |
| 7054 | CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType())); |
| 7055 | StoreVal = Builder.CreateBitCast(StoreVal, IntTy); |
| 7056 | StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int64Ty); |
| 7057 | } |
| 7058 | |
| 7059 | Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex |
| 7060 | ? Intrinsic::aarch64_stlxr |
| 7061 | : Intrinsic::aarch64_stxr, |
| 7062 | StoreAddr->getType()); |
| 7063 | return Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr"); |
| 7064 | } |
| 7065 | |
| 7066 | if (BuiltinID == AArch64::BI__getReg) { |
| 7067 | Expr::EvalResult Result; |
| 7068 | if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext())) |
| 7069 | llvm_unreachable("Sema will ensure that the parameter is constant")::llvm::llvm_unreachable_internal("Sema will ensure that the parameter is constant" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7069); |
| 7070 | |
| 7071 | llvm::APSInt Value = Result.Val.getInt(); |
| 7072 | LLVMContext &Context = CGM.getLLVMContext(); |
| 7073 | std::string Reg = Value == 31 ? "sp" : "x" + Value.toString(10); |
| 7074 | |
| 7075 | llvm::Metadata *Ops[] = {llvm::MDString::get(Context, Reg)}; |
| 7076 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); |
| 7077 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); |
| 7078 | |
| 7079 | llvm::Function *F = |
| 7080 | CGM.getIntrinsic(llvm::Intrinsic::read_register, {Int64Ty}); |
| 7081 | return Builder.CreateCall(F, Metadata); |
| 7082 | } |
| 7083 | |
| 7084 | if (BuiltinID == AArch64::BI__builtin_arm_clrex) { |
| 7085 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_clrex); |
| 7086 | return Builder.CreateCall(F); |
| 7087 | } |
| 7088 | |
| 7089 | if (BuiltinID == AArch64::BI_ReadWriteBarrier) |
| 7090 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, |
| 7091 | llvm::SyncScope::SingleThread); |
| 7092 | |
| 7093 | // CRC32 |
| 7094 | Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic; |
| 7095 | switch (BuiltinID) { |
| 7096 | case AArch64::BI__builtin_arm_crc32b: |
| 7097 | CRCIntrinsicID = Intrinsic::aarch64_crc32b; break; |
| 7098 | case AArch64::BI__builtin_arm_crc32cb: |
| 7099 | CRCIntrinsicID = Intrinsic::aarch64_crc32cb; break; |
| 7100 | case AArch64::BI__builtin_arm_crc32h: |
| 7101 | CRCIntrinsicID = Intrinsic::aarch64_crc32h; break; |
| 7102 | case AArch64::BI__builtin_arm_crc32ch: |
| 7103 | CRCIntrinsicID = Intrinsic::aarch64_crc32ch; break; |
| 7104 | case AArch64::BI__builtin_arm_crc32w: |
| 7105 | CRCIntrinsicID = Intrinsic::aarch64_crc32w; break; |
| 7106 | case AArch64::BI__builtin_arm_crc32cw: |
| 7107 | CRCIntrinsicID = Intrinsic::aarch64_crc32cw; break; |
| 7108 | case AArch64::BI__builtin_arm_crc32d: |
| 7109 | CRCIntrinsicID = Intrinsic::aarch64_crc32x; break; |
| 7110 | case AArch64::BI__builtin_arm_crc32cd: |
| 7111 | CRCIntrinsicID = Intrinsic::aarch64_crc32cx; break; |
| 7112 | } |
| 7113 | |
| 7114 | if (CRCIntrinsicID != Intrinsic::not_intrinsic) { |
| 7115 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); |
| 7116 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 7117 | Function *F = CGM.getIntrinsic(CRCIntrinsicID); |
| 7118 | |
| 7119 | llvm::Type *DataTy = F->getFunctionType()->getParamType(1); |
| 7120 | Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy); |
| 7121 | |
| 7122 | return Builder.CreateCall(F, {Arg0, Arg1}); |
| 7123 | } |
| 7124 | |
| 7125 | // Memory Tagging Extensions (MTE) Intrinsics |
| 7126 | Intrinsic::ID MTEIntrinsicID = Intrinsic::not_intrinsic; |
| 7127 | switch (BuiltinID) { |
| 7128 | case AArch64::BI__builtin_arm_irg: |
| 7129 | MTEIntrinsicID = Intrinsic::aarch64_irg; break; |
| 7130 | case AArch64::BI__builtin_arm_addg: |
| 7131 | MTEIntrinsicID = Intrinsic::aarch64_addg; break; |
| 7132 | case AArch64::BI__builtin_arm_gmi: |
| 7133 | MTEIntrinsicID = Intrinsic::aarch64_gmi; break; |
| 7134 | case AArch64::BI__builtin_arm_ldg: |
| 7135 | MTEIntrinsicID = Intrinsic::aarch64_ldg; break; |
| 7136 | case AArch64::BI__builtin_arm_stg: |
| 7137 | MTEIntrinsicID = Intrinsic::aarch64_stg; break; |
| 7138 | case AArch64::BI__builtin_arm_subp: |
| 7139 | MTEIntrinsicID = Intrinsic::aarch64_subp; break; |
| 7140 | } |
| 7141 | |
| 7142 | if (MTEIntrinsicID != Intrinsic::not_intrinsic) { |
| 7143 | llvm::Type *T = ConvertType(E->getType()); |
| 7144 | |
| 7145 | if (MTEIntrinsicID == Intrinsic::aarch64_irg) { |
| 7146 | Value *Pointer = EmitScalarExpr(E->getArg(0)); |
| 7147 | Value *Mask = EmitScalarExpr(E->getArg(1)); |
| 7148 | |
| 7149 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); |
| 7150 | Mask = Builder.CreateZExt(Mask, Int64Ty); |
| 7151 | Value *RV = Builder.CreateCall( |
| 7152 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, Mask}); |
| 7153 | return Builder.CreatePointerCast(RV, T); |
| 7154 | } |
| 7155 | if (MTEIntrinsicID == Intrinsic::aarch64_addg) { |
| 7156 | Value *Pointer = EmitScalarExpr(E->getArg(0)); |
| 7157 | Value *TagOffset = EmitScalarExpr(E->getArg(1)); |
| 7158 | |
| 7159 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); |
| 7160 | TagOffset = Builder.CreateZExt(TagOffset, Int64Ty); |
| 7161 | Value *RV = Builder.CreateCall( |
| 7162 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, TagOffset}); |
| 7163 | return Builder.CreatePointerCast(RV, T); |
| 7164 | } |
| 7165 | if (MTEIntrinsicID == Intrinsic::aarch64_gmi) { |
| 7166 | Value *Pointer = EmitScalarExpr(E->getArg(0)); |
| 7167 | Value *ExcludedMask = EmitScalarExpr(E->getArg(1)); |
| 7168 | |
| 7169 | ExcludedMask = Builder.CreateZExt(ExcludedMask, Int64Ty); |
| 7170 | Pointer = Builder.CreatePointerCast(Pointer, Int8PtrTy); |
| 7171 | return Builder.CreateCall( |
| 7172 | CGM.getIntrinsic(MTEIntrinsicID), {Pointer, ExcludedMask}); |
| 7173 | } |
| 7174 | // Although it is possible to supply a different return |
| 7175 | // address (first arg) to this intrinsic, for now we set |
| 7176 | // return address same as input address. |
| 7177 | if (MTEIntrinsicID == Intrinsic::aarch64_ldg) { |
| 7178 | Value *TagAddress = EmitScalarExpr(E->getArg(0)); |
| 7179 | TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy); |
| 7180 | Value *RV = Builder.CreateCall( |
| 7181 | CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress}); |
| 7182 | return Builder.CreatePointerCast(RV, T); |
| 7183 | } |
| 7184 | // Although it is possible to supply a different tag (to set) |
| 7185 | // to this intrinsic (as first arg), for now we supply |
| 7186 | // the tag that is in input address arg (common use case). |
| 7187 | if (MTEIntrinsicID == Intrinsic::aarch64_stg) { |
| 7188 | Value *TagAddress = EmitScalarExpr(E->getArg(0)); |
| 7189 | TagAddress = Builder.CreatePointerCast(TagAddress, Int8PtrTy); |
| 7190 | return Builder.CreateCall( |
| 7191 | CGM.getIntrinsic(MTEIntrinsicID), {TagAddress, TagAddress}); |
| 7192 | } |
| 7193 | if (MTEIntrinsicID == Intrinsic::aarch64_subp) { |
| 7194 | Value *PointerA = EmitScalarExpr(E->getArg(0)); |
| 7195 | Value *PointerB = EmitScalarExpr(E->getArg(1)); |
| 7196 | PointerA = Builder.CreatePointerCast(PointerA, Int8PtrTy); |
| 7197 | PointerB = Builder.CreatePointerCast(PointerB, Int8PtrTy); |
| 7198 | return Builder.CreateCall( |
| 7199 | CGM.getIntrinsic(MTEIntrinsicID), {PointerA, PointerB}); |
| 7200 | } |
| 7201 | } |
| 7202 | |
| 7203 | if (BuiltinID == AArch64::BI__builtin_arm_rsr || |
| 7204 | BuiltinID == AArch64::BI__builtin_arm_rsr64 || |
| 7205 | BuiltinID == AArch64::BI__builtin_arm_rsrp || |
| 7206 | BuiltinID == AArch64::BI__builtin_arm_wsr || |
| 7207 | BuiltinID == AArch64::BI__builtin_arm_wsr64 || |
| 7208 | BuiltinID == AArch64::BI__builtin_arm_wsrp) { |
| 7209 | |
| 7210 | bool IsRead = BuiltinID == AArch64::BI__builtin_arm_rsr || |
| 7211 | BuiltinID == AArch64::BI__builtin_arm_rsr64 || |
| 7212 | BuiltinID == AArch64::BI__builtin_arm_rsrp; |
| 7213 | |
| 7214 | bool IsPointerBuiltin = BuiltinID == AArch64::BI__builtin_arm_rsrp || |
| 7215 | BuiltinID == AArch64::BI__builtin_arm_wsrp; |
| 7216 | |
| 7217 | bool Is64Bit = BuiltinID != AArch64::BI__builtin_arm_rsr && |
| 7218 | BuiltinID != AArch64::BI__builtin_arm_wsr; |
| 7219 | |
| 7220 | llvm::Type *ValueType; |
| 7221 | llvm::Type *RegisterType = Int64Ty; |
| 7222 | if (IsPointerBuiltin) { |
| 7223 | ValueType = VoidPtrTy; |
| 7224 | } else if (Is64Bit) { |
| 7225 | ValueType = Int64Ty; |
| 7226 | } else { |
| 7227 | ValueType = Int32Ty; |
| 7228 | } |
| 7229 | |
| 7230 | return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead); |
| 7231 | } |
| 7232 | |
| 7233 | if (BuiltinID == AArch64::BI_ReadStatusReg || |
| 7234 | BuiltinID == AArch64::BI_WriteStatusReg) { |
| 7235 | LLVMContext &Context = CGM.getLLVMContext(); |
| 7236 | |
| 7237 | unsigned SysReg = |
| 7238 | E->getArg(0)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
| 7239 | |
| 7240 | std::string SysRegStr; |
| 7241 | llvm::raw_string_ostream(SysRegStr) << |
| 7242 | ((1 << 1) | ((SysReg >> 14) & 1)) << ":" << |
| 7243 | ((SysReg >> 11) & 7) << ":" << |
| 7244 | ((SysReg >> 7) & 15) << ":" << |
| 7245 | ((SysReg >> 3) & 15) << ":" << |
| 7246 | ( SysReg & 7); |
| 7247 | |
| 7248 | llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysRegStr) }; |
| 7249 | llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops); |
| 7250 | llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName); |
| 7251 | |
| 7252 | llvm::Type *RegisterType = Int64Ty; |
| 7253 | llvm::Type *Types[] = { RegisterType }; |
| 7254 | |
| 7255 | if (BuiltinID == AArch64::BI_ReadStatusReg) { |
| 7256 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types); |
| 7257 | |
| 7258 | return Builder.CreateCall(F, Metadata); |
| 7259 | } |
| 7260 | |
| 7261 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types); |
| 7262 | llvm::Value *ArgValue = EmitScalarExpr(E->getArg(1)); |
| 7263 | |
| 7264 | return Builder.CreateCall(F, { Metadata, ArgValue }); |
| 7265 | } |
| 7266 | |
| 7267 | if (BuiltinID == AArch64::BI_AddressOfReturnAddress) { |
| 7268 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::addressofreturnaddress); |
| 7269 | return Builder.CreateCall(F); |
| 7270 | } |
| 7271 | |
| 7272 | if (BuiltinID == AArch64::BI__builtin_sponentry) { |
| 7273 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::sponentry); |
| 7274 | return Builder.CreateCall(F); |
| 7275 | } |
| 7276 | |
| 7277 | // Find out if any arguments are required to be integer constant |
| 7278 | // expressions. |
| 7279 | unsigned ICEArguments = 0; |
| 7280 | ASTContext::GetBuiltinTypeError Error; |
| 7281 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); |
| 7282 | assert(Error == ASTContext::GE_None && "Should not codegen an error")((Error == ASTContext::GE_None && "Should not codegen an error" ) ? static_cast<void> (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7282, __PRETTY_FUNCTION__)); |
| 7283 | |
| 7284 | llvm::SmallVector<Value*, 4> Ops; |
| 7285 | for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++) { |
| 7286 | if ((ICEArguments & (1 << i)) == 0) { |
| 7287 | Ops.push_back(EmitScalarExpr(E->getArg(i))); |
| 7288 | } else { |
| 7289 | // If this is required to be a constant, constant fold it so that we know |
| 7290 | // that the generated intrinsic gets a ConstantInt. |
| 7291 | llvm::APSInt Result; |
| 7292 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext()); |
| 7293 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7293, __PRETTY_FUNCTION__)); |
| 7294 | (void)IsConst; |
| 7295 | Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result)); |
| 7296 | } |
| 7297 | } |
| 7298 | |
| 7299 | auto SISDMap = makeArrayRef(AArch64SISDIntrinsicMap); |
| 7300 | const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap( |
| 7301 | SISDMap, BuiltinID, AArch64SISDIntrinsicsProvenSorted); |
| 7302 | |
| 7303 | if (Builtin) { |
| 7304 | Ops.push_back(EmitScalarExpr(E->getArg(E->getNumArgs() - 1))); |
| 7305 | Value *Result = EmitCommonNeonSISDBuiltinExpr(*this, *Builtin, Ops, E); |
| 7306 | assert(Result && "SISD intrinsic should have been handled")((Result && "SISD intrinsic should have been handled" ) ? static_cast<void> (0) : __assert_fail ("Result && \"SISD intrinsic should have been handled\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7306, __PRETTY_FUNCTION__)); |
| 7307 | return Result; |
| 7308 | } |
| 7309 | |
| 7310 | llvm::APSInt Result; |
| 7311 | const Expr *Arg = E->getArg(E->getNumArgs()-1); |
| 7312 | NeonTypeFlags Type(0); |
| 7313 | if (Arg->isIntegerConstantExpr(Result, getContext())) |
| 7314 | // Determine the type of this overloaded NEON intrinsic. |
| 7315 | Type = NeonTypeFlags(Result.getZExtValue()); |
| 7316 | |
| 7317 | bool usgn = Type.isUnsigned(); |
| 7318 | bool quad = Type.isQuad(); |
| 7319 | |
| 7320 | // Handle non-overloaded intrinsics first. |
| 7321 | switch (BuiltinID) { |
| 7322 | default: break; |
| 7323 | case NEON::BI__builtin_neon_vabsh_f16: |
| 7324 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7325 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, HalfTy), Ops, "vabs"); |
| 7326 | case NEON::BI__builtin_neon_vldrq_p128: { |
| 7327 | llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128); |
| 7328 | llvm::Type *Int128PTy = llvm::PointerType::get(Int128Ty, 0); |
| 7329 | Value *Ptr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int128PTy); |
| 7330 | return Builder.CreateAlignedLoad(Int128Ty, Ptr, |
| 7331 | CharUnits::fromQuantity(16)); |
| 7332 | } |
| 7333 | case NEON::BI__builtin_neon_vstrq_p128: { |
| 7334 | llvm::Type *Int128PTy = llvm::Type::getIntNPtrTy(getLLVMContext(), 128); |
| 7335 | Value *Ptr = Builder.CreateBitCast(Ops[0], Int128PTy); |
| 7336 | return Builder.CreateDefaultAlignedStore(EmitScalarExpr(E->getArg(1)), Ptr); |
| 7337 | } |
| 7338 | case NEON::BI__builtin_neon_vcvts_u32_f32: |
| 7339 | case NEON::BI__builtin_neon_vcvtd_u64_f64: |
| 7340 | usgn = true; |
| 7341 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 7342 | case NEON::BI__builtin_neon_vcvts_s32_f32: |
| 7343 | case NEON::BI__builtin_neon_vcvtd_s64_f64: { |
| 7344 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7345 | bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64; |
| 7346 | llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty; |
| 7347 | llvm::Type *FTy = Is64 ? DoubleTy : FloatTy; |
| 7348 | Ops[0] = Builder.CreateBitCast(Ops[0], FTy); |
| 7349 | if (usgn) |
| 7350 | return Builder.CreateFPToUI(Ops[0], InTy); |
| 7351 | return Builder.CreateFPToSI(Ops[0], InTy); |
| 7352 | } |
| 7353 | case NEON::BI__builtin_neon_vcvts_f32_u32: |
| 7354 | case NEON::BI__builtin_neon_vcvtd_f64_u64: |
| 7355 | usgn = true; |
| 7356 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 7357 | case NEON::BI__builtin_neon_vcvts_f32_s32: |
| 7358 | case NEON::BI__builtin_neon_vcvtd_f64_s64: { |
| 7359 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7360 | bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64; |
| 7361 | llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty; |
| 7362 | llvm::Type *FTy = Is64 ? DoubleTy : FloatTy; |
| 7363 | Ops[0] = Builder.CreateBitCast(Ops[0], InTy); |
| 7364 | if (usgn) |
| 7365 | return Builder.CreateUIToFP(Ops[0], FTy); |
| 7366 | return Builder.CreateSIToFP(Ops[0], FTy); |
| 7367 | } |
| 7368 | case NEON::BI__builtin_neon_vcvth_f16_u16: |
| 7369 | case NEON::BI__builtin_neon_vcvth_f16_u32: |
| 7370 | case NEON::BI__builtin_neon_vcvth_f16_u64: |
| 7371 | usgn = true; |
| 7372 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 7373 | case NEON::BI__builtin_neon_vcvth_f16_s16: |
| 7374 | case NEON::BI__builtin_neon_vcvth_f16_s32: |
| 7375 | case NEON::BI__builtin_neon_vcvth_f16_s64: { |
| 7376 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7377 | llvm::Type *FTy = HalfTy; |
| 7378 | llvm::Type *InTy; |
| 7379 | if (Ops[0]->getType()->getPrimitiveSizeInBits() == 64) |
| 7380 | InTy = Int64Ty; |
| 7381 | else if (Ops[0]->getType()->getPrimitiveSizeInBits() == 32) |
| 7382 | InTy = Int32Ty; |
| 7383 | else |
| 7384 | InTy = Int16Ty; |
| 7385 | Ops[0] = Builder.CreateBitCast(Ops[0], InTy); |
| 7386 | if (usgn) |
| 7387 | return Builder.CreateUIToFP(Ops[0], FTy); |
| 7388 | return Builder.CreateSIToFP(Ops[0], FTy); |
| 7389 | } |
| 7390 | case NEON::BI__builtin_neon_vcvth_u16_f16: |
| 7391 | usgn = true; |
| 7392 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 7393 | case NEON::BI__builtin_neon_vcvth_s16_f16: { |
| 7394 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7395 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); |
| 7396 | if (usgn) |
| 7397 | return Builder.CreateFPToUI(Ops[0], Int16Ty); |
| 7398 | return Builder.CreateFPToSI(Ops[0], Int16Ty); |
| 7399 | } |
| 7400 | case NEON::BI__builtin_neon_vcvth_u32_f16: |
| 7401 | usgn = true; |
| 7402 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 7403 | case NEON::BI__builtin_neon_vcvth_s32_f16: { |
| 7404 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7405 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); |
| 7406 | if (usgn) |
| 7407 | return Builder.CreateFPToUI(Ops[0], Int32Ty); |
| 7408 | return Builder.CreateFPToSI(Ops[0], Int32Ty); |
| 7409 | } |
| 7410 | case NEON::BI__builtin_neon_vcvth_u64_f16: |
| 7411 | usgn = true; |
| 7412 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 7413 | case NEON::BI__builtin_neon_vcvth_s64_f16: { |
| 7414 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7415 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); |
| 7416 | if (usgn) |
| 7417 | return Builder.CreateFPToUI(Ops[0], Int64Ty); |
| 7418 | return Builder.CreateFPToSI(Ops[0], Int64Ty); |
| 7419 | } |
| 7420 | case NEON::BI__builtin_neon_vcvtah_u16_f16: |
| 7421 | case NEON::BI__builtin_neon_vcvtmh_u16_f16: |
| 7422 | case NEON::BI__builtin_neon_vcvtnh_u16_f16: |
| 7423 | case NEON::BI__builtin_neon_vcvtph_u16_f16: |
| 7424 | case NEON::BI__builtin_neon_vcvtah_s16_f16: |
| 7425 | case NEON::BI__builtin_neon_vcvtmh_s16_f16: |
| 7426 | case NEON::BI__builtin_neon_vcvtnh_s16_f16: |
| 7427 | case NEON::BI__builtin_neon_vcvtph_s16_f16: { |
| 7428 | unsigned Int; |
| 7429 | llvm::Type* InTy = Int32Ty; |
| 7430 | llvm::Type* FTy = HalfTy; |
| 7431 | llvm::Type *Tys[2] = {InTy, FTy}; |
| 7432 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7433 | switch (BuiltinID) { |
| 7434 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7434); |
| 7435 | case NEON::BI__builtin_neon_vcvtah_u16_f16: |
| 7436 | Int = Intrinsic::aarch64_neon_fcvtau; break; |
| 7437 | case NEON::BI__builtin_neon_vcvtmh_u16_f16: |
| 7438 | Int = Intrinsic::aarch64_neon_fcvtmu; break; |
| 7439 | case NEON::BI__builtin_neon_vcvtnh_u16_f16: |
| 7440 | Int = Intrinsic::aarch64_neon_fcvtnu; break; |
| 7441 | case NEON::BI__builtin_neon_vcvtph_u16_f16: |
| 7442 | Int = Intrinsic::aarch64_neon_fcvtpu; break; |
| 7443 | case NEON::BI__builtin_neon_vcvtah_s16_f16: |
| 7444 | Int = Intrinsic::aarch64_neon_fcvtas; break; |
| 7445 | case NEON::BI__builtin_neon_vcvtmh_s16_f16: |
| 7446 | Int = Intrinsic::aarch64_neon_fcvtms; break; |
| 7447 | case NEON::BI__builtin_neon_vcvtnh_s16_f16: |
| 7448 | Int = Intrinsic::aarch64_neon_fcvtns; break; |
| 7449 | case NEON::BI__builtin_neon_vcvtph_s16_f16: |
| 7450 | Int = Intrinsic::aarch64_neon_fcvtps; break; |
| 7451 | } |
| 7452 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvt"); |
| 7453 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 7454 | } |
| 7455 | case NEON::BI__builtin_neon_vcaleh_f16: |
| 7456 | case NEON::BI__builtin_neon_vcalth_f16: |
| 7457 | case NEON::BI__builtin_neon_vcageh_f16: |
| 7458 | case NEON::BI__builtin_neon_vcagth_f16: { |
| 7459 | unsigned Int; |
| 7460 | llvm::Type* InTy = Int32Ty; |
| 7461 | llvm::Type* FTy = HalfTy; |
| 7462 | llvm::Type *Tys[2] = {InTy, FTy}; |
| 7463 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7464 | switch (BuiltinID) { |
| 7465 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7465); |
| 7466 | case NEON::BI__builtin_neon_vcageh_f16: |
| 7467 | Int = Intrinsic::aarch64_neon_facge; break; |
| 7468 | case NEON::BI__builtin_neon_vcagth_f16: |
| 7469 | Int = Intrinsic::aarch64_neon_facgt; break; |
| 7470 | case NEON::BI__builtin_neon_vcaleh_f16: |
| 7471 | Int = Intrinsic::aarch64_neon_facge; std::swap(Ops[0], Ops[1]); break; |
| 7472 | case NEON::BI__builtin_neon_vcalth_f16: |
| 7473 | Int = Intrinsic::aarch64_neon_facgt; std::swap(Ops[0], Ops[1]); break; |
| 7474 | } |
| 7475 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "facg"); |
| 7476 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 7477 | } |
| 7478 | case NEON::BI__builtin_neon_vcvth_n_s16_f16: |
| 7479 | case NEON::BI__builtin_neon_vcvth_n_u16_f16: { |
| 7480 | unsigned Int; |
| 7481 | llvm::Type* InTy = Int32Ty; |
| 7482 | llvm::Type* FTy = HalfTy; |
| 7483 | llvm::Type *Tys[2] = {InTy, FTy}; |
| 7484 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7485 | switch (BuiltinID) { |
| 7486 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7486); |
| 7487 | case NEON::BI__builtin_neon_vcvth_n_s16_f16: |
| 7488 | Int = Intrinsic::aarch64_neon_vcvtfp2fxs; break; |
| 7489 | case NEON::BI__builtin_neon_vcvth_n_u16_f16: |
| 7490 | Int = Intrinsic::aarch64_neon_vcvtfp2fxu; break; |
| 7491 | } |
| 7492 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n"); |
| 7493 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 7494 | } |
| 7495 | case NEON::BI__builtin_neon_vcvth_n_f16_s16: |
| 7496 | case NEON::BI__builtin_neon_vcvth_n_f16_u16: { |
| 7497 | unsigned Int; |
| 7498 | llvm::Type* FTy = HalfTy; |
| 7499 | llvm::Type* InTy = Int32Ty; |
| 7500 | llvm::Type *Tys[2] = {FTy, InTy}; |
| 7501 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7502 | switch (BuiltinID) { |
| 7503 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7503); |
| 7504 | case NEON::BI__builtin_neon_vcvth_n_f16_s16: |
| 7505 | Int = Intrinsic::aarch64_neon_vcvtfxs2fp; |
| 7506 | Ops[0] = Builder.CreateSExt(Ops[0], InTy, "sext"); |
| 7507 | break; |
| 7508 | case NEON::BI__builtin_neon_vcvth_n_f16_u16: |
| 7509 | Int = Intrinsic::aarch64_neon_vcvtfxu2fp; |
| 7510 | Ops[0] = Builder.CreateZExt(Ops[0], InTy); |
| 7511 | break; |
| 7512 | } |
| 7513 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n"); |
| 7514 | } |
| 7515 | case NEON::BI__builtin_neon_vpaddd_s64: { |
| 7516 | llvm::Type *Ty = llvm::VectorType::get(Int64Ty, 2); |
| 7517 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 7518 | // The vector is v2f64, so make sure it's bitcast to that. |
| 7519 | Vec = Builder.CreateBitCast(Vec, Ty, "v2i64"); |
| 7520 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); |
| 7521 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); |
| 7522 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); |
| 7523 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); |
| 7524 | // Pairwise addition of a v2f64 into a scalar f64. |
| 7525 | return Builder.CreateAdd(Op0, Op1, "vpaddd"); |
| 7526 | } |
| 7527 | case NEON::BI__builtin_neon_vpaddd_f64: { |
| 7528 | llvm::Type *Ty = |
| 7529 | llvm::VectorType::get(DoubleTy, 2); |
| 7530 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 7531 | // The vector is v2f64, so make sure it's bitcast to that. |
| 7532 | Vec = Builder.CreateBitCast(Vec, Ty, "v2f64"); |
| 7533 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); |
| 7534 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); |
| 7535 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); |
| 7536 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); |
| 7537 | // Pairwise addition of a v2f64 into a scalar f64. |
| 7538 | return Builder.CreateFAdd(Op0, Op1, "vpaddd"); |
| 7539 | } |
| 7540 | case NEON::BI__builtin_neon_vpadds_f32: { |
| 7541 | llvm::Type *Ty = |
| 7542 | llvm::VectorType::get(FloatTy, 2); |
| 7543 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 7544 | // The vector is v2f32, so make sure it's bitcast to that. |
| 7545 | Vec = Builder.CreateBitCast(Vec, Ty, "v2f32"); |
| 7546 | llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0); |
| 7547 | llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1); |
| 7548 | Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0"); |
| 7549 | Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1"); |
| 7550 | // Pairwise addition of a v2f32 into a scalar f32. |
| 7551 | return Builder.CreateFAdd(Op0, Op1, "vpaddd"); |
| 7552 | } |
| 7553 | case NEON::BI__builtin_neon_vceqzd_s64: |
| 7554 | case NEON::BI__builtin_neon_vceqzd_f64: |
| 7555 | case NEON::BI__builtin_neon_vceqzs_f32: |
| 7556 | case NEON::BI__builtin_neon_vceqzh_f16: |
| 7557 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7558 | return EmitAArch64CompareBuiltinExpr( |
| 7559 | Ops[0], ConvertType(E->getCallReturnType(getContext())), |
| 7560 | ICmpInst::FCMP_OEQ, ICmpInst::ICMP_EQ, "vceqz"); |
| 7561 | case NEON::BI__builtin_neon_vcgezd_s64: |
| 7562 | case NEON::BI__builtin_neon_vcgezd_f64: |
| 7563 | case NEON::BI__builtin_neon_vcgezs_f32: |
| 7564 | case NEON::BI__builtin_neon_vcgezh_f16: |
| 7565 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7566 | return EmitAArch64CompareBuiltinExpr( |
| 7567 | Ops[0], ConvertType(E->getCallReturnType(getContext())), |
| 7568 | ICmpInst::FCMP_OGE, ICmpInst::ICMP_SGE, "vcgez"); |
| 7569 | case NEON::BI__builtin_neon_vclezd_s64: |
| 7570 | case NEON::BI__builtin_neon_vclezd_f64: |
| 7571 | case NEON::BI__builtin_neon_vclezs_f32: |
| 7572 | case NEON::BI__builtin_neon_vclezh_f16: |
| 7573 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7574 | return EmitAArch64CompareBuiltinExpr( |
| 7575 | Ops[0], ConvertType(E->getCallReturnType(getContext())), |
| 7576 | ICmpInst::FCMP_OLE, ICmpInst::ICMP_SLE, "vclez"); |
| 7577 | case NEON::BI__builtin_neon_vcgtzd_s64: |
| 7578 | case NEON::BI__builtin_neon_vcgtzd_f64: |
| 7579 | case NEON::BI__builtin_neon_vcgtzs_f32: |
| 7580 | case NEON::BI__builtin_neon_vcgtzh_f16: |
| 7581 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7582 | return EmitAArch64CompareBuiltinExpr( |
| 7583 | Ops[0], ConvertType(E->getCallReturnType(getContext())), |
| 7584 | ICmpInst::FCMP_OGT, ICmpInst::ICMP_SGT, "vcgtz"); |
| 7585 | case NEON::BI__builtin_neon_vcltzd_s64: |
| 7586 | case NEON::BI__builtin_neon_vcltzd_f64: |
| 7587 | case NEON::BI__builtin_neon_vcltzs_f32: |
| 7588 | case NEON::BI__builtin_neon_vcltzh_f16: |
| 7589 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7590 | return EmitAArch64CompareBuiltinExpr( |
| 7591 | Ops[0], ConvertType(E->getCallReturnType(getContext())), |
| 7592 | ICmpInst::FCMP_OLT, ICmpInst::ICMP_SLT, "vcltz"); |
| 7593 | |
| 7594 | case NEON::BI__builtin_neon_vceqzd_u64: { |
| 7595 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 7596 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); |
| 7597 | Ops[0] = |
| 7598 | Builder.CreateICmpEQ(Ops[0], llvm::Constant::getNullValue(Int64Ty)); |
| 7599 | return Builder.CreateSExt(Ops[0], Int64Ty, "vceqzd"); |
| 7600 | } |
| 7601 | case NEON::BI__builtin_neon_vceqd_f64: |
| 7602 | case NEON::BI__builtin_neon_vcled_f64: |
| 7603 | case NEON::BI__builtin_neon_vcltd_f64: |
| 7604 | case NEON::BI__builtin_neon_vcged_f64: |
| 7605 | case NEON::BI__builtin_neon_vcgtd_f64: { |
| 7606 | llvm::CmpInst::Predicate P; |
| 7607 | switch (BuiltinID) { |
| 7608 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7608); |
| 7609 | case NEON::BI__builtin_neon_vceqd_f64: P = llvm::FCmpInst::FCMP_OEQ; break; |
| 7610 | case NEON::BI__builtin_neon_vcled_f64: P = llvm::FCmpInst::FCMP_OLE; break; |
| 7611 | case NEON::BI__builtin_neon_vcltd_f64: P = llvm::FCmpInst::FCMP_OLT; break; |
| 7612 | case NEON::BI__builtin_neon_vcged_f64: P = llvm::FCmpInst::FCMP_OGE; break; |
| 7613 | case NEON::BI__builtin_neon_vcgtd_f64: P = llvm::FCmpInst::FCMP_OGT; break; |
| 7614 | } |
| 7615 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7616 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); |
| 7617 | Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy); |
| 7618 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); |
| 7619 | return Builder.CreateSExt(Ops[0], Int64Ty, "vcmpd"); |
| 7620 | } |
| 7621 | case NEON::BI__builtin_neon_vceqs_f32: |
| 7622 | case NEON::BI__builtin_neon_vcles_f32: |
| 7623 | case NEON::BI__builtin_neon_vclts_f32: |
| 7624 | case NEON::BI__builtin_neon_vcges_f32: |
| 7625 | case NEON::BI__builtin_neon_vcgts_f32: { |
| 7626 | llvm::CmpInst::Predicate P; |
| 7627 | switch (BuiltinID) { |
| 7628 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7628); |
| 7629 | case NEON::BI__builtin_neon_vceqs_f32: P = llvm::FCmpInst::FCMP_OEQ; break; |
| 7630 | case NEON::BI__builtin_neon_vcles_f32: P = llvm::FCmpInst::FCMP_OLE; break; |
| 7631 | case NEON::BI__builtin_neon_vclts_f32: P = llvm::FCmpInst::FCMP_OLT; break; |
| 7632 | case NEON::BI__builtin_neon_vcges_f32: P = llvm::FCmpInst::FCMP_OGE; break; |
| 7633 | case NEON::BI__builtin_neon_vcgts_f32: P = llvm::FCmpInst::FCMP_OGT; break; |
| 7634 | } |
| 7635 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7636 | Ops[0] = Builder.CreateBitCast(Ops[0], FloatTy); |
| 7637 | Ops[1] = Builder.CreateBitCast(Ops[1], FloatTy); |
| 7638 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); |
| 7639 | return Builder.CreateSExt(Ops[0], Int32Ty, "vcmpd"); |
| 7640 | } |
| 7641 | case NEON::BI__builtin_neon_vceqh_f16: |
| 7642 | case NEON::BI__builtin_neon_vcleh_f16: |
| 7643 | case NEON::BI__builtin_neon_vclth_f16: |
| 7644 | case NEON::BI__builtin_neon_vcgeh_f16: |
| 7645 | case NEON::BI__builtin_neon_vcgth_f16: { |
| 7646 | llvm::CmpInst::Predicate P; |
| 7647 | switch (BuiltinID) { |
| 7648 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7648); |
| 7649 | case NEON::BI__builtin_neon_vceqh_f16: P = llvm::FCmpInst::FCMP_OEQ; break; |
| 7650 | case NEON::BI__builtin_neon_vcleh_f16: P = llvm::FCmpInst::FCMP_OLE; break; |
| 7651 | case NEON::BI__builtin_neon_vclth_f16: P = llvm::FCmpInst::FCMP_OLT; break; |
| 7652 | case NEON::BI__builtin_neon_vcgeh_f16: P = llvm::FCmpInst::FCMP_OGE; break; |
| 7653 | case NEON::BI__builtin_neon_vcgth_f16: P = llvm::FCmpInst::FCMP_OGT; break; |
| 7654 | } |
| 7655 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7656 | Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy); |
| 7657 | Ops[1] = Builder.CreateBitCast(Ops[1], HalfTy); |
| 7658 | Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]); |
| 7659 | return Builder.CreateSExt(Ops[0], Int16Ty, "vcmpd"); |
| 7660 | } |
| 7661 | case NEON::BI__builtin_neon_vceqd_s64: |
| 7662 | case NEON::BI__builtin_neon_vceqd_u64: |
| 7663 | case NEON::BI__builtin_neon_vcgtd_s64: |
| 7664 | case NEON::BI__builtin_neon_vcgtd_u64: |
| 7665 | case NEON::BI__builtin_neon_vcltd_s64: |
| 7666 | case NEON::BI__builtin_neon_vcltd_u64: |
| 7667 | case NEON::BI__builtin_neon_vcged_u64: |
| 7668 | case NEON::BI__builtin_neon_vcged_s64: |
| 7669 | case NEON::BI__builtin_neon_vcled_u64: |
| 7670 | case NEON::BI__builtin_neon_vcled_s64: { |
| 7671 | llvm::CmpInst::Predicate P; |
| 7672 | switch (BuiltinID) { |
| 7673 | default: llvm_unreachable("missing builtin ID in switch!")::llvm::llvm_unreachable_internal("missing builtin ID in switch!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 7673); |
| 7674 | case NEON::BI__builtin_neon_vceqd_s64: |
| 7675 | case NEON::BI__builtin_neon_vceqd_u64:P = llvm::ICmpInst::ICMP_EQ;break; |
| 7676 | case NEON::BI__builtin_neon_vcgtd_s64:P = llvm::ICmpInst::ICMP_SGT;break; |
| 7677 | case NEON::BI__builtin_neon_vcgtd_u64:P = llvm::ICmpInst::ICMP_UGT;break; |
| 7678 | case NEON::BI__builtin_neon_vcltd_s64:P = llvm::ICmpInst::ICMP_SLT;break; |
| 7679 | case NEON::BI__builtin_neon_vcltd_u64:P = llvm::ICmpInst::ICMP_ULT;break; |
| 7680 | case NEON::BI__builtin_neon_vcged_u64:P = llvm::ICmpInst::ICMP_UGE;break; |
| 7681 | case NEON::BI__builtin_neon_vcged_s64:P = llvm::ICmpInst::ICMP_SGE;break; |
| 7682 | case NEON::BI__builtin_neon_vcled_u64:P = llvm::ICmpInst::ICMP_ULE;break; |
| 7683 | case NEON::BI__builtin_neon_vcled_s64:P = llvm::ICmpInst::ICMP_SLE;break; |
| 7684 | } |
| 7685 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7686 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); |
| 7687 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); |
| 7688 | Ops[0] = Builder.CreateICmp(P, Ops[0], Ops[1]); |
| 7689 | return Builder.CreateSExt(Ops[0], Int64Ty, "vceqd"); |
| 7690 | } |
| 7691 | case NEON::BI__builtin_neon_vtstd_s64: |
| 7692 | case NEON::BI__builtin_neon_vtstd_u64: { |
| 7693 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7694 | Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty); |
| 7695 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); |
| 7696 | Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]); |
| 7697 | Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0], |
| 7698 | llvm::Constant::getNullValue(Int64Ty)); |
| 7699 | return Builder.CreateSExt(Ops[0], Int64Ty, "vtstd"); |
| 7700 | } |
| 7701 | case NEON::BI__builtin_neon_vset_lane_i8: |
| 7702 | case NEON::BI__builtin_neon_vset_lane_i16: |
| 7703 | case NEON::BI__builtin_neon_vset_lane_i32: |
| 7704 | case NEON::BI__builtin_neon_vset_lane_i64: |
| 7705 | case NEON::BI__builtin_neon_vset_lane_f32: |
| 7706 | case NEON::BI__builtin_neon_vsetq_lane_i8: |
| 7707 | case NEON::BI__builtin_neon_vsetq_lane_i16: |
| 7708 | case NEON::BI__builtin_neon_vsetq_lane_i32: |
| 7709 | case NEON::BI__builtin_neon_vsetq_lane_i64: |
| 7710 | case NEON::BI__builtin_neon_vsetq_lane_f32: |
| 7711 | Ops.push_back(EmitScalarExpr(E->getArg(2))); |
| 7712 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); |
| 7713 | case NEON::BI__builtin_neon_vset_lane_f64: |
| 7714 | // The vector type needs a cast for the v1f64 variant. |
| 7715 | Ops[1] = Builder.CreateBitCast(Ops[1], |
| 7716 | llvm::VectorType::get(DoubleTy, 1)); |
| 7717 | Ops.push_back(EmitScalarExpr(E->getArg(2))); |
| 7718 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); |
| 7719 | case NEON::BI__builtin_neon_vsetq_lane_f64: |
| 7720 | // The vector type needs a cast for the v2f64 variant. |
| 7721 | Ops[1] = Builder.CreateBitCast(Ops[1], |
| 7722 | llvm::VectorType::get(DoubleTy, 2)); |
| 7723 | Ops.push_back(EmitScalarExpr(E->getArg(2))); |
| 7724 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane"); |
| 7725 | |
| 7726 | case NEON::BI__builtin_neon_vget_lane_i8: |
| 7727 | case NEON::BI__builtin_neon_vdupb_lane_i8: |
| 7728 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 8)); |
| 7729 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7730 | "vget_lane"); |
| 7731 | case NEON::BI__builtin_neon_vgetq_lane_i8: |
| 7732 | case NEON::BI__builtin_neon_vdupb_laneq_i8: |
| 7733 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 16)); |
| 7734 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7735 | "vgetq_lane"); |
| 7736 | case NEON::BI__builtin_neon_vget_lane_i16: |
| 7737 | case NEON::BI__builtin_neon_vduph_lane_i16: |
| 7738 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 4)); |
| 7739 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7740 | "vget_lane"); |
| 7741 | case NEON::BI__builtin_neon_vgetq_lane_i16: |
| 7742 | case NEON::BI__builtin_neon_vduph_laneq_i16: |
| 7743 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 8)); |
| 7744 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7745 | "vgetq_lane"); |
| 7746 | case NEON::BI__builtin_neon_vget_lane_i32: |
| 7747 | case NEON::BI__builtin_neon_vdups_lane_i32: |
| 7748 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 2)); |
| 7749 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7750 | "vget_lane"); |
| 7751 | case NEON::BI__builtin_neon_vdups_lane_f32: |
| 7752 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 7753 | llvm::VectorType::get(FloatTy, 2)); |
| 7754 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7755 | "vdups_lane"); |
| 7756 | case NEON::BI__builtin_neon_vgetq_lane_i32: |
| 7757 | case NEON::BI__builtin_neon_vdups_laneq_i32: |
| 7758 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4)); |
| 7759 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7760 | "vgetq_lane"); |
| 7761 | case NEON::BI__builtin_neon_vget_lane_i64: |
| 7762 | case NEON::BI__builtin_neon_vdupd_lane_i64: |
| 7763 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 1)); |
| 7764 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7765 | "vget_lane"); |
| 7766 | case NEON::BI__builtin_neon_vdupd_lane_f64: |
| 7767 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 7768 | llvm::VectorType::get(DoubleTy, 1)); |
| 7769 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7770 | "vdupd_lane"); |
| 7771 | case NEON::BI__builtin_neon_vgetq_lane_i64: |
| 7772 | case NEON::BI__builtin_neon_vdupd_laneq_i64: |
| 7773 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); |
| 7774 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7775 | "vgetq_lane"); |
| 7776 | case NEON::BI__builtin_neon_vget_lane_f32: |
| 7777 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 7778 | llvm::VectorType::get(FloatTy, 2)); |
| 7779 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7780 | "vget_lane"); |
| 7781 | case NEON::BI__builtin_neon_vget_lane_f64: |
| 7782 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 7783 | llvm::VectorType::get(DoubleTy, 1)); |
| 7784 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7785 | "vget_lane"); |
| 7786 | case NEON::BI__builtin_neon_vgetq_lane_f32: |
| 7787 | case NEON::BI__builtin_neon_vdups_laneq_f32: |
| 7788 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 7789 | llvm::VectorType::get(FloatTy, 4)); |
| 7790 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7791 | "vgetq_lane"); |
| 7792 | case NEON::BI__builtin_neon_vgetq_lane_f64: |
| 7793 | case NEON::BI__builtin_neon_vdupd_laneq_f64: |
| 7794 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 7795 | llvm::VectorType::get(DoubleTy, 2)); |
| 7796 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7797 | "vgetq_lane"); |
| 7798 | case NEON::BI__builtin_neon_vaddh_f16: |
| 7799 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7800 | return Builder.CreateFAdd(Ops[0], Ops[1], "vaddh"); |
| 7801 | case NEON::BI__builtin_neon_vsubh_f16: |
| 7802 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7803 | return Builder.CreateFSub(Ops[0], Ops[1], "vsubh"); |
| 7804 | case NEON::BI__builtin_neon_vmulh_f16: |
| 7805 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7806 | return Builder.CreateFMul(Ops[0], Ops[1], "vmulh"); |
| 7807 | case NEON::BI__builtin_neon_vdivh_f16: |
| 7808 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7809 | return Builder.CreateFDiv(Ops[0], Ops[1], "vdivh"); |
| 7810 | case NEON::BI__builtin_neon_vfmah_f16: { |
| 7811 | Function *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy); |
| 7812 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. |
| 7813 | return Builder.CreateCall(F, |
| 7814 | {EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), Ops[0]}); |
| 7815 | } |
| 7816 | case NEON::BI__builtin_neon_vfmsh_f16: { |
| 7817 | Function *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy); |
| 7818 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(HalfTy); |
| 7819 | Value* Sub = Builder.CreateFSub(Zero, EmitScalarExpr(E->getArg(1)), "vsubh"); |
| 7820 | // NEON intrinsic puts accumulator first, unlike the LLVM fma. |
| 7821 | return Builder.CreateCall(F, {Sub, EmitScalarExpr(E->getArg(2)), Ops[0]}); |
| 7822 | } |
| 7823 | case NEON::BI__builtin_neon_vaddd_s64: |
| 7824 | case NEON::BI__builtin_neon_vaddd_u64: |
| 7825 | return Builder.CreateAdd(Ops[0], EmitScalarExpr(E->getArg(1)), "vaddd"); |
| 7826 | case NEON::BI__builtin_neon_vsubd_s64: |
| 7827 | case NEON::BI__builtin_neon_vsubd_u64: |
| 7828 | return Builder.CreateSub(Ops[0], EmitScalarExpr(E->getArg(1)), "vsubd"); |
| 7829 | case NEON::BI__builtin_neon_vqdmlalh_s16: |
| 7830 | case NEON::BI__builtin_neon_vqdmlslh_s16: { |
| 7831 | SmallVector<Value *, 2> ProductOps; |
| 7832 | ProductOps.push_back(vectorWrapScalar16(Ops[1])); |
| 7833 | ProductOps.push_back(vectorWrapScalar16(EmitScalarExpr(E->getArg(2)))); |
| 7834 | llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4); |
| 7835 | Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy), |
| 7836 | ProductOps, "vqdmlXl"); |
| 7837 | Constant *CI = ConstantInt::get(SizeTy, 0); |
| 7838 | Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0"); |
| 7839 | |
| 7840 | unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlalh_s16 |
| 7841 | ? Intrinsic::aarch64_neon_sqadd |
| 7842 | : Intrinsic::aarch64_neon_sqsub; |
| 7843 | return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int32Ty), Ops, "vqdmlXl"); |
| 7844 | } |
| 7845 | case NEON::BI__builtin_neon_vqshlud_n_s64: { |
| 7846 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7847 | Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty); |
| 7848 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqshlu, Int64Ty), |
| 7849 | Ops, "vqshlu_n"); |
| 7850 | } |
| 7851 | case NEON::BI__builtin_neon_vqshld_n_u64: |
| 7852 | case NEON::BI__builtin_neon_vqshld_n_s64: { |
| 7853 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vqshld_n_u64 |
| 7854 | ? Intrinsic::aarch64_neon_uqshl |
| 7855 | : Intrinsic::aarch64_neon_sqshl; |
| 7856 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7857 | Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty); |
| 7858 | return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vqshl_n"); |
| 7859 | } |
| 7860 | case NEON::BI__builtin_neon_vrshrd_n_u64: |
| 7861 | case NEON::BI__builtin_neon_vrshrd_n_s64: { |
| 7862 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrshrd_n_u64 |
| 7863 | ? Intrinsic::aarch64_neon_urshl |
| 7864 | : Intrinsic::aarch64_neon_srshl; |
| 7865 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 7866 | int SV = cast<ConstantInt>(Ops[1])->getSExtValue(); |
| 7867 | Ops[1] = ConstantInt::get(Int64Ty, -SV); |
| 7868 | return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vrshr_n"); |
| 7869 | } |
| 7870 | case NEON::BI__builtin_neon_vrsrad_n_u64: |
| 7871 | case NEON::BI__builtin_neon_vrsrad_n_s64: { |
| 7872 | unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrsrad_n_u64 |
| 7873 | ? Intrinsic::aarch64_neon_urshl |
| 7874 | : Intrinsic::aarch64_neon_srshl; |
| 7875 | Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty); |
| 7876 | Ops.push_back(Builder.CreateNeg(EmitScalarExpr(E->getArg(2)))); |
| 7877 | Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Int64Ty), |
| 7878 | {Ops[1], Builder.CreateSExt(Ops[2], Int64Ty)}); |
| 7879 | return Builder.CreateAdd(Ops[0], Builder.CreateBitCast(Ops[1], Int64Ty)); |
| 7880 | } |
| 7881 | case NEON::BI__builtin_neon_vshld_n_s64: |
| 7882 | case NEON::BI__builtin_neon_vshld_n_u64: { |
| 7883 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); |
| 7884 | return Builder.CreateShl( |
| 7885 | Ops[0], ConstantInt::get(Int64Ty, Amt->getZExtValue()), "shld_n"); |
| 7886 | } |
| 7887 | case NEON::BI__builtin_neon_vshrd_n_s64: { |
| 7888 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); |
| 7889 | return Builder.CreateAShr( |
| 7890 | Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63), |
| 7891 | Amt->getZExtValue())), |
| 7892 | "shrd_n"); |
| 7893 | } |
| 7894 | case NEON::BI__builtin_neon_vshrd_n_u64: { |
| 7895 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1))); |
| 7896 | uint64_t ShiftAmt = Amt->getZExtValue(); |
| 7897 | // Right-shifting an unsigned value by its size yields 0. |
| 7898 | if (ShiftAmt == 64) |
| 7899 | return ConstantInt::get(Int64Ty, 0); |
| 7900 | return Builder.CreateLShr(Ops[0], ConstantInt::get(Int64Ty, ShiftAmt), |
| 7901 | "shrd_n"); |
| 7902 | } |
| 7903 | case NEON::BI__builtin_neon_vsrad_n_s64: { |
| 7904 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2))); |
| 7905 | Ops[1] = Builder.CreateAShr( |
| 7906 | Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63), |
| 7907 | Amt->getZExtValue())), |
| 7908 | "shrd_n"); |
| 7909 | return Builder.CreateAdd(Ops[0], Ops[1]); |
| 7910 | } |
| 7911 | case NEON::BI__builtin_neon_vsrad_n_u64: { |
| 7912 | llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2))); |
| 7913 | uint64_t ShiftAmt = Amt->getZExtValue(); |
| 7914 | // Right-shifting an unsigned value by its size yields 0. |
| 7915 | // As Op + 0 = Op, return Ops[0] directly. |
| 7916 | if (ShiftAmt == 64) |
| 7917 | return Ops[0]; |
| 7918 | Ops[1] = Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, ShiftAmt), |
| 7919 | "shrd_n"); |
| 7920 | return Builder.CreateAdd(Ops[0], Ops[1]); |
| 7921 | } |
| 7922 | case NEON::BI__builtin_neon_vqdmlalh_lane_s16: |
| 7923 | case NEON::BI__builtin_neon_vqdmlalh_laneq_s16: |
| 7924 | case NEON::BI__builtin_neon_vqdmlslh_lane_s16: |
| 7925 | case NEON::BI__builtin_neon_vqdmlslh_laneq_s16: { |
| 7926 | Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)), |
| 7927 | "lane"); |
| 7928 | SmallVector<Value *, 2> ProductOps; |
| 7929 | ProductOps.push_back(vectorWrapScalar16(Ops[1])); |
| 7930 | ProductOps.push_back(vectorWrapScalar16(Ops[2])); |
| 7931 | llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4); |
| 7932 | Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy), |
| 7933 | ProductOps, "vqdmlXl"); |
| 7934 | Constant *CI = ConstantInt::get(SizeTy, 0); |
| 7935 | Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0"); |
| 7936 | Ops.pop_back(); |
| 7937 | |
| 7938 | unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlalh_lane_s16 || |
| 7939 | BuiltinID == NEON::BI__builtin_neon_vqdmlalh_laneq_s16) |
| 7940 | ? Intrinsic::aarch64_neon_sqadd |
| 7941 | : Intrinsic::aarch64_neon_sqsub; |
| 7942 | return EmitNeonCall(CGM.getIntrinsic(AccInt, Int32Ty), Ops, "vqdmlXl"); |
| 7943 | } |
| 7944 | case NEON::BI__builtin_neon_vqdmlals_s32: |
| 7945 | case NEON::BI__builtin_neon_vqdmlsls_s32: { |
| 7946 | SmallVector<Value *, 2> ProductOps; |
| 7947 | ProductOps.push_back(Ops[1]); |
| 7948 | ProductOps.push_back(EmitScalarExpr(E->getArg(2))); |
| 7949 | Ops[1] = |
| 7950 | EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar), |
| 7951 | ProductOps, "vqdmlXl"); |
| 7952 | |
| 7953 | unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlals_s32 |
| 7954 | ? Intrinsic::aarch64_neon_sqadd |
| 7955 | : Intrinsic::aarch64_neon_sqsub; |
| 7956 | return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int64Ty), Ops, "vqdmlXl"); |
| 7957 | } |
| 7958 | case NEON::BI__builtin_neon_vqdmlals_lane_s32: |
| 7959 | case NEON::BI__builtin_neon_vqdmlals_laneq_s32: |
| 7960 | case NEON::BI__builtin_neon_vqdmlsls_lane_s32: |
| 7961 | case NEON::BI__builtin_neon_vqdmlsls_laneq_s32: { |
| 7962 | Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)), |
| 7963 | "lane"); |
| 7964 | SmallVector<Value *, 2> ProductOps; |
| 7965 | ProductOps.push_back(Ops[1]); |
| 7966 | ProductOps.push_back(Ops[2]); |
| 7967 | Ops[1] = |
| 7968 | EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar), |
| 7969 | ProductOps, "vqdmlXl"); |
| 7970 | Ops.pop_back(); |
| 7971 | |
| 7972 | unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlals_lane_s32 || |
| 7973 | BuiltinID == NEON::BI__builtin_neon_vqdmlals_laneq_s32) |
| 7974 | ? Intrinsic::aarch64_neon_sqadd |
| 7975 | : Intrinsic::aarch64_neon_sqsub; |
| 7976 | return EmitNeonCall(CGM.getIntrinsic(AccInt, Int64Ty), Ops, "vqdmlXl"); |
| 7977 | } |
| 7978 | case NEON::BI__builtin_neon_vduph_lane_f16: { |
| 7979 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7980 | "vget_lane"); |
| 7981 | } |
| 7982 | case NEON::BI__builtin_neon_vduph_laneq_f16: { |
| 7983 | return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)), |
| 7984 | "vgetq_lane"); |
| 7985 | } |
| 7986 | } |
| 7987 | |
| 7988 | llvm::VectorType *VTy = GetNeonType(this, Type); |
| 7989 | llvm::Type *Ty = VTy; |
| 7990 | if (!Ty) |
| 7991 | return nullptr; |
| 7992 | |
| 7993 | // Not all intrinsics handled by the common case work for AArch64 yet, so only |
| 7994 | // defer to common code if it's been added to our special map. |
| 7995 | Builtin = findNeonIntrinsicInMap(AArch64SIMDIntrinsicMap, BuiltinID, |
| 7996 | AArch64SIMDIntrinsicsProvenSorted); |
| 7997 | |
| 7998 | if (Builtin) |
| 7999 | return EmitCommonNeonBuiltinExpr( |
| 8000 | Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic, |
| 8001 | Builtin->NameHint, Builtin->TypeModifier, E, Ops, |
| 8002 | /*never use addresses*/ Address::invalid(), Address::invalid(), Arch); |
| 8003 | |
| 8004 | if (Value *V = EmitAArch64TblBuiltinExpr(*this, BuiltinID, E, Ops, Arch)) |
| 8005 | return V; |
| 8006 | |
| 8007 | unsigned Int; |
| 8008 | switch (BuiltinID) { |
| 8009 | default: return nullptr; |
| 8010 | case NEON::BI__builtin_neon_vbsl_v: |
| 8011 | case NEON::BI__builtin_neon_vbslq_v: { |
| 8012 | llvm::Type *BitTy = llvm::VectorType::getInteger(VTy); |
| 8013 | Ops[0] = Builder.CreateBitCast(Ops[0], BitTy, "vbsl"); |
| 8014 | Ops[1] = Builder.CreateBitCast(Ops[1], BitTy, "vbsl"); |
| 8015 | Ops[2] = Builder.CreateBitCast(Ops[2], BitTy, "vbsl"); |
| 8016 | |
| 8017 | Ops[1] = Builder.CreateAnd(Ops[0], Ops[1], "vbsl"); |
| 8018 | Ops[2] = Builder.CreateAnd(Builder.CreateNot(Ops[0]), Ops[2], "vbsl"); |
| 8019 | Ops[0] = Builder.CreateOr(Ops[1], Ops[2], "vbsl"); |
| 8020 | return Builder.CreateBitCast(Ops[0], Ty); |
| 8021 | } |
| 8022 | case NEON::BI__builtin_neon_vfma_lane_v: |
| 8023 | case NEON::BI__builtin_neon_vfmaq_lane_v: { // Only used for FP types |
| 8024 | // The ARM builtins (and instructions) have the addend as the first |
| 8025 | // operand, but the 'fma' intrinsics have it last. Swap it around here. |
| 8026 | Value *Addend = Ops[0]; |
| 8027 | Value *Multiplicand = Ops[1]; |
| 8028 | Value *LaneSource = Ops[2]; |
| 8029 | Ops[0] = Multiplicand; |
| 8030 | Ops[1] = LaneSource; |
| 8031 | Ops[2] = Addend; |
| 8032 | |
| 8033 | // Now adjust things to handle the lane access. |
| 8034 | llvm::Type *SourceTy = BuiltinID == NEON::BI__builtin_neon_vfmaq_lane_v ? |
| 8035 | llvm::VectorType::get(VTy->getElementType(), VTy->getNumElements() / 2) : |
| 8036 | VTy; |
| 8037 | llvm::Constant *cst = cast<Constant>(Ops[3]); |
| 8038 | Value *SV = llvm::ConstantVector::getSplat(VTy->getNumElements(), cst); |
| 8039 | Ops[1] = Builder.CreateBitCast(Ops[1], SourceTy); |
| 8040 | Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV, "lane"); |
| 8041 | |
| 8042 | Ops.pop_back(); |
| 8043 | Int = Intrinsic::fma; |
| 8044 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "fmla"); |
| 8045 | } |
| 8046 | case NEON::BI__builtin_neon_vfma_laneq_v: { |
| 8047 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); |
| 8048 | // v1f64 fma should be mapped to Neon scalar f64 fma |
| 8049 | if (VTy && VTy->getElementType() == DoubleTy) { |
| 8050 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); |
| 8051 | Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy); |
| 8052 | llvm::Type *VTy = GetNeonType(this, |
| 8053 | NeonTypeFlags(NeonTypeFlags::Float64, false, true)); |
| 8054 | Ops[2] = Builder.CreateBitCast(Ops[2], VTy); |
| 8055 | Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract"); |
| 8056 | Function *F = CGM.getIntrinsic(Intrinsic::fma, DoubleTy); |
| 8057 | Value *Result = Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]}); |
| 8058 | return Builder.CreateBitCast(Result, Ty); |
| 8059 | } |
| 8060 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); |
| 8061 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8062 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 8063 | |
| 8064 | llvm::Type *STy = llvm::VectorType::get(VTy->getElementType(), |
| 8065 | VTy->getNumElements() * 2); |
| 8066 | Ops[2] = Builder.CreateBitCast(Ops[2], STy); |
| 8067 | Value* SV = llvm::ConstantVector::getSplat(VTy->getNumElements(), |
| 8068 | cast<ConstantInt>(Ops[3])); |
| 8069 | Ops[2] = Builder.CreateShuffleVector(Ops[2], Ops[2], SV, "lane"); |
| 8070 | |
| 8071 | return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]}); |
| 8072 | } |
| 8073 | case NEON::BI__builtin_neon_vfmaq_laneq_v: { |
| 8074 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); |
| 8075 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8076 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 8077 | |
| 8078 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 8079 | Ops[2] = EmitNeonSplat(Ops[2], cast<ConstantInt>(Ops[3])); |
| 8080 | return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]}); |
| 8081 | } |
| 8082 | case NEON::BI__builtin_neon_vfmah_lane_f16: |
| 8083 | case NEON::BI__builtin_neon_vfmas_lane_f32: |
| 8084 | case NEON::BI__builtin_neon_vfmah_laneq_f16: |
| 8085 | case NEON::BI__builtin_neon_vfmas_laneq_f32: |
| 8086 | case NEON::BI__builtin_neon_vfmad_lane_f64: |
| 8087 | case NEON::BI__builtin_neon_vfmad_laneq_f64: { |
| 8088 | Ops.push_back(EmitScalarExpr(E->getArg(3))); |
| 8089 | llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext())); |
| 8090 | Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty); |
| 8091 | Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract"); |
| 8092 | return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]}); |
| 8093 | } |
| 8094 | case NEON::BI__builtin_neon_vmull_v: |
| 8095 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. |
| 8096 | Int = usgn ? Intrinsic::aarch64_neon_umull : Intrinsic::aarch64_neon_smull; |
| 8097 | if (Type.isPoly()) Int = Intrinsic::aarch64_neon_pmull; |
| 8098 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull"); |
| 8099 | case NEON::BI__builtin_neon_vmax_v: |
| 8100 | case NEON::BI__builtin_neon_vmaxq_v: |
| 8101 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. |
| 8102 | Int = usgn ? Intrinsic::aarch64_neon_umax : Intrinsic::aarch64_neon_smax; |
| 8103 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmax; |
| 8104 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmax"); |
| 8105 | case NEON::BI__builtin_neon_vmaxh_f16: { |
| 8106 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 8107 | Int = Intrinsic::aarch64_neon_fmax; |
| 8108 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmax"); |
| 8109 | } |
| 8110 | case NEON::BI__builtin_neon_vmin_v: |
| 8111 | case NEON::BI__builtin_neon_vminq_v: |
| 8112 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. |
| 8113 | Int = usgn ? Intrinsic::aarch64_neon_umin : Intrinsic::aarch64_neon_smin; |
| 8114 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmin; |
| 8115 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmin"); |
| 8116 | case NEON::BI__builtin_neon_vminh_f16: { |
| 8117 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 8118 | Int = Intrinsic::aarch64_neon_fmin; |
| 8119 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmin"); |
| 8120 | } |
| 8121 | case NEON::BI__builtin_neon_vabd_v: |
| 8122 | case NEON::BI__builtin_neon_vabdq_v: |
| 8123 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. |
| 8124 | Int = usgn ? Intrinsic::aarch64_neon_uabd : Intrinsic::aarch64_neon_sabd; |
| 8125 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fabd; |
| 8126 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vabd"); |
| 8127 | case NEON::BI__builtin_neon_vpadal_v: |
| 8128 | case NEON::BI__builtin_neon_vpadalq_v: { |
| 8129 | unsigned ArgElts = VTy->getNumElements(); |
| 8130 | llvm::IntegerType *EltTy = cast<IntegerType>(VTy->getElementType()); |
| 8131 | unsigned BitWidth = EltTy->getBitWidth(); |
| 8132 | llvm::Type *ArgTy = llvm::VectorType::get( |
| 8133 | llvm::IntegerType::get(getLLVMContext(), BitWidth/2), 2*ArgElts); |
| 8134 | llvm::Type* Tys[2] = { VTy, ArgTy }; |
| 8135 | Int = usgn ? Intrinsic::aarch64_neon_uaddlp : Intrinsic::aarch64_neon_saddlp; |
| 8136 | SmallVector<llvm::Value*, 1> TmpOps; |
| 8137 | TmpOps.push_back(Ops[1]); |
| 8138 | Function *F = CGM.getIntrinsic(Int, Tys); |
| 8139 | llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vpadal"); |
| 8140 | llvm::Value *addend = Builder.CreateBitCast(Ops[0], tmp->getType()); |
| 8141 | return Builder.CreateAdd(tmp, addend); |
| 8142 | } |
| 8143 | case NEON::BI__builtin_neon_vpmin_v: |
| 8144 | case NEON::BI__builtin_neon_vpminq_v: |
| 8145 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. |
| 8146 | Int = usgn ? Intrinsic::aarch64_neon_uminp : Intrinsic::aarch64_neon_sminp; |
| 8147 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fminp; |
| 8148 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmin"); |
| 8149 | case NEON::BI__builtin_neon_vpmax_v: |
| 8150 | case NEON::BI__builtin_neon_vpmaxq_v: |
| 8151 | // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics. |
| 8152 | Int = usgn ? Intrinsic::aarch64_neon_umaxp : Intrinsic::aarch64_neon_smaxp; |
| 8153 | if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmaxp; |
| 8154 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmax"); |
| 8155 | case NEON::BI__builtin_neon_vminnm_v: |
| 8156 | case NEON::BI__builtin_neon_vminnmq_v: |
| 8157 | Int = Intrinsic::aarch64_neon_fminnm; |
| 8158 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vminnm"); |
| 8159 | case NEON::BI__builtin_neon_vminnmh_f16: |
| 8160 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 8161 | Int = Intrinsic::aarch64_neon_fminnm; |
| 8162 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vminnm"); |
| 8163 | case NEON::BI__builtin_neon_vmaxnm_v: |
| 8164 | case NEON::BI__builtin_neon_vmaxnmq_v: |
| 8165 | Int = Intrinsic::aarch64_neon_fmaxnm; |
| 8166 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmaxnm"); |
| 8167 | case NEON::BI__builtin_neon_vmaxnmh_f16: |
| 8168 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 8169 | Int = Intrinsic::aarch64_neon_fmaxnm; |
| 8170 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmaxnm"); |
| 8171 | case NEON::BI__builtin_neon_vrecpss_f32: { |
| 8172 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 8173 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, FloatTy), |
| 8174 | Ops, "vrecps"); |
| 8175 | } |
| 8176 | case NEON::BI__builtin_neon_vrecpsd_f64: |
| 8177 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 8178 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, DoubleTy), |
| 8179 | Ops, "vrecps"); |
| 8180 | case NEON::BI__builtin_neon_vrecpsh_f16: |
| 8181 | Ops.push_back(EmitScalarExpr(E->getArg(1))); |
| 8182 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, HalfTy), |
| 8183 | Ops, "vrecps"); |
| 8184 | case NEON::BI__builtin_neon_vqshrun_n_v: |
| 8185 | Int = Intrinsic::aarch64_neon_sqshrun; |
| 8186 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrun_n"); |
| 8187 | case NEON::BI__builtin_neon_vqrshrun_n_v: |
| 8188 | Int = Intrinsic::aarch64_neon_sqrshrun; |
| 8189 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrun_n"); |
| 8190 | case NEON::BI__builtin_neon_vqshrn_n_v: |
| 8191 | Int = usgn ? Intrinsic::aarch64_neon_uqshrn : Intrinsic::aarch64_neon_sqshrn; |
| 8192 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n"); |
| 8193 | case NEON::BI__builtin_neon_vrshrn_n_v: |
| 8194 | Int = Intrinsic::aarch64_neon_rshrn; |
| 8195 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshrn_n"); |
| 8196 | case NEON::BI__builtin_neon_vqrshrn_n_v: |
| 8197 | Int = usgn ? Intrinsic::aarch64_neon_uqrshrn : Intrinsic::aarch64_neon_sqrshrn; |
| 8198 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n"); |
| 8199 | case NEON::BI__builtin_neon_vrndah_f16: { |
| 8200 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8201 | Int = Intrinsic::round; |
| 8202 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrnda"); |
| 8203 | } |
| 8204 | case NEON::BI__builtin_neon_vrnda_v: |
| 8205 | case NEON::BI__builtin_neon_vrndaq_v: { |
| 8206 | Int = Intrinsic::round; |
| 8207 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnda"); |
| 8208 | } |
| 8209 | case NEON::BI__builtin_neon_vrndih_f16: { |
| 8210 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8211 | Int = Intrinsic::nearbyint; |
| 8212 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndi"); |
| 8213 | } |
| 8214 | case NEON::BI__builtin_neon_vrndmh_f16: { |
| 8215 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8216 | Int = Intrinsic::floor; |
| 8217 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndm"); |
| 8218 | } |
| 8219 | case NEON::BI__builtin_neon_vrndm_v: |
| 8220 | case NEON::BI__builtin_neon_vrndmq_v: { |
| 8221 | Int = Intrinsic::floor; |
| 8222 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndm"); |
| 8223 | } |
| 8224 | case NEON::BI__builtin_neon_vrndnh_f16: { |
| 8225 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8226 | Int = Intrinsic::aarch64_neon_frintn; |
| 8227 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndn"); |
| 8228 | } |
| 8229 | case NEON::BI__builtin_neon_vrndn_v: |
| 8230 | case NEON::BI__builtin_neon_vrndnq_v: { |
| 8231 | Int = Intrinsic::aarch64_neon_frintn; |
| 8232 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndn"); |
| 8233 | } |
| 8234 | case NEON::BI__builtin_neon_vrndns_f32: { |
| 8235 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8236 | Int = Intrinsic::aarch64_neon_frintn; |
| 8237 | return EmitNeonCall(CGM.getIntrinsic(Int, FloatTy), Ops, "vrndn"); |
| 8238 | } |
| 8239 | case NEON::BI__builtin_neon_vrndph_f16: { |
| 8240 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8241 | Int = Intrinsic::ceil; |
| 8242 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndp"); |
| 8243 | } |
| 8244 | case NEON::BI__builtin_neon_vrndp_v: |
| 8245 | case NEON::BI__builtin_neon_vrndpq_v: { |
| 8246 | Int = Intrinsic::ceil; |
| 8247 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndp"); |
| 8248 | } |
| 8249 | case NEON::BI__builtin_neon_vrndxh_f16: { |
| 8250 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8251 | Int = Intrinsic::rint; |
| 8252 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndx"); |
| 8253 | } |
| 8254 | case NEON::BI__builtin_neon_vrndx_v: |
| 8255 | case NEON::BI__builtin_neon_vrndxq_v: { |
| 8256 | Int = Intrinsic::rint; |
| 8257 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndx"); |
| 8258 | } |
| 8259 | case NEON::BI__builtin_neon_vrndh_f16: { |
| 8260 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8261 | Int = Intrinsic::trunc; |
| 8262 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndz"); |
| 8263 | } |
| 8264 | case NEON::BI__builtin_neon_vrnd_v: |
| 8265 | case NEON::BI__builtin_neon_vrndq_v: { |
| 8266 | Int = Intrinsic::trunc; |
| 8267 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndz"); |
| 8268 | } |
| 8269 | case NEON::BI__builtin_neon_vcvt_f64_v: |
| 8270 | case NEON::BI__builtin_neon_vcvtq_f64_v: |
| 8271 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8272 | Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, quad)); |
| 8273 | return usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt") |
| 8274 | : Builder.CreateSIToFP(Ops[0], Ty, "vcvt"); |
| 8275 | case NEON::BI__builtin_neon_vcvt_f64_f32: { |
| 8276 | assert(Type.getEltType() == NeonTypeFlags::Float64 && quad &&((Type.getEltType() == NeonTypeFlags::Float64 && quad && "unexpected vcvt_f64_f32 builtin") ? static_cast< void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8277, __PRETTY_FUNCTION__)) |
| 8277 | "unexpected vcvt_f64_f32 builtin")((Type.getEltType() == NeonTypeFlags::Float64 && quad && "unexpected vcvt_f64_f32 builtin") ? static_cast< void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8277, __PRETTY_FUNCTION__)); |
| 8278 | NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float32, false, false); |
| 8279 | Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag)); |
| 8280 | |
| 8281 | return Builder.CreateFPExt(Ops[0], Ty, "vcvt"); |
| 8282 | } |
| 8283 | case NEON::BI__builtin_neon_vcvt_f32_f64: { |
| 8284 | assert(Type.getEltType() == NeonTypeFlags::Float32 &&((Type.getEltType() == NeonTypeFlags::Float32 && "unexpected vcvt_f32_f64 builtin" ) ? static_cast<void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8285, __PRETTY_FUNCTION__)) |
| 8285 | "unexpected vcvt_f32_f64 builtin")((Type.getEltType() == NeonTypeFlags::Float32 && "unexpected vcvt_f32_f64 builtin" ) ? static_cast<void> (0) : __assert_fail ("Type.getEltType() == NeonTypeFlags::Float32 && \"unexpected vcvt_f32_f64 builtin\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 8285, __PRETTY_FUNCTION__)); |
| 8286 | NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float64, false, true); |
| 8287 | Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag)); |
| 8288 | |
| 8289 | return Builder.CreateFPTrunc(Ops[0], Ty, "vcvt"); |
| 8290 | } |
| 8291 | case NEON::BI__builtin_neon_vcvt_s32_v: |
| 8292 | case NEON::BI__builtin_neon_vcvt_u32_v: |
| 8293 | case NEON::BI__builtin_neon_vcvt_s64_v: |
| 8294 | case NEON::BI__builtin_neon_vcvt_u64_v: |
| 8295 | case NEON::BI__builtin_neon_vcvt_s16_v: |
| 8296 | case NEON::BI__builtin_neon_vcvt_u16_v: |
| 8297 | case NEON::BI__builtin_neon_vcvtq_s32_v: |
| 8298 | case NEON::BI__builtin_neon_vcvtq_u32_v: |
| 8299 | case NEON::BI__builtin_neon_vcvtq_s64_v: |
| 8300 | case NEON::BI__builtin_neon_vcvtq_u64_v: |
| 8301 | case NEON::BI__builtin_neon_vcvtq_s16_v: |
| 8302 | case NEON::BI__builtin_neon_vcvtq_u16_v: { |
| 8303 | Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type)); |
| 8304 | if (usgn) |
| 8305 | return Builder.CreateFPToUI(Ops[0], Ty); |
| 8306 | return Builder.CreateFPToSI(Ops[0], Ty); |
| 8307 | } |
| 8308 | case NEON::BI__builtin_neon_vcvta_s16_v: |
| 8309 | case NEON::BI__builtin_neon_vcvta_u16_v: |
| 8310 | case NEON::BI__builtin_neon_vcvta_s32_v: |
| 8311 | case NEON::BI__builtin_neon_vcvtaq_s16_v: |
| 8312 | case NEON::BI__builtin_neon_vcvtaq_s32_v: |
| 8313 | case NEON::BI__builtin_neon_vcvta_u32_v: |
| 8314 | case NEON::BI__builtin_neon_vcvtaq_u16_v: |
| 8315 | case NEON::BI__builtin_neon_vcvtaq_u32_v: |
| 8316 | case NEON::BI__builtin_neon_vcvta_s64_v: |
| 8317 | case NEON::BI__builtin_neon_vcvtaq_s64_v: |
| 8318 | case NEON::BI__builtin_neon_vcvta_u64_v: |
| 8319 | case NEON::BI__builtin_neon_vcvtaq_u64_v: { |
| 8320 | Int = usgn ? Intrinsic::aarch64_neon_fcvtau : Intrinsic::aarch64_neon_fcvtas; |
| 8321 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; |
| 8322 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvta"); |
| 8323 | } |
| 8324 | case NEON::BI__builtin_neon_vcvtm_s16_v: |
| 8325 | case NEON::BI__builtin_neon_vcvtm_s32_v: |
| 8326 | case NEON::BI__builtin_neon_vcvtmq_s16_v: |
| 8327 | case NEON::BI__builtin_neon_vcvtmq_s32_v: |
| 8328 | case NEON::BI__builtin_neon_vcvtm_u16_v: |
| 8329 | case NEON::BI__builtin_neon_vcvtm_u32_v: |
| 8330 | case NEON::BI__builtin_neon_vcvtmq_u16_v: |
| 8331 | case NEON::BI__builtin_neon_vcvtmq_u32_v: |
| 8332 | case NEON::BI__builtin_neon_vcvtm_s64_v: |
| 8333 | case NEON::BI__builtin_neon_vcvtmq_s64_v: |
| 8334 | case NEON::BI__builtin_neon_vcvtm_u64_v: |
| 8335 | case NEON::BI__builtin_neon_vcvtmq_u64_v: { |
| 8336 | Int = usgn ? Intrinsic::aarch64_neon_fcvtmu : Intrinsic::aarch64_neon_fcvtms; |
| 8337 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; |
| 8338 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtm"); |
| 8339 | } |
| 8340 | case NEON::BI__builtin_neon_vcvtn_s16_v: |
| 8341 | case NEON::BI__builtin_neon_vcvtn_s32_v: |
| 8342 | case NEON::BI__builtin_neon_vcvtnq_s16_v: |
| 8343 | case NEON::BI__builtin_neon_vcvtnq_s32_v: |
| 8344 | case NEON::BI__builtin_neon_vcvtn_u16_v: |
| 8345 | case NEON::BI__builtin_neon_vcvtn_u32_v: |
| 8346 | case NEON::BI__builtin_neon_vcvtnq_u16_v: |
| 8347 | case NEON::BI__builtin_neon_vcvtnq_u32_v: |
| 8348 | case NEON::BI__builtin_neon_vcvtn_s64_v: |
| 8349 | case NEON::BI__builtin_neon_vcvtnq_s64_v: |
| 8350 | case NEON::BI__builtin_neon_vcvtn_u64_v: |
| 8351 | case NEON::BI__builtin_neon_vcvtnq_u64_v: { |
| 8352 | Int = usgn ? Intrinsic::aarch64_neon_fcvtnu : Intrinsic::aarch64_neon_fcvtns; |
| 8353 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; |
| 8354 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtn"); |
| 8355 | } |
| 8356 | case NEON::BI__builtin_neon_vcvtp_s16_v: |
| 8357 | case NEON::BI__builtin_neon_vcvtp_s32_v: |
| 8358 | case NEON::BI__builtin_neon_vcvtpq_s16_v: |
| 8359 | case NEON::BI__builtin_neon_vcvtpq_s32_v: |
| 8360 | case NEON::BI__builtin_neon_vcvtp_u16_v: |
| 8361 | case NEON::BI__builtin_neon_vcvtp_u32_v: |
| 8362 | case NEON::BI__builtin_neon_vcvtpq_u16_v: |
| 8363 | case NEON::BI__builtin_neon_vcvtpq_u32_v: |
| 8364 | case NEON::BI__builtin_neon_vcvtp_s64_v: |
| 8365 | case NEON::BI__builtin_neon_vcvtpq_s64_v: |
| 8366 | case NEON::BI__builtin_neon_vcvtp_u64_v: |
| 8367 | case NEON::BI__builtin_neon_vcvtpq_u64_v: { |
| 8368 | Int = usgn ? Intrinsic::aarch64_neon_fcvtpu : Intrinsic::aarch64_neon_fcvtps; |
| 8369 | llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) }; |
| 8370 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtp"); |
| 8371 | } |
| 8372 | case NEON::BI__builtin_neon_vmulx_v: |
| 8373 | case NEON::BI__builtin_neon_vmulxq_v: { |
| 8374 | Int = Intrinsic::aarch64_neon_fmulx; |
| 8375 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmulx"); |
| 8376 | } |
| 8377 | case NEON::BI__builtin_neon_vmulxh_lane_f16: |
| 8378 | case NEON::BI__builtin_neon_vmulxh_laneq_f16: { |
| 8379 | // vmulx_lane should be mapped to Neon scalar mulx after |
| 8380 | // extracting the scalar element |
| 8381 | Ops.push_back(EmitScalarExpr(E->getArg(2))); |
| 8382 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract"); |
| 8383 | Ops.pop_back(); |
| 8384 | Int = Intrinsic::aarch64_neon_fmulx; |
| 8385 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmulx"); |
| 8386 | } |
| 8387 | case NEON::BI__builtin_neon_vmul_lane_v: |
| 8388 | case NEON::BI__builtin_neon_vmul_laneq_v: { |
| 8389 | // v1f64 vmul_lane should be mapped to Neon scalar mul lane |
| 8390 | bool Quad = false; |
| 8391 | if (BuiltinID == NEON::BI__builtin_neon_vmul_laneq_v) |
| 8392 | Quad = true; |
| 8393 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); |
| 8394 | llvm::Type *VTy = GetNeonType(this, |
| 8395 | NeonTypeFlags(NeonTypeFlags::Float64, false, Quad)); |
| 8396 | Ops[1] = Builder.CreateBitCast(Ops[1], VTy); |
| 8397 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract"); |
| 8398 | Value *Result = Builder.CreateFMul(Ops[0], Ops[1]); |
| 8399 | return Builder.CreateBitCast(Result, Ty); |
| 8400 | } |
| 8401 | case NEON::BI__builtin_neon_vnegd_s64: |
| 8402 | return Builder.CreateNeg(EmitScalarExpr(E->getArg(0)), "vnegd"); |
| 8403 | case NEON::BI__builtin_neon_vnegh_f16: |
| 8404 | return Builder.CreateFNeg(EmitScalarExpr(E->getArg(0)), "vnegh"); |
| 8405 | case NEON::BI__builtin_neon_vpmaxnm_v: |
| 8406 | case NEON::BI__builtin_neon_vpmaxnmq_v: { |
| 8407 | Int = Intrinsic::aarch64_neon_fmaxnmp; |
| 8408 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmaxnm"); |
| 8409 | } |
| 8410 | case NEON::BI__builtin_neon_vpminnm_v: |
| 8411 | case NEON::BI__builtin_neon_vpminnmq_v: { |
| 8412 | Int = Intrinsic::aarch64_neon_fminnmp; |
| 8413 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpminnm"); |
| 8414 | } |
| 8415 | case NEON::BI__builtin_neon_vsqrth_f16: { |
| 8416 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8417 | Int = Intrinsic::sqrt; |
| 8418 | return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vsqrt"); |
| 8419 | } |
| 8420 | case NEON::BI__builtin_neon_vsqrt_v: |
| 8421 | case NEON::BI__builtin_neon_vsqrtq_v: { |
| 8422 | Int = Intrinsic::sqrt; |
| 8423 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8424 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqrt"); |
| 8425 | } |
| 8426 | case NEON::BI__builtin_neon_vrbit_v: |
| 8427 | case NEON::BI__builtin_neon_vrbitq_v: { |
| 8428 | Int = Intrinsic::aarch64_neon_rbit; |
| 8429 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrbit"); |
| 8430 | } |
| 8431 | case NEON::BI__builtin_neon_vaddv_u8: |
| 8432 | // FIXME: These are handled by the AArch64 scalar code. |
| 8433 | usgn = true; |
| 8434 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 8435 | case NEON::BI__builtin_neon_vaddv_s8: { |
| 8436 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; |
| 8437 | Ty = Int32Ty; |
| 8438 | VTy = llvm::VectorType::get(Int8Ty, 8); |
| 8439 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8440 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8441 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); |
| 8442 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8443 | } |
| 8444 | case NEON::BI__builtin_neon_vaddv_u16: |
| 8445 | usgn = true; |
| 8446 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 8447 | case NEON::BI__builtin_neon_vaddv_s16: { |
| 8448 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; |
| 8449 | Ty = Int32Ty; |
| 8450 | VTy = llvm::VectorType::get(Int16Ty, 4); |
| 8451 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8452 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8453 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); |
| 8454 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8455 | } |
| 8456 | case NEON::BI__builtin_neon_vaddvq_u8: |
| 8457 | usgn = true; |
| 8458 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 8459 | case NEON::BI__builtin_neon_vaddvq_s8: { |
| 8460 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; |
| 8461 | Ty = Int32Ty; |
| 8462 | VTy = llvm::VectorType::get(Int8Ty, 16); |
| 8463 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8464 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8465 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); |
| 8466 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8467 | } |
| 8468 | case NEON::BI__builtin_neon_vaddvq_u16: |
| 8469 | usgn = true; |
| 8470 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 8471 | case NEON::BI__builtin_neon_vaddvq_s16: { |
| 8472 | Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv; |
| 8473 | Ty = Int32Ty; |
| 8474 | VTy = llvm::VectorType::get(Int16Ty, 8); |
| 8475 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8476 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8477 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv"); |
| 8478 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8479 | } |
| 8480 | case NEON::BI__builtin_neon_vmaxv_u8: { |
| 8481 | Int = Intrinsic::aarch64_neon_umaxv; |
| 8482 | Ty = Int32Ty; |
| 8483 | VTy = llvm::VectorType::get(Int8Ty, 8); |
| 8484 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8485 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8486 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8487 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8488 | } |
| 8489 | case NEON::BI__builtin_neon_vmaxv_u16: { |
| 8490 | Int = Intrinsic::aarch64_neon_umaxv; |
| 8491 | Ty = Int32Ty; |
| 8492 | VTy = llvm::VectorType::get(Int16Ty, 4); |
| 8493 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8494 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8495 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8496 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8497 | } |
| 8498 | case NEON::BI__builtin_neon_vmaxvq_u8: { |
| 8499 | Int = Intrinsic::aarch64_neon_umaxv; |
| 8500 | Ty = Int32Ty; |
| 8501 | VTy = llvm::VectorType::get(Int8Ty, 16); |
| 8502 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8503 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8504 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8505 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8506 | } |
| 8507 | case NEON::BI__builtin_neon_vmaxvq_u16: { |
| 8508 | Int = Intrinsic::aarch64_neon_umaxv; |
| 8509 | Ty = Int32Ty; |
| 8510 | VTy = llvm::VectorType::get(Int16Ty, 8); |
| 8511 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8512 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8513 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8514 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8515 | } |
| 8516 | case NEON::BI__builtin_neon_vmaxv_s8: { |
| 8517 | Int = Intrinsic::aarch64_neon_smaxv; |
| 8518 | Ty = Int32Ty; |
| 8519 | VTy = llvm::VectorType::get(Int8Ty, 8); |
| 8520 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8521 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8522 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8523 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8524 | } |
| 8525 | case NEON::BI__builtin_neon_vmaxv_s16: { |
| 8526 | Int = Intrinsic::aarch64_neon_smaxv; |
| 8527 | Ty = Int32Ty; |
| 8528 | VTy = llvm::VectorType::get(Int16Ty, 4); |
| 8529 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8530 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8531 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8532 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8533 | } |
| 8534 | case NEON::BI__builtin_neon_vmaxvq_s8: { |
| 8535 | Int = Intrinsic::aarch64_neon_smaxv; |
| 8536 | Ty = Int32Ty; |
| 8537 | VTy = llvm::VectorType::get(Int8Ty, 16); |
| 8538 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8539 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8540 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8541 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8542 | } |
| 8543 | case NEON::BI__builtin_neon_vmaxvq_s16: { |
| 8544 | Int = Intrinsic::aarch64_neon_smaxv; |
| 8545 | Ty = Int32Ty; |
| 8546 | VTy = llvm::VectorType::get(Int16Ty, 8); |
| 8547 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8548 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8549 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8550 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8551 | } |
| 8552 | case NEON::BI__builtin_neon_vmaxv_f16: { |
| 8553 | Int = Intrinsic::aarch64_neon_fmaxv; |
| 8554 | Ty = HalfTy; |
| 8555 | VTy = llvm::VectorType::get(HalfTy, 4); |
| 8556 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8557 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8558 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8559 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8560 | } |
| 8561 | case NEON::BI__builtin_neon_vmaxvq_f16: { |
| 8562 | Int = Intrinsic::aarch64_neon_fmaxv; |
| 8563 | Ty = HalfTy; |
| 8564 | VTy = llvm::VectorType::get(HalfTy, 8); |
| 8565 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8566 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8567 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv"); |
| 8568 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8569 | } |
| 8570 | case NEON::BI__builtin_neon_vminv_u8: { |
| 8571 | Int = Intrinsic::aarch64_neon_uminv; |
| 8572 | Ty = Int32Ty; |
| 8573 | VTy = llvm::VectorType::get(Int8Ty, 8); |
| 8574 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8575 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8576 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8577 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8578 | } |
| 8579 | case NEON::BI__builtin_neon_vminv_u16: { |
| 8580 | Int = Intrinsic::aarch64_neon_uminv; |
| 8581 | Ty = Int32Ty; |
| 8582 | VTy = llvm::VectorType::get(Int16Ty, 4); |
| 8583 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8584 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8585 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8586 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8587 | } |
| 8588 | case NEON::BI__builtin_neon_vminvq_u8: { |
| 8589 | Int = Intrinsic::aarch64_neon_uminv; |
| 8590 | Ty = Int32Ty; |
| 8591 | VTy = llvm::VectorType::get(Int8Ty, 16); |
| 8592 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8593 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8594 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8595 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8596 | } |
| 8597 | case NEON::BI__builtin_neon_vminvq_u16: { |
| 8598 | Int = Intrinsic::aarch64_neon_uminv; |
| 8599 | Ty = Int32Ty; |
| 8600 | VTy = llvm::VectorType::get(Int16Ty, 8); |
| 8601 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8602 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8603 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8604 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8605 | } |
| 8606 | case NEON::BI__builtin_neon_vminv_s8: { |
| 8607 | Int = Intrinsic::aarch64_neon_sminv; |
| 8608 | Ty = Int32Ty; |
| 8609 | VTy = llvm::VectorType::get(Int8Ty, 8); |
| 8610 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8611 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8612 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8613 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8614 | } |
| 8615 | case NEON::BI__builtin_neon_vminv_s16: { |
| 8616 | Int = Intrinsic::aarch64_neon_sminv; |
| 8617 | Ty = Int32Ty; |
| 8618 | VTy = llvm::VectorType::get(Int16Ty, 4); |
| 8619 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8620 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8621 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8622 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8623 | } |
| 8624 | case NEON::BI__builtin_neon_vminvq_s8: { |
| 8625 | Int = Intrinsic::aarch64_neon_sminv; |
| 8626 | Ty = Int32Ty; |
| 8627 | VTy = llvm::VectorType::get(Int8Ty, 16); |
| 8628 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8629 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8630 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8631 | return Builder.CreateTrunc(Ops[0], Int8Ty); |
| 8632 | } |
| 8633 | case NEON::BI__builtin_neon_vminvq_s16: { |
| 8634 | Int = Intrinsic::aarch64_neon_sminv; |
| 8635 | Ty = Int32Ty; |
| 8636 | VTy = llvm::VectorType::get(Int16Ty, 8); |
| 8637 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8638 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8639 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8640 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8641 | } |
| 8642 | case NEON::BI__builtin_neon_vminv_f16: { |
| 8643 | Int = Intrinsic::aarch64_neon_fminv; |
| 8644 | Ty = HalfTy; |
| 8645 | VTy = llvm::VectorType::get(HalfTy, 4); |
| 8646 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8647 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8648 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8649 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8650 | } |
| 8651 | case NEON::BI__builtin_neon_vminvq_f16: { |
| 8652 | Int = Intrinsic::aarch64_neon_fminv; |
| 8653 | Ty = HalfTy; |
| 8654 | VTy = llvm::VectorType::get(HalfTy, 8); |
| 8655 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8656 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8657 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv"); |
| 8658 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8659 | } |
| 8660 | case NEON::BI__builtin_neon_vmaxnmv_f16: { |
| 8661 | Int = Intrinsic::aarch64_neon_fmaxnmv; |
| 8662 | Ty = HalfTy; |
| 8663 | VTy = llvm::VectorType::get(HalfTy, 4); |
| 8664 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8665 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8666 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv"); |
| 8667 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8668 | } |
| 8669 | case NEON::BI__builtin_neon_vmaxnmvq_f16: { |
| 8670 | Int = Intrinsic::aarch64_neon_fmaxnmv; |
| 8671 | Ty = HalfTy; |
| 8672 | VTy = llvm::VectorType::get(HalfTy, 8); |
| 8673 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8674 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8675 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv"); |
| 8676 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8677 | } |
| 8678 | case NEON::BI__builtin_neon_vminnmv_f16: { |
| 8679 | Int = Intrinsic::aarch64_neon_fminnmv; |
| 8680 | Ty = HalfTy; |
| 8681 | VTy = llvm::VectorType::get(HalfTy, 4); |
| 8682 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8683 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8684 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv"); |
| 8685 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8686 | } |
| 8687 | case NEON::BI__builtin_neon_vminnmvq_f16: { |
| 8688 | Int = Intrinsic::aarch64_neon_fminnmv; |
| 8689 | Ty = HalfTy; |
| 8690 | VTy = llvm::VectorType::get(HalfTy, 8); |
| 8691 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8692 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8693 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv"); |
| 8694 | return Builder.CreateTrunc(Ops[0], HalfTy); |
| 8695 | } |
| 8696 | case NEON::BI__builtin_neon_vmul_n_f64: { |
| 8697 | Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy); |
| 8698 | Value *RHS = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), DoubleTy); |
| 8699 | return Builder.CreateFMul(Ops[0], RHS); |
| 8700 | } |
| 8701 | case NEON::BI__builtin_neon_vaddlv_u8: { |
| 8702 | Int = Intrinsic::aarch64_neon_uaddlv; |
| 8703 | Ty = Int32Ty; |
| 8704 | VTy = llvm::VectorType::get(Int8Ty, 8); |
| 8705 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8706 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8707 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8708 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8709 | } |
| 8710 | case NEON::BI__builtin_neon_vaddlv_u16: { |
| 8711 | Int = Intrinsic::aarch64_neon_uaddlv; |
| 8712 | Ty = Int32Ty; |
| 8713 | VTy = llvm::VectorType::get(Int16Ty, 4); |
| 8714 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8715 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8716 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8717 | } |
| 8718 | case NEON::BI__builtin_neon_vaddlvq_u8: { |
| 8719 | Int = Intrinsic::aarch64_neon_uaddlv; |
| 8720 | Ty = Int32Ty; |
| 8721 | VTy = llvm::VectorType::get(Int8Ty, 16); |
| 8722 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8723 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8724 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8725 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8726 | } |
| 8727 | case NEON::BI__builtin_neon_vaddlvq_u16: { |
| 8728 | Int = Intrinsic::aarch64_neon_uaddlv; |
| 8729 | Ty = Int32Ty; |
| 8730 | VTy = llvm::VectorType::get(Int16Ty, 8); |
| 8731 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8732 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8733 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8734 | } |
| 8735 | case NEON::BI__builtin_neon_vaddlv_s8: { |
| 8736 | Int = Intrinsic::aarch64_neon_saddlv; |
| 8737 | Ty = Int32Ty; |
| 8738 | VTy = llvm::VectorType::get(Int8Ty, 8); |
| 8739 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8740 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8741 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8742 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8743 | } |
| 8744 | case NEON::BI__builtin_neon_vaddlv_s16: { |
| 8745 | Int = Intrinsic::aarch64_neon_saddlv; |
| 8746 | Ty = Int32Ty; |
| 8747 | VTy = llvm::VectorType::get(Int16Ty, 4); |
| 8748 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8749 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8750 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8751 | } |
| 8752 | case NEON::BI__builtin_neon_vaddlvq_s8: { |
| 8753 | Int = Intrinsic::aarch64_neon_saddlv; |
| 8754 | Ty = Int32Ty; |
| 8755 | VTy = llvm::VectorType::get(Int8Ty, 16); |
| 8756 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8757 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8758 | Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8759 | return Builder.CreateTrunc(Ops[0], Int16Ty); |
| 8760 | } |
| 8761 | case NEON::BI__builtin_neon_vaddlvq_s16: { |
| 8762 | Int = Intrinsic::aarch64_neon_saddlv; |
| 8763 | Ty = Int32Ty; |
| 8764 | VTy = llvm::VectorType::get(Int16Ty, 8); |
| 8765 | llvm::Type *Tys[2] = { Ty, VTy }; |
| 8766 | Ops.push_back(EmitScalarExpr(E->getArg(0))); |
| 8767 | return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv"); |
| 8768 | } |
| 8769 | case NEON::BI__builtin_neon_vsri_n_v: |
| 8770 | case NEON::BI__builtin_neon_vsriq_n_v: { |
| 8771 | Int = Intrinsic::aarch64_neon_vsri; |
| 8772 | llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty); |
| 8773 | return EmitNeonCall(Intrin, Ops, "vsri_n"); |
| 8774 | } |
| 8775 | case NEON::BI__builtin_neon_vsli_n_v: |
| 8776 | case NEON::BI__builtin_neon_vsliq_n_v: { |
| 8777 | Int = Intrinsic::aarch64_neon_vsli; |
| 8778 | llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty); |
| 8779 | return EmitNeonCall(Intrin, Ops, "vsli_n"); |
| 8780 | } |
| 8781 | case NEON::BI__builtin_neon_vsra_n_v: |
| 8782 | case NEON::BI__builtin_neon_vsraq_n_v: |
| 8783 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8784 | Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n"); |
| 8785 | return Builder.CreateAdd(Ops[0], Ops[1]); |
| 8786 | case NEON::BI__builtin_neon_vrsra_n_v: |
| 8787 | case NEON::BI__builtin_neon_vrsraq_n_v: { |
| 8788 | Int = usgn ? Intrinsic::aarch64_neon_urshl : Intrinsic::aarch64_neon_srshl; |
| 8789 | SmallVector<llvm::Value*,2> TmpOps; |
| 8790 | TmpOps.push_back(Ops[1]); |
| 8791 | TmpOps.push_back(Ops[2]); |
| 8792 | Function* F = CGM.getIntrinsic(Int, Ty); |
| 8793 | llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vrshr_n", 1, true); |
| 8794 | Ops[0] = Builder.CreateBitCast(Ops[0], VTy); |
| 8795 | return Builder.CreateAdd(Ops[0], tmp); |
| 8796 | } |
| 8797 | case NEON::BI__builtin_neon_vld1_v: |
| 8798 | case NEON::BI__builtin_neon_vld1q_v: { |
| 8799 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy)); |
| 8800 | auto Alignment = CharUnits::fromQuantity( |
| 8801 | BuiltinID == NEON::BI__builtin_neon_vld1_v ? 8 : 16); |
| 8802 | return Builder.CreateAlignedLoad(VTy, Ops[0], Alignment); |
| 8803 | } |
| 8804 | case NEON::BI__builtin_neon_vst1_v: |
| 8805 | case NEON::BI__builtin_neon_vst1q_v: |
| 8806 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy)); |
| 8807 | Ops[1] = Builder.CreateBitCast(Ops[1], VTy); |
| 8808 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8809 | case NEON::BI__builtin_neon_vld1_lane_v: |
| 8810 | case NEON::BI__builtin_neon_vld1q_lane_v: { |
| 8811 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 8812 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); |
| 8813 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8814 | auto Alignment = CharUnits::fromQuantity( |
| 8815 | BuiltinID == NEON::BI__builtin_neon_vld1_lane_v ? 8 : 16); |
| 8816 | Ops[0] = |
| 8817 | Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment); |
| 8818 | return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vld1_lane"); |
| 8819 | } |
| 8820 | case NEON::BI__builtin_neon_vld1_dup_v: |
| 8821 | case NEON::BI__builtin_neon_vld1q_dup_v: { |
| 8822 | Value *V = UndefValue::get(Ty); |
| 8823 | Ty = llvm::PointerType::getUnqual(VTy->getElementType()); |
| 8824 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8825 | auto Alignment = CharUnits::fromQuantity( |
| 8826 | BuiltinID == NEON::BI__builtin_neon_vld1_dup_v ? 8 : 16); |
| 8827 | Ops[0] = |
| 8828 | Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment); |
| 8829 | llvm::Constant *CI = ConstantInt::get(Int32Ty, 0); |
| 8830 | Ops[0] = Builder.CreateInsertElement(V, Ops[0], CI); |
| 8831 | return EmitNeonSplat(Ops[0], CI); |
| 8832 | } |
| 8833 | case NEON::BI__builtin_neon_vst1_lane_v: |
| 8834 | case NEON::BI__builtin_neon_vst1q_lane_v: |
| 8835 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 8836 | Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]); |
| 8837 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 8838 | return Builder.CreateDefaultAlignedStore(Ops[1], |
| 8839 | Builder.CreateBitCast(Ops[0], Ty)); |
| 8840 | case NEON::BI__builtin_neon_vld2_v: |
| 8841 | case NEON::BI__builtin_neon_vld2q_v: { |
| 8842 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); |
| 8843 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); |
| 8844 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 8845 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2, Tys); |
| 8846 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld2"); |
| 8847 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 8848 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 8849 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8850 | } |
| 8851 | case NEON::BI__builtin_neon_vld3_v: |
| 8852 | case NEON::BI__builtin_neon_vld3q_v: { |
| 8853 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); |
| 8854 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); |
| 8855 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 8856 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3, Tys); |
| 8857 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld3"); |
| 8858 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 8859 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 8860 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8861 | } |
| 8862 | case NEON::BI__builtin_neon_vld4_v: |
| 8863 | case NEON::BI__builtin_neon_vld4q_v: { |
| 8864 | llvm::Type *PTy = llvm::PointerType::getUnqual(VTy); |
| 8865 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); |
| 8866 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 8867 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4, Tys); |
| 8868 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld4"); |
| 8869 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 8870 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 8871 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8872 | } |
| 8873 | case NEON::BI__builtin_neon_vld2_dup_v: |
| 8874 | case NEON::BI__builtin_neon_vld2q_dup_v: { |
| 8875 | llvm::Type *PTy = |
| 8876 | llvm::PointerType::getUnqual(VTy->getElementType()); |
| 8877 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); |
| 8878 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 8879 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2r, Tys); |
| 8880 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld2"); |
| 8881 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 8882 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 8883 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8884 | } |
| 8885 | case NEON::BI__builtin_neon_vld3_dup_v: |
| 8886 | case NEON::BI__builtin_neon_vld3q_dup_v: { |
| 8887 | llvm::Type *PTy = |
| 8888 | llvm::PointerType::getUnqual(VTy->getElementType()); |
| 8889 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); |
| 8890 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 8891 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3r, Tys); |
| 8892 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld3"); |
| 8893 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 8894 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 8895 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8896 | } |
| 8897 | case NEON::BI__builtin_neon_vld4_dup_v: |
| 8898 | case NEON::BI__builtin_neon_vld4q_dup_v: { |
| 8899 | llvm::Type *PTy = |
| 8900 | llvm::PointerType::getUnqual(VTy->getElementType()); |
| 8901 | Ops[1] = Builder.CreateBitCast(Ops[1], PTy); |
| 8902 | llvm::Type *Tys[2] = { VTy, PTy }; |
| 8903 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4r, Tys); |
| 8904 | Ops[1] = Builder.CreateCall(F, Ops[1], "vld4"); |
| 8905 | Ops[0] = Builder.CreateBitCast(Ops[0], |
| 8906 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 8907 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8908 | } |
| 8909 | case NEON::BI__builtin_neon_vld2_lane_v: |
| 8910 | case NEON::BI__builtin_neon_vld2q_lane_v: { |
| 8911 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; |
| 8912 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2lane, Tys); |
| 8913 | Ops.push_back(Ops[1]); |
| 8914 | Ops.erase(Ops.begin()+1); |
| 8915 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 8916 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 8917 | Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty); |
| 8918 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld2_lane"); |
| 8919 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 8920 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8921 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8922 | } |
| 8923 | case NEON::BI__builtin_neon_vld3_lane_v: |
| 8924 | case NEON::BI__builtin_neon_vld3q_lane_v: { |
| 8925 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; |
| 8926 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3lane, Tys); |
| 8927 | Ops.push_back(Ops[1]); |
| 8928 | Ops.erase(Ops.begin()+1); |
| 8929 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 8930 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 8931 | Ops[3] = Builder.CreateBitCast(Ops[3], Ty); |
| 8932 | Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty); |
| 8933 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld3_lane"); |
| 8934 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 8935 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8936 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8937 | } |
| 8938 | case NEON::BI__builtin_neon_vld4_lane_v: |
| 8939 | case NEON::BI__builtin_neon_vld4q_lane_v: { |
| 8940 | llvm::Type *Tys[2] = { VTy, Ops[1]->getType() }; |
| 8941 | Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4lane, Tys); |
| 8942 | Ops.push_back(Ops[1]); |
| 8943 | Ops.erase(Ops.begin()+1); |
| 8944 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 8945 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 8946 | Ops[3] = Builder.CreateBitCast(Ops[3], Ty); |
| 8947 | Ops[4] = Builder.CreateBitCast(Ops[4], Ty); |
| 8948 | Ops[5] = Builder.CreateZExt(Ops[5], Int64Ty); |
| 8949 | Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld4_lane"); |
| 8950 | Ty = llvm::PointerType::getUnqual(Ops[1]->getType()); |
| 8951 | Ops[0] = Builder.CreateBitCast(Ops[0], Ty); |
| 8952 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 8953 | } |
| 8954 | case NEON::BI__builtin_neon_vst2_v: |
| 8955 | case NEON::BI__builtin_neon_vst2q_v: { |
| 8956 | Ops.push_back(Ops[0]); |
| 8957 | Ops.erase(Ops.begin()); |
| 8958 | llvm::Type *Tys[2] = { VTy, Ops[2]->getType() }; |
| 8959 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2, Tys), |
| 8960 | Ops, ""); |
| 8961 | } |
| 8962 | case NEON::BI__builtin_neon_vst2_lane_v: |
| 8963 | case NEON::BI__builtin_neon_vst2q_lane_v: { |
| 8964 | Ops.push_back(Ops[0]); |
| 8965 | Ops.erase(Ops.begin()); |
| 8966 | Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); |
| 8967 | llvm::Type *Tys[2] = { VTy, Ops[3]->getType() }; |
| 8968 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2lane, Tys), |
| 8969 | Ops, ""); |
| 8970 | } |
| 8971 | case NEON::BI__builtin_neon_vst3_v: |
| 8972 | case NEON::BI__builtin_neon_vst3q_v: { |
| 8973 | Ops.push_back(Ops[0]); |
| 8974 | Ops.erase(Ops.begin()); |
| 8975 | llvm::Type *Tys[2] = { VTy, Ops[3]->getType() }; |
| 8976 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3, Tys), |
| 8977 | Ops, ""); |
| 8978 | } |
| 8979 | case NEON::BI__builtin_neon_vst3_lane_v: |
| 8980 | case NEON::BI__builtin_neon_vst3q_lane_v: { |
| 8981 | Ops.push_back(Ops[0]); |
| 8982 | Ops.erase(Ops.begin()); |
| 8983 | Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty); |
| 8984 | llvm::Type *Tys[2] = { VTy, Ops[4]->getType() }; |
| 8985 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3lane, Tys), |
| 8986 | Ops, ""); |
| 8987 | } |
| 8988 | case NEON::BI__builtin_neon_vst4_v: |
| 8989 | case NEON::BI__builtin_neon_vst4q_v: { |
| 8990 | Ops.push_back(Ops[0]); |
| 8991 | Ops.erase(Ops.begin()); |
| 8992 | llvm::Type *Tys[2] = { VTy, Ops[4]->getType() }; |
| 8993 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4, Tys), |
| 8994 | Ops, ""); |
| 8995 | } |
| 8996 | case NEON::BI__builtin_neon_vst4_lane_v: |
| 8997 | case NEON::BI__builtin_neon_vst4q_lane_v: { |
| 8998 | Ops.push_back(Ops[0]); |
| 8999 | Ops.erase(Ops.begin()); |
| 9000 | Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty); |
| 9001 | llvm::Type *Tys[2] = { VTy, Ops[5]->getType() }; |
| 9002 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4lane, Tys), |
| 9003 | Ops, ""); |
| 9004 | } |
| 9005 | case NEON::BI__builtin_neon_vtrn_v: |
| 9006 | case NEON::BI__builtin_neon_vtrnq_v: { |
| 9007 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); |
| 9008 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 9009 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 9010 | Value *SV = nullptr; |
| 9011 | |
| 9012 | for (unsigned vi = 0; vi != 2; ++vi) { |
| 9013 | SmallVector<uint32_t, 16> Indices; |
| 9014 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { |
| 9015 | Indices.push_back(i+vi); |
| 9016 | Indices.push_back(i+e+vi); |
| 9017 | } |
| 9018 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); |
| 9019 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn"); |
| 9020 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); |
| 9021 | } |
| 9022 | return SV; |
| 9023 | } |
| 9024 | case NEON::BI__builtin_neon_vuzp_v: |
| 9025 | case NEON::BI__builtin_neon_vuzpq_v: { |
| 9026 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); |
| 9027 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 9028 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 9029 | Value *SV = nullptr; |
| 9030 | |
| 9031 | for (unsigned vi = 0; vi != 2; ++vi) { |
| 9032 | SmallVector<uint32_t, 16> Indices; |
| 9033 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) |
| 9034 | Indices.push_back(2*i+vi); |
| 9035 | |
| 9036 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); |
| 9037 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp"); |
| 9038 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); |
| 9039 | } |
| 9040 | return SV; |
| 9041 | } |
| 9042 | case NEON::BI__builtin_neon_vzip_v: |
| 9043 | case NEON::BI__builtin_neon_vzipq_v: { |
| 9044 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty)); |
| 9045 | Ops[1] = Builder.CreateBitCast(Ops[1], Ty); |
| 9046 | Ops[2] = Builder.CreateBitCast(Ops[2], Ty); |
| 9047 | Value *SV = nullptr; |
| 9048 | |
| 9049 | for (unsigned vi = 0; vi != 2; ++vi) { |
| 9050 | SmallVector<uint32_t, 16> Indices; |
| 9051 | for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) { |
| 9052 | Indices.push_back((i + vi*e) >> 1); |
| 9053 | Indices.push_back(((i + vi*e) >> 1)+e); |
| 9054 | } |
| 9055 | Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi); |
| 9056 | SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip"); |
| 9057 | SV = Builder.CreateDefaultAlignedStore(SV, Addr); |
| 9058 | } |
| 9059 | return SV; |
| 9060 | } |
| 9061 | case NEON::BI__builtin_neon_vqtbl1q_v: { |
| 9062 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl1, Ty), |
| 9063 | Ops, "vtbl1"); |
| 9064 | } |
| 9065 | case NEON::BI__builtin_neon_vqtbl2q_v: { |
| 9066 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl2, Ty), |
| 9067 | Ops, "vtbl2"); |
| 9068 | } |
| 9069 | case NEON::BI__builtin_neon_vqtbl3q_v: { |
| 9070 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl3, Ty), |
| 9071 | Ops, "vtbl3"); |
| 9072 | } |
| 9073 | case NEON::BI__builtin_neon_vqtbl4q_v: { |
| 9074 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl4, Ty), |
| 9075 | Ops, "vtbl4"); |
| 9076 | } |
| 9077 | case NEON::BI__builtin_neon_vqtbx1q_v: { |
| 9078 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx1, Ty), |
| 9079 | Ops, "vtbx1"); |
| 9080 | } |
| 9081 | case NEON::BI__builtin_neon_vqtbx2q_v: { |
| 9082 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx2, Ty), |
| 9083 | Ops, "vtbx2"); |
| 9084 | } |
| 9085 | case NEON::BI__builtin_neon_vqtbx3q_v: { |
| 9086 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx3, Ty), |
| 9087 | Ops, "vtbx3"); |
| 9088 | } |
| 9089 | case NEON::BI__builtin_neon_vqtbx4q_v: { |
| 9090 | return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx4, Ty), |
| 9091 | Ops, "vtbx4"); |
| 9092 | } |
| 9093 | case NEON::BI__builtin_neon_vsqadd_v: |
| 9094 | case NEON::BI__builtin_neon_vsqaddq_v: { |
| 9095 | Int = Intrinsic::aarch64_neon_usqadd; |
| 9096 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqadd"); |
| 9097 | } |
| 9098 | case NEON::BI__builtin_neon_vuqadd_v: |
| 9099 | case NEON::BI__builtin_neon_vuqaddq_v: { |
| 9100 | Int = Intrinsic::aarch64_neon_suqadd; |
| 9101 | return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vuqadd"); |
| 9102 | } |
| 9103 | case AArch64::BI_BitScanForward: |
| 9104 | case AArch64::BI_BitScanForward64: |
| 9105 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E); |
| 9106 | case AArch64::BI_BitScanReverse: |
| 9107 | case AArch64::BI_BitScanReverse64: |
| 9108 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E); |
| 9109 | case AArch64::BI_InterlockedAnd64: |
| 9110 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E); |
| 9111 | case AArch64::BI_InterlockedExchange64: |
| 9112 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E); |
| 9113 | case AArch64::BI_InterlockedExchangeAdd64: |
| 9114 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E); |
| 9115 | case AArch64::BI_InterlockedExchangeSub64: |
| 9116 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E); |
| 9117 | case AArch64::BI_InterlockedOr64: |
| 9118 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E); |
| 9119 | case AArch64::BI_InterlockedXor64: |
| 9120 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E); |
| 9121 | case AArch64::BI_InterlockedDecrement64: |
| 9122 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E); |
| 9123 | case AArch64::BI_InterlockedIncrement64: |
| 9124 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E); |
| 9125 | case AArch64::BI_InterlockedExchangeAdd8_acq: |
| 9126 | case AArch64::BI_InterlockedExchangeAdd16_acq: |
| 9127 | case AArch64::BI_InterlockedExchangeAdd_acq: |
| 9128 | case AArch64::BI_InterlockedExchangeAdd64_acq: |
| 9129 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_acq, E); |
| 9130 | case AArch64::BI_InterlockedExchangeAdd8_rel: |
| 9131 | case AArch64::BI_InterlockedExchangeAdd16_rel: |
| 9132 | case AArch64::BI_InterlockedExchangeAdd_rel: |
| 9133 | case AArch64::BI_InterlockedExchangeAdd64_rel: |
| 9134 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_rel, E); |
| 9135 | case AArch64::BI_InterlockedExchangeAdd8_nf: |
| 9136 | case AArch64::BI_InterlockedExchangeAdd16_nf: |
| 9137 | case AArch64::BI_InterlockedExchangeAdd_nf: |
| 9138 | case AArch64::BI_InterlockedExchangeAdd64_nf: |
| 9139 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_nf, E); |
| 9140 | case AArch64::BI_InterlockedExchange8_acq: |
| 9141 | case AArch64::BI_InterlockedExchange16_acq: |
| 9142 | case AArch64::BI_InterlockedExchange_acq: |
| 9143 | case AArch64::BI_InterlockedExchange64_acq: |
| 9144 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_acq, E); |
| 9145 | case AArch64::BI_InterlockedExchange8_rel: |
| 9146 | case AArch64::BI_InterlockedExchange16_rel: |
| 9147 | case AArch64::BI_InterlockedExchange_rel: |
| 9148 | case AArch64::BI_InterlockedExchange64_rel: |
| 9149 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_rel, E); |
| 9150 | case AArch64::BI_InterlockedExchange8_nf: |
| 9151 | case AArch64::BI_InterlockedExchange16_nf: |
| 9152 | case AArch64::BI_InterlockedExchange_nf: |
| 9153 | case AArch64::BI_InterlockedExchange64_nf: |
| 9154 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_nf, E); |
| 9155 | case AArch64::BI_InterlockedCompareExchange8_acq: |
| 9156 | case AArch64::BI_InterlockedCompareExchange16_acq: |
| 9157 | case AArch64::BI_InterlockedCompareExchange_acq: |
| 9158 | case AArch64::BI_InterlockedCompareExchange64_acq: |
| 9159 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_acq, E); |
| 9160 | case AArch64::BI_InterlockedCompareExchange8_rel: |
| 9161 | case AArch64::BI_InterlockedCompareExchange16_rel: |
| 9162 | case AArch64::BI_InterlockedCompareExchange_rel: |
| 9163 | case AArch64::BI_InterlockedCompareExchange64_rel: |
| 9164 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_rel, E); |
| 9165 | case AArch64::BI_InterlockedCompareExchange8_nf: |
| 9166 | case AArch64::BI_InterlockedCompareExchange16_nf: |
| 9167 | case AArch64::BI_InterlockedCompareExchange_nf: |
| 9168 | case AArch64::BI_InterlockedCompareExchange64_nf: |
| 9169 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_nf, E); |
| 9170 | case AArch64::BI_InterlockedOr8_acq: |
| 9171 | case AArch64::BI_InterlockedOr16_acq: |
| 9172 | case AArch64::BI_InterlockedOr_acq: |
| 9173 | case AArch64::BI_InterlockedOr64_acq: |
| 9174 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_acq, E); |
| 9175 | case AArch64::BI_InterlockedOr8_rel: |
| 9176 | case AArch64::BI_InterlockedOr16_rel: |
| 9177 | case AArch64::BI_InterlockedOr_rel: |
| 9178 | case AArch64::BI_InterlockedOr64_rel: |
| 9179 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_rel, E); |
| 9180 | case AArch64::BI_InterlockedOr8_nf: |
| 9181 | case AArch64::BI_InterlockedOr16_nf: |
| 9182 | case AArch64::BI_InterlockedOr_nf: |
| 9183 | case AArch64::BI_InterlockedOr64_nf: |
| 9184 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_nf, E); |
| 9185 | case AArch64::BI_InterlockedXor8_acq: |
| 9186 | case AArch64::BI_InterlockedXor16_acq: |
| 9187 | case AArch64::BI_InterlockedXor_acq: |
| 9188 | case AArch64::BI_InterlockedXor64_acq: |
| 9189 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_acq, E); |
| 9190 | case AArch64::BI_InterlockedXor8_rel: |
| 9191 | case AArch64::BI_InterlockedXor16_rel: |
| 9192 | case AArch64::BI_InterlockedXor_rel: |
| 9193 | case AArch64::BI_InterlockedXor64_rel: |
| 9194 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_rel, E); |
| 9195 | case AArch64::BI_InterlockedXor8_nf: |
| 9196 | case AArch64::BI_InterlockedXor16_nf: |
| 9197 | case AArch64::BI_InterlockedXor_nf: |
| 9198 | case AArch64::BI_InterlockedXor64_nf: |
| 9199 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_nf, E); |
| 9200 | case AArch64::BI_InterlockedAnd8_acq: |
| 9201 | case AArch64::BI_InterlockedAnd16_acq: |
| 9202 | case AArch64::BI_InterlockedAnd_acq: |
| 9203 | case AArch64::BI_InterlockedAnd64_acq: |
| 9204 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_acq, E); |
| 9205 | case AArch64::BI_InterlockedAnd8_rel: |
| 9206 | case AArch64::BI_InterlockedAnd16_rel: |
| 9207 | case AArch64::BI_InterlockedAnd_rel: |
| 9208 | case AArch64::BI_InterlockedAnd64_rel: |
| 9209 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_rel, E); |
| 9210 | case AArch64::BI_InterlockedAnd8_nf: |
| 9211 | case AArch64::BI_InterlockedAnd16_nf: |
| 9212 | case AArch64::BI_InterlockedAnd_nf: |
| 9213 | case AArch64::BI_InterlockedAnd64_nf: |
| 9214 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E); |
| 9215 | case AArch64::BI_InterlockedIncrement16_acq: |
| 9216 | case AArch64::BI_InterlockedIncrement_acq: |
| 9217 | case AArch64::BI_InterlockedIncrement64_acq: |
| 9218 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E); |
| 9219 | case AArch64::BI_InterlockedIncrement16_rel: |
| 9220 | case AArch64::BI_InterlockedIncrement_rel: |
| 9221 | case AArch64::BI_InterlockedIncrement64_rel: |
| 9222 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E); |
| 9223 | case AArch64::BI_InterlockedIncrement16_nf: |
| 9224 | case AArch64::BI_InterlockedIncrement_nf: |
| 9225 | case AArch64::BI_InterlockedIncrement64_nf: |
| 9226 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E); |
| 9227 | case AArch64::BI_InterlockedDecrement16_acq: |
| 9228 | case AArch64::BI_InterlockedDecrement_acq: |
| 9229 | case AArch64::BI_InterlockedDecrement64_acq: |
| 9230 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_acq, E); |
| 9231 | case AArch64::BI_InterlockedDecrement16_rel: |
| 9232 | case AArch64::BI_InterlockedDecrement_rel: |
| 9233 | case AArch64::BI_InterlockedDecrement64_rel: |
| 9234 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_rel, E); |
| 9235 | case AArch64::BI_InterlockedDecrement16_nf: |
| 9236 | case AArch64::BI_InterlockedDecrement_nf: |
| 9237 | case AArch64::BI_InterlockedDecrement64_nf: |
| 9238 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_nf, E); |
| 9239 | |
| 9240 | case AArch64::BI_InterlockedAdd: { |
| 9241 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); |
| 9242 | Value *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 9243 | AtomicRMWInst *RMWI = Builder.CreateAtomicRMW( |
| 9244 | AtomicRMWInst::Add, Arg0, Arg1, |
| 9245 | llvm::AtomicOrdering::SequentiallyConsistent); |
| 9246 | return Builder.CreateAdd(RMWI, Arg1); |
| 9247 | } |
| 9248 | } |
| 9249 | } |
| 9250 | |
| 9251 | llvm::Value *CodeGenFunction:: |
| 9252 | BuildVector(ArrayRef<llvm::Value*> Ops) { |
| 9253 | assert((Ops.size() & (Ops.size() - 1)) == 0 &&(((Ops.size() & (Ops.size() - 1)) == 0 && "Not a power-of-two sized vector!" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9254, __PRETTY_FUNCTION__)) |
| 9254 | "Not a power-of-two sized vector!")(((Ops.size() & (Ops.size() - 1)) == 0 && "Not a power-of-two sized vector!" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9254, __PRETTY_FUNCTION__)); |
| 9255 | bool AllConstants = true; |
| 9256 | for (unsigned i = 0, e = Ops.size(); i != e && AllConstants; ++i) |
| 9257 | AllConstants &= isa<Constant>(Ops[i]); |
| 9258 | |
| 9259 | // If this is a constant vector, create a ConstantVector. |
| 9260 | if (AllConstants) { |
| 9261 | SmallVector<llvm::Constant*, 16> CstOps; |
| 9262 | for (unsigned i = 0, e = Ops.size(); i != e; ++i) |
| 9263 | CstOps.push_back(cast<Constant>(Ops[i])); |
| 9264 | return llvm::ConstantVector::get(CstOps); |
| 9265 | } |
| 9266 | |
| 9267 | // Otherwise, insertelement the values to build the vector. |
| 9268 | Value *Result = |
| 9269 | llvm::UndefValue::get(llvm::VectorType::get(Ops[0]->getType(), Ops.size())); |
| 9270 | |
| 9271 | for (unsigned i = 0, e = Ops.size(); i != e; ++i) |
| 9272 | Result = Builder.CreateInsertElement(Result, Ops[i], Builder.getInt32(i)); |
| 9273 | |
| 9274 | return Result; |
| 9275 | } |
| 9276 | |
| 9277 | // Convert the mask from an integer type to a vector of i1. |
| 9278 | static Value *getMaskVecValue(CodeGenFunction &CGF, Value *Mask, |
| 9279 | unsigned NumElts) { |
| 9280 | |
| 9281 | llvm::VectorType *MaskTy = llvm::VectorType::get(CGF.Builder.getInt1Ty(), |
| 9282 | cast<IntegerType>(Mask->getType())->getBitWidth()); |
| 9283 | Value *MaskVec = CGF.Builder.CreateBitCast(Mask, MaskTy); |
| 9284 | |
| 9285 | // If we have less than 8 elements, then the starting mask was an i8 and |
| 9286 | // we need to extract down to the right number of elements. |
| 9287 | if (NumElts < 8) { |
| 9288 | uint32_t Indices[4]; |
| 9289 | for (unsigned i = 0; i != NumElts; ++i) |
| 9290 | Indices[i] = i; |
| 9291 | MaskVec = CGF.Builder.CreateShuffleVector(MaskVec, MaskVec, |
| 9292 | makeArrayRef(Indices, NumElts), |
| 9293 | "extract"); |
| 9294 | } |
| 9295 | return MaskVec; |
| 9296 | } |
| 9297 | |
| 9298 | static Value *EmitX86MaskedStore(CodeGenFunction &CGF, |
| 9299 | ArrayRef<Value *> Ops, |
| 9300 | unsigned Align) { |
| 9301 | // Cast the pointer to right type. |
| 9302 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], |
| 9303 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 9304 | |
| 9305 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], |
| 9306 | Ops[1]->getType()->getVectorNumElements()); |
| 9307 | |
| 9308 | return CGF.Builder.CreateMaskedStore(Ops[1], Ptr, Align, MaskVec); |
| 9309 | } |
| 9310 | |
| 9311 | static Value *EmitX86MaskedLoad(CodeGenFunction &CGF, |
| 9312 | ArrayRef<Value *> Ops, unsigned Align) { |
| 9313 | // Cast the pointer to right type. |
| 9314 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], |
| 9315 | llvm::PointerType::getUnqual(Ops[1]->getType())); |
| 9316 | |
| 9317 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], |
| 9318 | Ops[1]->getType()->getVectorNumElements()); |
| 9319 | |
| 9320 | return CGF.Builder.CreateMaskedLoad(Ptr, Align, MaskVec, Ops[1]); |
| 9321 | } |
| 9322 | |
| 9323 | static Value *EmitX86ExpandLoad(CodeGenFunction &CGF, |
| 9324 | ArrayRef<Value *> Ops) { |
| 9325 | llvm::Type *ResultTy = Ops[1]->getType(); |
| 9326 | llvm::Type *PtrTy = ResultTy->getVectorElementType(); |
| 9327 | |
| 9328 | // Cast the pointer to element type. |
| 9329 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], |
| 9330 | llvm::PointerType::getUnqual(PtrTy)); |
| 9331 | |
| 9332 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], |
| 9333 | ResultTy->getVectorNumElements()); |
| 9334 | |
| 9335 | llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_expandload, |
| 9336 | ResultTy); |
| 9337 | return CGF.Builder.CreateCall(F, { Ptr, MaskVec, Ops[1] }); |
| 9338 | } |
| 9339 | |
| 9340 | static Value *EmitX86CompressExpand(CodeGenFunction &CGF, |
| 9341 | ArrayRef<Value *> Ops, |
| 9342 | bool IsCompress) { |
| 9343 | llvm::Type *ResultTy = Ops[1]->getType(); |
| 9344 | |
| 9345 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], |
| 9346 | ResultTy->getVectorNumElements()); |
| 9347 | |
| 9348 | Intrinsic::ID IID = IsCompress ? Intrinsic::x86_avx512_mask_compress |
| 9349 | : Intrinsic::x86_avx512_mask_expand; |
| 9350 | llvm::Function *F = CGF.CGM.getIntrinsic(IID, ResultTy); |
| 9351 | return CGF.Builder.CreateCall(F, { Ops[0], Ops[1], MaskVec }); |
| 9352 | } |
| 9353 | |
| 9354 | static Value *EmitX86CompressStore(CodeGenFunction &CGF, |
| 9355 | ArrayRef<Value *> Ops) { |
| 9356 | llvm::Type *ResultTy = Ops[1]->getType(); |
| 9357 | llvm::Type *PtrTy = ResultTy->getVectorElementType(); |
| 9358 | |
| 9359 | // Cast the pointer to element type. |
| 9360 | Value *Ptr = CGF.Builder.CreateBitCast(Ops[0], |
| 9361 | llvm::PointerType::getUnqual(PtrTy)); |
| 9362 | |
| 9363 | Value *MaskVec = getMaskVecValue(CGF, Ops[2], |
| 9364 | ResultTy->getVectorNumElements()); |
| 9365 | |
| 9366 | llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_compressstore, |
| 9367 | ResultTy); |
| 9368 | return CGF.Builder.CreateCall(F, { Ops[1], Ptr, MaskVec }); |
| 9369 | } |
| 9370 | |
| 9371 | static Value *EmitX86MaskLogic(CodeGenFunction &CGF, Instruction::BinaryOps Opc, |
| 9372 | ArrayRef<Value *> Ops, |
| 9373 | bool InvertLHS = false) { |
| 9374 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 9375 | Value *LHS = getMaskVecValue(CGF, Ops[0], NumElts); |
| 9376 | Value *RHS = getMaskVecValue(CGF, Ops[1], NumElts); |
| 9377 | |
| 9378 | if (InvertLHS) |
| 9379 | LHS = CGF.Builder.CreateNot(LHS); |
| 9380 | |
| 9381 | return CGF.Builder.CreateBitCast(CGF.Builder.CreateBinOp(Opc, LHS, RHS), |
| 9382 | Ops[0]->getType()); |
| 9383 | } |
| 9384 | |
| 9385 | static Value *EmitX86FunnelShift(CodeGenFunction &CGF, Value *Op0, Value *Op1, |
| 9386 | Value *Amt, bool IsRight) { |
| 9387 | llvm::Type *Ty = Op0->getType(); |
| 9388 | |
| 9389 | // Amount may be scalar immediate, in which case create a splat vector. |
| 9390 | // Funnel shifts amounts are treated as modulo and types are all power-of-2 so |
| 9391 | // we only care about the lowest log2 bits anyway. |
| 9392 | if (Amt->getType() != Ty) { |
| 9393 | unsigned NumElts = Ty->getVectorNumElements(); |
| 9394 | Amt = CGF.Builder.CreateIntCast(Amt, Ty->getScalarType(), false); |
| 9395 | Amt = CGF.Builder.CreateVectorSplat(NumElts, Amt); |
| 9396 | } |
| 9397 | |
| 9398 | unsigned IID = IsRight ? Intrinsic::fshr : Intrinsic::fshl; |
| 9399 | Function *F = CGF.CGM.getIntrinsic(IID, Ty); |
| 9400 | return CGF.Builder.CreateCall(F, {Op0, Op1, Amt}); |
| 9401 | } |
| 9402 | |
| 9403 | static Value *EmitX86vpcom(CodeGenFunction &CGF, ArrayRef<Value *> Ops, |
| 9404 | bool IsSigned) { |
| 9405 | Value *Op0 = Ops[0]; |
| 9406 | Value *Op1 = Ops[1]; |
| 9407 | llvm::Type *Ty = Op0->getType(); |
| 9408 | uint64_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; |
| 9409 | |
| 9410 | CmpInst::Predicate Pred; |
| 9411 | switch (Imm) { |
| 9412 | case 0x0: |
| 9413 | Pred = IsSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; |
| 9414 | break; |
| 9415 | case 0x1: |
| 9416 | Pred = IsSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; |
| 9417 | break; |
| 9418 | case 0x2: |
| 9419 | Pred = IsSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; |
| 9420 | break; |
| 9421 | case 0x3: |
| 9422 | Pred = IsSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; |
| 9423 | break; |
| 9424 | case 0x4: |
| 9425 | Pred = ICmpInst::ICMP_EQ; |
| 9426 | break; |
| 9427 | case 0x5: |
| 9428 | Pred = ICmpInst::ICMP_NE; |
| 9429 | break; |
| 9430 | case 0x6: |
| 9431 | return llvm::Constant::getNullValue(Ty); // FALSE |
| 9432 | case 0x7: |
| 9433 | return llvm::Constant::getAllOnesValue(Ty); // TRUE |
| 9434 | default: |
| 9435 | llvm_unreachable("Unexpected XOP vpcom/vpcomu predicate")::llvm::llvm_unreachable_internal("Unexpected XOP vpcom/vpcomu predicate" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9435); |
| 9436 | } |
| 9437 | |
| 9438 | Value *Cmp = CGF.Builder.CreateICmp(Pred, Op0, Op1); |
| 9439 | Value *Res = CGF.Builder.CreateSExt(Cmp, Ty); |
| 9440 | return Res; |
| 9441 | } |
| 9442 | |
| 9443 | static Value *EmitX86Select(CodeGenFunction &CGF, |
| 9444 | Value *Mask, Value *Op0, Value *Op1) { |
| 9445 | |
| 9446 | // If the mask is all ones just return first argument. |
| 9447 | if (const auto *C = dyn_cast<Constant>(Mask)) |
| 9448 | if (C->isAllOnesValue()) |
| 9449 | return Op0; |
| 9450 | |
| 9451 | Mask = getMaskVecValue(CGF, Mask, Op0->getType()->getVectorNumElements()); |
| 9452 | |
| 9453 | return CGF.Builder.CreateSelect(Mask, Op0, Op1); |
| 9454 | } |
| 9455 | |
| 9456 | static Value *EmitX86ScalarSelect(CodeGenFunction &CGF, |
| 9457 | Value *Mask, Value *Op0, Value *Op1) { |
| 9458 | // If the mask is all ones just return first argument. |
| 9459 | if (const auto *C = dyn_cast<Constant>(Mask)) |
| 9460 | if (C->isAllOnesValue()) |
| 9461 | return Op0; |
| 9462 | |
| 9463 | llvm::VectorType *MaskTy = |
| 9464 | llvm::VectorType::get(CGF.Builder.getInt1Ty(), |
| 9465 | Mask->getType()->getIntegerBitWidth()); |
| 9466 | Mask = CGF.Builder.CreateBitCast(Mask, MaskTy); |
| 9467 | Mask = CGF.Builder.CreateExtractElement(Mask, (uint64_t)0); |
| 9468 | return CGF.Builder.CreateSelect(Mask, Op0, Op1); |
| 9469 | } |
| 9470 | |
| 9471 | static Value *EmitX86MaskedCompareResult(CodeGenFunction &CGF, Value *Cmp, |
| 9472 | unsigned NumElts, Value *MaskIn) { |
| 9473 | if (MaskIn) { |
| 9474 | const auto *C = dyn_cast<Constant>(MaskIn); |
| 9475 | if (!C || !C->isAllOnesValue()) |
| 9476 | Cmp = CGF.Builder.CreateAnd(Cmp, getMaskVecValue(CGF, MaskIn, NumElts)); |
| 9477 | } |
| 9478 | |
| 9479 | if (NumElts < 8) { |
| 9480 | uint32_t Indices[8]; |
| 9481 | for (unsigned i = 0; i != NumElts; ++i) |
| 9482 | Indices[i] = i; |
| 9483 | for (unsigned i = NumElts; i != 8; ++i) |
| 9484 | Indices[i] = i % NumElts + NumElts; |
| 9485 | Cmp = CGF.Builder.CreateShuffleVector( |
| 9486 | Cmp, llvm::Constant::getNullValue(Cmp->getType()), Indices); |
| 9487 | } |
| 9488 | |
| 9489 | return CGF.Builder.CreateBitCast(Cmp, |
| 9490 | IntegerType::get(CGF.getLLVMContext(), |
| 9491 | std::max(NumElts, 8U))); |
| 9492 | } |
| 9493 | |
| 9494 | static Value *EmitX86MaskedCompare(CodeGenFunction &CGF, unsigned CC, |
| 9495 | bool Signed, ArrayRef<Value *> Ops) { |
| 9496 | assert((Ops.size() == 2 || Ops.size() == 4) &&(((Ops.size() == 2 || Ops.size() == 4) && "Unexpected number of arguments" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9497, __PRETTY_FUNCTION__)) |
| 9497 | "Unexpected number of arguments")(((Ops.size() == 2 || Ops.size() == 4) && "Unexpected number of arguments" ) ? static_cast<void> (0) : __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9497, __PRETTY_FUNCTION__)); |
| 9498 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 9499 | Value *Cmp; |
| 9500 | |
| 9501 | if (CC == 3) { |
| 9502 | Cmp = Constant::getNullValue( |
| 9503 | llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts)); |
| 9504 | } else if (CC == 7) { |
| 9505 | Cmp = Constant::getAllOnesValue( |
| 9506 | llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts)); |
| 9507 | } else { |
| 9508 | ICmpInst::Predicate Pred; |
| 9509 | switch (CC) { |
| 9510 | default: llvm_unreachable("Unknown condition code")::llvm::llvm_unreachable_internal("Unknown condition code", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9510); |
| 9511 | case 0: Pred = ICmpInst::ICMP_EQ; break; |
| 9512 | case 1: Pred = Signed ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; break; |
| 9513 | case 2: Pred = Signed ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; break; |
| 9514 | case 4: Pred = ICmpInst::ICMP_NE; break; |
| 9515 | case 5: Pred = Signed ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; break; |
| 9516 | case 6: Pred = Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; break; |
| 9517 | } |
| 9518 | Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]); |
| 9519 | } |
| 9520 | |
| 9521 | Value *MaskIn = nullptr; |
| 9522 | if (Ops.size() == 4) |
| 9523 | MaskIn = Ops[3]; |
| 9524 | |
| 9525 | return EmitX86MaskedCompareResult(CGF, Cmp, NumElts, MaskIn); |
| 9526 | } |
| 9527 | |
| 9528 | static Value *EmitX86ConvertToMask(CodeGenFunction &CGF, Value *In) { |
| 9529 | Value *Zero = Constant::getNullValue(In->getType()); |
| 9530 | return EmitX86MaskedCompare(CGF, 1, true, { In, Zero }); |
| 9531 | } |
| 9532 | |
| 9533 | static Value *EmitX86ConvertIntToFp(CodeGenFunction &CGF, |
| 9534 | ArrayRef<Value *> Ops, bool IsSigned) { |
| 9535 | unsigned Rnd = cast<llvm::ConstantInt>(Ops[3])->getZExtValue(); |
| 9536 | llvm::Type *Ty = Ops[1]->getType(); |
| 9537 | |
| 9538 | Value *Res; |
| 9539 | if (Rnd != 4) { |
| 9540 | Intrinsic::ID IID = IsSigned ? Intrinsic::x86_avx512_sitofp_round |
| 9541 | : Intrinsic::x86_avx512_uitofp_round; |
| 9542 | Function *F = CGF.CGM.getIntrinsic(IID, { Ty, Ops[0]->getType() }); |
| 9543 | Res = CGF.Builder.CreateCall(F, { Ops[0], Ops[3] }); |
| 9544 | } else { |
| 9545 | Res = IsSigned ? CGF.Builder.CreateSIToFP(Ops[0], Ty) |
| 9546 | : CGF.Builder.CreateUIToFP(Ops[0], Ty); |
| 9547 | } |
| 9548 | |
| 9549 | return EmitX86Select(CGF, Ops[2], Res, Ops[1]); |
| 9550 | } |
| 9551 | |
| 9552 | static Value *EmitX86Abs(CodeGenFunction &CGF, ArrayRef<Value *> Ops) { |
| 9553 | |
| 9554 | llvm::Type *Ty = Ops[0]->getType(); |
| 9555 | Value *Zero = llvm::Constant::getNullValue(Ty); |
| 9556 | Value *Sub = CGF.Builder.CreateSub(Zero, Ops[0]); |
| 9557 | Value *Cmp = CGF.Builder.CreateICmp(ICmpInst::ICMP_SGT, Ops[0], Zero); |
| 9558 | Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Sub); |
| 9559 | return Res; |
| 9560 | } |
| 9561 | |
| 9562 | static Value *EmitX86MinMax(CodeGenFunction &CGF, ICmpInst::Predicate Pred, |
| 9563 | ArrayRef<Value *> Ops) { |
| 9564 | Value *Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]); |
| 9565 | Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Ops[1]); |
| 9566 | |
| 9567 | assert(Ops.size() == 2)((Ops.size() == 2) ? static_cast<void> (0) : __assert_fail ("Ops.size() == 2", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9567, __PRETTY_FUNCTION__)); |
| 9568 | return Res; |
| 9569 | } |
| 9570 | |
| 9571 | // Lowers X86 FMA intrinsics to IR. |
| 9572 | static Value *EmitX86FMAExpr(CodeGenFunction &CGF, ArrayRef<Value *> Ops, |
| 9573 | unsigned BuiltinID, bool IsAddSub) { |
| 9574 | |
| 9575 | bool Subtract = false; |
| 9576 | Intrinsic::ID IID = Intrinsic::not_intrinsic; |
| 9577 | switch (BuiltinID) { |
| 9578 | default: break; |
| 9579 | case clang::X86::BI__builtin_ia32_vfmsubps512_mask3: |
| 9580 | Subtract = true; |
| 9581 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 9582 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask: |
| 9583 | case clang::X86::BI__builtin_ia32_vfmaddps512_maskz: |
| 9584 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask3: |
| 9585 | IID = llvm::Intrinsic::x86_avx512_vfmadd_ps_512; break; |
| 9586 | case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3: |
| 9587 | Subtract = true; |
| 9588 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 9589 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask: |
| 9590 | case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz: |
| 9591 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3: |
| 9592 | IID = llvm::Intrinsic::x86_avx512_vfmadd_pd_512; break; |
| 9593 | case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3: |
| 9594 | Subtract = true; |
| 9595 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 9596 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask: |
| 9597 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz: |
| 9598 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3: |
| 9599 | IID = llvm::Intrinsic::x86_avx512_vfmaddsub_ps_512; |
| 9600 | break; |
| 9601 | case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3: |
| 9602 | Subtract = true; |
| 9603 | LLVM_FALLTHROUGH[[clang::fallthrough]]; |
| 9604 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask: |
| 9605 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz: |
| 9606 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3: |
| 9607 | IID = llvm::Intrinsic::x86_avx512_vfmaddsub_pd_512; |
| 9608 | break; |
| 9609 | } |
| 9610 | |
| 9611 | Value *A = Ops[0]; |
| 9612 | Value *B = Ops[1]; |
| 9613 | Value *C = Ops[2]; |
| 9614 | |
| 9615 | if (Subtract) |
| 9616 | C = CGF.Builder.CreateFNeg(C); |
| 9617 | |
| 9618 | Value *Res; |
| 9619 | |
| 9620 | // Only handle in case of _MM_FROUND_CUR_DIRECTION/4 (no rounding). |
| 9621 | if (IID != Intrinsic::not_intrinsic && |
| 9622 | cast<llvm::ConstantInt>(Ops.back())->getZExtValue() != (uint64_t)4) { |
| 9623 | Function *Intr = CGF.CGM.getIntrinsic(IID); |
| 9624 | Res = CGF.Builder.CreateCall(Intr, {A, B, C, Ops.back() }); |
| 9625 | } else { |
| 9626 | llvm::Type *Ty = A->getType(); |
| 9627 | Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ty); |
| 9628 | Res = CGF.Builder.CreateCall(FMA, {A, B, C} ); |
| 9629 | |
| 9630 | if (IsAddSub) { |
| 9631 | // Negate even elts in C using a mask. |
| 9632 | unsigned NumElts = Ty->getVectorNumElements(); |
| 9633 | SmallVector<uint32_t, 16> Indices(NumElts); |
| 9634 | for (unsigned i = 0; i != NumElts; ++i) |
| 9635 | Indices[i] = i + (i % 2) * NumElts; |
| 9636 | |
| 9637 | Value *NegC = CGF.Builder.CreateFNeg(C); |
| 9638 | Value *FMSub = CGF.Builder.CreateCall(FMA, {A, B, NegC} ); |
| 9639 | Res = CGF.Builder.CreateShuffleVector(FMSub, Res, Indices); |
| 9640 | } |
| 9641 | } |
| 9642 | |
| 9643 | // Handle any required masking. |
| 9644 | Value *MaskFalseVal = nullptr; |
| 9645 | switch (BuiltinID) { |
| 9646 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask: |
| 9647 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask: |
| 9648 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask: |
| 9649 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask: |
| 9650 | MaskFalseVal = Ops[0]; |
| 9651 | break; |
| 9652 | case clang::X86::BI__builtin_ia32_vfmaddps512_maskz: |
| 9653 | case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz: |
| 9654 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz: |
| 9655 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz: |
| 9656 | MaskFalseVal = Constant::getNullValue(Ops[0]->getType()); |
| 9657 | break; |
| 9658 | case clang::X86::BI__builtin_ia32_vfmsubps512_mask3: |
| 9659 | case clang::X86::BI__builtin_ia32_vfmaddps512_mask3: |
| 9660 | case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3: |
| 9661 | case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3: |
| 9662 | case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3: |
| 9663 | case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3: |
| 9664 | case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3: |
| 9665 | case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3: |
| 9666 | MaskFalseVal = Ops[2]; |
| 9667 | break; |
| 9668 | } |
| 9669 | |
| 9670 | if (MaskFalseVal) |
| 9671 | return EmitX86Select(CGF, Ops[3], Res, MaskFalseVal); |
| 9672 | |
| 9673 | return Res; |
| 9674 | } |
| 9675 | |
| 9676 | static Value * |
| 9677 | EmitScalarFMAExpr(CodeGenFunction &CGF, MutableArrayRef<Value *> Ops, |
| 9678 | Value *Upper, bool ZeroMask = false, unsigned PTIdx = 0, |
| 9679 | bool NegAcc = false) { |
| 9680 | unsigned Rnd = 4; |
| 9681 | if (Ops.size() > 4) |
| 9682 | Rnd = cast<llvm::ConstantInt>(Ops[4])->getZExtValue(); |
| 9683 | |
| 9684 | if (NegAcc) |
| 9685 | Ops[2] = CGF.Builder.CreateFNeg(Ops[2]); |
| 9686 | |
| 9687 | Ops[0] = CGF.Builder.CreateExtractElement(Ops[0], (uint64_t)0); |
| 9688 | Ops[1] = CGF.Builder.CreateExtractElement(Ops[1], (uint64_t)0); |
| 9689 | Ops[2] = CGF.Builder.CreateExtractElement(Ops[2], (uint64_t)0); |
| 9690 | Value *Res; |
| 9691 | if (Rnd != 4) { |
| 9692 | Intrinsic::ID IID = Ops[0]->getType()->getPrimitiveSizeInBits() == 32 ? |
| 9693 | Intrinsic::x86_avx512_vfmadd_f32 : |
| 9694 | Intrinsic::x86_avx512_vfmadd_f64; |
| 9695 | Res = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID), |
| 9696 | {Ops[0], Ops[1], Ops[2], Ops[4]}); |
| 9697 | } else { |
| 9698 | Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ops[0]->getType()); |
| 9699 | Res = CGF.Builder.CreateCall(FMA, Ops.slice(0, 3)); |
| 9700 | } |
| 9701 | // If we have more than 3 arguments, we need to do masking. |
| 9702 | if (Ops.size() > 3) { |
| 9703 | Value *PassThru = ZeroMask ? Constant::getNullValue(Res->getType()) |
| 9704 | : Ops[PTIdx]; |
| 9705 | |
| 9706 | // If we negated the accumulator and the its the PassThru value we need to |
| 9707 | // bypass the negate. Conveniently Upper should be the same thing in this |
| 9708 | // case. |
| 9709 | if (NegAcc && PTIdx == 2) |
| 9710 | PassThru = CGF.Builder.CreateExtractElement(Upper, (uint64_t)0); |
| 9711 | |
| 9712 | Res = EmitX86ScalarSelect(CGF, Ops[3], Res, PassThru); |
| 9713 | } |
| 9714 | return CGF.Builder.CreateInsertElement(Upper, Res, (uint64_t)0); |
| 9715 | } |
| 9716 | |
| 9717 | static Value *EmitX86Muldq(CodeGenFunction &CGF, bool IsSigned, |
| 9718 | ArrayRef<Value *> Ops) { |
| 9719 | llvm::Type *Ty = Ops[0]->getType(); |
| 9720 | // Arguments have a vXi32 type so cast to vXi64. |
| 9721 | Ty = llvm::VectorType::get(CGF.Int64Ty, |
| 9722 | Ty->getPrimitiveSizeInBits() / 64); |
| 9723 | Value *LHS = CGF.Builder.CreateBitCast(Ops[0], Ty); |
| 9724 | Value *RHS = CGF.Builder.CreateBitCast(Ops[1], Ty); |
| 9725 | |
| 9726 | if (IsSigned) { |
| 9727 | // Shift left then arithmetic shift right. |
| 9728 | Constant *ShiftAmt = ConstantInt::get(Ty, 32); |
| 9729 | LHS = CGF.Builder.CreateShl(LHS, ShiftAmt); |
| 9730 | LHS = CGF.Builder.CreateAShr(LHS, ShiftAmt); |
| 9731 | RHS = CGF.Builder.CreateShl(RHS, ShiftAmt); |
| 9732 | RHS = CGF.Builder.CreateAShr(RHS, ShiftAmt); |
| 9733 | } else { |
| 9734 | // Clear the upper bits. |
| 9735 | Constant *Mask = ConstantInt::get(Ty, 0xffffffff); |
| 9736 | LHS = CGF.Builder.CreateAnd(LHS, Mask); |
| 9737 | RHS = CGF.Builder.CreateAnd(RHS, Mask); |
| 9738 | } |
| 9739 | |
| 9740 | return CGF.Builder.CreateMul(LHS, RHS); |
| 9741 | } |
| 9742 | |
| 9743 | // Emit a masked pternlog intrinsic. This only exists because the header has to |
| 9744 | // use a macro and we aren't able to pass the input argument to a pternlog |
| 9745 | // builtin and a select builtin without evaluating it twice. |
| 9746 | static Value *EmitX86Ternlog(CodeGenFunction &CGF, bool ZeroMask, |
| 9747 | ArrayRef<Value *> Ops) { |
| 9748 | llvm::Type *Ty = Ops[0]->getType(); |
| 9749 | |
| 9750 | unsigned VecWidth = Ty->getPrimitiveSizeInBits(); |
| 9751 | unsigned EltWidth = Ty->getScalarSizeInBits(); |
| 9752 | Intrinsic::ID IID; |
| 9753 | if (VecWidth == 128 && EltWidth == 32) |
| 9754 | IID = Intrinsic::x86_avx512_pternlog_d_128; |
| 9755 | else if (VecWidth == 256 && EltWidth == 32) |
| 9756 | IID = Intrinsic::x86_avx512_pternlog_d_256; |
| 9757 | else if (VecWidth == 512 && EltWidth == 32) |
| 9758 | IID = Intrinsic::x86_avx512_pternlog_d_512; |
| 9759 | else if (VecWidth == 128 && EltWidth == 64) |
| 9760 | IID = Intrinsic::x86_avx512_pternlog_q_128; |
| 9761 | else if (VecWidth == 256 && EltWidth == 64) |
| 9762 | IID = Intrinsic::x86_avx512_pternlog_q_256; |
| 9763 | else if (VecWidth == 512 && EltWidth == 64) |
| 9764 | IID = Intrinsic::x86_avx512_pternlog_q_512; |
| 9765 | else |
| 9766 | llvm_unreachable("Unexpected intrinsic")::llvm::llvm_unreachable_internal("Unexpected intrinsic", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9766); |
| 9767 | |
| 9768 | Value *Ternlog = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID), |
| 9769 | Ops.drop_back()); |
| 9770 | Value *PassThru = ZeroMask ? ConstantAggregateZero::get(Ty) : Ops[0]; |
| 9771 | return EmitX86Select(CGF, Ops[4], Ternlog, PassThru); |
| 9772 | } |
| 9773 | |
| 9774 | static Value *EmitX86SExtMask(CodeGenFunction &CGF, Value *Op, |
| 9775 | llvm::Type *DstTy) { |
| 9776 | unsigned NumberOfElements = DstTy->getVectorNumElements(); |
| 9777 | Value *Mask = getMaskVecValue(CGF, Op, NumberOfElements); |
| 9778 | return CGF.Builder.CreateSExt(Mask, DstTy, "vpmovm2"); |
| 9779 | } |
| 9780 | |
| 9781 | // Emit addition or subtraction with signed/unsigned saturation. |
| 9782 | static Value *EmitX86AddSubSatExpr(CodeGenFunction &CGF, |
| 9783 | ArrayRef<Value *> Ops, bool IsSigned, |
| 9784 | bool IsAddition) { |
| 9785 | Intrinsic::ID IID = |
| 9786 | IsSigned ? (IsAddition ? Intrinsic::sadd_sat : Intrinsic::ssub_sat) |
| 9787 | : (IsAddition ? Intrinsic::uadd_sat : Intrinsic::usub_sat); |
| 9788 | llvm::Function *F = CGF.CGM.getIntrinsic(IID, Ops[0]->getType()); |
| 9789 | return CGF.Builder.CreateCall(F, {Ops[0], Ops[1]}); |
| 9790 | } |
| 9791 | |
| 9792 | Value *CodeGenFunction::EmitX86CpuIs(const CallExpr *E) { |
| 9793 | const Expr *CPUExpr = E->getArg(0)->IgnoreParenCasts(); |
| 9794 | StringRef CPUStr = cast<clang::StringLiteral>(CPUExpr)->getString(); |
| 9795 | return EmitX86CpuIs(CPUStr); |
| 9796 | } |
| 9797 | |
| 9798 | Value *CodeGenFunction::EmitX86CpuIs(StringRef CPUStr) { |
| 9799 | |
| 9800 | llvm::Type *Int32Ty = Builder.getInt32Ty(); |
| 9801 | |
| 9802 | // Matching the struct layout from the compiler-rt/libgcc structure that is |
| 9803 | // filled in: |
| 9804 | // unsigned int __cpu_vendor; |
| 9805 | // unsigned int __cpu_type; |
| 9806 | // unsigned int __cpu_subtype; |
| 9807 | // unsigned int __cpu_features[1]; |
| 9808 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty, |
| 9809 | llvm::ArrayType::get(Int32Ty, 1)); |
| 9810 | |
| 9811 | // Grab the global __cpu_model. |
| 9812 | llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model"); |
| 9813 | cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true); |
| 9814 | |
| 9815 | // Calculate the index needed to access the correct field based on the |
| 9816 | // range. Also adjust the expected value. |
| 9817 | unsigned Index; |
| 9818 | unsigned Value; |
| 9819 | std::tie(Index, Value) = StringSwitch<std::pair<unsigned, unsigned>>(CPUStr) |
| 9820 | #define X86_VENDOR(ENUM, STRING) \ |
| 9821 | .Case(STRING, {0u, static_cast<unsigned>(llvm::X86::ENUM)}) |
| 9822 | #define X86_CPU_TYPE_COMPAT_WITH_ALIAS(ARCHNAME, ENUM, STR, ALIAS) \ |
| 9823 | .Cases(STR, ALIAS, {1u, static_cast<unsigned>(llvm::X86::ENUM)}) |
| 9824 | #define X86_CPU_TYPE_COMPAT(ARCHNAME, ENUM, STR) \ |
| 9825 | .Case(STR, {1u, static_cast<unsigned>(llvm::X86::ENUM)}) |
| 9826 | #define X86_CPU_SUBTYPE_COMPAT(ARCHNAME, ENUM, STR) \ |
| 9827 | .Case(STR, {2u, static_cast<unsigned>(llvm::X86::ENUM)}) |
| 9828 | #include "llvm/Support/X86TargetParser.def" |
| 9829 | .Default({0, 0}); |
| 9830 | assert(Value != 0 && "Invalid CPUStr passed to CpuIs")((Value != 0 && "Invalid CPUStr passed to CpuIs") ? static_cast <void> (0) : __assert_fail ("Value != 0 && \"Invalid CPUStr passed to CpuIs\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9830, __PRETTY_FUNCTION__)); |
| 9831 | |
| 9832 | // Grab the appropriate field from __cpu_model. |
| 9833 | llvm::Value *Idxs[] = {ConstantInt::get(Int32Ty, 0), |
| 9834 | ConstantInt::get(Int32Ty, Index)}; |
| 9835 | llvm::Value *CpuValue = Builder.CreateGEP(STy, CpuModel, Idxs); |
| 9836 | CpuValue = Builder.CreateAlignedLoad(CpuValue, CharUnits::fromQuantity(4)); |
| 9837 | |
| 9838 | // Check the value of the field against the requested value. |
| 9839 | return Builder.CreateICmpEQ(CpuValue, |
| 9840 | llvm::ConstantInt::get(Int32Ty, Value)); |
| 9841 | } |
| 9842 | |
| 9843 | Value *CodeGenFunction::EmitX86CpuSupports(const CallExpr *E) { |
| 9844 | const Expr *FeatureExpr = E->getArg(0)->IgnoreParenCasts(); |
| 9845 | StringRef FeatureStr = cast<StringLiteral>(FeatureExpr)->getString(); |
| 9846 | return EmitX86CpuSupports(FeatureStr); |
| 9847 | } |
| 9848 | |
| 9849 | uint64_t |
| 9850 | CodeGenFunction::GetX86CpuSupportsMask(ArrayRef<StringRef> FeatureStrs) { |
| 9851 | // Processor features and mapping to processor feature value. |
| 9852 | uint64_t FeaturesMask = 0; |
| 9853 | for (const StringRef &FeatureStr : FeatureStrs) { |
| 9854 | unsigned Feature = |
| 9855 | StringSwitch<unsigned>(FeatureStr) |
| 9856 | #define X86_FEATURE_COMPAT(VAL, ENUM, STR) .Case(STR, VAL) |
| 9857 | #include "llvm/Support/X86TargetParser.def" |
| 9858 | ; |
| 9859 | FeaturesMask |= (1ULL << Feature); |
| 9860 | } |
| 9861 | return FeaturesMask; |
| 9862 | } |
| 9863 | |
| 9864 | Value *CodeGenFunction::EmitX86CpuSupports(ArrayRef<StringRef> FeatureStrs) { |
| 9865 | return EmitX86CpuSupports(GetX86CpuSupportsMask(FeatureStrs)); |
| 9866 | } |
| 9867 | |
| 9868 | llvm::Value *CodeGenFunction::EmitX86CpuSupports(uint64_t FeaturesMask) { |
| 9869 | uint32_t Features1 = Lo_32(FeaturesMask); |
| 9870 | uint32_t Features2 = Hi_32(FeaturesMask); |
| 9871 | |
| 9872 | Value *Result = Builder.getTrue(); |
| 9873 | |
| 9874 | if (Features1 != 0) { |
| 9875 | // Matching the struct layout from the compiler-rt/libgcc structure that is |
| 9876 | // filled in: |
| 9877 | // unsigned int __cpu_vendor; |
| 9878 | // unsigned int __cpu_type; |
| 9879 | // unsigned int __cpu_subtype; |
| 9880 | // unsigned int __cpu_features[1]; |
| 9881 | llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty, |
| 9882 | llvm::ArrayType::get(Int32Ty, 1)); |
| 9883 | |
| 9884 | // Grab the global __cpu_model. |
| 9885 | llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model"); |
| 9886 | cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true); |
| 9887 | |
| 9888 | // Grab the first (0th) element from the field __cpu_features off of the |
| 9889 | // global in the struct STy. |
| 9890 | Value *Idxs[] = {Builder.getInt32(0), Builder.getInt32(3), |
| 9891 | Builder.getInt32(0)}; |
| 9892 | Value *CpuFeatures = Builder.CreateGEP(STy, CpuModel, Idxs); |
| 9893 | Value *Features = |
| 9894 | Builder.CreateAlignedLoad(CpuFeatures, CharUnits::fromQuantity(4)); |
| 9895 | |
| 9896 | // Check the value of the bit corresponding to the feature requested. |
| 9897 | Value *Mask = Builder.getInt32(Features1); |
| 9898 | Value *Bitset = Builder.CreateAnd(Features, Mask); |
| 9899 | Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask); |
| 9900 | Result = Builder.CreateAnd(Result, Cmp); |
| 9901 | } |
| 9902 | |
| 9903 | if (Features2 != 0) { |
| 9904 | llvm::Constant *CpuFeatures2 = CGM.CreateRuntimeVariable(Int32Ty, |
| 9905 | "__cpu_features2"); |
| 9906 | cast<llvm::GlobalValue>(CpuFeatures2)->setDSOLocal(true); |
| 9907 | |
| 9908 | Value *Features = |
| 9909 | Builder.CreateAlignedLoad(CpuFeatures2, CharUnits::fromQuantity(4)); |
| 9910 | |
| 9911 | // Check the value of the bit corresponding to the feature requested. |
| 9912 | Value *Mask = Builder.getInt32(Features2); |
| 9913 | Value *Bitset = Builder.CreateAnd(Features, Mask); |
| 9914 | Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask); |
| 9915 | Result = Builder.CreateAnd(Result, Cmp); |
| 9916 | } |
| 9917 | |
| 9918 | return Result; |
| 9919 | } |
| 9920 | |
| 9921 | Value *CodeGenFunction::EmitX86CpuInit() { |
| 9922 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, |
| 9923 | /*Variadic*/ false); |
| 9924 | llvm::FunctionCallee Func = |
| 9925 | CGM.CreateRuntimeFunction(FTy, "__cpu_indicator_init"); |
| 9926 | cast<llvm::GlobalValue>(Func.getCallee())->setDSOLocal(true); |
| 9927 | cast<llvm::GlobalValue>(Func.getCallee()) |
| 9928 | ->setDLLStorageClass(llvm::GlobalValue::DefaultStorageClass); |
| 9929 | return Builder.CreateCall(Func); |
| 9930 | } |
| 9931 | |
| 9932 | Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID, |
| 9933 | const CallExpr *E) { |
| 9934 | if (BuiltinID == X86::BI__builtin_cpu_is) |
| 9935 | return EmitX86CpuIs(E); |
| 9936 | if (BuiltinID == X86::BI__builtin_cpu_supports) |
| 9937 | return EmitX86CpuSupports(E); |
| 9938 | if (BuiltinID == X86::BI__builtin_cpu_init) |
| 9939 | return EmitX86CpuInit(); |
| 9940 | |
| 9941 | SmallVector<Value*, 4> Ops; |
| 9942 | |
| 9943 | // Find out if any arguments are required to be integer constant expressions. |
| 9944 | unsigned ICEArguments = 0; |
| 9945 | ASTContext::GetBuiltinTypeError Error; |
| 9946 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); |
| 9947 | assert(Error == ASTContext::GE_None && "Should not codegen an error")((Error == ASTContext::GE_None && "Should not codegen an error" ) ? static_cast<void> (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9947, __PRETTY_FUNCTION__)); |
| 9948 | |
| 9949 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { |
| 9950 | // If this is a normal argument, just emit it as a scalar. |
| 9951 | if ((ICEArguments & (1 << i)) == 0) { |
| 9952 | Ops.push_back(EmitScalarExpr(E->getArg(i))); |
| 9953 | continue; |
| 9954 | } |
| 9955 | |
| 9956 | // If this is required to be a constant, constant fold it so that we know |
| 9957 | // that the generated intrinsic gets a ConstantInt. |
| 9958 | llvm::APSInt Result; |
| 9959 | bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext()); |
| 9960 | assert(IsConst && "Constant arg isn't actually constant?")((IsConst && "Constant arg isn't actually constant?") ? static_cast<void> (0) : __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 9960, __PRETTY_FUNCTION__)); (void)IsConst; |
| 9961 | Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result)); |
| 9962 | } |
| 9963 | |
| 9964 | // These exist so that the builtin that takes an immediate can be bounds |
| 9965 | // checked by clang to avoid passing bad immediates to the backend. Since |
| 9966 | // AVX has a larger immediate than SSE we would need separate builtins to |
| 9967 | // do the different bounds checking. Rather than create a clang specific |
| 9968 | // SSE only builtin, this implements eight separate builtins to match gcc |
| 9969 | // implementation. |
| 9970 | auto getCmpIntrinsicCall = [this, &Ops](Intrinsic::ID ID, unsigned Imm) { |
| 9971 | Ops.push_back(llvm::ConstantInt::get(Int8Ty, Imm)); |
| 9972 | llvm::Function *F = CGM.getIntrinsic(ID); |
| 9973 | return Builder.CreateCall(F, Ops); |
| 9974 | }; |
| 9975 | |
| 9976 | // For the vector forms of FP comparisons, translate the builtins directly to |
| 9977 | // IR. |
| 9978 | // TODO: The builtins could be removed if the SSE header files used vector |
| 9979 | // extension comparisons directly (vector ordered/unordered may need |
| 9980 | // additional support via __builtin_isnan()). |
| 9981 | auto getVectorFCmpIR = [this, &Ops](CmpInst::Predicate Pred) { |
| 9982 | Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]); |
| 9983 | llvm::VectorType *FPVecTy = cast<llvm::VectorType>(Ops[0]->getType()); |
| 9984 | llvm::VectorType *IntVecTy = llvm::VectorType::getInteger(FPVecTy); |
| 9985 | Value *Sext = Builder.CreateSExt(Cmp, IntVecTy); |
| 9986 | return Builder.CreateBitCast(Sext, FPVecTy); |
| 9987 | }; |
| 9988 | |
| 9989 | switch (BuiltinID) { |
| 9990 | default: return nullptr; |
| 9991 | case X86::BI_mm_prefetch: { |
| 9992 | Value *Address = Ops[0]; |
| 9993 | ConstantInt *C = cast<ConstantInt>(Ops[1]); |
| 9994 | Value *RW = ConstantInt::get(Int32Ty, (C->getZExtValue() >> 2) & 0x1); |
| 9995 | Value *Locality = ConstantInt::get(Int32Ty, C->getZExtValue() & 0x3); |
| 9996 | Value *Data = ConstantInt::get(Int32Ty, 1); |
| 9997 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch); |
| 9998 | return Builder.CreateCall(F, {Address, RW, Locality, Data}); |
| 9999 | } |
| 10000 | case X86::BI_mm_clflush: { |
| 10001 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_clflush), |
| 10002 | Ops[0]); |
| 10003 | } |
| 10004 | case X86::BI_mm_lfence: { |
| 10005 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_lfence)); |
| 10006 | } |
| 10007 | case X86::BI_mm_mfence: { |
| 10008 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_mfence)); |
| 10009 | } |
| 10010 | case X86::BI_mm_sfence: { |
| 10011 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_sfence)); |
| 10012 | } |
| 10013 | case X86::BI_mm_pause: { |
| 10014 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_pause)); |
| 10015 | } |
| 10016 | case X86::BI__rdtsc: { |
| 10017 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtsc)); |
| 10018 | } |
| 10019 | case X86::BI__builtin_ia32_rdtscp: { |
| 10020 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtscp)); |
| 10021 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1), |
| 10022 | Ops[0]); |
| 10023 | return Builder.CreateExtractValue(Call, 0); |
| 10024 | } |
| 10025 | case X86::BI__builtin_ia32_lzcnt_u16: |
| 10026 | case X86::BI__builtin_ia32_lzcnt_u32: |
| 10027 | case X86::BI__builtin_ia32_lzcnt_u64: { |
| 10028 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType()); |
| 10029 | return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)}); |
| 10030 | } |
| 10031 | case X86::BI__builtin_ia32_tzcnt_u16: |
| 10032 | case X86::BI__builtin_ia32_tzcnt_u32: |
| 10033 | case X86::BI__builtin_ia32_tzcnt_u64: { |
| 10034 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, Ops[0]->getType()); |
| 10035 | return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)}); |
| 10036 | } |
| 10037 | case X86::BI__builtin_ia32_undef128: |
| 10038 | case X86::BI__builtin_ia32_undef256: |
| 10039 | case X86::BI__builtin_ia32_undef512: |
| 10040 | // The x86 definition of "undef" is not the same as the LLVM definition |
| 10041 | // (PR32176). We leave optimizing away an unnecessary zero constant to the |
| 10042 | // IR optimizer and backend. |
| 10043 | // TODO: If we had a "freeze" IR instruction to generate a fixed undef |
| 10044 | // value, we should use that here instead of a zero. |
| 10045 | return llvm::Constant::getNullValue(ConvertType(E->getType())); |
| 10046 | case X86::BI__builtin_ia32_vec_init_v8qi: |
| 10047 | case X86::BI__builtin_ia32_vec_init_v4hi: |
| 10048 | case X86::BI__builtin_ia32_vec_init_v2si: |
| 10049 | return Builder.CreateBitCast(BuildVector(Ops), |
| 10050 | llvm::Type::getX86_MMXTy(getLLVMContext())); |
| 10051 | case X86::BI__builtin_ia32_vec_ext_v2si: |
| 10052 | case X86::BI__builtin_ia32_vec_ext_v16qi: |
| 10053 | case X86::BI__builtin_ia32_vec_ext_v8hi: |
| 10054 | case X86::BI__builtin_ia32_vec_ext_v4si: |
| 10055 | case X86::BI__builtin_ia32_vec_ext_v4sf: |
| 10056 | case X86::BI__builtin_ia32_vec_ext_v2di: |
| 10057 | case X86::BI__builtin_ia32_vec_ext_v32qi: |
| 10058 | case X86::BI__builtin_ia32_vec_ext_v16hi: |
| 10059 | case X86::BI__builtin_ia32_vec_ext_v8si: |
| 10060 | case X86::BI__builtin_ia32_vec_ext_v4di: { |
| 10061 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10062 | uint64_t Index = cast<ConstantInt>(Ops[1])->getZExtValue(); |
| 10063 | Index &= NumElts - 1; |
| 10064 | // These builtins exist so we can ensure the index is an ICE and in range. |
| 10065 | // Otherwise we could just do this in the header file. |
| 10066 | return Builder.CreateExtractElement(Ops[0], Index); |
| 10067 | } |
| 10068 | case X86::BI__builtin_ia32_vec_set_v16qi: |
| 10069 | case X86::BI__builtin_ia32_vec_set_v8hi: |
| 10070 | case X86::BI__builtin_ia32_vec_set_v4si: |
| 10071 | case X86::BI__builtin_ia32_vec_set_v2di: |
| 10072 | case X86::BI__builtin_ia32_vec_set_v32qi: |
| 10073 | case X86::BI__builtin_ia32_vec_set_v16hi: |
| 10074 | case X86::BI__builtin_ia32_vec_set_v8si: |
| 10075 | case X86::BI__builtin_ia32_vec_set_v4di: { |
| 10076 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10077 | unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue(); |
| 10078 | Index &= NumElts - 1; |
| 10079 | // These builtins exist so we can ensure the index is an ICE and in range. |
| 10080 | // Otherwise we could just do this in the header file. |
| 10081 | return Builder.CreateInsertElement(Ops[0], Ops[1], Index); |
| 10082 | } |
| 10083 | case X86::BI_mm_setcsr: |
| 10084 | case X86::BI__builtin_ia32_ldmxcsr: { |
| 10085 | Address Tmp = CreateMemTemp(E->getArg(0)->getType()); |
| 10086 | Builder.CreateStore(Ops[0], Tmp); |
| 10087 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_ldmxcsr), |
| 10088 | Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy)); |
| 10089 | } |
| 10090 | case X86::BI_mm_getcsr: |
| 10091 | case X86::BI__builtin_ia32_stmxcsr: { |
| 10092 | Address Tmp = CreateMemTemp(E->getType()); |
| 10093 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_stmxcsr), |
| 10094 | Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy)); |
| 10095 | return Builder.CreateLoad(Tmp, "stmxcsr"); |
| 10096 | } |
| 10097 | case X86::BI__builtin_ia32_xsave: |
| 10098 | case X86::BI__builtin_ia32_xsave64: |
| 10099 | case X86::BI__builtin_ia32_xrstor: |
| 10100 | case X86::BI__builtin_ia32_xrstor64: |
| 10101 | case X86::BI__builtin_ia32_xsaveopt: |
| 10102 | case X86::BI__builtin_ia32_xsaveopt64: |
| 10103 | case X86::BI__builtin_ia32_xrstors: |
| 10104 | case X86::BI__builtin_ia32_xrstors64: |
| 10105 | case X86::BI__builtin_ia32_xsavec: |
| 10106 | case X86::BI__builtin_ia32_xsavec64: |
| 10107 | case X86::BI__builtin_ia32_xsaves: |
| 10108 | case X86::BI__builtin_ia32_xsaves64: |
| 10109 | case X86::BI__builtin_ia32_xsetbv: |
| 10110 | case X86::BI_xsetbv: { |
| 10111 | Intrinsic::ID ID; |
| 10112 | #define INTRINSIC_X86_XSAVE_ID(NAME) \ |
| 10113 | case X86::BI__builtin_ia32_##NAME: \ |
| 10114 | ID = Intrinsic::x86_##NAME; \ |
| 10115 | break |
| 10116 | switch (BuiltinID) { |
| 10117 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10117); |
| 10118 | INTRINSIC_X86_XSAVE_ID(xsave); |
| 10119 | INTRINSIC_X86_XSAVE_ID(xsave64); |
| 10120 | INTRINSIC_X86_XSAVE_ID(xrstor); |
| 10121 | INTRINSIC_X86_XSAVE_ID(xrstor64); |
| 10122 | INTRINSIC_X86_XSAVE_ID(xsaveopt); |
| 10123 | INTRINSIC_X86_XSAVE_ID(xsaveopt64); |
| 10124 | INTRINSIC_X86_XSAVE_ID(xrstors); |
| 10125 | INTRINSIC_X86_XSAVE_ID(xrstors64); |
| 10126 | INTRINSIC_X86_XSAVE_ID(xsavec); |
| 10127 | INTRINSIC_X86_XSAVE_ID(xsavec64); |
| 10128 | INTRINSIC_X86_XSAVE_ID(xsaves); |
| 10129 | INTRINSIC_X86_XSAVE_ID(xsaves64); |
| 10130 | INTRINSIC_X86_XSAVE_ID(xsetbv); |
| 10131 | case X86::BI_xsetbv: |
| 10132 | ID = Intrinsic::x86_xsetbv; |
| 10133 | break; |
| 10134 | } |
| 10135 | #undef INTRINSIC_X86_XSAVE_ID |
| 10136 | Value *Mhi = Builder.CreateTrunc( |
| 10137 | Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, 32)), Int32Ty); |
| 10138 | Value *Mlo = Builder.CreateTrunc(Ops[1], Int32Ty); |
| 10139 | Ops[1] = Mhi; |
| 10140 | Ops.push_back(Mlo); |
| 10141 | return Builder.CreateCall(CGM.getIntrinsic(ID), Ops); |
| 10142 | } |
| 10143 | case X86::BI__builtin_ia32_xgetbv: |
| 10144 | case X86::BI_xgetbv: |
| 10145 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_xgetbv), Ops); |
| 10146 | case X86::BI__builtin_ia32_storedqudi128_mask: |
| 10147 | case X86::BI__builtin_ia32_storedqusi128_mask: |
| 10148 | case X86::BI__builtin_ia32_storedquhi128_mask: |
| 10149 | case X86::BI__builtin_ia32_storedquqi128_mask: |
| 10150 | case X86::BI__builtin_ia32_storeupd128_mask: |
| 10151 | case X86::BI__builtin_ia32_storeups128_mask: |
| 10152 | case X86::BI__builtin_ia32_storedqudi256_mask: |
| 10153 | case X86::BI__builtin_ia32_storedqusi256_mask: |
| 10154 | case X86::BI__builtin_ia32_storedquhi256_mask: |
| 10155 | case X86::BI__builtin_ia32_storedquqi256_mask: |
| 10156 | case X86::BI__builtin_ia32_storeupd256_mask: |
| 10157 | case X86::BI__builtin_ia32_storeups256_mask: |
| 10158 | case X86::BI__builtin_ia32_storedqudi512_mask: |
| 10159 | case X86::BI__builtin_ia32_storedqusi512_mask: |
| 10160 | case X86::BI__builtin_ia32_storedquhi512_mask: |
| 10161 | case X86::BI__builtin_ia32_storedquqi512_mask: |
| 10162 | case X86::BI__builtin_ia32_storeupd512_mask: |
| 10163 | case X86::BI__builtin_ia32_storeups512_mask: |
| 10164 | return EmitX86MaskedStore(*this, Ops, 1); |
| 10165 | |
| 10166 | case X86::BI__builtin_ia32_storess128_mask: |
| 10167 | case X86::BI__builtin_ia32_storesd128_mask: { |
| 10168 | return EmitX86MaskedStore(*this, Ops, 1); |
| 10169 | } |
| 10170 | case X86::BI__builtin_ia32_vpopcntb_128: |
| 10171 | case X86::BI__builtin_ia32_vpopcntd_128: |
| 10172 | case X86::BI__builtin_ia32_vpopcntq_128: |
| 10173 | case X86::BI__builtin_ia32_vpopcntw_128: |
| 10174 | case X86::BI__builtin_ia32_vpopcntb_256: |
| 10175 | case X86::BI__builtin_ia32_vpopcntd_256: |
| 10176 | case X86::BI__builtin_ia32_vpopcntq_256: |
| 10177 | case X86::BI__builtin_ia32_vpopcntw_256: |
| 10178 | case X86::BI__builtin_ia32_vpopcntb_512: |
| 10179 | case X86::BI__builtin_ia32_vpopcntd_512: |
| 10180 | case X86::BI__builtin_ia32_vpopcntq_512: |
| 10181 | case X86::BI__builtin_ia32_vpopcntw_512: { |
| 10182 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 10183 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); |
| 10184 | return Builder.CreateCall(F, Ops); |
| 10185 | } |
| 10186 | case X86::BI__builtin_ia32_cvtmask2b128: |
| 10187 | case X86::BI__builtin_ia32_cvtmask2b256: |
| 10188 | case X86::BI__builtin_ia32_cvtmask2b512: |
| 10189 | case X86::BI__builtin_ia32_cvtmask2w128: |
| 10190 | case X86::BI__builtin_ia32_cvtmask2w256: |
| 10191 | case X86::BI__builtin_ia32_cvtmask2w512: |
| 10192 | case X86::BI__builtin_ia32_cvtmask2d128: |
| 10193 | case X86::BI__builtin_ia32_cvtmask2d256: |
| 10194 | case X86::BI__builtin_ia32_cvtmask2d512: |
| 10195 | case X86::BI__builtin_ia32_cvtmask2q128: |
| 10196 | case X86::BI__builtin_ia32_cvtmask2q256: |
| 10197 | case X86::BI__builtin_ia32_cvtmask2q512: |
| 10198 | return EmitX86SExtMask(*this, Ops[0], ConvertType(E->getType())); |
| 10199 | |
| 10200 | case X86::BI__builtin_ia32_cvtb2mask128: |
| 10201 | case X86::BI__builtin_ia32_cvtb2mask256: |
| 10202 | case X86::BI__builtin_ia32_cvtb2mask512: |
| 10203 | case X86::BI__builtin_ia32_cvtw2mask128: |
| 10204 | case X86::BI__builtin_ia32_cvtw2mask256: |
| 10205 | case X86::BI__builtin_ia32_cvtw2mask512: |
| 10206 | case X86::BI__builtin_ia32_cvtd2mask128: |
| 10207 | case X86::BI__builtin_ia32_cvtd2mask256: |
| 10208 | case X86::BI__builtin_ia32_cvtd2mask512: |
| 10209 | case X86::BI__builtin_ia32_cvtq2mask128: |
| 10210 | case X86::BI__builtin_ia32_cvtq2mask256: |
| 10211 | case X86::BI__builtin_ia32_cvtq2mask512: |
| 10212 | return EmitX86ConvertToMask(*this, Ops[0]); |
| 10213 | |
| 10214 | case X86::BI__builtin_ia32_cvtdq2ps512_mask: |
| 10215 | case X86::BI__builtin_ia32_cvtqq2ps512_mask: |
| 10216 | case X86::BI__builtin_ia32_cvtqq2pd512_mask: |
| 10217 | return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/true); |
| 10218 | case X86::BI__builtin_ia32_cvtudq2ps512_mask: |
| 10219 | case X86::BI__builtin_ia32_cvtuqq2ps512_mask: |
| 10220 | case X86::BI__builtin_ia32_cvtuqq2pd512_mask: |
| 10221 | return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/false); |
| 10222 | |
| 10223 | case X86::BI__builtin_ia32_vfmaddss3: |
| 10224 | case X86::BI__builtin_ia32_vfmaddsd3: |
| 10225 | case X86::BI__builtin_ia32_vfmaddss3_mask: |
| 10226 | case X86::BI__builtin_ia32_vfmaddsd3_mask: |
| 10227 | return EmitScalarFMAExpr(*this, Ops, Ops[0]); |
| 10228 | case X86::BI__builtin_ia32_vfmaddss: |
| 10229 | case X86::BI__builtin_ia32_vfmaddsd: |
| 10230 | return EmitScalarFMAExpr(*this, Ops, |
| 10231 | Constant::getNullValue(Ops[0]->getType())); |
| 10232 | case X86::BI__builtin_ia32_vfmaddss3_maskz: |
| 10233 | case X86::BI__builtin_ia32_vfmaddsd3_maskz: |
| 10234 | return EmitScalarFMAExpr(*this, Ops, Ops[0], /*ZeroMask*/true); |
| 10235 | case X86::BI__builtin_ia32_vfmaddss3_mask3: |
| 10236 | case X86::BI__builtin_ia32_vfmaddsd3_mask3: |
| 10237 | return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2); |
| 10238 | case X86::BI__builtin_ia32_vfmsubss3_mask3: |
| 10239 | case X86::BI__builtin_ia32_vfmsubsd3_mask3: |
| 10240 | return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2, |
| 10241 | /*NegAcc*/true); |
| 10242 | case X86::BI__builtin_ia32_vfmaddps: |
| 10243 | case X86::BI__builtin_ia32_vfmaddpd: |
| 10244 | case X86::BI__builtin_ia32_vfmaddps256: |
| 10245 | case X86::BI__builtin_ia32_vfmaddpd256: |
| 10246 | case X86::BI__builtin_ia32_vfmaddps512_mask: |
| 10247 | case X86::BI__builtin_ia32_vfmaddps512_maskz: |
| 10248 | case X86::BI__builtin_ia32_vfmaddps512_mask3: |
| 10249 | case X86::BI__builtin_ia32_vfmsubps512_mask3: |
| 10250 | case X86::BI__builtin_ia32_vfmaddpd512_mask: |
| 10251 | case X86::BI__builtin_ia32_vfmaddpd512_maskz: |
| 10252 | case X86::BI__builtin_ia32_vfmaddpd512_mask3: |
| 10253 | case X86::BI__builtin_ia32_vfmsubpd512_mask3: |
| 10254 | return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/false); |
| 10255 | case X86::BI__builtin_ia32_vfmaddsubps: |
| 10256 | case X86::BI__builtin_ia32_vfmaddsubpd: |
| 10257 | case X86::BI__builtin_ia32_vfmaddsubps256: |
| 10258 | case X86::BI__builtin_ia32_vfmaddsubpd256: |
| 10259 | case X86::BI__builtin_ia32_vfmaddsubps512_mask: |
| 10260 | case X86::BI__builtin_ia32_vfmaddsubps512_maskz: |
| 10261 | case X86::BI__builtin_ia32_vfmaddsubps512_mask3: |
| 10262 | case X86::BI__builtin_ia32_vfmsubaddps512_mask3: |
| 10263 | case X86::BI__builtin_ia32_vfmaddsubpd512_mask: |
| 10264 | case X86::BI__builtin_ia32_vfmaddsubpd512_maskz: |
| 10265 | case X86::BI__builtin_ia32_vfmaddsubpd512_mask3: |
| 10266 | case X86::BI__builtin_ia32_vfmsubaddpd512_mask3: |
| 10267 | return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/true); |
| 10268 | |
| 10269 | case X86::BI__builtin_ia32_movdqa32store128_mask: |
| 10270 | case X86::BI__builtin_ia32_movdqa64store128_mask: |
| 10271 | case X86::BI__builtin_ia32_storeaps128_mask: |
| 10272 | case X86::BI__builtin_ia32_storeapd128_mask: |
| 10273 | case X86::BI__builtin_ia32_movdqa32store256_mask: |
| 10274 | case X86::BI__builtin_ia32_movdqa64store256_mask: |
| 10275 | case X86::BI__builtin_ia32_storeaps256_mask: |
| 10276 | case X86::BI__builtin_ia32_storeapd256_mask: |
| 10277 | case X86::BI__builtin_ia32_movdqa32store512_mask: |
| 10278 | case X86::BI__builtin_ia32_movdqa64store512_mask: |
| 10279 | case X86::BI__builtin_ia32_storeaps512_mask: |
| 10280 | case X86::BI__builtin_ia32_storeapd512_mask: { |
| 10281 | unsigned Align = |
| 10282 | getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity(); |
| 10283 | return EmitX86MaskedStore(*this, Ops, Align); |
| 10284 | } |
| 10285 | case X86::BI__builtin_ia32_loadups128_mask: |
| 10286 | case X86::BI__builtin_ia32_loadups256_mask: |
| 10287 | case X86::BI__builtin_ia32_loadups512_mask: |
| 10288 | case X86::BI__builtin_ia32_loadupd128_mask: |
| 10289 | case X86::BI__builtin_ia32_loadupd256_mask: |
| 10290 | case X86::BI__builtin_ia32_loadupd512_mask: |
| 10291 | case X86::BI__builtin_ia32_loaddquqi128_mask: |
| 10292 | case X86::BI__builtin_ia32_loaddquqi256_mask: |
| 10293 | case X86::BI__builtin_ia32_loaddquqi512_mask: |
| 10294 | case X86::BI__builtin_ia32_loaddquhi128_mask: |
| 10295 | case X86::BI__builtin_ia32_loaddquhi256_mask: |
| 10296 | case X86::BI__builtin_ia32_loaddquhi512_mask: |
| 10297 | case X86::BI__builtin_ia32_loaddqusi128_mask: |
| 10298 | case X86::BI__builtin_ia32_loaddqusi256_mask: |
| 10299 | case X86::BI__builtin_ia32_loaddqusi512_mask: |
| 10300 | case X86::BI__builtin_ia32_loaddqudi128_mask: |
| 10301 | case X86::BI__builtin_ia32_loaddqudi256_mask: |
| 10302 | case X86::BI__builtin_ia32_loaddqudi512_mask: |
| 10303 | return EmitX86MaskedLoad(*this, Ops, 1); |
| 10304 | |
| 10305 | case X86::BI__builtin_ia32_loadss128_mask: |
| 10306 | case X86::BI__builtin_ia32_loadsd128_mask: |
| 10307 | return EmitX86MaskedLoad(*this, Ops, 1); |
| 10308 | |
| 10309 | case X86::BI__builtin_ia32_loadaps128_mask: |
| 10310 | case X86::BI__builtin_ia32_loadaps256_mask: |
| 10311 | case X86::BI__builtin_ia32_loadaps512_mask: |
| 10312 | case X86::BI__builtin_ia32_loadapd128_mask: |
| 10313 | case X86::BI__builtin_ia32_loadapd256_mask: |
| 10314 | case X86::BI__builtin_ia32_loadapd512_mask: |
| 10315 | case X86::BI__builtin_ia32_movdqa32load128_mask: |
| 10316 | case X86::BI__builtin_ia32_movdqa32load256_mask: |
| 10317 | case X86::BI__builtin_ia32_movdqa32load512_mask: |
| 10318 | case X86::BI__builtin_ia32_movdqa64load128_mask: |
| 10319 | case X86::BI__builtin_ia32_movdqa64load256_mask: |
| 10320 | case X86::BI__builtin_ia32_movdqa64load512_mask: { |
| 10321 | unsigned Align = |
| 10322 | getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity(); |
| 10323 | return EmitX86MaskedLoad(*this, Ops, Align); |
| 10324 | } |
| 10325 | |
| 10326 | case X86::BI__builtin_ia32_expandloaddf128_mask: |
| 10327 | case X86::BI__builtin_ia32_expandloaddf256_mask: |
| 10328 | case X86::BI__builtin_ia32_expandloaddf512_mask: |
| 10329 | case X86::BI__builtin_ia32_expandloadsf128_mask: |
| 10330 | case X86::BI__builtin_ia32_expandloadsf256_mask: |
| 10331 | case X86::BI__builtin_ia32_expandloadsf512_mask: |
| 10332 | case X86::BI__builtin_ia32_expandloaddi128_mask: |
| 10333 | case X86::BI__builtin_ia32_expandloaddi256_mask: |
| 10334 | case X86::BI__builtin_ia32_expandloaddi512_mask: |
| 10335 | case X86::BI__builtin_ia32_expandloadsi128_mask: |
| 10336 | case X86::BI__builtin_ia32_expandloadsi256_mask: |
| 10337 | case X86::BI__builtin_ia32_expandloadsi512_mask: |
| 10338 | case X86::BI__builtin_ia32_expandloadhi128_mask: |
| 10339 | case X86::BI__builtin_ia32_expandloadhi256_mask: |
| 10340 | case X86::BI__builtin_ia32_expandloadhi512_mask: |
| 10341 | case X86::BI__builtin_ia32_expandloadqi128_mask: |
| 10342 | case X86::BI__builtin_ia32_expandloadqi256_mask: |
| 10343 | case X86::BI__builtin_ia32_expandloadqi512_mask: |
| 10344 | return EmitX86ExpandLoad(*this, Ops); |
| 10345 | |
| 10346 | case X86::BI__builtin_ia32_compressstoredf128_mask: |
| 10347 | case X86::BI__builtin_ia32_compressstoredf256_mask: |
| 10348 | case X86::BI__builtin_ia32_compressstoredf512_mask: |
| 10349 | case X86::BI__builtin_ia32_compressstoresf128_mask: |
| 10350 | case X86::BI__builtin_ia32_compressstoresf256_mask: |
| 10351 | case X86::BI__builtin_ia32_compressstoresf512_mask: |
| 10352 | case X86::BI__builtin_ia32_compressstoredi128_mask: |
| 10353 | case X86::BI__builtin_ia32_compressstoredi256_mask: |
| 10354 | case X86::BI__builtin_ia32_compressstoredi512_mask: |
| 10355 | case X86::BI__builtin_ia32_compressstoresi128_mask: |
| 10356 | case X86::BI__builtin_ia32_compressstoresi256_mask: |
| 10357 | case X86::BI__builtin_ia32_compressstoresi512_mask: |
| 10358 | case X86::BI__builtin_ia32_compressstorehi128_mask: |
| 10359 | case X86::BI__builtin_ia32_compressstorehi256_mask: |
| 10360 | case X86::BI__builtin_ia32_compressstorehi512_mask: |
| 10361 | case X86::BI__builtin_ia32_compressstoreqi128_mask: |
| 10362 | case X86::BI__builtin_ia32_compressstoreqi256_mask: |
| 10363 | case X86::BI__builtin_ia32_compressstoreqi512_mask: |
| 10364 | return EmitX86CompressStore(*this, Ops); |
| 10365 | |
| 10366 | case X86::BI__builtin_ia32_expanddf128_mask: |
| 10367 | case X86::BI__builtin_ia32_expanddf256_mask: |
| 10368 | case X86::BI__builtin_ia32_expanddf512_mask: |
| 10369 | case X86::BI__builtin_ia32_expandsf128_mask: |
| 10370 | case X86::BI__builtin_ia32_expandsf256_mask: |
| 10371 | case X86::BI__builtin_ia32_expandsf512_mask: |
| 10372 | case X86::BI__builtin_ia32_expanddi128_mask: |
| 10373 | case X86::BI__builtin_ia32_expanddi256_mask: |
| 10374 | case X86::BI__builtin_ia32_expanddi512_mask: |
| 10375 | case X86::BI__builtin_ia32_expandsi128_mask: |
| 10376 | case X86::BI__builtin_ia32_expandsi256_mask: |
| 10377 | case X86::BI__builtin_ia32_expandsi512_mask: |
| 10378 | case X86::BI__builtin_ia32_expandhi128_mask: |
| 10379 | case X86::BI__builtin_ia32_expandhi256_mask: |
| 10380 | case X86::BI__builtin_ia32_expandhi512_mask: |
| 10381 | case X86::BI__builtin_ia32_expandqi128_mask: |
| 10382 | case X86::BI__builtin_ia32_expandqi256_mask: |
| 10383 | case X86::BI__builtin_ia32_expandqi512_mask: |
| 10384 | return EmitX86CompressExpand(*this, Ops, /*IsCompress*/false); |
| 10385 | |
| 10386 | case X86::BI__builtin_ia32_compressdf128_mask: |
| 10387 | case X86::BI__builtin_ia32_compressdf256_mask: |
| 10388 | case X86::BI__builtin_ia32_compressdf512_mask: |
| 10389 | case X86::BI__builtin_ia32_compresssf128_mask: |
| 10390 | case X86::BI__builtin_ia32_compresssf256_mask: |
| 10391 | case X86::BI__builtin_ia32_compresssf512_mask: |
| 10392 | case X86::BI__builtin_ia32_compressdi128_mask: |
| 10393 | case X86::BI__builtin_ia32_compressdi256_mask: |
| 10394 | case X86::BI__builtin_ia32_compressdi512_mask: |
| 10395 | case X86::BI__builtin_ia32_compresssi128_mask: |
| 10396 | case X86::BI__builtin_ia32_compresssi256_mask: |
| 10397 | case X86::BI__builtin_ia32_compresssi512_mask: |
| 10398 | case X86::BI__builtin_ia32_compresshi128_mask: |
| 10399 | case X86::BI__builtin_ia32_compresshi256_mask: |
| 10400 | case X86::BI__builtin_ia32_compresshi512_mask: |
| 10401 | case X86::BI__builtin_ia32_compressqi128_mask: |
| 10402 | case X86::BI__builtin_ia32_compressqi256_mask: |
| 10403 | case X86::BI__builtin_ia32_compressqi512_mask: |
| 10404 | return EmitX86CompressExpand(*this, Ops, /*IsCompress*/true); |
| 10405 | |
| 10406 | case X86::BI__builtin_ia32_gather3div2df: |
| 10407 | case X86::BI__builtin_ia32_gather3div2di: |
| 10408 | case X86::BI__builtin_ia32_gather3div4df: |
| 10409 | case X86::BI__builtin_ia32_gather3div4di: |
| 10410 | case X86::BI__builtin_ia32_gather3div4sf: |
| 10411 | case X86::BI__builtin_ia32_gather3div4si: |
| 10412 | case X86::BI__builtin_ia32_gather3div8sf: |
| 10413 | case X86::BI__builtin_ia32_gather3div8si: |
| 10414 | case X86::BI__builtin_ia32_gather3siv2df: |
| 10415 | case X86::BI__builtin_ia32_gather3siv2di: |
| 10416 | case X86::BI__builtin_ia32_gather3siv4df: |
| 10417 | case X86::BI__builtin_ia32_gather3siv4di: |
| 10418 | case X86::BI__builtin_ia32_gather3siv4sf: |
| 10419 | case X86::BI__builtin_ia32_gather3siv4si: |
| 10420 | case X86::BI__builtin_ia32_gather3siv8sf: |
| 10421 | case X86::BI__builtin_ia32_gather3siv8si: |
| 10422 | case X86::BI__builtin_ia32_gathersiv8df: |
| 10423 | case X86::BI__builtin_ia32_gathersiv16sf: |
| 10424 | case X86::BI__builtin_ia32_gatherdiv8df: |
| 10425 | case X86::BI__builtin_ia32_gatherdiv16sf: |
| 10426 | case X86::BI__builtin_ia32_gathersiv8di: |
| 10427 | case X86::BI__builtin_ia32_gathersiv16si: |
| 10428 | case X86::BI__builtin_ia32_gatherdiv8di: |
| 10429 | case X86::BI__builtin_ia32_gatherdiv16si: { |
| 10430 | Intrinsic::ID IID; |
| 10431 | switch (BuiltinID) { |
| 10432 | default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10432); |
| 10433 | case X86::BI__builtin_ia32_gather3div2df: |
| 10434 | IID = Intrinsic::x86_avx512_mask_gather3div2_df; |
| 10435 | break; |
| 10436 | case X86::BI__builtin_ia32_gather3div2di: |
| 10437 | IID = Intrinsic::x86_avx512_mask_gather3div2_di; |
| 10438 | break; |
| 10439 | case X86::BI__builtin_ia32_gather3div4df: |
| 10440 | IID = Intrinsic::x86_avx512_mask_gather3div4_df; |
| 10441 | break; |
| 10442 | case X86::BI__builtin_ia32_gather3div4di: |
| 10443 | IID = Intrinsic::x86_avx512_mask_gather3div4_di; |
| 10444 | break; |
| 10445 | case X86::BI__builtin_ia32_gather3div4sf: |
| 10446 | IID = Intrinsic::x86_avx512_mask_gather3div4_sf; |
| 10447 | break; |
| 10448 | case X86::BI__builtin_ia32_gather3div4si: |
| 10449 | IID = Intrinsic::x86_avx512_mask_gather3div4_si; |
| 10450 | break; |
| 10451 | case X86::BI__builtin_ia32_gather3div8sf: |
| 10452 | IID = Intrinsic::x86_avx512_mask_gather3div8_sf; |
| 10453 | break; |
| 10454 | case X86::BI__builtin_ia32_gather3div8si: |
| 10455 | IID = Intrinsic::x86_avx512_mask_gather3div8_si; |
| 10456 | break; |
| 10457 | case X86::BI__builtin_ia32_gather3siv2df: |
| 10458 | IID = Intrinsic::x86_avx512_mask_gather3siv2_df; |
| 10459 | break; |
| 10460 | case X86::BI__builtin_ia32_gather3siv2di: |
| 10461 | IID = Intrinsic::x86_avx512_mask_gather3siv2_di; |
| 10462 | break; |
| 10463 | case X86::BI__builtin_ia32_gather3siv4df: |
| 10464 | IID = Intrinsic::x86_avx512_mask_gather3siv4_df; |
| 10465 | break; |
| 10466 | case X86::BI__builtin_ia32_gather3siv4di: |
| 10467 | IID = Intrinsic::x86_avx512_mask_gather3siv4_di; |
| 10468 | break; |
| 10469 | case X86::BI__builtin_ia32_gather3siv4sf: |
| 10470 | IID = Intrinsic::x86_avx512_mask_gather3siv4_sf; |
| 10471 | break; |
| 10472 | case X86::BI__builtin_ia32_gather3siv4si: |
| 10473 | IID = Intrinsic::x86_avx512_mask_gather3siv4_si; |
| 10474 | break; |
| 10475 | case X86::BI__builtin_ia32_gather3siv8sf: |
| 10476 | IID = Intrinsic::x86_avx512_mask_gather3siv8_sf; |
| 10477 | break; |
| 10478 | case X86::BI__builtin_ia32_gather3siv8si: |
| 10479 | IID = Intrinsic::x86_avx512_mask_gather3siv8_si; |
| 10480 | break; |
| 10481 | case X86::BI__builtin_ia32_gathersiv8df: |
| 10482 | IID = Intrinsic::x86_avx512_mask_gather_dpd_512; |
| 10483 | break; |
| 10484 | case X86::BI__builtin_ia32_gathersiv16sf: |
| 10485 | IID = Intrinsic::x86_avx512_mask_gather_dps_512; |
| 10486 | break; |
| 10487 | case X86::BI__builtin_ia32_gatherdiv8df: |
| 10488 | IID = Intrinsic::x86_avx512_mask_gather_qpd_512; |
| 10489 | break; |
| 10490 | case X86::BI__builtin_ia32_gatherdiv16sf: |
| 10491 | IID = Intrinsic::x86_avx512_mask_gather_qps_512; |
| 10492 | break; |
| 10493 | case X86::BI__builtin_ia32_gathersiv8di: |
| 10494 | IID = Intrinsic::x86_avx512_mask_gather_dpq_512; |
| 10495 | break; |
| 10496 | case X86::BI__builtin_ia32_gathersiv16si: |
| 10497 | IID = Intrinsic::x86_avx512_mask_gather_dpi_512; |
| 10498 | break; |
| 10499 | case X86::BI__builtin_ia32_gatherdiv8di: |
| 10500 | IID = Intrinsic::x86_avx512_mask_gather_qpq_512; |
| 10501 | break; |
| 10502 | case X86::BI__builtin_ia32_gatherdiv16si: |
| 10503 | IID = Intrinsic::x86_avx512_mask_gather_qpi_512; |
| 10504 | break; |
| 10505 | } |
| 10506 | |
| 10507 | unsigned MinElts = std::min(Ops[0]->getType()->getVectorNumElements(), |
| 10508 | Ops[2]->getType()->getVectorNumElements()); |
| 10509 | Ops[3] = getMaskVecValue(*this, Ops[3], MinElts); |
| 10510 | Function *Intr = CGM.getIntrinsic(IID); |
| 10511 | return Builder.CreateCall(Intr, Ops); |
| 10512 | } |
| 10513 | |
| 10514 | case X86::BI__builtin_ia32_scattersiv8df: |
| 10515 | case X86::BI__builtin_ia32_scattersiv16sf: |
| 10516 | case X86::BI__builtin_ia32_scatterdiv8df: |
| 10517 | case X86::BI__builtin_ia32_scatterdiv16sf: |
| 10518 | case X86::BI__builtin_ia32_scattersiv8di: |
| 10519 | case X86::BI__builtin_ia32_scattersiv16si: |
| 10520 | case X86::BI__builtin_ia32_scatterdiv8di: |
| 10521 | case X86::BI__builtin_ia32_scatterdiv16si: |
| 10522 | case X86::BI__builtin_ia32_scatterdiv2df: |
| 10523 | case X86::BI__builtin_ia32_scatterdiv2di: |
| 10524 | case X86::BI__builtin_ia32_scatterdiv4df: |
| 10525 | case X86::BI__builtin_ia32_scatterdiv4di: |
| 10526 | case X86::BI__builtin_ia32_scatterdiv4sf: |
| 10527 | case X86::BI__builtin_ia32_scatterdiv4si: |
| 10528 | case X86::BI__builtin_ia32_scatterdiv8sf: |
| 10529 | case X86::BI__builtin_ia32_scatterdiv8si: |
| 10530 | case X86::BI__builtin_ia32_scattersiv2df: |
| 10531 | case X86::BI__builtin_ia32_scattersiv2di: |
| 10532 | case X86::BI__builtin_ia32_scattersiv4df: |
| 10533 | case X86::BI__builtin_ia32_scattersiv4di: |
| 10534 | case X86::BI__builtin_ia32_scattersiv4sf: |
| 10535 | case X86::BI__builtin_ia32_scattersiv4si: |
| 10536 | case X86::BI__builtin_ia32_scattersiv8sf: |
| 10537 | case X86::BI__builtin_ia32_scattersiv8si: { |
| 10538 | Intrinsic::ID IID; |
| 10539 | switch (BuiltinID) { |
| 10540 | default: llvm_unreachable("Unexpected builtin")::llvm::llvm_unreachable_internal("Unexpected builtin", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10540); |
| 10541 | case X86::BI__builtin_ia32_scattersiv8df: |
| 10542 | IID = Intrinsic::x86_avx512_mask_scatter_dpd_512; |
| 10543 | break; |
| 10544 | case X86::BI__builtin_ia32_scattersiv16sf: |
| 10545 | IID = Intrinsic::x86_avx512_mask_scatter_dps_512; |
| 10546 | break; |
| 10547 | case X86::BI__builtin_ia32_scatterdiv8df: |
| 10548 | IID = Intrinsic::x86_avx512_mask_scatter_qpd_512; |
| 10549 | break; |
| 10550 | case X86::BI__builtin_ia32_scatterdiv16sf: |
| 10551 | IID = Intrinsic::x86_avx512_mask_scatter_qps_512; |
| 10552 | break; |
| 10553 | case X86::BI__builtin_ia32_scattersiv8di: |
| 10554 | IID = Intrinsic::x86_avx512_mask_scatter_dpq_512; |
| 10555 | break; |
| 10556 | case X86::BI__builtin_ia32_scattersiv16si: |
| 10557 | IID = Intrinsic::x86_avx512_mask_scatter_dpi_512; |
| 10558 | break; |
| 10559 | case X86::BI__builtin_ia32_scatterdiv8di: |
| 10560 | IID = Intrinsic::x86_avx512_mask_scatter_qpq_512; |
| 10561 | break; |
| 10562 | case X86::BI__builtin_ia32_scatterdiv16si: |
| 10563 | IID = Intrinsic::x86_avx512_mask_scatter_qpi_512; |
| 10564 | break; |
| 10565 | case X86::BI__builtin_ia32_scatterdiv2df: |
| 10566 | IID = Intrinsic::x86_avx512_mask_scatterdiv2_df; |
| 10567 | break; |
| 10568 | case X86::BI__builtin_ia32_scatterdiv2di: |
| 10569 | IID = Intrinsic::x86_avx512_mask_scatterdiv2_di; |
| 10570 | break; |
| 10571 | case X86::BI__builtin_ia32_scatterdiv4df: |
| 10572 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_df; |
| 10573 | break; |
| 10574 | case X86::BI__builtin_ia32_scatterdiv4di: |
| 10575 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_di; |
| 10576 | break; |
| 10577 | case X86::BI__builtin_ia32_scatterdiv4sf: |
| 10578 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_sf; |
| 10579 | break; |
| 10580 | case X86::BI__builtin_ia32_scatterdiv4si: |
| 10581 | IID = Intrinsic::x86_avx512_mask_scatterdiv4_si; |
| 10582 | break; |
| 10583 | case X86::BI__builtin_ia32_scatterdiv8sf: |
| 10584 | IID = Intrinsic::x86_avx512_mask_scatterdiv8_sf; |
| 10585 | break; |
| 10586 | case X86::BI__builtin_ia32_scatterdiv8si: |
| 10587 | IID = Intrinsic::x86_avx512_mask_scatterdiv8_si; |
| 10588 | break; |
| 10589 | case X86::BI__builtin_ia32_scattersiv2df: |
| 10590 | IID = Intrinsic::x86_avx512_mask_scattersiv2_df; |
| 10591 | break; |
| 10592 | case X86::BI__builtin_ia32_scattersiv2di: |
| 10593 | IID = Intrinsic::x86_avx512_mask_scattersiv2_di; |
| 10594 | break; |
| 10595 | case X86::BI__builtin_ia32_scattersiv4df: |
| 10596 | IID = Intrinsic::x86_avx512_mask_scattersiv4_df; |
| 10597 | break; |
| 10598 | case X86::BI__builtin_ia32_scattersiv4di: |
| 10599 | IID = Intrinsic::x86_avx512_mask_scattersiv4_di; |
| 10600 | break; |
| 10601 | case X86::BI__builtin_ia32_scattersiv4sf: |
| 10602 | IID = Intrinsic::x86_avx512_mask_scattersiv4_sf; |
| 10603 | break; |
| 10604 | case X86::BI__builtin_ia32_scattersiv4si: |
| 10605 | IID = Intrinsic::x86_avx512_mask_scattersiv4_si; |
| 10606 | break; |
| 10607 | case X86::BI__builtin_ia32_scattersiv8sf: |
| 10608 | IID = Intrinsic::x86_avx512_mask_scattersiv8_sf; |
| 10609 | break; |
| 10610 | case X86::BI__builtin_ia32_scattersiv8si: |
| 10611 | IID = Intrinsic::x86_avx512_mask_scattersiv8_si; |
| 10612 | break; |
| 10613 | } |
| 10614 | |
| 10615 | unsigned MinElts = std::min(Ops[2]->getType()->getVectorNumElements(), |
| 10616 | Ops[3]->getType()->getVectorNumElements()); |
| 10617 | Ops[1] = getMaskVecValue(*this, Ops[1], MinElts); |
| 10618 | Function *Intr = CGM.getIntrinsic(IID); |
| 10619 | return Builder.CreateCall(Intr, Ops); |
| 10620 | } |
| 10621 | |
| 10622 | case X86::BI__builtin_ia32_storehps: |
| 10623 | case X86::BI__builtin_ia32_storelps: { |
| 10624 | llvm::Type *PtrTy = llvm::PointerType::getUnqual(Int64Ty); |
| 10625 | llvm::Type *VecTy = llvm::VectorType::get(Int64Ty, 2); |
| 10626 | |
| 10627 | // cast val v2i64 |
| 10628 | Ops[1] = Builder.CreateBitCast(Ops[1], VecTy, "cast"); |
| 10629 | |
| 10630 | // extract (0, 1) |
| 10631 | unsigned Index = BuiltinID == X86::BI__builtin_ia32_storelps ? 0 : 1; |
| 10632 | Ops[1] = Builder.CreateExtractElement(Ops[1], Index, "extract"); |
| 10633 | |
| 10634 | // cast pointer to i64 & store |
| 10635 | Ops[0] = Builder.CreateBitCast(Ops[0], PtrTy); |
| 10636 | return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]); |
| 10637 | } |
| 10638 | case X86::BI__builtin_ia32_vextractf128_pd256: |
| 10639 | case X86::BI__builtin_ia32_vextractf128_ps256: |
| 10640 | case X86::BI__builtin_ia32_vextractf128_si256: |
| 10641 | case X86::BI__builtin_ia32_extract128i256: |
| 10642 | case X86::BI__builtin_ia32_extractf64x4_mask: |
| 10643 | case X86::BI__builtin_ia32_extractf32x4_mask: |
| 10644 | case X86::BI__builtin_ia32_extracti64x4_mask: |
| 10645 | case X86::BI__builtin_ia32_extracti32x4_mask: |
| 10646 | case X86::BI__builtin_ia32_extractf32x8_mask: |
| 10647 | case X86::BI__builtin_ia32_extracti32x8_mask: |
| 10648 | case X86::BI__builtin_ia32_extractf32x4_256_mask: |
| 10649 | case X86::BI__builtin_ia32_extracti32x4_256_mask: |
| 10650 | case X86::BI__builtin_ia32_extractf64x2_256_mask: |
| 10651 | case X86::BI__builtin_ia32_extracti64x2_256_mask: |
| 10652 | case X86::BI__builtin_ia32_extractf64x2_512_mask: |
| 10653 | case X86::BI__builtin_ia32_extracti64x2_512_mask: { |
| 10654 | llvm::Type *DstTy = ConvertType(E->getType()); |
| 10655 | unsigned NumElts = DstTy->getVectorNumElements(); |
| 10656 | unsigned SrcNumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10657 | unsigned SubVectors = SrcNumElts / NumElts; |
| 10658 | unsigned Index = cast<ConstantInt>(Ops[1])->getZExtValue(); |
| 10659 | assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")((llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors" ) ? static_cast<void> (0) : __assert_fail ("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10659, __PRETTY_FUNCTION__)); |
| 10660 | Index &= SubVectors - 1; // Remove any extra bits. |
| 10661 | Index *= NumElts; |
| 10662 | |
| 10663 | uint32_t Indices[16]; |
| 10664 | for (unsigned i = 0; i != NumElts; ++i) |
| 10665 | Indices[i] = i + Index; |
| 10666 | |
| 10667 | Value *Res = Builder.CreateShuffleVector(Ops[0], |
| 10668 | UndefValue::get(Ops[0]->getType()), |
| 10669 | makeArrayRef(Indices, NumElts), |
| 10670 | "extract"); |
| 10671 | |
| 10672 | if (Ops.size() == 4) |
| 10673 | Res = EmitX86Select(*this, Ops[3], Res, Ops[2]); |
| 10674 | |
| 10675 | return Res; |
| 10676 | } |
| 10677 | case X86::BI__builtin_ia32_vinsertf128_pd256: |
| 10678 | case X86::BI__builtin_ia32_vinsertf128_ps256: |
| 10679 | case X86::BI__builtin_ia32_vinsertf128_si256: |
| 10680 | case X86::BI__builtin_ia32_insert128i256: |
| 10681 | case X86::BI__builtin_ia32_insertf64x4: |
| 10682 | case X86::BI__builtin_ia32_insertf32x4: |
| 10683 | case X86::BI__builtin_ia32_inserti64x4: |
| 10684 | case X86::BI__builtin_ia32_inserti32x4: |
| 10685 | case X86::BI__builtin_ia32_insertf32x8: |
| 10686 | case X86::BI__builtin_ia32_inserti32x8: |
| 10687 | case X86::BI__builtin_ia32_insertf32x4_256: |
| 10688 | case X86::BI__builtin_ia32_inserti32x4_256: |
| 10689 | case X86::BI__builtin_ia32_insertf64x2_256: |
| 10690 | case X86::BI__builtin_ia32_inserti64x2_256: |
| 10691 | case X86::BI__builtin_ia32_insertf64x2_512: |
| 10692 | case X86::BI__builtin_ia32_inserti64x2_512: { |
| 10693 | unsigned DstNumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10694 | unsigned SrcNumElts = Ops[1]->getType()->getVectorNumElements(); |
| 10695 | unsigned SubVectors = DstNumElts / SrcNumElts; |
| 10696 | unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue(); |
| 10697 | assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors")((llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors" ) ? static_cast<void> (0) : __assert_fail ("llvm::isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10697, __PRETTY_FUNCTION__)); |
| 10698 | Index &= SubVectors - 1; // Remove any extra bits. |
| 10699 | Index *= SrcNumElts; |
| 10700 | |
| 10701 | uint32_t Indices[16]; |
| 10702 | for (unsigned i = 0; i != DstNumElts; ++i) |
| 10703 | Indices[i] = (i >= SrcNumElts) ? SrcNumElts + (i % SrcNumElts) : i; |
| 10704 | |
| 10705 | Value *Op1 = Builder.CreateShuffleVector(Ops[1], |
| 10706 | UndefValue::get(Ops[1]->getType()), |
| 10707 | makeArrayRef(Indices, DstNumElts), |
| 10708 | "widen"); |
| 10709 | |
| 10710 | for (unsigned i = 0; i != DstNumElts; ++i) { |
| 10711 | if (i >= Index && i < (Index + SrcNumElts)) |
| 10712 | Indices[i] = (i - Index) + DstNumElts; |
| 10713 | else |
| 10714 | Indices[i] = i; |
| 10715 | } |
| 10716 | |
| 10717 | return Builder.CreateShuffleVector(Ops[0], Op1, |
| 10718 | makeArrayRef(Indices, DstNumElts), |
| 10719 | "insert"); |
| 10720 | } |
| 10721 | case X86::BI__builtin_ia32_pmovqd512_mask: |
| 10722 | case X86::BI__builtin_ia32_pmovwb512_mask: { |
| 10723 | Value *Res = Builder.CreateTrunc(Ops[0], Ops[1]->getType()); |
| 10724 | return EmitX86Select(*this, Ops[2], Res, Ops[1]); |
| 10725 | } |
| 10726 | case X86::BI__builtin_ia32_pmovdb512_mask: |
| 10727 | case X86::BI__builtin_ia32_pmovdw512_mask: |
| 10728 | case X86::BI__builtin_ia32_pmovqw512_mask: { |
| 10729 | if (const auto *C = dyn_cast<Constant>(Ops[2])) |
| 10730 | if (C->isAllOnesValue()) |
| 10731 | return Builder.CreateTrunc(Ops[0], Ops[1]->getType()); |
| 10732 | |
| 10733 | Intrinsic::ID IID; |
| 10734 | switch (BuiltinID) { |
| 10735 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10735); |
| 10736 | case X86::BI__builtin_ia32_pmovdb512_mask: |
| 10737 | IID = Intrinsic::x86_avx512_mask_pmov_db_512; |
| 10738 | break; |
| 10739 | case X86::BI__builtin_ia32_pmovdw512_mask: |
| 10740 | IID = Intrinsic::x86_avx512_mask_pmov_dw_512; |
| 10741 | break; |
| 10742 | case X86::BI__builtin_ia32_pmovqw512_mask: |
| 10743 | IID = Intrinsic::x86_avx512_mask_pmov_qw_512; |
| 10744 | break; |
| 10745 | } |
| 10746 | |
| 10747 | Function *Intr = CGM.getIntrinsic(IID); |
| 10748 | return Builder.CreateCall(Intr, Ops); |
| 10749 | } |
| 10750 | case X86::BI__builtin_ia32_pblendw128: |
| 10751 | case X86::BI__builtin_ia32_blendpd: |
| 10752 | case X86::BI__builtin_ia32_blendps: |
| 10753 | case X86::BI__builtin_ia32_blendpd256: |
| 10754 | case X86::BI__builtin_ia32_blendps256: |
| 10755 | case X86::BI__builtin_ia32_pblendw256: |
| 10756 | case X86::BI__builtin_ia32_pblendd128: |
| 10757 | case X86::BI__builtin_ia32_pblendd256: { |
| 10758 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10759 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); |
| 10760 | |
| 10761 | uint32_t Indices[16]; |
| 10762 | // If there are more than 8 elements, the immediate is used twice so make |
| 10763 | // sure we handle that. |
| 10764 | for (unsigned i = 0; i != NumElts; ++i) |
| 10765 | Indices[i] = ((Imm >> (i % 8)) & 0x1) ? NumElts + i : i; |
| 10766 | |
| 10767 | return Builder.CreateShuffleVector(Ops[0], Ops[1], |
| 10768 | makeArrayRef(Indices, NumElts), |
| 10769 | "blend"); |
| 10770 | } |
| 10771 | case X86::BI__builtin_ia32_pshuflw: |
| 10772 | case X86::BI__builtin_ia32_pshuflw256: |
| 10773 | case X86::BI__builtin_ia32_pshuflw512: { |
| 10774 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); |
| 10775 | llvm::Type *Ty = Ops[0]->getType(); |
| 10776 | unsigned NumElts = Ty->getVectorNumElements(); |
| 10777 | |
| 10778 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. |
| 10779 | Imm = (Imm & 0xff) * 0x01010101; |
| 10780 | |
| 10781 | uint32_t Indices[32]; |
| 10782 | for (unsigned l = 0; l != NumElts; l += 8) { |
| 10783 | for (unsigned i = 0; i != 4; ++i) { |
| 10784 | Indices[l + i] = l + (Imm & 3); |
| 10785 | Imm >>= 2; |
| 10786 | } |
| 10787 | for (unsigned i = 4; i != 8; ++i) |
| 10788 | Indices[l + i] = l + i; |
| 10789 | } |
| 10790 | |
| 10791 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), |
| 10792 | makeArrayRef(Indices, NumElts), |
| 10793 | "pshuflw"); |
| 10794 | } |
| 10795 | case X86::BI__builtin_ia32_pshufhw: |
| 10796 | case X86::BI__builtin_ia32_pshufhw256: |
| 10797 | case X86::BI__builtin_ia32_pshufhw512: { |
| 10798 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); |
| 10799 | llvm::Type *Ty = Ops[0]->getType(); |
| 10800 | unsigned NumElts = Ty->getVectorNumElements(); |
| 10801 | |
| 10802 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. |
| 10803 | Imm = (Imm & 0xff) * 0x01010101; |
| 10804 | |
| 10805 | uint32_t Indices[32]; |
| 10806 | for (unsigned l = 0; l != NumElts; l += 8) { |
| 10807 | for (unsigned i = 0; i != 4; ++i) |
| 10808 | Indices[l + i] = l + i; |
| 10809 | for (unsigned i = 4; i != 8; ++i) { |
| 10810 | Indices[l + i] = l + 4 + (Imm & 3); |
| 10811 | Imm >>= 2; |
| 10812 | } |
| 10813 | } |
| 10814 | |
| 10815 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), |
| 10816 | makeArrayRef(Indices, NumElts), |
| 10817 | "pshufhw"); |
| 10818 | } |
| 10819 | case X86::BI__builtin_ia32_pshufd: |
| 10820 | case X86::BI__builtin_ia32_pshufd256: |
| 10821 | case X86::BI__builtin_ia32_pshufd512: |
| 10822 | case X86::BI__builtin_ia32_vpermilpd: |
| 10823 | case X86::BI__builtin_ia32_vpermilps: |
| 10824 | case X86::BI__builtin_ia32_vpermilpd256: |
| 10825 | case X86::BI__builtin_ia32_vpermilps256: |
| 10826 | case X86::BI__builtin_ia32_vpermilpd512: |
| 10827 | case X86::BI__builtin_ia32_vpermilps512: { |
| 10828 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); |
| 10829 | llvm::Type *Ty = Ops[0]->getType(); |
| 10830 | unsigned NumElts = Ty->getVectorNumElements(); |
| 10831 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128; |
| 10832 | unsigned NumLaneElts = NumElts / NumLanes; |
| 10833 | |
| 10834 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. |
| 10835 | Imm = (Imm & 0xff) * 0x01010101; |
| 10836 | |
| 10837 | uint32_t Indices[16]; |
| 10838 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { |
| 10839 | for (unsigned i = 0; i != NumLaneElts; ++i) { |
| 10840 | Indices[i + l] = (Imm % NumLaneElts) + l; |
| 10841 | Imm /= NumLaneElts; |
| 10842 | } |
| 10843 | } |
| 10844 | |
| 10845 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), |
| 10846 | makeArrayRef(Indices, NumElts), |
| 10847 | "permil"); |
| 10848 | } |
| 10849 | case X86::BI__builtin_ia32_shufpd: |
| 10850 | case X86::BI__builtin_ia32_shufpd256: |
| 10851 | case X86::BI__builtin_ia32_shufpd512: |
| 10852 | case X86::BI__builtin_ia32_shufps: |
| 10853 | case X86::BI__builtin_ia32_shufps256: |
| 10854 | case X86::BI__builtin_ia32_shufps512: { |
| 10855 | uint32_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); |
| 10856 | llvm::Type *Ty = Ops[0]->getType(); |
| 10857 | unsigned NumElts = Ty->getVectorNumElements(); |
| 10858 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128; |
| 10859 | unsigned NumLaneElts = NumElts / NumLanes; |
| 10860 | |
| 10861 | // Splat the 8-bits of immediate 4 times to help the loop wrap around. |
| 10862 | Imm = (Imm & 0xff) * 0x01010101; |
| 10863 | |
| 10864 | uint32_t Indices[16]; |
| 10865 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { |
| 10866 | for (unsigned i = 0; i != NumLaneElts; ++i) { |
| 10867 | unsigned Index = Imm % NumLaneElts; |
| 10868 | Imm /= NumLaneElts; |
| 10869 | if (i >= (NumLaneElts / 2)) |
| 10870 | Index += NumElts; |
| 10871 | Indices[l + i] = l + Index; |
| 10872 | } |
| 10873 | } |
| 10874 | |
| 10875 | return Builder.CreateShuffleVector(Ops[0], Ops[1], |
| 10876 | makeArrayRef(Indices, NumElts), |
| 10877 | "shufp"); |
| 10878 | } |
| 10879 | case X86::BI__builtin_ia32_permdi256: |
| 10880 | case X86::BI__builtin_ia32_permdf256: |
| 10881 | case X86::BI__builtin_ia32_permdi512: |
| 10882 | case X86::BI__builtin_ia32_permdf512: { |
| 10883 | unsigned Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); |
| 10884 | llvm::Type *Ty = Ops[0]->getType(); |
| 10885 | unsigned NumElts = Ty->getVectorNumElements(); |
| 10886 | |
| 10887 | // These intrinsics operate on 256-bit lanes of four 64-bit elements. |
| 10888 | uint32_t Indices[8]; |
| 10889 | for (unsigned l = 0; l != NumElts; l += 4) |
| 10890 | for (unsigned i = 0; i != 4; ++i) |
| 10891 | Indices[l + i] = l + ((Imm >> (2 * i)) & 0x3); |
| 10892 | |
| 10893 | return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty), |
| 10894 | makeArrayRef(Indices, NumElts), |
| 10895 | "perm"); |
| 10896 | } |
| 10897 | case X86::BI__builtin_ia32_palignr128: |
| 10898 | case X86::BI__builtin_ia32_palignr256: |
| 10899 | case X86::BI__builtin_ia32_palignr512: { |
| 10900 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff; |
| 10901 | |
| 10902 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10903 | assert(NumElts % 16 == 0)((NumElts % 16 == 0) ? static_cast<void> (0) : __assert_fail ("NumElts % 16 == 0", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 10903, __PRETTY_FUNCTION__)); |
| 10904 | |
| 10905 | // If palignr is shifting the pair of vectors more than the size of two |
| 10906 | // lanes, emit zero. |
| 10907 | if (ShiftVal >= 32) |
| 10908 | return llvm::Constant::getNullValue(ConvertType(E->getType())); |
| 10909 | |
| 10910 | // If palignr is shifting the pair of input vectors more than one lane, |
| 10911 | // but less than two lanes, convert to shifting in zeroes. |
| 10912 | if (ShiftVal > 16) { |
| 10913 | ShiftVal -= 16; |
| 10914 | Ops[1] = Ops[0]; |
| 10915 | Ops[0] = llvm::Constant::getNullValue(Ops[0]->getType()); |
| 10916 | } |
| 10917 | |
| 10918 | uint32_t Indices[64]; |
| 10919 | // 256-bit palignr operates on 128-bit lanes so we need to handle that |
| 10920 | for (unsigned l = 0; l != NumElts; l += 16) { |
| 10921 | for (unsigned i = 0; i != 16; ++i) { |
| 10922 | unsigned Idx = ShiftVal + i; |
| 10923 | if (Idx >= 16) |
| 10924 | Idx += NumElts - 16; // End of lane, switch operand. |
| 10925 | Indices[l + i] = Idx + l; |
| 10926 | } |
| 10927 | } |
| 10928 | |
| 10929 | return Builder.CreateShuffleVector(Ops[1], Ops[0], |
| 10930 | makeArrayRef(Indices, NumElts), |
| 10931 | "palignr"); |
| 10932 | } |
| 10933 | case X86::BI__builtin_ia32_alignd128: |
| 10934 | case X86::BI__builtin_ia32_alignd256: |
| 10935 | case X86::BI__builtin_ia32_alignd512: |
| 10936 | case X86::BI__builtin_ia32_alignq128: |
| 10937 | case X86::BI__builtin_ia32_alignq256: |
| 10938 | case X86::BI__builtin_ia32_alignq512: { |
| 10939 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10940 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff; |
| 10941 | |
| 10942 | // Mask the shift amount to width of two vectors. |
| 10943 | ShiftVal &= (2 * NumElts) - 1; |
| 10944 | |
| 10945 | uint32_t Indices[16]; |
| 10946 | for (unsigned i = 0; i != NumElts; ++i) |
| 10947 | Indices[i] = i + ShiftVal; |
| 10948 | |
| 10949 | return Builder.CreateShuffleVector(Ops[1], Ops[0], |
| 10950 | makeArrayRef(Indices, NumElts), |
| 10951 | "valign"); |
| 10952 | } |
| 10953 | case X86::BI__builtin_ia32_shuf_f32x4_256: |
| 10954 | case X86::BI__builtin_ia32_shuf_f64x2_256: |
| 10955 | case X86::BI__builtin_ia32_shuf_i32x4_256: |
| 10956 | case X86::BI__builtin_ia32_shuf_i64x2_256: |
| 10957 | case X86::BI__builtin_ia32_shuf_f32x4: |
| 10958 | case X86::BI__builtin_ia32_shuf_f64x2: |
| 10959 | case X86::BI__builtin_ia32_shuf_i32x4: |
| 10960 | case X86::BI__builtin_ia32_shuf_i64x2: { |
| 10961 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); |
| 10962 | llvm::Type *Ty = Ops[0]->getType(); |
| 10963 | unsigned NumElts = Ty->getVectorNumElements(); |
| 10964 | unsigned NumLanes = Ty->getPrimitiveSizeInBits() == 512 ? 4 : 2; |
| 10965 | unsigned NumLaneElts = NumElts / NumLanes; |
| 10966 | |
| 10967 | uint32_t Indices[16]; |
| 10968 | for (unsigned l = 0; l != NumElts; l += NumLaneElts) { |
| 10969 | unsigned Index = (Imm % NumLanes) * NumLaneElts; |
| 10970 | Imm /= NumLanes; // Discard the bits we just used. |
| 10971 | if (l >= (NumElts / 2)) |
| 10972 | Index += NumElts; // Switch to other source. |
| 10973 | for (unsigned i = 0; i != NumLaneElts; ++i) { |
| 10974 | Indices[l + i] = Index + i; |
| 10975 | } |
| 10976 | } |
| 10977 | |
| 10978 | return Builder.CreateShuffleVector(Ops[0], Ops[1], |
| 10979 | makeArrayRef(Indices, NumElts), |
| 10980 | "shuf"); |
| 10981 | } |
| 10982 | |
| 10983 | case X86::BI__builtin_ia32_vperm2f128_pd256: |
| 10984 | case X86::BI__builtin_ia32_vperm2f128_ps256: |
| 10985 | case X86::BI__builtin_ia32_vperm2f128_si256: |
| 10986 | case X86::BI__builtin_ia32_permti256: { |
| 10987 | unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue(); |
| 10988 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 10989 | |
| 10990 | // This takes a very simple approach since there are two lanes and a |
| 10991 | // shuffle can have 2 inputs. So we reserve the first input for the first |
| 10992 | // lane and the second input for the second lane. This may result in |
| 10993 | // duplicate sources, but this can be dealt with in the backend. |
| 10994 | |
| 10995 | Value *OutOps[2]; |
| 10996 | uint32_t Indices[8]; |
| 10997 | for (unsigned l = 0; l != 2; ++l) { |
| 10998 | // Determine the source for this lane. |
| 10999 | if (Imm & (1 << ((l * 4) + 3))) |
| 11000 | OutOps[l] = llvm::ConstantAggregateZero::get(Ops[0]->getType()); |
| 11001 | else if (Imm & (1 << ((l * 4) + 1))) |
| 11002 | OutOps[l] = Ops[1]; |
| 11003 | else |
| 11004 | OutOps[l] = Ops[0]; |
| 11005 | |
| 11006 | for (unsigned i = 0; i != NumElts/2; ++i) { |
| 11007 | // Start with ith element of the source for this lane. |
| 11008 | unsigned Idx = (l * NumElts) + i; |
| 11009 | // If bit 0 of the immediate half is set, switch to the high half of |
| 11010 | // the source. |
| 11011 | if (Imm & (1 << (l * 4))) |
| 11012 | Idx += NumElts/2; |
| 11013 | Indices[(l * (NumElts/2)) + i] = Idx; |
| 11014 | } |
| 11015 | } |
| 11016 | |
| 11017 | return Builder.CreateShuffleVector(OutOps[0], OutOps[1], |
| 11018 | makeArrayRef(Indices, NumElts), |
| 11019 | "vperm"); |
| 11020 | } |
| 11021 | |
| 11022 | case X86::BI__builtin_ia32_pslldqi128_byteshift: |
| 11023 | case X86::BI__builtin_ia32_pslldqi256_byteshift: |
| 11024 | case X86::BI__builtin_ia32_pslldqi512_byteshift: { |
| 11025 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; |
| 11026 | llvm::Type *ResultType = Ops[0]->getType(); |
| 11027 | // Builtin type is vXi64 so multiply by 8 to get bytes. |
| 11028 | unsigned NumElts = ResultType->getVectorNumElements() * 8; |
| 11029 | |
| 11030 | // If pslldq is shifting the vector more than 15 bytes, emit zero. |
| 11031 | if (ShiftVal >= 16) |
| 11032 | return llvm::Constant::getNullValue(ResultType); |
| 11033 | |
| 11034 | uint32_t Indices[64]; |
| 11035 | // 256/512-bit pslldq operates on 128-bit lanes so we need to handle that |
| 11036 | for (unsigned l = 0; l != NumElts; l += 16) { |
| 11037 | for (unsigned i = 0; i != 16; ++i) { |
| 11038 | unsigned Idx = NumElts + i - ShiftVal; |
| 11039 | if (Idx < NumElts) Idx -= NumElts - 16; // end of lane, switch operand. |
| 11040 | Indices[l + i] = Idx + l; |
| 11041 | } |
| 11042 | } |
| 11043 | |
| 11044 | llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts); |
| 11045 | Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast"); |
| 11046 | Value *Zero = llvm::Constant::getNullValue(VecTy); |
| 11047 | Value *SV = Builder.CreateShuffleVector(Zero, Cast, |
| 11048 | makeArrayRef(Indices, NumElts), |
| 11049 | "pslldq"); |
| 11050 | return Builder.CreateBitCast(SV, Ops[0]->getType(), "cast"); |
| 11051 | } |
| 11052 | case X86::BI__builtin_ia32_psrldqi128_byteshift: |
| 11053 | case X86::BI__builtin_ia32_psrldqi256_byteshift: |
| 11054 | case X86::BI__builtin_ia32_psrldqi512_byteshift: { |
| 11055 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; |
| 11056 | llvm::Type *ResultType = Ops[0]->getType(); |
| 11057 | // Builtin type is vXi64 so multiply by 8 to get bytes. |
| 11058 | unsigned NumElts = ResultType->getVectorNumElements() * 8; |
| 11059 | |
| 11060 | // If psrldq is shifting the vector more than 15 bytes, emit zero. |
| 11061 | if (ShiftVal >= 16) |
| 11062 | return llvm::Constant::getNullValue(ResultType); |
| 11063 | |
| 11064 | uint32_t Indices[64]; |
| 11065 | // 256/512-bit psrldq operates on 128-bit lanes so we need to handle that |
| 11066 | for (unsigned l = 0; l != NumElts; l += 16) { |
| 11067 | for (unsigned i = 0; i != 16; ++i) { |
| 11068 | unsigned Idx = i + ShiftVal; |
| 11069 | if (Idx >= 16) Idx += NumElts - 16; // end of lane, switch operand. |
| 11070 | Indices[l + i] = Idx + l; |
| 11071 | } |
| 11072 | } |
| 11073 | |
| 11074 | llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts); |
| 11075 | Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast"); |
| 11076 | Value *Zero = llvm::Constant::getNullValue(VecTy); |
| 11077 | Value *SV = Builder.CreateShuffleVector(Cast, Zero, |
| 11078 | makeArrayRef(Indices, NumElts), |
| 11079 | "psrldq"); |
| 11080 | return Builder.CreateBitCast(SV, ResultType, "cast"); |
| 11081 | } |
| 11082 | case X86::BI__builtin_ia32_kshiftliqi: |
| 11083 | case X86::BI__builtin_ia32_kshiftlihi: |
| 11084 | case X86::BI__builtin_ia32_kshiftlisi: |
| 11085 | case X86::BI__builtin_ia32_kshiftlidi: { |
| 11086 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; |
| 11087 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 11088 | |
| 11089 | if (ShiftVal >= NumElts) |
| 11090 | return llvm::Constant::getNullValue(Ops[0]->getType()); |
| 11091 | |
| 11092 | Value *In = getMaskVecValue(*this, Ops[0], NumElts); |
| 11093 | |
| 11094 | uint32_t Indices[64]; |
| 11095 | for (unsigned i = 0; i != NumElts; ++i) |
| 11096 | Indices[i] = NumElts + i - ShiftVal; |
| 11097 | |
| 11098 | Value *Zero = llvm::Constant::getNullValue(In->getType()); |
| 11099 | Value *SV = Builder.CreateShuffleVector(Zero, In, |
| 11100 | makeArrayRef(Indices, NumElts), |
| 11101 | "kshiftl"); |
| 11102 | return Builder.CreateBitCast(SV, Ops[0]->getType()); |
| 11103 | } |
| 11104 | case X86::BI__builtin_ia32_kshiftriqi: |
| 11105 | case X86::BI__builtin_ia32_kshiftrihi: |
| 11106 | case X86::BI__builtin_ia32_kshiftrisi: |
| 11107 | case X86::BI__builtin_ia32_kshiftridi: { |
| 11108 | unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff; |
| 11109 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 11110 | |
| 11111 | if (ShiftVal >= NumElts) |
| 11112 | return llvm::Constant::getNullValue(Ops[0]->getType()); |
| 11113 | |
| 11114 | Value *In = getMaskVecValue(*this, Ops[0], NumElts); |
| 11115 | |
| 11116 | uint32_t Indices[64]; |
| 11117 | for (unsigned i = 0; i != NumElts; ++i) |
| 11118 | Indices[i] = i + ShiftVal; |
| 11119 | |
| 11120 | Value *Zero = llvm::Constant::getNullValue(In->getType()); |
| 11121 | Value *SV = Builder.CreateShuffleVector(In, Zero, |
| 11122 | makeArrayRef(Indices, NumElts), |
| 11123 | "kshiftr"); |
| 11124 | return Builder.CreateBitCast(SV, Ops[0]->getType()); |
| 11125 | } |
| 11126 | case X86::BI__builtin_ia32_movnti: |
| 11127 | case X86::BI__builtin_ia32_movnti64: |
| 11128 | case X86::BI__builtin_ia32_movntsd: |
| 11129 | case X86::BI__builtin_ia32_movntss: { |
| 11130 | llvm::MDNode *Node = llvm::MDNode::get( |
| 11131 | getLLVMContext(), llvm::ConstantAsMetadata::get(Builder.getInt32(1))); |
| 11132 | |
| 11133 | Value *Ptr = Ops[0]; |
| 11134 | Value *Src = Ops[1]; |
| 11135 | |
| 11136 | // Extract the 0'th element of the source vector. |
| 11137 | if (BuiltinID == X86::BI__builtin_ia32_movntsd || |
| 11138 | BuiltinID == X86::BI__builtin_ia32_movntss) |
| 11139 | Src = Builder.CreateExtractElement(Src, (uint64_t)0, "extract"); |
| 11140 | |
| 11141 | // Convert the type of the pointer to a pointer to the stored type. |
| 11142 | Value *BC = Builder.CreateBitCast( |
| 11143 | Ptr, llvm::PointerType::getUnqual(Src->getType()), "cast"); |
| 11144 | |
| 11145 | // Unaligned nontemporal store of the scalar value. |
| 11146 | StoreInst *SI = Builder.CreateDefaultAlignedStore(Src, BC); |
| 11147 | SI->setMetadata(CGM.getModule().getMDKindID("nontemporal"), Node); |
| 11148 | SI->setAlignment(1); |
| 11149 | return SI; |
| 11150 | } |
| 11151 | // Rotate is a special case of funnel shift - 1st 2 args are the same. |
| 11152 | case X86::BI__builtin_ia32_vprotb: |
| 11153 | case X86::BI__builtin_ia32_vprotw: |
| 11154 | case X86::BI__builtin_ia32_vprotd: |
| 11155 | case X86::BI__builtin_ia32_vprotq: |
| 11156 | case X86::BI__builtin_ia32_vprotbi: |
| 11157 | case X86::BI__builtin_ia32_vprotwi: |
| 11158 | case X86::BI__builtin_ia32_vprotdi: |
| 11159 | case X86::BI__builtin_ia32_vprotqi: |
| 11160 | case X86::BI__builtin_ia32_prold128: |
| 11161 | case X86::BI__builtin_ia32_prold256: |
| 11162 | case X86::BI__builtin_ia32_prold512: |
| 11163 | case X86::BI__builtin_ia32_prolq128: |
| 11164 | case X86::BI__builtin_ia32_prolq256: |
| 11165 | case X86::BI__builtin_ia32_prolq512: |
| 11166 | case X86::BI__builtin_ia32_prolvd128: |
| 11167 | case X86::BI__builtin_ia32_prolvd256: |
| 11168 | case X86::BI__builtin_ia32_prolvd512: |
| 11169 | case X86::BI__builtin_ia32_prolvq128: |
| 11170 | case X86::BI__builtin_ia32_prolvq256: |
| 11171 | case X86::BI__builtin_ia32_prolvq512: |
| 11172 | return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], false); |
| 11173 | case X86::BI__builtin_ia32_prord128: |
| 11174 | case X86::BI__builtin_ia32_prord256: |
| 11175 | case X86::BI__builtin_ia32_prord512: |
| 11176 | case X86::BI__builtin_ia32_prorq128: |
| 11177 | case X86::BI__builtin_ia32_prorq256: |
| 11178 | case X86::BI__builtin_ia32_prorq512: |
| 11179 | case X86::BI__builtin_ia32_prorvd128: |
| 11180 | case X86::BI__builtin_ia32_prorvd256: |
| 11181 | case X86::BI__builtin_ia32_prorvd512: |
| 11182 | case X86::BI__builtin_ia32_prorvq128: |
| 11183 | case X86::BI__builtin_ia32_prorvq256: |
| 11184 | case X86::BI__builtin_ia32_prorvq512: |
| 11185 | return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], true); |
| 11186 | case X86::BI__builtin_ia32_selectb_128: |
| 11187 | case X86::BI__builtin_ia32_selectb_256: |
| 11188 | case X86::BI__builtin_ia32_selectb_512: |
| 11189 | case X86::BI__builtin_ia32_selectw_128: |
| 11190 | case X86::BI__builtin_ia32_selectw_256: |
| 11191 | case X86::BI__builtin_ia32_selectw_512: |
| 11192 | case X86::BI__builtin_ia32_selectd_128: |
| 11193 | case X86::BI__builtin_ia32_selectd_256: |
| 11194 | case X86::BI__builtin_ia32_selectd_512: |
| 11195 | case X86::BI__builtin_ia32_selectq_128: |
| 11196 | case X86::BI__builtin_ia32_selectq_256: |
| 11197 | case X86::BI__builtin_ia32_selectq_512: |
| 11198 | case X86::BI__builtin_ia32_selectps_128: |
| 11199 | case X86::BI__builtin_ia32_selectps_256: |
| 11200 | case X86::BI__builtin_ia32_selectps_512: |
| 11201 | case X86::BI__builtin_ia32_selectpd_128: |
| 11202 | case X86::BI__builtin_ia32_selectpd_256: |
| 11203 | case X86::BI__builtin_ia32_selectpd_512: |
| 11204 | return EmitX86Select(*this, Ops[0], Ops[1], Ops[2]); |
| 11205 | case X86::BI__builtin_ia32_selectss_128: |
| 11206 | case X86::BI__builtin_ia32_selectsd_128: { |
| 11207 | Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0); |
| 11208 | Value *B = Builder.CreateExtractElement(Ops[2], (uint64_t)0); |
| 11209 | A = EmitX86ScalarSelect(*this, Ops[0], A, B); |
| 11210 | return Builder.CreateInsertElement(Ops[1], A, (uint64_t)0); |
| 11211 | } |
| 11212 | case X86::BI__builtin_ia32_cmpb128_mask: |
| 11213 | case X86::BI__builtin_ia32_cmpb256_mask: |
| 11214 | case X86::BI__builtin_ia32_cmpb512_mask: |
| 11215 | case X86::BI__builtin_ia32_cmpw128_mask: |
| 11216 | case X86::BI__builtin_ia32_cmpw256_mask: |
| 11217 | case X86::BI__builtin_ia32_cmpw512_mask: |
| 11218 | case X86::BI__builtin_ia32_cmpd128_mask: |
| 11219 | case X86::BI__builtin_ia32_cmpd256_mask: |
| 11220 | case X86::BI__builtin_ia32_cmpd512_mask: |
| 11221 | case X86::BI__builtin_ia32_cmpq128_mask: |
| 11222 | case X86::BI__builtin_ia32_cmpq256_mask: |
| 11223 | case X86::BI__builtin_ia32_cmpq512_mask: { |
| 11224 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; |
| 11225 | return EmitX86MaskedCompare(*this, CC, true, Ops); |
| 11226 | } |
| 11227 | case X86::BI__builtin_ia32_ucmpb128_mask: |
| 11228 | case X86::BI__builtin_ia32_ucmpb256_mask: |
| 11229 | case X86::BI__builtin_ia32_ucmpb512_mask: |
| 11230 | case X86::BI__builtin_ia32_ucmpw128_mask: |
| 11231 | case X86::BI__builtin_ia32_ucmpw256_mask: |
| 11232 | case X86::BI__builtin_ia32_ucmpw512_mask: |
| 11233 | case X86::BI__builtin_ia32_ucmpd128_mask: |
| 11234 | case X86::BI__builtin_ia32_ucmpd256_mask: |
| 11235 | case X86::BI__builtin_ia32_ucmpd512_mask: |
| 11236 | case X86::BI__builtin_ia32_ucmpq128_mask: |
| 11237 | case X86::BI__builtin_ia32_ucmpq256_mask: |
| 11238 | case X86::BI__builtin_ia32_ucmpq512_mask: { |
| 11239 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7; |
| 11240 | return EmitX86MaskedCompare(*this, CC, false, Ops); |
| 11241 | } |
| 11242 | case X86::BI__builtin_ia32_vpcomb: |
| 11243 | case X86::BI__builtin_ia32_vpcomw: |
| 11244 | case X86::BI__builtin_ia32_vpcomd: |
| 11245 | case X86::BI__builtin_ia32_vpcomq: |
| 11246 | return EmitX86vpcom(*this, Ops, true); |
| 11247 | case X86::BI__builtin_ia32_vpcomub: |
| 11248 | case X86::BI__builtin_ia32_vpcomuw: |
| 11249 | case X86::BI__builtin_ia32_vpcomud: |
| 11250 | case X86::BI__builtin_ia32_vpcomuq: |
| 11251 | return EmitX86vpcom(*this, Ops, false); |
| 11252 | |
| 11253 | case X86::BI__builtin_ia32_kortestcqi: |
| 11254 | case X86::BI__builtin_ia32_kortestchi: |
| 11255 | case X86::BI__builtin_ia32_kortestcsi: |
| 11256 | case X86::BI__builtin_ia32_kortestcdi: { |
| 11257 | Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops); |
| 11258 | Value *C = llvm::Constant::getAllOnesValue(Ops[0]->getType()); |
| 11259 | Value *Cmp = Builder.CreateICmpEQ(Or, C); |
| 11260 | return Builder.CreateZExt(Cmp, ConvertType(E->getType())); |
| 11261 | } |
| 11262 | case X86::BI__builtin_ia32_kortestzqi: |
| 11263 | case X86::BI__builtin_ia32_kortestzhi: |
| 11264 | case X86::BI__builtin_ia32_kortestzsi: |
| 11265 | case X86::BI__builtin_ia32_kortestzdi: { |
| 11266 | Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops); |
| 11267 | Value *C = llvm::Constant::getNullValue(Ops[0]->getType()); |
| 11268 | Value *Cmp = Builder.CreateICmpEQ(Or, C); |
| 11269 | return Builder.CreateZExt(Cmp, ConvertType(E->getType())); |
| 11270 | } |
| 11271 | |
| 11272 | case X86::BI__builtin_ia32_ktestcqi: |
| 11273 | case X86::BI__builtin_ia32_ktestzqi: |
| 11274 | case X86::BI__builtin_ia32_ktestchi: |
| 11275 | case X86::BI__builtin_ia32_ktestzhi: |
| 11276 | case X86::BI__builtin_ia32_ktestcsi: |
| 11277 | case X86::BI__builtin_ia32_ktestzsi: |
| 11278 | case X86::BI__builtin_ia32_ktestcdi: |
| 11279 | case X86::BI__builtin_ia32_ktestzdi: { |
| 11280 | Intrinsic::ID IID; |
| 11281 | switch (BuiltinID) { |
| 11282 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11282); |
| 11283 | case X86::BI__builtin_ia32_ktestcqi: |
| 11284 | IID = Intrinsic::x86_avx512_ktestc_b; |
| 11285 | break; |
| 11286 | case X86::BI__builtin_ia32_ktestzqi: |
| 11287 | IID = Intrinsic::x86_avx512_ktestz_b; |
| 11288 | break; |
| 11289 | case X86::BI__builtin_ia32_ktestchi: |
| 11290 | IID = Intrinsic::x86_avx512_ktestc_w; |
| 11291 | break; |
| 11292 | case X86::BI__builtin_ia32_ktestzhi: |
| 11293 | IID = Intrinsic::x86_avx512_ktestz_w; |
| 11294 | break; |
| 11295 | case X86::BI__builtin_ia32_ktestcsi: |
| 11296 | IID = Intrinsic::x86_avx512_ktestc_d; |
| 11297 | break; |
| 11298 | case X86::BI__builtin_ia32_ktestzsi: |
| 11299 | IID = Intrinsic::x86_avx512_ktestz_d; |
| 11300 | break; |
| 11301 | case X86::BI__builtin_ia32_ktestcdi: |
| 11302 | IID = Intrinsic::x86_avx512_ktestc_q; |
| 11303 | break; |
| 11304 | case X86::BI__builtin_ia32_ktestzdi: |
| 11305 | IID = Intrinsic::x86_avx512_ktestz_q; |
| 11306 | break; |
| 11307 | } |
| 11308 | |
| 11309 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 11310 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); |
| 11311 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); |
| 11312 | Function *Intr = CGM.getIntrinsic(IID); |
| 11313 | return Builder.CreateCall(Intr, {LHS, RHS}); |
| 11314 | } |
| 11315 | |
| 11316 | case X86::BI__builtin_ia32_kaddqi: |
| 11317 | case X86::BI__builtin_ia32_kaddhi: |
| 11318 | case X86::BI__builtin_ia32_kaddsi: |
| 11319 | case X86::BI__builtin_ia32_kadddi: { |
| 11320 | Intrinsic::ID IID; |
| 11321 | switch (BuiltinID) { |
| 11322 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11322); |
| 11323 | case X86::BI__builtin_ia32_kaddqi: |
| 11324 | IID = Intrinsic::x86_avx512_kadd_b; |
| 11325 | break; |
| 11326 | case X86::BI__builtin_ia32_kaddhi: |
| 11327 | IID = Intrinsic::x86_avx512_kadd_w; |
| 11328 | break; |
| 11329 | case X86::BI__builtin_ia32_kaddsi: |
| 11330 | IID = Intrinsic::x86_avx512_kadd_d; |
| 11331 | break; |
| 11332 | case X86::BI__builtin_ia32_kadddi: |
| 11333 | IID = Intrinsic::x86_avx512_kadd_q; |
| 11334 | break; |
| 11335 | } |
| 11336 | |
| 11337 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 11338 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); |
| 11339 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); |
| 11340 | Function *Intr = CGM.getIntrinsic(IID); |
| 11341 | Value *Res = Builder.CreateCall(Intr, {LHS, RHS}); |
| 11342 | return Builder.CreateBitCast(Res, Ops[0]->getType()); |
| 11343 | } |
| 11344 | case X86::BI__builtin_ia32_kandqi: |
| 11345 | case X86::BI__builtin_ia32_kandhi: |
| 11346 | case X86::BI__builtin_ia32_kandsi: |
| 11347 | case X86::BI__builtin_ia32_kanddi: |
| 11348 | return EmitX86MaskLogic(*this, Instruction::And, Ops); |
| 11349 | case X86::BI__builtin_ia32_kandnqi: |
| 11350 | case X86::BI__builtin_ia32_kandnhi: |
| 11351 | case X86::BI__builtin_ia32_kandnsi: |
| 11352 | case X86::BI__builtin_ia32_kandndi: |
| 11353 | return EmitX86MaskLogic(*this, Instruction::And, Ops, true); |
| 11354 | case X86::BI__builtin_ia32_korqi: |
| 11355 | case X86::BI__builtin_ia32_korhi: |
| 11356 | case X86::BI__builtin_ia32_korsi: |
| 11357 | case X86::BI__builtin_ia32_kordi: |
| 11358 | return EmitX86MaskLogic(*this, Instruction::Or, Ops); |
| 11359 | case X86::BI__builtin_ia32_kxnorqi: |
| 11360 | case X86::BI__builtin_ia32_kxnorhi: |
| 11361 | case X86::BI__builtin_ia32_kxnorsi: |
| 11362 | case X86::BI__builtin_ia32_kxnordi: |
| 11363 | return EmitX86MaskLogic(*this, Instruction::Xor, Ops, true); |
| 11364 | case X86::BI__builtin_ia32_kxorqi: |
| 11365 | case X86::BI__builtin_ia32_kxorhi: |
| 11366 | case X86::BI__builtin_ia32_kxorsi: |
| 11367 | case X86::BI__builtin_ia32_kxordi: |
| 11368 | return EmitX86MaskLogic(*this, Instruction::Xor, Ops); |
| 11369 | case X86::BI__builtin_ia32_knotqi: |
| 11370 | case X86::BI__builtin_ia32_knothi: |
| 11371 | case X86::BI__builtin_ia32_knotsi: |
| 11372 | case X86::BI__builtin_ia32_knotdi: { |
| 11373 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 11374 | Value *Res = getMaskVecValue(*this, Ops[0], NumElts); |
| 11375 | return Builder.CreateBitCast(Builder.CreateNot(Res), |
| 11376 | Ops[0]->getType()); |
| 11377 | } |
| 11378 | case X86::BI__builtin_ia32_kmovb: |
| 11379 | case X86::BI__builtin_ia32_kmovw: |
| 11380 | case X86::BI__builtin_ia32_kmovd: |
| 11381 | case X86::BI__builtin_ia32_kmovq: { |
| 11382 | // Bitcast to vXi1 type and then back to integer. This gets the mask |
| 11383 | // register type into the IR, but might be optimized out depending on |
| 11384 | // what's around it. |
| 11385 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 11386 | Value *Res = getMaskVecValue(*this, Ops[0], NumElts); |
| 11387 | return Builder.CreateBitCast(Res, Ops[0]->getType()); |
| 11388 | } |
| 11389 | |
| 11390 | case X86::BI__builtin_ia32_kunpckdi: |
| 11391 | case X86::BI__builtin_ia32_kunpcksi: |
| 11392 | case X86::BI__builtin_ia32_kunpckhi: { |
| 11393 | unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth(); |
| 11394 | Value *LHS = getMaskVecValue(*this, Ops[0], NumElts); |
| 11395 | Value *RHS = getMaskVecValue(*this, Ops[1], NumElts); |
| 11396 | uint32_t Indices[64]; |
| 11397 | for (unsigned i = 0; i != NumElts; ++i) |
| 11398 | Indices[i] = i; |
| 11399 | |
| 11400 | // First extract half of each vector. This gives better codegen than |
| 11401 | // doing it in a single shuffle. |
| 11402 | LHS = Builder.CreateShuffleVector(LHS, LHS, |
| 11403 | makeArrayRef(Indices, NumElts / 2)); |
| 11404 | RHS = Builder.CreateShuffleVector(RHS, RHS, |
| 11405 | makeArrayRef(Indices, NumElts / 2)); |
| 11406 | // Concat the vectors. |
| 11407 | // NOTE: Operands are swapped to match the intrinsic definition. |
| 11408 | Value *Res = Builder.CreateShuffleVector(RHS, LHS, |
| 11409 | makeArrayRef(Indices, NumElts)); |
| 11410 | return Builder.CreateBitCast(Res, Ops[0]->getType()); |
| 11411 | } |
| 11412 | |
| 11413 | case X86::BI__builtin_ia32_vplzcntd_128: |
| 11414 | case X86::BI__builtin_ia32_vplzcntd_256: |
| 11415 | case X86::BI__builtin_ia32_vplzcntd_512: |
| 11416 | case X86::BI__builtin_ia32_vplzcntq_128: |
| 11417 | case X86::BI__builtin_ia32_vplzcntq_256: |
| 11418 | case X86::BI__builtin_ia32_vplzcntq_512: { |
| 11419 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType()); |
| 11420 | return Builder.CreateCall(F, {Ops[0],Builder.getInt1(false)}); |
| 11421 | } |
| 11422 | case X86::BI__builtin_ia32_sqrtss: |
| 11423 | case X86::BI__builtin_ia32_sqrtsd: { |
| 11424 | Value *A = Builder.CreateExtractElement(Ops[0], (uint64_t)0); |
| 11425 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType()); |
| 11426 | A = Builder.CreateCall(F, {A}); |
| 11427 | return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0); |
| 11428 | } |
| 11429 | case X86::BI__builtin_ia32_sqrtsd_round_mask: |
| 11430 | case X86::BI__builtin_ia32_sqrtss_round_mask: { |
| 11431 | unsigned CC = cast<llvm::ConstantInt>(Ops[4])->getZExtValue(); |
| 11432 | // Support only if the rounding mode is 4 (AKA CUR_DIRECTION), |
| 11433 | // otherwise keep the intrinsic. |
| 11434 | if (CC != 4) { |
| 11435 | Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtsd_round_mask ? |
| 11436 | Intrinsic::x86_avx512_mask_sqrt_sd : |
| 11437 | Intrinsic::x86_avx512_mask_sqrt_ss; |
| 11438 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); |
| 11439 | } |
| 11440 | Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0); |
| 11441 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType()); |
| 11442 | A = Builder.CreateCall(F, A); |
| 11443 | Value *Src = Builder.CreateExtractElement(Ops[2], (uint64_t)0); |
| 11444 | A = EmitX86ScalarSelect(*this, Ops[3], A, Src); |
| 11445 | return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0); |
| 11446 | } |
| 11447 | case X86::BI__builtin_ia32_sqrtpd256: |
| 11448 | case X86::BI__builtin_ia32_sqrtpd: |
| 11449 | case X86::BI__builtin_ia32_sqrtps256: |
| 11450 | case X86::BI__builtin_ia32_sqrtps: |
| 11451 | case X86::BI__builtin_ia32_sqrtps512: |
| 11452 | case X86::BI__builtin_ia32_sqrtpd512: { |
| 11453 | if (Ops.size() == 2) { |
| 11454 | unsigned CC = cast<llvm::ConstantInt>(Ops[1])->getZExtValue(); |
| 11455 | // Support only if the rounding mode is 4 (AKA CUR_DIRECTION), |
| 11456 | // otherwise keep the intrinsic. |
| 11457 | if (CC != 4) { |
| 11458 | Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtps512 ? |
| 11459 | Intrinsic::x86_avx512_sqrt_ps_512 : |
| 11460 | Intrinsic::x86_avx512_sqrt_pd_512; |
| 11461 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); |
| 11462 | } |
| 11463 | } |
| 11464 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, Ops[0]->getType()); |
| 11465 | return Builder.CreateCall(F, Ops[0]); |
| 11466 | } |
| 11467 | case X86::BI__builtin_ia32_pabsb128: |
| 11468 | case X86::BI__builtin_ia32_pabsw128: |
| 11469 | case X86::BI__builtin_ia32_pabsd128: |
| 11470 | case X86::BI__builtin_ia32_pabsb256: |
| 11471 | case X86::BI__builtin_ia32_pabsw256: |
| 11472 | case X86::BI__builtin_ia32_pabsd256: |
| 11473 | case X86::BI__builtin_ia32_pabsq128: |
| 11474 | case X86::BI__builtin_ia32_pabsq256: |
| 11475 | case X86::BI__builtin_ia32_pabsb512: |
| 11476 | case X86::BI__builtin_ia32_pabsw512: |
| 11477 | case X86::BI__builtin_ia32_pabsd512: |
| 11478 | case X86::BI__builtin_ia32_pabsq512: |
| 11479 | return EmitX86Abs(*this, Ops); |
| 11480 | |
| 11481 | case X86::BI__builtin_ia32_pmaxsb128: |
| 11482 | case X86::BI__builtin_ia32_pmaxsw128: |
| 11483 | case X86::BI__builtin_ia32_pmaxsd128: |
| 11484 | case X86::BI__builtin_ia32_pmaxsq128: |
| 11485 | case X86::BI__builtin_ia32_pmaxsb256: |
| 11486 | case X86::BI__builtin_ia32_pmaxsw256: |
| 11487 | case X86::BI__builtin_ia32_pmaxsd256: |
| 11488 | case X86::BI__builtin_ia32_pmaxsq256: |
| 11489 | case X86::BI__builtin_ia32_pmaxsb512: |
| 11490 | case X86::BI__builtin_ia32_pmaxsw512: |
| 11491 | case X86::BI__builtin_ia32_pmaxsd512: |
| 11492 | case X86::BI__builtin_ia32_pmaxsq512: |
| 11493 | return EmitX86MinMax(*this, ICmpInst::ICMP_SGT, Ops); |
| 11494 | case X86::BI__builtin_ia32_pmaxub128: |
| 11495 | case X86::BI__builtin_ia32_pmaxuw128: |
| 11496 | case X86::BI__builtin_ia32_pmaxud128: |
| 11497 | case X86::BI__builtin_ia32_pmaxuq128: |
| 11498 | case X86::BI__builtin_ia32_pmaxub256: |
| 11499 | case X86::BI__builtin_ia32_pmaxuw256: |
| 11500 | case X86::BI__builtin_ia32_pmaxud256: |
| 11501 | case X86::BI__builtin_ia32_pmaxuq256: |
| 11502 | case X86::BI__builtin_ia32_pmaxub512: |
| 11503 | case X86::BI__builtin_ia32_pmaxuw512: |
| 11504 | case X86::BI__builtin_ia32_pmaxud512: |
| 11505 | case X86::BI__builtin_ia32_pmaxuq512: |
| 11506 | return EmitX86MinMax(*this, ICmpInst::ICMP_UGT, Ops); |
| 11507 | case X86::BI__builtin_ia32_pminsb128: |
| 11508 | case X86::BI__builtin_ia32_pminsw128: |
| 11509 | case X86::BI__builtin_ia32_pminsd128: |
| 11510 | case X86::BI__builtin_ia32_pminsq128: |
| 11511 | case X86::BI__builtin_ia32_pminsb256: |
| 11512 | case X86::BI__builtin_ia32_pminsw256: |
| 11513 | case X86::BI__builtin_ia32_pminsd256: |
| 11514 | case X86::BI__builtin_ia32_pminsq256: |
| 11515 | case X86::BI__builtin_ia32_pminsb512: |
| 11516 | case X86::BI__builtin_ia32_pminsw512: |
| 11517 | case X86::BI__builtin_ia32_pminsd512: |
| 11518 | case X86::BI__builtin_ia32_pminsq512: |
| 11519 | return EmitX86MinMax(*this, ICmpInst::ICMP_SLT, Ops); |
| 11520 | case X86::BI__builtin_ia32_pminub128: |
| 11521 | case X86::BI__builtin_ia32_pminuw128: |
| 11522 | case X86::BI__builtin_ia32_pminud128: |
| 11523 | case X86::BI__builtin_ia32_pminuq128: |
| 11524 | case X86::BI__builtin_ia32_pminub256: |
| 11525 | case X86::BI__builtin_ia32_pminuw256: |
| 11526 | case X86::BI__builtin_ia32_pminud256: |
| 11527 | case X86::BI__builtin_ia32_pminuq256: |
| 11528 | case X86::BI__builtin_ia32_pminub512: |
| 11529 | case X86::BI__builtin_ia32_pminuw512: |
| 11530 | case X86::BI__builtin_ia32_pminud512: |
| 11531 | case X86::BI__builtin_ia32_pminuq512: |
| 11532 | return EmitX86MinMax(*this, ICmpInst::ICMP_ULT, Ops); |
| 11533 | |
| 11534 | case X86::BI__builtin_ia32_pmuludq128: |
| 11535 | case X86::BI__builtin_ia32_pmuludq256: |
| 11536 | case X86::BI__builtin_ia32_pmuludq512: |
| 11537 | return EmitX86Muldq(*this, /*IsSigned*/false, Ops); |
| 11538 | |
| 11539 | case X86::BI__builtin_ia32_pmuldq128: |
| 11540 | case X86::BI__builtin_ia32_pmuldq256: |
| 11541 | case X86::BI__builtin_ia32_pmuldq512: |
| 11542 | return EmitX86Muldq(*this, /*IsSigned*/true, Ops); |
| 11543 | |
| 11544 | case X86::BI__builtin_ia32_pternlogd512_mask: |
| 11545 | case X86::BI__builtin_ia32_pternlogq512_mask: |
| 11546 | case X86::BI__builtin_ia32_pternlogd128_mask: |
| 11547 | case X86::BI__builtin_ia32_pternlogd256_mask: |
| 11548 | case X86::BI__builtin_ia32_pternlogq128_mask: |
| 11549 | case X86::BI__builtin_ia32_pternlogq256_mask: |
| 11550 | return EmitX86Ternlog(*this, /*ZeroMask*/false, Ops); |
| 11551 | |
| 11552 | case X86::BI__builtin_ia32_pternlogd512_maskz: |
| 11553 | case X86::BI__builtin_ia32_pternlogq512_maskz: |
| 11554 | case X86::BI__builtin_ia32_pternlogd128_maskz: |
| 11555 | case X86::BI__builtin_ia32_pternlogd256_maskz: |
| 11556 | case X86::BI__builtin_ia32_pternlogq128_maskz: |
| 11557 | case X86::BI__builtin_ia32_pternlogq256_maskz: |
| 11558 | return EmitX86Ternlog(*this, /*ZeroMask*/true, Ops); |
| 11559 | |
| 11560 | case X86::BI__builtin_ia32_vpshldd128: |
| 11561 | case X86::BI__builtin_ia32_vpshldd256: |
| 11562 | case X86::BI__builtin_ia32_vpshldd512: |
| 11563 | case X86::BI__builtin_ia32_vpshldq128: |
| 11564 | case X86::BI__builtin_ia32_vpshldq256: |
| 11565 | case X86::BI__builtin_ia32_vpshldq512: |
| 11566 | case X86::BI__builtin_ia32_vpshldw128: |
| 11567 | case X86::BI__builtin_ia32_vpshldw256: |
| 11568 | case X86::BI__builtin_ia32_vpshldw512: |
| 11569 | return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); |
| 11570 | |
| 11571 | case X86::BI__builtin_ia32_vpshrdd128: |
| 11572 | case X86::BI__builtin_ia32_vpshrdd256: |
| 11573 | case X86::BI__builtin_ia32_vpshrdd512: |
| 11574 | case X86::BI__builtin_ia32_vpshrdq128: |
| 11575 | case X86::BI__builtin_ia32_vpshrdq256: |
| 11576 | case X86::BI__builtin_ia32_vpshrdq512: |
| 11577 | case X86::BI__builtin_ia32_vpshrdw128: |
| 11578 | case X86::BI__builtin_ia32_vpshrdw256: |
| 11579 | case X86::BI__builtin_ia32_vpshrdw512: |
| 11580 | // Ops 0 and 1 are swapped. |
| 11581 | return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); |
| 11582 | |
| 11583 | case X86::BI__builtin_ia32_vpshldvd128: |
| 11584 | case X86::BI__builtin_ia32_vpshldvd256: |
| 11585 | case X86::BI__builtin_ia32_vpshldvd512: |
| 11586 | case X86::BI__builtin_ia32_vpshldvq128: |
| 11587 | case X86::BI__builtin_ia32_vpshldvq256: |
| 11588 | case X86::BI__builtin_ia32_vpshldvq512: |
| 11589 | case X86::BI__builtin_ia32_vpshldvw128: |
| 11590 | case X86::BI__builtin_ia32_vpshldvw256: |
| 11591 | case X86::BI__builtin_ia32_vpshldvw512: |
| 11592 | return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); |
| 11593 | |
| 11594 | case X86::BI__builtin_ia32_vpshrdvd128: |
| 11595 | case X86::BI__builtin_ia32_vpshrdvd256: |
| 11596 | case X86::BI__builtin_ia32_vpshrdvd512: |
| 11597 | case X86::BI__builtin_ia32_vpshrdvq128: |
| 11598 | case X86::BI__builtin_ia32_vpshrdvq256: |
| 11599 | case X86::BI__builtin_ia32_vpshrdvq512: |
| 11600 | case X86::BI__builtin_ia32_vpshrdvw128: |
| 11601 | case X86::BI__builtin_ia32_vpshrdvw256: |
| 11602 | case X86::BI__builtin_ia32_vpshrdvw512: |
| 11603 | // Ops 0 and 1 are swapped. |
| 11604 | return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); |
| 11605 | |
| 11606 | // 3DNow! |
| 11607 | case X86::BI__builtin_ia32_pswapdsf: |
| 11608 | case X86::BI__builtin_ia32_pswapdsi: { |
| 11609 | llvm::Type *MMXTy = llvm::Type::getX86_MMXTy(getLLVMContext()); |
| 11610 | Ops[0] = Builder.CreateBitCast(Ops[0], MMXTy, "cast"); |
| 11611 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_3dnowa_pswapd); |
| 11612 | return Builder.CreateCall(F, Ops, "pswapd"); |
| 11613 | } |
| 11614 | case X86::BI__builtin_ia32_rdrand16_step: |
| 11615 | case X86::BI__builtin_ia32_rdrand32_step: |
| 11616 | case X86::BI__builtin_ia32_rdrand64_step: |
| 11617 | case X86::BI__builtin_ia32_rdseed16_step: |
| 11618 | case X86::BI__builtin_ia32_rdseed32_step: |
| 11619 | case X86::BI__builtin_ia32_rdseed64_step: { |
| 11620 | Intrinsic::ID ID; |
| 11621 | switch (BuiltinID) { |
| 11622 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11622); |
| 11623 | case X86::BI__builtin_ia32_rdrand16_step: |
| 11624 | ID = Intrinsic::x86_rdrand_16; |
| 11625 | break; |
| 11626 | case X86::BI__builtin_ia32_rdrand32_step: |
| 11627 | ID = Intrinsic::x86_rdrand_32; |
| 11628 | break; |
| 11629 | case X86::BI__builtin_ia32_rdrand64_step: |
| 11630 | ID = Intrinsic::x86_rdrand_64; |
| 11631 | break; |
| 11632 | case X86::BI__builtin_ia32_rdseed16_step: |
| 11633 | ID = Intrinsic::x86_rdseed_16; |
| 11634 | break; |
| 11635 | case X86::BI__builtin_ia32_rdseed32_step: |
| 11636 | ID = Intrinsic::x86_rdseed_32; |
| 11637 | break; |
| 11638 | case X86::BI__builtin_ia32_rdseed64_step: |
| 11639 | ID = Intrinsic::x86_rdseed_64; |
| 11640 | break; |
| 11641 | } |
| 11642 | |
| 11643 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID)); |
| 11644 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 0), |
| 11645 | Ops[0]); |
| 11646 | return Builder.CreateExtractValue(Call, 1); |
| 11647 | } |
| 11648 | case X86::BI__builtin_ia32_addcarryx_u32: |
| 11649 | case X86::BI__builtin_ia32_addcarryx_u64: |
| 11650 | case X86::BI__builtin_ia32_subborrow_u32: |
| 11651 | case X86::BI__builtin_ia32_subborrow_u64: { |
| 11652 | Intrinsic::ID IID; |
| 11653 | switch (BuiltinID) { |
| 11654 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11654); |
| 11655 | case X86::BI__builtin_ia32_addcarryx_u32: |
| 11656 | IID = Intrinsic::x86_addcarry_32; |
| 11657 | break; |
| 11658 | case X86::BI__builtin_ia32_addcarryx_u64: |
| 11659 | IID = Intrinsic::x86_addcarry_64; |
| 11660 | break; |
| 11661 | case X86::BI__builtin_ia32_subborrow_u32: |
| 11662 | IID = Intrinsic::x86_subborrow_32; |
| 11663 | break; |
| 11664 | case X86::BI__builtin_ia32_subborrow_u64: |
| 11665 | IID = Intrinsic::x86_subborrow_64; |
| 11666 | break; |
| 11667 | } |
| 11668 | |
| 11669 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID), |
| 11670 | { Ops[0], Ops[1], Ops[2] }); |
| 11671 | Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1), |
| 11672 | Ops[3]); |
| 11673 | return Builder.CreateExtractValue(Call, 0); |
| 11674 | } |
| 11675 | |
| 11676 | case X86::BI__builtin_ia32_fpclassps128_mask: |
| 11677 | case X86::BI__builtin_ia32_fpclassps256_mask: |
| 11678 | case X86::BI__builtin_ia32_fpclassps512_mask: |
| 11679 | case X86::BI__builtin_ia32_fpclasspd128_mask: |
| 11680 | case X86::BI__builtin_ia32_fpclasspd256_mask: |
| 11681 | case X86::BI__builtin_ia32_fpclasspd512_mask: { |
| 11682 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 11683 | Value *MaskIn = Ops[2]; |
| 11684 | Ops.erase(&Ops[2]); |
| 11685 | |
| 11686 | Intrinsic::ID ID; |
| 11687 | switch (BuiltinID) { |
| 11688 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11688); |
| 11689 | case X86::BI__builtin_ia32_fpclassps128_mask: |
| 11690 | ID = Intrinsic::x86_avx512_fpclass_ps_128; |
| 11691 | break; |
| 11692 | case X86::BI__builtin_ia32_fpclassps256_mask: |
| 11693 | ID = Intrinsic::x86_avx512_fpclass_ps_256; |
| 11694 | break; |
| 11695 | case X86::BI__builtin_ia32_fpclassps512_mask: |
| 11696 | ID = Intrinsic::x86_avx512_fpclass_ps_512; |
| 11697 | break; |
| 11698 | case X86::BI__builtin_ia32_fpclasspd128_mask: |
| 11699 | ID = Intrinsic::x86_avx512_fpclass_pd_128; |
| 11700 | break; |
| 11701 | case X86::BI__builtin_ia32_fpclasspd256_mask: |
| 11702 | ID = Intrinsic::x86_avx512_fpclass_pd_256; |
| 11703 | break; |
| 11704 | case X86::BI__builtin_ia32_fpclasspd512_mask: |
| 11705 | ID = Intrinsic::x86_avx512_fpclass_pd_512; |
| 11706 | break; |
| 11707 | } |
| 11708 | |
| 11709 | Value *Fpclass = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); |
| 11710 | return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn); |
| 11711 | } |
| 11712 | |
| 11713 | case X86::BI__builtin_ia32_vp2intersect_q_512: |
| 11714 | case X86::BI__builtin_ia32_vp2intersect_q_256: |
| 11715 | case X86::BI__builtin_ia32_vp2intersect_q_128: |
| 11716 | case X86::BI__builtin_ia32_vp2intersect_d_512: |
| 11717 | case X86::BI__builtin_ia32_vp2intersect_d_256: |
| 11718 | case X86::BI__builtin_ia32_vp2intersect_d_128: { |
| 11719 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 11720 | Intrinsic::ID ID; |
| 11721 | |
| 11722 | switch (BuiltinID) { |
| 11723 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11723); |
| 11724 | case X86::BI__builtin_ia32_vp2intersect_q_512: |
| 11725 | ID = Intrinsic::x86_avx512_vp2intersect_q_512; |
| 11726 | break; |
| 11727 | case X86::BI__builtin_ia32_vp2intersect_q_256: |
| 11728 | ID = Intrinsic::x86_avx512_vp2intersect_q_256; |
| 11729 | break; |
| 11730 | case X86::BI__builtin_ia32_vp2intersect_q_128: |
| 11731 | ID = Intrinsic::x86_avx512_vp2intersect_q_128; |
| 11732 | break; |
| 11733 | case X86::BI__builtin_ia32_vp2intersect_d_512: |
| 11734 | ID = Intrinsic::x86_avx512_vp2intersect_d_512; |
| 11735 | break; |
| 11736 | case X86::BI__builtin_ia32_vp2intersect_d_256: |
| 11737 | ID = Intrinsic::x86_avx512_vp2intersect_d_256; |
| 11738 | break; |
| 11739 | case X86::BI__builtin_ia32_vp2intersect_d_128: |
| 11740 | ID = Intrinsic::x86_avx512_vp2intersect_d_128; |
| 11741 | break; |
| 11742 | } |
| 11743 | |
| 11744 | Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID), {Ops[0], Ops[1]}); |
| 11745 | Value *Result = Builder.CreateExtractValue(Call, 0); |
| 11746 | Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr); |
| 11747 | Value *Store = Builder.CreateDefaultAlignedStore(Result, Ops[2]); |
Value stored to 'Store' during its initialization is never read | |
| 11748 | |
| 11749 | Result = Builder.CreateExtractValue(Call, 1); |
| 11750 | Result = EmitX86MaskedCompareResult(*this, Result, NumElts, nullptr); |
| 11751 | Store = Builder.CreateDefaultAlignedStore(Result, Ops[3]); |
| 11752 | return Store; |
| 11753 | } |
| 11754 | |
| 11755 | case X86::BI__builtin_ia32_vpmultishiftqb128: |
| 11756 | case X86::BI__builtin_ia32_vpmultishiftqb256: |
| 11757 | case X86::BI__builtin_ia32_vpmultishiftqb512: { |
| 11758 | Intrinsic::ID ID; |
| 11759 | switch (BuiltinID) { |
| 11760 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11760); |
| 11761 | case X86::BI__builtin_ia32_vpmultishiftqb128: |
| 11762 | ID = Intrinsic::x86_avx512_pmultishift_qb_128; |
| 11763 | break; |
| 11764 | case X86::BI__builtin_ia32_vpmultishiftqb256: |
| 11765 | ID = Intrinsic::x86_avx512_pmultishift_qb_256; |
| 11766 | break; |
| 11767 | case X86::BI__builtin_ia32_vpmultishiftqb512: |
| 11768 | ID = Intrinsic::x86_avx512_pmultishift_qb_512; |
| 11769 | break; |
| 11770 | } |
| 11771 | |
| 11772 | return Builder.CreateCall(CGM.getIntrinsic(ID), Ops); |
| 11773 | } |
| 11774 | |
| 11775 | case X86::BI__builtin_ia32_vpshufbitqmb128_mask: |
| 11776 | case X86::BI__builtin_ia32_vpshufbitqmb256_mask: |
| 11777 | case X86::BI__builtin_ia32_vpshufbitqmb512_mask: { |
| 11778 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 11779 | Value *MaskIn = Ops[2]; |
| 11780 | Ops.erase(&Ops[2]); |
| 11781 | |
| 11782 | Intrinsic::ID ID; |
| 11783 | switch (BuiltinID) { |
| 11784 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11784); |
| 11785 | case X86::BI__builtin_ia32_vpshufbitqmb128_mask: |
| 11786 | ID = Intrinsic::x86_avx512_vpshufbitqmb_128; |
| 11787 | break; |
| 11788 | case X86::BI__builtin_ia32_vpshufbitqmb256_mask: |
| 11789 | ID = Intrinsic::x86_avx512_vpshufbitqmb_256; |
| 11790 | break; |
| 11791 | case X86::BI__builtin_ia32_vpshufbitqmb512_mask: |
| 11792 | ID = Intrinsic::x86_avx512_vpshufbitqmb_512; |
| 11793 | break; |
| 11794 | } |
| 11795 | |
| 11796 | Value *Shufbit = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); |
| 11797 | return EmitX86MaskedCompareResult(*this, Shufbit, NumElts, MaskIn); |
| 11798 | } |
| 11799 | |
| 11800 | // packed comparison intrinsics |
| 11801 | case X86::BI__builtin_ia32_cmpeqps: |
| 11802 | case X86::BI__builtin_ia32_cmpeqpd: |
| 11803 | return getVectorFCmpIR(CmpInst::FCMP_OEQ); |
| 11804 | case X86::BI__builtin_ia32_cmpltps: |
| 11805 | case X86::BI__builtin_ia32_cmpltpd: |
| 11806 | return getVectorFCmpIR(CmpInst::FCMP_OLT); |
| 11807 | case X86::BI__builtin_ia32_cmpleps: |
| 11808 | case X86::BI__builtin_ia32_cmplepd: |
| 11809 | return getVectorFCmpIR(CmpInst::FCMP_OLE); |
| 11810 | case X86::BI__builtin_ia32_cmpunordps: |
| 11811 | case X86::BI__builtin_ia32_cmpunordpd: |
| 11812 | return getVectorFCmpIR(CmpInst::FCMP_UNO); |
| 11813 | case X86::BI__builtin_ia32_cmpneqps: |
| 11814 | case X86::BI__builtin_ia32_cmpneqpd: |
| 11815 | return getVectorFCmpIR(CmpInst::FCMP_UNE); |
| 11816 | case X86::BI__builtin_ia32_cmpnltps: |
| 11817 | case X86::BI__builtin_ia32_cmpnltpd: |
| 11818 | return getVectorFCmpIR(CmpInst::FCMP_UGE); |
| 11819 | case X86::BI__builtin_ia32_cmpnleps: |
| 11820 | case X86::BI__builtin_ia32_cmpnlepd: |
| 11821 | return getVectorFCmpIR(CmpInst::FCMP_UGT); |
| 11822 | case X86::BI__builtin_ia32_cmpordps: |
| 11823 | case X86::BI__builtin_ia32_cmpordpd: |
| 11824 | return getVectorFCmpIR(CmpInst::FCMP_ORD); |
| 11825 | case X86::BI__builtin_ia32_cmpps: |
| 11826 | case X86::BI__builtin_ia32_cmpps256: |
| 11827 | case X86::BI__builtin_ia32_cmppd: |
| 11828 | case X86::BI__builtin_ia32_cmppd256: |
| 11829 | case X86::BI__builtin_ia32_cmpps128_mask: |
| 11830 | case X86::BI__builtin_ia32_cmpps256_mask: |
| 11831 | case X86::BI__builtin_ia32_cmpps512_mask: |
| 11832 | case X86::BI__builtin_ia32_cmppd128_mask: |
| 11833 | case X86::BI__builtin_ia32_cmppd256_mask: |
| 11834 | case X86::BI__builtin_ia32_cmppd512_mask: { |
| 11835 | // Lowering vector comparisons to fcmp instructions, while |
| 11836 | // ignoring signalling behaviour requested |
| 11837 | // ignoring rounding mode requested |
| 11838 | // This is is only possible as long as FENV_ACCESS is not implemented. |
| 11839 | // See also: https://reviews.llvm.org/D45616 |
| 11840 | |
| 11841 | // The third argument is the comparison condition, and integer in the |
| 11842 | // range [0, 31] |
| 11843 | unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x1f; |
| 11844 | |
| 11845 | // Lowering to IR fcmp instruction. |
| 11846 | // Ignoring requested signaling behaviour, |
| 11847 | // e.g. both _CMP_GT_OS & _CMP_GT_OQ are translated to FCMP_OGT. |
| 11848 | FCmpInst::Predicate Pred; |
| 11849 | switch (CC) { |
| 11850 | case 0x00: Pred = FCmpInst::FCMP_OEQ; break; |
| 11851 | case 0x01: Pred = FCmpInst::FCMP_OLT; break; |
| 11852 | case 0x02: Pred = FCmpInst::FCMP_OLE; break; |
| 11853 | case 0x03: Pred = FCmpInst::FCMP_UNO; break; |
| 11854 | case 0x04: Pred = FCmpInst::FCMP_UNE; break; |
| 11855 | case 0x05: Pred = FCmpInst::FCMP_UGE; break; |
| 11856 | case 0x06: Pred = FCmpInst::FCMP_UGT; break; |
| 11857 | case 0x07: Pred = FCmpInst::FCMP_ORD; break; |
| 11858 | case 0x08: Pred = FCmpInst::FCMP_UEQ; break; |
| 11859 | case 0x09: Pred = FCmpInst::FCMP_ULT; break; |
| 11860 | case 0x0a: Pred = FCmpInst::FCMP_ULE; break; |
| 11861 | case 0x0b: Pred = FCmpInst::FCMP_FALSE; break; |
| 11862 | case 0x0c: Pred = FCmpInst::FCMP_ONE; break; |
| 11863 | case 0x0d: Pred = FCmpInst::FCMP_OGE; break; |
| 11864 | case 0x0e: Pred = FCmpInst::FCMP_OGT; break; |
| 11865 | case 0x0f: Pred = FCmpInst::FCMP_TRUE; break; |
| 11866 | case 0x10: Pred = FCmpInst::FCMP_OEQ; break; |
| 11867 | case 0x11: Pred = FCmpInst::FCMP_OLT; break; |
| 11868 | case 0x12: Pred = FCmpInst::FCMP_OLE; break; |
| 11869 | case 0x13: Pred = FCmpInst::FCMP_UNO; break; |
| 11870 | case 0x14: Pred = FCmpInst::FCMP_UNE; break; |
| 11871 | case 0x15: Pred = FCmpInst::FCMP_UGE; break; |
| 11872 | case 0x16: Pred = FCmpInst::FCMP_UGT; break; |
| 11873 | case 0x17: Pred = FCmpInst::FCMP_ORD; break; |
| 11874 | case 0x18: Pred = FCmpInst::FCMP_UEQ; break; |
| 11875 | case 0x19: Pred = FCmpInst::FCMP_ULT; break; |
| 11876 | case 0x1a: Pred = FCmpInst::FCMP_ULE; break; |
| 11877 | case 0x1b: Pred = FCmpInst::FCMP_FALSE; break; |
| 11878 | case 0x1c: Pred = FCmpInst::FCMP_ONE; break; |
| 11879 | case 0x1d: Pred = FCmpInst::FCMP_OGE; break; |
| 11880 | case 0x1e: Pred = FCmpInst::FCMP_OGT; break; |
| 11881 | case 0x1f: Pred = FCmpInst::FCMP_TRUE; break; |
| 11882 | default: llvm_unreachable("Unhandled CC")::llvm::llvm_unreachable_internal("Unhandled CC", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11882); |
| 11883 | } |
| 11884 | |
| 11885 | // Builtins without the _mask suffix return a vector of integers |
| 11886 | // of the same width as the input vectors |
| 11887 | switch (BuiltinID) { |
| 11888 | case X86::BI__builtin_ia32_cmpps512_mask: |
| 11889 | case X86::BI__builtin_ia32_cmppd512_mask: |
| 11890 | case X86::BI__builtin_ia32_cmpps128_mask: |
| 11891 | case X86::BI__builtin_ia32_cmpps256_mask: |
| 11892 | case X86::BI__builtin_ia32_cmppd128_mask: |
| 11893 | case X86::BI__builtin_ia32_cmppd256_mask: { |
| 11894 | unsigned NumElts = Ops[0]->getType()->getVectorNumElements(); |
| 11895 | Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]); |
| 11896 | return EmitX86MaskedCompareResult(*this, Cmp, NumElts, Ops[3]); |
| 11897 | } |
| 11898 | default: |
| 11899 | return getVectorFCmpIR(Pred); |
| 11900 | } |
| 11901 | } |
| 11902 | |
| 11903 | // SSE scalar comparison intrinsics |
| 11904 | case X86::BI__builtin_ia32_cmpeqss: |
| 11905 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 0); |
| 11906 | case X86::BI__builtin_ia32_cmpltss: |
| 11907 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 1); |
| 11908 | case X86::BI__builtin_ia32_cmpless: |
| 11909 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 2); |
| 11910 | case X86::BI__builtin_ia32_cmpunordss: |
| 11911 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 3); |
| 11912 | case X86::BI__builtin_ia32_cmpneqss: |
| 11913 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 4); |
| 11914 | case X86::BI__builtin_ia32_cmpnltss: |
| 11915 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 5); |
| 11916 | case X86::BI__builtin_ia32_cmpnless: |
| 11917 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 6); |
| 11918 | case X86::BI__builtin_ia32_cmpordss: |
| 11919 | return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 7); |
| 11920 | case X86::BI__builtin_ia32_cmpeqsd: |
| 11921 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 0); |
| 11922 | case X86::BI__builtin_ia32_cmpltsd: |
| 11923 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 1); |
| 11924 | case X86::BI__builtin_ia32_cmplesd: |
| 11925 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 2); |
| 11926 | case X86::BI__builtin_ia32_cmpunordsd: |
| 11927 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 3); |
| 11928 | case X86::BI__builtin_ia32_cmpneqsd: |
| 11929 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 4); |
| 11930 | case X86::BI__builtin_ia32_cmpnltsd: |
| 11931 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 5); |
| 11932 | case X86::BI__builtin_ia32_cmpnlesd: |
| 11933 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 6); |
| 11934 | case X86::BI__builtin_ia32_cmpordsd: |
| 11935 | return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 7); |
| 11936 | |
| 11937 | // AVX512 bf16 intrinsics |
| 11938 | case X86::BI__builtin_ia32_cvtneps2bf16_128_mask: { |
| 11939 | Ops[2] = getMaskVecValue(*this, Ops[2], |
| 11940 | Ops[0]->getType()->getVectorNumElements()); |
| 11941 | Intrinsic::ID IID = Intrinsic::x86_avx512bf16_mask_cvtneps2bf16_128; |
| 11942 | return Builder.CreateCall(CGM.getIntrinsic(IID), Ops); |
| 11943 | } |
| 11944 | |
| 11945 | case X86::BI__builtin_ia32_cvtneps2bf16_256_mask: |
| 11946 | case X86::BI__builtin_ia32_cvtneps2bf16_512_mask: { |
| 11947 | Intrinsic::ID IID; |
| 11948 | switch (BuiltinID) { |
| 11949 | default: llvm_unreachable("Unsupported intrinsic!")::llvm::llvm_unreachable_internal("Unsupported intrinsic!", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 11949); |
| 11950 | case X86::BI__builtin_ia32_cvtneps2bf16_256_mask: |
| 11951 | IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_256; |
| 11952 | break; |
| 11953 | case X86::BI__builtin_ia32_cvtneps2bf16_512_mask: |
| 11954 | IID = Intrinsic::x86_avx512bf16_cvtneps2bf16_512; |
| 11955 | break; |
| 11956 | } |
| 11957 | Value *Res = Builder.CreateCall(CGM.getIntrinsic(IID), Ops[0]); |
| 11958 | return EmitX86Select(*this, Ops[2], Res, Ops[1]); |
| 11959 | } |
| 11960 | |
| 11961 | case X86::BI__emul: |
| 11962 | case X86::BI__emulu: { |
| 11963 | llvm::Type *Int64Ty = llvm::IntegerType::get(getLLVMContext(), 64); |
| 11964 | bool isSigned = (BuiltinID == X86::BI__emul); |
| 11965 | Value *LHS = Builder.CreateIntCast(Ops[0], Int64Ty, isSigned); |
| 11966 | Value *RHS = Builder.CreateIntCast(Ops[1], Int64Ty, isSigned); |
| 11967 | return Builder.CreateMul(LHS, RHS, "", !isSigned, isSigned); |
| 11968 | } |
| 11969 | case X86::BI__mulh: |
| 11970 | case X86::BI__umulh: |
| 11971 | case X86::BI_mul128: |
| 11972 | case X86::BI_umul128: { |
| 11973 | llvm::Type *ResType = ConvertType(E->getType()); |
| 11974 | llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128); |
| 11975 | |
| 11976 | bool IsSigned = (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI_mul128); |
| 11977 | Value *LHS = Builder.CreateIntCast(Ops[0], Int128Ty, IsSigned); |
| 11978 | Value *RHS = Builder.CreateIntCast(Ops[1], Int128Ty, IsSigned); |
| 11979 | |
| 11980 | Value *MulResult, *HigherBits; |
| 11981 | if (IsSigned) { |
| 11982 | MulResult = Builder.CreateNSWMul(LHS, RHS); |
| 11983 | HigherBits = Builder.CreateAShr(MulResult, 64); |
| 11984 | } else { |
| 11985 | MulResult = Builder.CreateNUWMul(LHS, RHS); |
| 11986 | HigherBits = Builder.CreateLShr(MulResult, 64); |
| 11987 | } |
| 11988 | HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned); |
| 11989 | |
| 11990 | if (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI__umulh) |
| 11991 | return HigherBits; |
| 11992 | |
| 11993 | Address HighBitsAddress = EmitPointerWithAlignment(E->getArg(2)); |
| 11994 | Builder.CreateStore(HigherBits, HighBitsAddress); |
| 11995 | return Builder.CreateIntCast(MulResult, ResType, IsSigned); |
| 11996 | } |
| 11997 | |
| 11998 | case X86::BI__faststorefence: { |
| 11999 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, |
| 12000 | llvm::SyncScope::System); |
| 12001 | } |
| 12002 | case X86::BI__shiftleft128: |
| 12003 | case X86::BI__shiftright128: { |
| 12004 | // FIXME: Once fshl/fshr no longer add an unneeded and and cmov, do this: |
| 12005 | // llvm::Function *F = CGM.getIntrinsic( |
| 12006 | // BuiltinID == X86::BI__shiftleft128 ? Intrinsic::fshl : Intrinsic::fshr, |
| 12007 | // Int64Ty); |
| 12008 | // Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty); |
| 12009 | // return Builder.CreateCall(F, Ops); |
| 12010 | llvm::Type *Int128Ty = Builder.getInt128Ty(); |
| 12011 | Value *HighPart128 = |
| 12012 | Builder.CreateShl(Builder.CreateZExt(Ops[1], Int128Ty), 64); |
| 12013 | Value *LowPart128 = Builder.CreateZExt(Ops[0], Int128Ty); |
| 12014 | Value *Val = Builder.CreateOr(HighPart128, LowPart128); |
| 12015 | Value *Amt = Builder.CreateAnd(Builder.CreateZExt(Ops[2], Int128Ty), |
| 12016 | llvm::ConstantInt::get(Int128Ty, 0x3f)); |
| 12017 | Value *Res; |
| 12018 | if (BuiltinID == X86::BI__shiftleft128) |
| 12019 | Res = Builder.CreateLShr(Builder.CreateShl(Val, Amt), 64); |
| 12020 | else |
| 12021 | Res = Builder.CreateLShr(Val, Amt); |
| 12022 | return Builder.CreateTrunc(Res, Int64Ty); |
| 12023 | } |
| 12024 | case X86::BI_ReadWriteBarrier: |
| 12025 | case X86::BI_ReadBarrier: |
| 12026 | case X86::BI_WriteBarrier: { |
| 12027 | return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, |
| 12028 | llvm::SyncScope::SingleThread); |
| 12029 | } |
| 12030 | case X86::BI_BitScanForward: |
| 12031 | case X86::BI_BitScanForward64: |
| 12032 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E); |
| 12033 | case X86::BI_BitScanReverse: |
| 12034 | case X86::BI_BitScanReverse64: |
| 12035 | return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E); |
| 12036 | |
| 12037 | case X86::BI_InterlockedAnd64: |
| 12038 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E); |
| 12039 | case X86::BI_InterlockedExchange64: |
| 12040 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E); |
| 12041 | case X86::BI_InterlockedExchangeAdd64: |
| 12042 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E); |
| 12043 | case X86::BI_InterlockedExchangeSub64: |
| 12044 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E); |
| 12045 | case X86::BI_InterlockedOr64: |
| 12046 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E); |
| 12047 | case X86::BI_InterlockedXor64: |
| 12048 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E); |
| 12049 | case X86::BI_InterlockedDecrement64: |
| 12050 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E); |
| 12051 | case X86::BI_InterlockedIncrement64: |
| 12052 | return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E); |
| 12053 | case X86::BI_InterlockedCompareExchange128: { |
| 12054 | // InterlockedCompareExchange128 doesn't directly refer to 128bit ints, |
| 12055 | // instead it takes pointers to 64bit ints for Destination and |
| 12056 | // ComparandResult, and exchange is taken as two 64bit ints (high & low). |
| 12057 | // The previous value is written to ComparandResult, and success is |
| 12058 | // returned. |
| 12059 | |
| 12060 | llvm::Type *Int128Ty = Builder.getInt128Ty(); |
| 12061 | llvm::Type *Int128PtrTy = Int128Ty->getPointerTo(); |
| 12062 | |
| 12063 | Value *Destination = |
| 12064 | Builder.CreateBitCast(Ops[0], Int128PtrTy); |
| 12065 | Value *ExchangeHigh128 = Builder.CreateZExt(Ops[1], Int128Ty); |
| 12066 | Value *ExchangeLow128 = Builder.CreateZExt(Ops[2], Int128Ty); |
| 12067 | Address ComparandResult(Builder.CreateBitCast(Ops[3], Int128PtrTy), |
| 12068 | getContext().toCharUnitsFromBits(128)); |
| 12069 | |
| 12070 | Value *Exchange = Builder.CreateOr( |
| 12071 | Builder.CreateShl(ExchangeHigh128, 64, "", false, false), |
| 12072 | ExchangeLow128); |
| 12073 | |
| 12074 | Value *Comparand = Builder.CreateLoad(ComparandResult); |
| 12075 | |
| 12076 | AtomicCmpXchgInst *CXI = |
| 12077 | Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
| 12078 | AtomicOrdering::SequentiallyConsistent, |
| 12079 | AtomicOrdering::SequentiallyConsistent); |
| 12080 | CXI->setVolatile(true); |
| 12081 | |
| 12082 | // Write the result back to the inout pointer. |
| 12083 | Builder.CreateStore(Builder.CreateExtractValue(CXI, 0), ComparandResult); |
| 12084 | |
| 12085 | // Get the success boolean and zero extend it to i8. |
| 12086 | Value *Success = Builder.CreateExtractValue(CXI, 1); |
| 12087 | return Builder.CreateZExt(Success, ConvertType(E->getType())); |
| 12088 | } |
| 12089 | |
| 12090 | case X86::BI_AddressOfReturnAddress: { |
| 12091 | Function *F = CGM.getIntrinsic(Intrinsic::addressofreturnaddress); |
| 12092 | return Builder.CreateCall(F); |
| 12093 | } |
| 12094 | case X86::BI__stosb: { |
| 12095 | // We treat __stosb as a volatile memset - it may not generate "rep stosb" |
| 12096 | // instruction, but it will create a memset that won't be optimized away. |
| 12097 | return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], 1, true); |
| 12098 | } |
| 12099 | case X86::BI__ud2: |
| 12100 | // llvm.trap makes a ud2a instruction on x86. |
| 12101 | return EmitTrapCall(Intrinsic::trap); |
| 12102 | case X86::BI__int2c: { |
| 12103 | // This syscall signals a driver assertion failure in x86 NT kernels. |
| 12104 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false); |
| 12105 | llvm::InlineAsm *IA = |
| 12106 | llvm::InlineAsm::get(FTy, "int $$0x2c", "", /*SideEffects=*/true); |
| 12107 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( |
| 12108 | getLLVMContext(), llvm::AttributeList::FunctionIndex, |
| 12109 | llvm::Attribute::NoReturn); |
| 12110 | llvm::CallInst *CI = Builder.CreateCall(IA); |
| 12111 | CI->setAttributes(NoReturnAttr); |
| 12112 | return CI; |
| 12113 | } |
| 12114 | case X86::BI__readfsbyte: |
| 12115 | case X86::BI__readfsword: |
| 12116 | case X86::BI__readfsdword: |
| 12117 | case X86::BI__readfsqword: { |
| 12118 | llvm::Type *IntTy = ConvertType(E->getType()); |
| 12119 | Value *Ptr = |
| 12120 | Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 257)); |
| 12121 | LoadInst *Load = Builder.CreateAlignedLoad( |
| 12122 | IntTy, Ptr, getContext().getTypeAlignInChars(E->getType())); |
| 12123 | Load->setVolatile(true); |
| 12124 | return Load; |
| 12125 | } |
| 12126 | case X86::BI__readgsbyte: |
| 12127 | case X86::BI__readgsword: |
| 12128 | case X86::BI__readgsdword: |
| 12129 | case X86::BI__readgsqword: { |
| 12130 | llvm::Type *IntTy = ConvertType(E->getType()); |
| 12131 | Value *Ptr = |
| 12132 | Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 256)); |
| 12133 | LoadInst *Load = Builder.CreateAlignedLoad( |
| 12134 | IntTy, Ptr, getContext().getTypeAlignInChars(E->getType())); |
| 12135 | Load->setVolatile(true); |
| 12136 | return Load; |
| 12137 | } |
| 12138 | case X86::BI__builtin_ia32_paddsb512: |
| 12139 | case X86::BI__builtin_ia32_paddsw512: |
| 12140 | case X86::BI__builtin_ia32_paddsb256: |
| 12141 | case X86::BI__builtin_ia32_paddsw256: |
| 12142 | case X86::BI__builtin_ia32_paddsb128: |
| 12143 | case X86::BI__builtin_ia32_paddsw128: |
| 12144 | return EmitX86AddSubSatExpr(*this, Ops, true, true); |
| 12145 | case X86::BI__builtin_ia32_paddusb512: |
| 12146 | case X86::BI__builtin_ia32_paddusw512: |
| 12147 | case X86::BI__builtin_ia32_paddusb256: |
| 12148 | case X86::BI__builtin_ia32_paddusw256: |
| 12149 | case X86::BI__builtin_ia32_paddusb128: |
| 12150 | case X86::BI__builtin_ia32_paddusw128: |
| 12151 | return EmitX86AddSubSatExpr(*this, Ops, false, true); |
| 12152 | case X86::BI__builtin_ia32_psubsb512: |
| 12153 | case X86::BI__builtin_ia32_psubsw512: |
| 12154 | case X86::BI__builtin_ia32_psubsb256: |
| 12155 | case X86::BI__builtin_ia32_psubsw256: |
| 12156 | case X86::BI__builtin_ia32_psubsb128: |
| 12157 | case X86::BI__builtin_ia32_psubsw128: |
| 12158 | return EmitX86AddSubSatExpr(*this, Ops, true, false); |
| 12159 | case X86::BI__builtin_ia32_psubusb512: |
| 12160 | case X86::BI__builtin_ia32_psubusw512: |
| 12161 | case X86::BI__builtin_ia32_psubusb256: |
| 12162 | case X86::BI__builtin_ia32_psubusw256: |
| 12163 | case X86::BI__builtin_ia32_psubusb128: |
| 12164 | case X86::BI__builtin_ia32_psubusw128: |
| 12165 | return EmitX86AddSubSatExpr(*this, Ops, false, false); |
| 12166 | } |
| 12167 | } |
| 12168 | |
| 12169 | Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID, |
| 12170 | const CallExpr *E) { |
| 12171 | SmallVector<Value*, 4> Ops; |
| 12172 | |
| 12173 | for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) |
| 12174 | Ops.push_back(EmitScalarExpr(E->getArg(i))); |
| 12175 | |
| 12176 | Intrinsic::ID ID = Intrinsic::not_intrinsic; |
| 12177 | |
| 12178 | switch (BuiltinID) { |
| 12179 | default: return nullptr; |
| 12180 | |
| 12181 | // __builtin_ppc_get_timebase is GCC 4.8+'s PowerPC-specific name for what we |
| 12182 | // call __builtin_readcyclecounter. |
| 12183 | case PPC::BI__builtin_ppc_get_timebase: |
| 12184 | return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::readcyclecounter)); |
| 12185 | |
| 12186 | // vec_ld, vec_xl_be, vec_lvsl, vec_lvsr |
| 12187 | case PPC::BI__builtin_altivec_lvx: |
| 12188 | case PPC::BI__builtin_altivec_lvxl: |
| 12189 | case PPC::BI__builtin_altivec_lvebx: |
| 12190 | case PPC::BI__builtin_altivec_lvehx: |
| 12191 | case PPC::BI__builtin_altivec_lvewx: |
| 12192 | case PPC::BI__builtin_altivec_lvsl: |
| 12193 | case PPC::BI__builtin_altivec_lvsr: |
| 12194 | case PPC::BI__builtin_vsx_lxvd2x: |
| 12195 | case PPC::BI__builtin_vsx_lxvw4x: |
| 12196 | case PPC::BI__builtin_vsx_lxvd2x_be: |
| 12197 | case PPC::BI__builtin_vsx_lxvw4x_be: |
| 12198 | case PPC::BI__builtin_vsx_lxvl: |
| 12199 | case PPC::BI__builtin_vsx_lxvll: |
| 12200 | { |
| 12201 | if(BuiltinID == PPC::BI__builtin_vsx_lxvl || |
| 12202 | BuiltinID == PPC::BI__builtin_vsx_lxvll){ |
| 12203 | Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy); |
| 12204 | }else { |
| 12205 | Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy); |
| 12206 | Ops[0] = Builder.CreateGEP(Ops[1], Ops[0]); |
| 12207 | Ops.pop_back(); |
| 12208 | } |
| 12209 | |
| 12210 | switch (BuiltinID) { |
| 12211 | default: llvm_unreachable("Unsupported ld/lvsl/lvsr intrinsic!")::llvm::llvm_unreachable_internal("Unsupported ld/lvsl/lvsr intrinsic!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12211); |
| 12212 | case PPC::BI__builtin_altivec_lvx: |
| 12213 | ID = Intrinsic::ppc_altivec_lvx; |
| 12214 | break; |
| 12215 | case PPC::BI__builtin_altivec_lvxl: |
| 12216 | ID = Intrinsic::ppc_altivec_lvxl; |
| 12217 | break; |
| 12218 | case PPC::BI__builtin_altivec_lvebx: |
| 12219 | ID = Intrinsic::ppc_altivec_lvebx; |
| 12220 | break; |
| 12221 | case PPC::BI__builtin_altivec_lvehx: |
| 12222 | ID = Intrinsic::ppc_altivec_lvehx; |
| 12223 | break; |
| 12224 | case PPC::BI__builtin_altivec_lvewx: |
| 12225 | ID = Intrinsic::ppc_altivec_lvewx; |
| 12226 | break; |
| 12227 | case PPC::BI__builtin_altivec_lvsl: |
| 12228 | ID = Intrinsic::ppc_altivec_lvsl; |
| 12229 | break; |
| 12230 | case PPC::BI__builtin_altivec_lvsr: |
| 12231 | ID = Intrinsic::ppc_altivec_lvsr; |
| 12232 | break; |
| 12233 | case PPC::BI__builtin_vsx_lxvd2x: |
| 12234 | ID = Intrinsic::ppc_vsx_lxvd2x; |
| 12235 | break; |
| 12236 | case PPC::BI__builtin_vsx_lxvw4x: |
| 12237 | ID = Intrinsic::ppc_vsx_lxvw4x; |
| 12238 | break; |
| 12239 | case PPC::BI__builtin_vsx_lxvd2x_be: |
| 12240 | ID = Intrinsic::ppc_vsx_lxvd2x_be; |
| 12241 | break; |
| 12242 | case PPC::BI__builtin_vsx_lxvw4x_be: |
| 12243 | ID = Intrinsic::ppc_vsx_lxvw4x_be; |
| 12244 | break; |
| 12245 | case PPC::BI__builtin_vsx_lxvl: |
| 12246 | ID = Intrinsic::ppc_vsx_lxvl; |
| 12247 | break; |
| 12248 | case PPC::BI__builtin_vsx_lxvll: |
| 12249 | ID = Intrinsic::ppc_vsx_lxvll; |
| 12250 | break; |
| 12251 | } |
| 12252 | llvm::Function *F = CGM.getIntrinsic(ID); |
| 12253 | return Builder.CreateCall(F, Ops, ""); |
| 12254 | } |
| 12255 | |
| 12256 | // vec_st, vec_xst_be |
| 12257 | case PPC::BI__builtin_altivec_stvx: |
| 12258 | case PPC::BI__builtin_altivec_stvxl: |
| 12259 | case PPC::BI__builtin_altivec_stvebx: |
| 12260 | case PPC::BI__builtin_altivec_stvehx: |
| 12261 | case PPC::BI__builtin_altivec_stvewx: |
| 12262 | case PPC::BI__builtin_vsx_stxvd2x: |
| 12263 | case PPC::BI__builtin_vsx_stxvw4x: |
| 12264 | case PPC::BI__builtin_vsx_stxvd2x_be: |
| 12265 | case PPC::BI__builtin_vsx_stxvw4x_be: |
| 12266 | case PPC::BI__builtin_vsx_stxvl: |
| 12267 | case PPC::BI__builtin_vsx_stxvll: |
| 12268 | { |
| 12269 | if(BuiltinID == PPC::BI__builtin_vsx_stxvl || |
| 12270 | BuiltinID == PPC::BI__builtin_vsx_stxvll ){ |
| 12271 | Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy); |
| 12272 | }else { |
| 12273 | Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy); |
| 12274 | Ops[1] = Builder.CreateGEP(Ops[2], Ops[1]); |
| 12275 | Ops.pop_back(); |
| 12276 | } |
| 12277 | |
| 12278 | switch (BuiltinID) { |
| 12279 | default: llvm_unreachable("Unsupported st intrinsic!")::llvm::llvm_unreachable_internal("Unsupported st intrinsic!" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12279); |
| 12280 | case PPC::BI__builtin_altivec_stvx: |
| 12281 | ID = Intrinsic::ppc_altivec_stvx; |
| 12282 | break; |
| 12283 | case PPC::BI__builtin_altivec_stvxl: |
| 12284 | ID = Intrinsic::ppc_altivec_stvxl; |
| 12285 | break; |
| 12286 | case PPC::BI__builtin_altivec_stvebx: |
| 12287 | ID = Intrinsic::ppc_altivec_stvebx; |
| 12288 | break; |
| 12289 | case PPC::BI__builtin_altivec_stvehx: |
| 12290 | ID = Intrinsic::ppc_altivec_stvehx; |
| 12291 | break; |
| 12292 | case PPC::BI__builtin_altivec_stvewx: |
| 12293 | ID = Intrinsic::ppc_altivec_stvewx; |
| 12294 | break; |
| 12295 | case PPC::BI__builtin_vsx_stxvd2x: |
| 12296 | ID = Intrinsic::ppc_vsx_stxvd2x; |
| 12297 | break; |
| 12298 | case PPC::BI__builtin_vsx_stxvw4x: |
| 12299 | ID = Intrinsic::ppc_vsx_stxvw4x; |
| 12300 | break; |
| 12301 | case PPC::BI__builtin_vsx_stxvd2x_be: |
| 12302 | ID = Intrinsic::ppc_vsx_stxvd2x_be; |
| 12303 | break; |
| 12304 | case PPC::BI__builtin_vsx_stxvw4x_be: |
| 12305 | ID = Intrinsic::ppc_vsx_stxvw4x_be; |
| 12306 | break; |
| 12307 | case PPC::BI__builtin_vsx_stxvl: |
| 12308 | ID = Intrinsic::ppc_vsx_stxvl; |
| 12309 | break; |
| 12310 | case PPC::BI__builtin_vsx_stxvll: |
| 12311 | ID = Intrinsic::ppc_vsx_stxvll; |
| 12312 | break; |
| 12313 | } |
| 12314 | llvm::Function *F = CGM.getIntrinsic(ID); |
| 12315 | return Builder.CreateCall(F, Ops, ""); |
| 12316 | } |
| 12317 | // Square root |
| 12318 | case PPC::BI__builtin_vsx_xvsqrtsp: |
| 12319 | case PPC::BI__builtin_vsx_xvsqrtdp: { |
| 12320 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12321 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12322 | ID = Intrinsic::sqrt; |
| 12323 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); |
| 12324 | return Builder.CreateCall(F, X); |
| 12325 | } |
| 12326 | // Count leading zeros |
| 12327 | case PPC::BI__builtin_altivec_vclzb: |
| 12328 | case PPC::BI__builtin_altivec_vclzh: |
| 12329 | case PPC::BI__builtin_altivec_vclzw: |
| 12330 | case PPC::BI__builtin_altivec_vclzd: { |
| 12331 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12332 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12333 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); |
| 12334 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType); |
| 12335 | return Builder.CreateCall(F, {X, Undef}); |
| 12336 | } |
| 12337 | case PPC::BI__builtin_altivec_vctzb: |
| 12338 | case PPC::BI__builtin_altivec_vctzh: |
| 12339 | case PPC::BI__builtin_altivec_vctzw: |
| 12340 | case PPC::BI__builtin_altivec_vctzd: { |
| 12341 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12342 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12343 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); |
| 12344 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType); |
| 12345 | return Builder.CreateCall(F, {X, Undef}); |
| 12346 | } |
| 12347 | case PPC::BI__builtin_altivec_vpopcntb: |
| 12348 | case PPC::BI__builtin_altivec_vpopcnth: |
| 12349 | case PPC::BI__builtin_altivec_vpopcntw: |
| 12350 | case PPC::BI__builtin_altivec_vpopcntd: { |
| 12351 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12352 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12353 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); |
| 12354 | return Builder.CreateCall(F, X); |
| 12355 | } |
| 12356 | // Copy sign |
| 12357 | case PPC::BI__builtin_vsx_xvcpsgnsp: |
| 12358 | case PPC::BI__builtin_vsx_xvcpsgndp: { |
| 12359 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12360 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12361 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12362 | ID = Intrinsic::copysign; |
| 12363 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); |
| 12364 | return Builder.CreateCall(F, {X, Y}); |
| 12365 | } |
| 12366 | // Rounding/truncation |
| 12367 | case PPC::BI__builtin_vsx_xvrspip: |
| 12368 | case PPC::BI__builtin_vsx_xvrdpip: |
| 12369 | case PPC::BI__builtin_vsx_xvrdpim: |
| 12370 | case PPC::BI__builtin_vsx_xvrspim: |
| 12371 | case PPC::BI__builtin_vsx_xvrdpi: |
| 12372 | case PPC::BI__builtin_vsx_xvrspi: |
| 12373 | case PPC::BI__builtin_vsx_xvrdpic: |
| 12374 | case PPC::BI__builtin_vsx_xvrspic: |
| 12375 | case PPC::BI__builtin_vsx_xvrdpiz: |
| 12376 | case PPC::BI__builtin_vsx_xvrspiz: { |
| 12377 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12378 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12379 | if (BuiltinID == PPC::BI__builtin_vsx_xvrdpim || |
| 12380 | BuiltinID == PPC::BI__builtin_vsx_xvrspim) |
| 12381 | ID = Intrinsic::floor; |
| 12382 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpi || |
| 12383 | BuiltinID == PPC::BI__builtin_vsx_xvrspi) |
| 12384 | ID = Intrinsic::round; |
| 12385 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpic || |
| 12386 | BuiltinID == PPC::BI__builtin_vsx_xvrspic) |
| 12387 | ID = Intrinsic::nearbyint; |
| 12388 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpip || |
| 12389 | BuiltinID == PPC::BI__builtin_vsx_xvrspip) |
| 12390 | ID = Intrinsic::ceil; |
| 12391 | else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpiz || |
| 12392 | BuiltinID == PPC::BI__builtin_vsx_xvrspiz) |
| 12393 | ID = Intrinsic::trunc; |
| 12394 | llvm::Function *F = CGM.getIntrinsic(ID, ResultType); |
| 12395 | return Builder.CreateCall(F, X); |
| 12396 | } |
| 12397 | |
| 12398 | // Absolute value |
| 12399 | case PPC::BI__builtin_vsx_xvabsdp: |
| 12400 | case PPC::BI__builtin_vsx_xvabssp: { |
| 12401 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12402 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12403 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); |
| 12404 | return Builder.CreateCall(F, X); |
| 12405 | } |
| 12406 | |
| 12407 | // FMA variations |
| 12408 | case PPC::BI__builtin_vsx_xvmaddadp: |
| 12409 | case PPC::BI__builtin_vsx_xvmaddasp: |
| 12410 | case PPC::BI__builtin_vsx_xvnmaddadp: |
| 12411 | case PPC::BI__builtin_vsx_xvnmaddasp: |
| 12412 | case PPC::BI__builtin_vsx_xvmsubadp: |
| 12413 | case PPC::BI__builtin_vsx_xvmsubasp: |
| 12414 | case PPC::BI__builtin_vsx_xvnmsubadp: |
| 12415 | case PPC::BI__builtin_vsx_xvnmsubasp: { |
| 12416 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12417 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12418 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12419 | Value *Z = EmitScalarExpr(E->getArg(2)); |
| 12420 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); |
| 12421 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); |
| 12422 | switch (BuiltinID) { |
| 12423 | case PPC::BI__builtin_vsx_xvmaddadp: |
| 12424 | case PPC::BI__builtin_vsx_xvmaddasp: |
| 12425 | return Builder.CreateCall(F, {X, Y, Z}); |
| 12426 | case PPC::BI__builtin_vsx_xvnmaddadp: |
| 12427 | case PPC::BI__builtin_vsx_xvnmaddasp: |
| 12428 | return Builder.CreateFSub(Zero, |
| 12429 | Builder.CreateCall(F, {X, Y, Z}), "sub"); |
| 12430 | case PPC::BI__builtin_vsx_xvmsubadp: |
| 12431 | case PPC::BI__builtin_vsx_xvmsubasp: |
| 12432 | return Builder.CreateCall(F, |
| 12433 | {X, Y, Builder.CreateFSub(Zero, Z, "sub")}); |
| 12434 | case PPC::BI__builtin_vsx_xvnmsubadp: |
| 12435 | case PPC::BI__builtin_vsx_xvnmsubasp: |
| 12436 | Value *FsubRes = |
| 12437 | Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")}); |
| 12438 | return Builder.CreateFSub(Zero, FsubRes, "sub"); |
| 12439 | } |
| 12440 | llvm_unreachable("Unknown FMA operation")::llvm::llvm_unreachable_internal("Unknown FMA operation", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12440); |
| 12441 | return nullptr; // Suppress no-return warning |
| 12442 | } |
| 12443 | |
| 12444 | case PPC::BI__builtin_vsx_insertword: { |
| 12445 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxinsertw); |
| 12446 | |
| 12447 | // Third argument is a compile time constant int. It must be clamped to |
| 12448 | // to the range [0, 12]. |
| 12449 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); |
| 12450 | assert(ArgCI &&((ArgCI && "Third arg to xxinsertw intrinsic must be constant integer" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12451, __PRETTY_FUNCTION__)) |
| 12451 | "Third arg to xxinsertw intrinsic must be constant integer")((ArgCI && "Third arg to xxinsertw intrinsic must be constant integer" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12451, __PRETTY_FUNCTION__)); |
| 12452 | const int64_t MaxIndex = 12; |
| 12453 | int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex); |
| 12454 | |
| 12455 | // The builtin semantics don't exactly match the xxinsertw instructions |
| 12456 | // semantics (which ppc_vsx_xxinsertw follows). The builtin extracts the |
| 12457 | // word from the first argument, and inserts it in the second argument. The |
| 12458 | // instruction extracts the word from its second input register and inserts |
| 12459 | // it into its first input register, so swap the first and second arguments. |
| 12460 | std::swap(Ops[0], Ops[1]); |
| 12461 | |
| 12462 | // Need to cast the second argument from a vector of unsigned int to a |
| 12463 | // vector of long long. |
| 12464 | Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2)); |
| 12465 | |
| 12466 | if (getTarget().isLittleEndian()) { |
| 12467 | // Create a shuffle mask of (1, 0) |
| 12468 | Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1), |
| 12469 | ConstantInt::get(Int32Ty, 0) |
| 12470 | }; |
| 12471 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); |
| 12472 | |
| 12473 | // Reverse the double words in the vector we will extract from. |
| 12474 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); |
| 12475 | Ops[0] = Builder.CreateShuffleVector(Ops[0], Ops[0], ShuffleMask); |
| 12476 | |
| 12477 | // Reverse the index. |
| 12478 | Index = MaxIndex - Index; |
| 12479 | } |
| 12480 | |
| 12481 | // Intrinsic expects the first arg to be a vector of int. |
| 12482 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4)); |
| 12483 | Ops[2] = ConstantInt::getSigned(Int32Ty, Index); |
| 12484 | return Builder.CreateCall(F, Ops); |
| 12485 | } |
| 12486 | |
| 12487 | case PPC::BI__builtin_vsx_extractuword: { |
| 12488 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxextractuw); |
| 12489 | |
| 12490 | // Intrinsic expects the first argument to be a vector of doublewords. |
| 12491 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); |
| 12492 | |
| 12493 | // The second argument is a compile time constant int that needs to |
| 12494 | // be clamped to the range [0, 12]. |
| 12495 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[1]); |
| 12496 | assert(ArgCI &&((ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12497, __PRETTY_FUNCTION__)) |
| 12497 | "Second Arg to xxextractuw intrinsic must be a constant integer!")((ArgCI && "Second Arg to xxextractuw intrinsic must be a constant integer!" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12497, __PRETTY_FUNCTION__)); |
| 12498 | const int64_t MaxIndex = 12; |
| 12499 | int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex); |
| 12500 | |
| 12501 | if (getTarget().isLittleEndian()) { |
| 12502 | // Reverse the index. |
| 12503 | Index = MaxIndex - Index; |
| 12504 | Ops[1] = ConstantInt::getSigned(Int32Ty, Index); |
| 12505 | |
| 12506 | // Emit the call, then reverse the double words of the results vector. |
| 12507 | Value *Call = Builder.CreateCall(F, Ops); |
| 12508 | |
| 12509 | // Create a shuffle mask of (1, 0) |
| 12510 | Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1), |
| 12511 | ConstantInt::get(Int32Ty, 0) |
| 12512 | }; |
| 12513 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); |
| 12514 | |
| 12515 | Value *ShuffleCall = Builder.CreateShuffleVector(Call, Call, ShuffleMask); |
| 12516 | return ShuffleCall; |
| 12517 | } else { |
| 12518 | Ops[1] = ConstantInt::getSigned(Int32Ty, Index); |
| 12519 | return Builder.CreateCall(F, Ops); |
| 12520 | } |
| 12521 | } |
| 12522 | |
| 12523 | case PPC::BI__builtin_vsx_xxpermdi: { |
| 12524 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); |
| 12525 | assert(ArgCI && "Third arg must be constant integer!")((ArgCI && "Third arg must be constant integer!") ? static_cast <void> (0) : __assert_fail ("ArgCI && \"Third arg must be constant integer!\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12525, __PRETTY_FUNCTION__)); |
| 12526 | |
| 12527 | unsigned Index = ArgCI->getZExtValue(); |
| 12528 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2)); |
| 12529 | Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2)); |
| 12530 | |
| 12531 | // Account for endianness by treating this as just a shuffle. So we use the |
| 12532 | // same indices for both LE and BE in order to produce expected results in |
| 12533 | // both cases. |
| 12534 | unsigned ElemIdx0 = (Index & 2) >> 1; |
| 12535 | unsigned ElemIdx1 = 2 + (Index & 1); |
| 12536 | |
| 12537 | Constant *ShuffleElts[2] = {ConstantInt::get(Int32Ty, ElemIdx0), |
| 12538 | ConstantInt::get(Int32Ty, ElemIdx1)}; |
| 12539 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); |
| 12540 | |
| 12541 | Value *ShuffleCall = |
| 12542 | Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask); |
| 12543 | QualType BIRetType = E->getType(); |
| 12544 | auto RetTy = ConvertType(BIRetType); |
| 12545 | return Builder.CreateBitCast(ShuffleCall, RetTy); |
| 12546 | } |
| 12547 | |
| 12548 | case PPC::BI__builtin_vsx_xxsldwi: { |
| 12549 | ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]); |
| 12550 | assert(ArgCI && "Third argument must be a compile time constant")((ArgCI && "Third argument must be a compile time constant" ) ? static_cast<void> (0) : __assert_fail ("ArgCI && \"Third argument must be a compile time constant\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12550, __PRETTY_FUNCTION__)); |
| 12551 | unsigned Index = ArgCI->getZExtValue() & 0x3; |
| 12552 | Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4)); |
| 12553 | Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int32Ty, 4)); |
| 12554 | |
| 12555 | // Create a shuffle mask |
| 12556 | unsigned ElemIdx0; |
| 12557 | unsigned ElemIdx1; |
| 12558 | unsigned ElemIdx2; |
| 12559 | unsigned ElemIdx3; |
| 12560 | if (getTarget().isLittleEndian()) { |
| 12561 | // Little endian element N comes from element 8+N-Index of the |
| 12562 | // concatenated wide vector (of course, using modulo arithmetic on |
| 12563 | // the total number of elements). |
| 12564 | ElemIdx0 = (8 - Index) % 8; |
| 12565 | ElemIdx1 = (9 - Index) % 8; |
| 12566 | ElemIdx2 = (10 - Index) % 8; |
| 12567 | ElemIdx3 = (11 - Index) % 8; |
| 12568 | } else { |
| 12569 | // Big endian ElemIdx<N> = Index + N |
| 12570 | ElemIdx0 = Index; |
| 12571 | ElemIdx1 = Index + 1; |
| 12572 | ElemIdx2 = Index + 2; |
| 12573 | ElemIdx3 = Index + 3; |
| 12574 | } |
| 12575 | |
| 12576 | Constant *ShuffleElts[4] = {ConstantInt::get(Int32Ty, ElemIdx0), |
| 12577 | ConstantInt::get(Int32Ty, ElemIdx1), |
| 12578 | ConstantInt::get(Int32Ty, ElemIdx2), |
| 12579 | ConstantInt::get(Int32Ty, ElemIdx3)}; |
| 12580 | |
| 12581 | Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts); |
| 12582 | Value *ShuffleCall = |
| 12583 | Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask); |
| 12584 | QualType BIRetType = E->getType(); |
| 12585 | auto RetTy = ConvertType(BIRetType); |
| 12586 | return Builder.CreateBitCast(ShuffleCall, RetTy); |
| 12587 | } |
| 12588 | |
| 12589 | case PPC::BI__builtin_pack_vector_int128: { |
| 12590 | bool isLittleEndian = getTarget().isLittleEndian(); |
| 12591 | Value *UndefValue = |
| 12592 | llvm::UndefValue::get(llvm::VectorType::get(Ops[0]->getType(), 2)); |
| 12593 | Value *Res = Builder.CreateInsertElement( |
| 12594 | UndefValue, Ops[0], (uint64_t)(isLittleEndian ? 1 : 0)); |
| 12595 | Res = Builder.CreateInsertElement(Res, Ops[1], |
| 12596 | (uint64_t)(isLittleEndian ? 0 : 1)); |
| 12597 | return Builder.CreateBitCast(Res, ConvertType(E->getType())); |
| 12598 | } |
| 12599 | |
| 12600 | case PPC::BI__builtin_unpack_vector_int128: { |
| 12601 | ConstantInt *Index = cast<ConstantInt>(Ops[1]); |
| 12602 | Value *Unpacked = Builder.CreateBitCast( |
| 12603 | Ops[0], llvm::VectorType::get(ConvertType(E->getType()), 2)); |
| 12604 | |
| 12605 | if (getTarget().isLittleEndian()) |
| 12606 | Index = ConstantInt::get(Index->getType(), 1 - Index->getZExtValue()); |
| 12607 | |
| 12608 | return Builder.CreateExtractElement(Unpacked, Index); |
| 12609 | } |
| 12610 | } |
| 12611 | } |
| 12612 | |
| 12613 | Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID, |
| 12614 | const CallExpr *E) { |
| 12615 | switch (BuiltinID) { |
| 12616 | case AMDGPU::BI__builtin_amdgcn_div_scale: |
| 12617 | case AMDGPU::BI__builtin_amdgcn_div_scalef: { |
| 12618 | // Translate from the intrinsics's struct return to the builtin's out |
| 12619 | // argument. |
| 12620 | |
| 12621 | Address FlagOutPtr = EmitPointerWithAlignment(E->getArg(3)); |
| 12622 | |
| 12623 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
| 12624 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12625 | llvm::Value *Z = EmitScalarExpr(E->getArg(2)); |
| 12626 | |
| 12627 | llvm::Function *Callee = CGM.getIntrinsic(Intrinsic::amdgcn_div_scale, |
| 12628 | X->getType()); |
| 12629 | |
| 12630 | llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z}); |
| 12631 | |
| 12632 | llvm::Value *Result = Builder.CreateExtractValue(Tmp, 0); |
| 12633 | llvm::Value *Flag = Builder.CreateExtractValue(Tmp, 1); |
| 12634 | |
| 12635 | llvm::Type *RealFlagType |
| 12636 | = FlagOutPtr.getPointer()->getType()->getPointerElementType(); |
| 12637 | |
| 12638 | llvm::Value *FlagExt = Builder.CreateZExt(Flag, RealFlagType); |
| 12639 | Builder.CreateStore(FlagExt, FlagOutPtr); |
| 12640 | return Result; |
| 12641 | } |
| 12642 | case AMDGPU::BI__builtin_amdgcn_div_fmas: |
| 12643 | case AMDGPU::BI__builtin_amdgcn_div_fmasf: { |
| 12644 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); |
| 12645 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); |
| 12646 | llvm::Value *Src2 = EmitScalarExpr(E->getArg(2)); |
| 12647 | llvm::Value *Src3 = EmitScalarExpr(E->getArg(3)); |
| 12648 | |
| 12649 | llvm::Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_div_fmas, |
| 12650 | Src0->getType()); |
| 12651 | llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3); |
| 12652 | return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool}); |
| 12653 | } |
| 12654 | |
| 12655 | case AMDGPU::BI__builtin_amdgcn_ds_swizzle: |
| 12656 | return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_ds_swizzle); |
| 12657 | case AMDGPU::BI__builtin_amdgcn_mov_dpp: |
| 12658 | case AMDGPU::BI__builtin_amdgcn_update_dpp: { |
| 12659 | llvm::SmallVector<llvm::Value *, 6> Args; |
| 12660 | for (unsigned I = 0; I != E->getNumArgs(); ++I) |
| 12661 | Args.push_back(EmitScalarExpr(E->getArg(I))); |
| 12662 | assert(Args.size() == 5 || Args.size() == 6)((Args.size() == 5 || Args.size() == 6) ? static_cast<void > (0) : __assert_fail ("Args.size() == 5 || Args.size() == 6" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12662, __PRETTY_FUNCTION__)); |
| 12663 | if (Args.size() == 5) |
| 12664 | Args.insert(Args.begin(), llvm::UndefValue::get(Args[0]->getType())); |
| 12665 | Function *F = |
| 12666 | CGM.getIntrinsic(Intrinsic::amdgcn_update_dpp, Args[0]->getType()); |
| 12667 | return Builder.CreateCall(F, Args); |
| 12668 | } |
| 12669 | case AMDGPU::BI__builtin_amdgcn_div_fixup: |
| 12670 | case AMDGPU::BI__builtin_amdgcn_div_fixupf: |
| 12671 | case AMDGPU::BI__builtin_amdgcn_div_fixuph: |
| 12672 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_div_fixup); |
| 12673 | case AMDGPU::BI__builtin_amdgcn_trig_preop: |
| 12674 | case AMDGPU::BI__builtin_amdgcn_trig_preopf: |
| 12675 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_trig_preop); |
| 12676 | case AMDGPU::BI__builtin_amdgcn_rcp: |
| 12677 | case AMDGPU::BI__builtin_amdgcn_rcpf: |
| 12678 | case AMDGPU::BI__builtin_amdgcn_rcph: |
| 12679 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rcp); |
| 12680 | case AMDGPU::BI__builtin_amdgcn_rsq: |
| 12681 | case AMDGPU::BI__builtin_amdgcn_rsqf: |
| 12682 | case AMDGPU::BI__builtin_amdgcn_rsqh: |
| 12683 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq); |
| 12684 | case AMDGPU::BI__builtin_amdgcn_rsq_clamp: |
| 12685 | case AMDGPU::BI__builtin_amdgcn_rsq_clampf: |
| 12686 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq_clamp); |
| 12687 | case AMDGPU::BI__builtin_amdgcn_sinf: |
| 12688 | case AMDGPU::BI__builtin_amdgcn_sinh: |
| 12689 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sin); |
| 12690 | case AMDGPU::BI__builtin_amdgcn_cosf: |
| 12691 | case AMDGPU::BI__builtin_amdgcn_cosh: |
| 12692 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_cos); |
| 12693 | case AMDGPU::BI__builtin_amdgcn_log_clampf: |
| 12694 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_log_clamp); |
| 12695 | case AMDGPU::BI__builtin_amdgcn_ldexp: |
| 12696 | case AMDGPU::BI__builtin_amdgcn_ldexpf: |
| 12697 | case AMDGPU::BI__builtin_amdgcn_ldexph: |
| 12698 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_ldexp); |
| 12699 | case AMDGPU::BI__builtin_amdgcn_frexp_mant: |
| 12700 | case AMDGPU::BI__builtin_amdgcn_frexp_mantf: |
| 12701 | case AMDGPU::BI__builtin_amdgcn_frexp_manth: |
| 12702 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_frexp_mant); |
| 12703 | case AMDGPU::BI__builtin_amdgcn_frexp_exp: |
| 12704 | case AMDGPU::BI__builtin_amdgcn_frexp_expf: { |
| 12705 | Value *Src0 = EmitScalarExpr(E->getArg(0)); |
| 12706 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp, |
| 12707 | { Builder.getInt32Ty(), Src0->getType() }); |
| 12708 | return Builder.CreateCall(F, Src0); |
| 12709 | } |
| 12710 | case AMDGPU::BI__builtin_amdgcn_frexp_exph: { |
| 12711 | Value *Src0 = EmitScalarExpr(E->getArg(0)); |
| 12712 | Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp, |
| 12713 | { Builder.getInt16Ty(), Src0->getType() }); |
| 12714 | return Builder.CreateCall(F, Src0); |
| 12715 | } |
| 12716 | case AMDGPU::BI__builtin_amdgcn_fract: |
| 12717 | case AMDGPU::BI__builtin_amdgcn_fractf: |
| 12718 | case AMDGPU::BI__builtin_amdgcn_fracth: |
| 12719 | return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_fract); |
| 12720 | case AMDGPU::BI__builtin_amdgcn_lerp: |
| 12721 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_lerp); |
| 12722 | case AMDGPU::BI__builtin_amdgcn_uicmp: |
| 12723 | case AMDGPU::BI__builtin_amdgcn_uicmpl: |
| 12724 | case AMDGPU::BI__builtin_amdgcn_sicmp: |
| 12725 | case AMDGPU::BI__builtin_amdgcn_sicmpl: |
| 12726 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_icmp); |
| 12727 | case AMDGPU::BI__builtin_amdgcn_fcmp: |
| 12728 | case AMDGPU::BI__builtin_amdgcn_fcmpf: |
| 12729 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fcmp); |
| 12730 | case AMDGPU::BI__builtin_amdgcn_class: |
| 12731 | case AMDGPU::BI__builtin_amdgcn_classf: |
| 12732 | case AMDGPU::BI__builtin_amdgcn_classh: |
| 12733 | return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_class); |
| 12734 | case AMDGPU::BI__builtin_amdgcn_fmed3f: |
| 12735 | case AMDGPU::BI__builtin_amdgcn_fmed3h: |
| 12736 | return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fmed3); |
| 12737 | case AMDGPU::BI__builtin_amdgcn_ds_append: |
| 12738 | case AMDGPU::BI__builtin_amdgcn_ds_consume: { |
| 12739 | Intrinsic::ID Intrin = BuiltinID == AMDGPU::BI__builtin_amdgcn_ds_append ? |
| 12740 | Intrinsic::amdgcn_ds_append : Intrinsic::amdgcn_ds_consume; |
| 12741 | Value *Src0 = EmitScalarExpr(E->getArg(0)); |
| 12742 | Function *F = CGM.getIntrinsic(Intrin, { Src0->getType() }); |
| 12743 | return Builder.CreateCall(F, { Src0, Builder.getFalse() }); |
| 12744 | } |
| 12745 | case AMDGPU::BI__builtin_amdgcn_read_exec: { |
| 12746 | CallInst *CI = cast<CallInst>( |
| 12747 | EmitSpecialRegisterBuiltin(*this, E, Int64Ty, Int64Ty, true, "exec")); |
| 12748 | CI->setConvergent(); |
| 12749 | return CI; |
| 12750 | } |
| 12751 | case AMDGPU::BI__builtin_amdgcn_read_exec_lo: |
| 12752 | case AMDGPU::BI__builtin_amdgcn_read_exec_hi: { |
| 12753 | StringRef RegName = BuiltinID == AMDGPU::BI__builtin_amdgcn_read_exec_lo ? |
| 12754 | "exec_lo" : "exec_hi"; |
| 12755 | CallInst *CI = cast<CallInst>( |
| 12756 | EmitSpecialRegisterBuiltin(*this, E, Int32Ty, Int32Ty, true, RegName)); |
| 12757 | CI->setConvergent(); |
| 12758 | return CI; |
| 12759 | } |
| 12760 | // amdgcn workitem |
| 12761 | case AMDGPU::BI__builtin_amdgcn_workitem_id_x: |
| 12762 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_x, 0, 1024); |
| 12763 | case AMDGPU::BI__builtin_amdgcn_workitem_id_y: |
| 12764 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_y, 0, 1024); |
| 12765 | case AMDGPU::BI__builtin_amdgcn_workitem_id_z: |
| 12766 | return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_z, 0, 1024); |
| 12767 | |
| 12768 | // r600 intrinsics |
| 12769 | case AMDGPU::BI__builtin_r600_recipsqrt_ieee: |
| 12770 | case AMDGPU::BI__builtin_r600_recipsqrt_ieeef: |
| 12771 | return emitUnaryBuiltin(*this, E, Intrinsic::r600_recipsqrt_ieee); |
| 12772 | case AMDGPU::BI__builtin_r600_read_tidig_x: |
| 12773 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_x, 0, 1024); |
| 12774 | case AMDGPU::BI__builtin_r600_read_tidig_y: |
| 12775 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_y, 0, 1024); |
| 12776 | case AMDGPU::BI__builtin_r600_read_tidig_z: |
| 12777 | return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_z, 0, 1024); |
| 12778 | default: |
| 12779 | return nullptr; |
| 12780 | } |
| 12781 | } |
| 12782 | |
| 12783 | /// Handle a SystemZ function in which the final argument is a pointer |
| 12784 | /// to an int that receives the post-instruction CC value. At the LLVM level |
| 12785 | /// this is represented as a function that returns a {result, cc} pair. |
| 12786 | static Value *EmitSystemZIntrinsicWithCC(CodeGenFunction &CGF, |
| 12787 | unsigned IntrinsicID, |
| 12788 | const CallExpr *E) { |
| 12789 | unsigned NumArgs = E->getNumArgs() - 1; |
| 12790 | SmallVector<Value *, 8> Args(NumArgs); |
| 12791 | for (unsigned I = 0; I < NumArgs; ++I) |
| 12792 | Args[I] = CGF.EmitScalarExpr(E->getArg(I)); |
| 12793 | Address CCPtr = CGF.EmitPointerWithAlignment(E->getArg(NumArgs)); |
| 12794 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID); |
| 12795 | Value *Call = CGF.Builder.CreateCall(F, Args); |
| 12796 | Value *CC = CGF.Builder.CreateExtractValue(Call, 1); |
| 12797 | CGF.Builder.CreateStore(CC, CCPtr); |
| 12798 | return CGF.Builder.CreateExtractValue(Call, 0); |
| 12799 | } |
| 12800 | |
| 12801 | Value *CodeGenFunction::EmitSystemZBuiltinExpr(unsigned BuiltinID, |
| 12802 | const CallExpr *E) { |
| 12803 | switch (BuiltinID) { |
| 12804 | case SystemZ::BI__builtin_tbegin: { |
| 12805 | Value *TDB = EmitScalarExpr(E->getArg(0)); |
| 12806 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c); |
| 12807 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin); |
| 12808 | return Builder.CreateCall(F, {TDB, Control}); |
| 12809 | } |
| 12810 | case SystemZ::BI__builtin_tbegin_nofloat: { |
| 12811 | Value *TDB = EmitScalarExpr(E->getArg(0)); |
| 12812 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c); |
| 12813 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin_nofloat); |
| 12814 | return Builder.CreateCall(F, {TDB, Control}); |
| 12815 | } |
| 12816 | case SystemZ::BI__builtin_tbeginc: { |
| 12817 | Value *TDB = llvm::ConstantPointerNull::get(Int8PtrTy); |
| 12818 | Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff08); |
| 12819 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tbeginc); |
| 12820 | return Builder.CreateCall(F, {TDB, Control}); |
| 12821 | } |
| 12822 | case SystemZ::BI__builtin_tabort: { |
| 12823 | Value *Data = EmitScalarExpr(E->getArg(0)); |
| 12824 | Function *F = CGM.getIntrinsic(Intrinsic::s390_tabort); |
| 12825 | return Builder.CreateCall(F, Builder.CreateSExt(Data, Int64Ty, "tabort")); |
| 12826 | } |
| 12827 | case SystemZ::BI__builtin_non_tx_store: { |
| 12828 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 12829 | Value *Data = EmitScalarExpr(E->getArg(1)); |
| 12830 | Function *F = CGM.getIntrinsic(Intrinsic::s390_ntstg); |
| 12831 | return Builder.CreateCall(F, {Data, Address}); |
| 12832 | } |
| 12833 | |
| 12834 | // Vector builtins. Note that most vector builtins are mapped automatically |
| 12835 | // to target-specific LLVM intrinsics. The ones handled specially here can |
| 12836 | // be represented via standard LLVM IR, which is preferable to enable common |
| 12837 | // LLVM optimizations. |
| 12838 | |
| 12839 | case SystemZ::BI__builtin_s390_vpopctb: |
| 12840 | case SystemZ::BI__builtin_s390_vpopcth: |
| 12841 | case SystemZ::BI__builtin_s390_vpopctf: |
| 12842 | case SystemZ::BI__builtin_s390_vpopctg: { |
| 12843 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12844 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12845 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType); |
| 12846 | return Builder.CreateCall(F, X); |
| 12847 | } |
| 12848 | |
| 12849 | case SystemZ::BI__builtin_s390_vclzb: |
| 12850 | case SystemZ::BI__builtin_s390_vclzh: |
| 12851 | case SystemZ::BI__builtin_s390_vclzf: |
| 12852 | case SystemZ::BI__builtin_s390_vclzg: { |
| 12853 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12854 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12855 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); |
| 12856 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType); |
| 12857 | return Builder.CreateCall(F, {X, Undef}); |
| 12858 | } |
| 12859 | |
| 12860 | case SystemZ::BI__builtin_s390_vctzb: |
| 12861 | case SystemZ::BI__builtin_s390_vctzh: |
| 12862 | case SystemZ::BI__builtin_s390_vctzf: |
| 12863 | case SystemZ::BI__builtin_s390_vctzg: { |
| 12864 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12865 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12866 | Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false); |
| 12867 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType); |
| 12868 | return Builder.CreateCall(F, {X, Undef}); |
| 12869 | } |
| 12870 | |
| 12871 | case SystemZ::BI__builtin_s390_vfsqsb: |
| 12872 | case SystemZ::BI__builtin_s390_vfsqdb: { |
| 12873 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12874 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12875 | Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType); |
| 12876 | return Builder.CreateCall(F, X); |
| 12877 | } |
| 12878 | case SystemZ::BI__builtin_s390_vfmasb: |
| 12879 | case SystemZ::BI__builtin_s390_vfmadb: { |
| 12880 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12881 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12882 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12883 | Value *Z = EmitScalarExpr(E->getArg(2)); |
| 12884 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); |
| 12885 | return Builder.CreateCall(F, {X, Y, Z}); |
| 12886 | } |
| 12887 | case SystemZ::BI__builtin_s390_vfmssb: |
| 12888 | case SystemZ::BI__builtin_s390_vfmsdb: { |
| 12889 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12890 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12891 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12892 | Value *Z = EmitScalarExpr(E->getArg(2)); |
| 12893 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); |
| 12894 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); |
| 12895 | return Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")}); |
| 12896 | } |
| 12897 | case SystemZ::BI__builtin_s390_vfnmasb: |
| 12898 | case SystemZ::BI__builtin_s390_vfnmadb: { |
| 12899 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12900 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12901 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12902 | Value *Z = EmitScalarExpr(E->getArg(2)); |
| 12903 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); |
| 12904 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); |
| 12905 | return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, Z}), "sub"); |
| 12906 | } |
| 12907 | case SystemZ::BI__builtin_s390_vfnmssb: |
| 12908 | case SystemZ::BI__builtin_s390_vfnmsdb: { |
| 12909 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12910 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12911 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12912 | Value *Z = EmitScalarExpr(E->getArg(2)); |
| 12913 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); |
| 12914 | Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType); |
| 12915 | Value *NegZ = Builder.CreateFSub(Zero, Z, "sub"); |
| 12916 | return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, NegZ})); |
| 12917 | } |
| 12918 | case SystemZ::BI__builtin_s390_vflpsb: |
| 12919 | case SystemZ::BI__builtin_s390_vflpdb: { |
| 12920 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12921 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12922 | Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); |
| 12923 | return Builder.CreateCall(F, X); |
| 12924 | } |
| 12925 | case SystemZ::BI__builtin_s390_vflnsb: |
| 12926 | case SystemZ::BI__builtin_s390_vflndb: { |
| 12927 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12928 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12929 | Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType); |
| 12930 | Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType); |
| 12931 | return Builder.CreateFSub(Zero, Builder.CreateCall(F, X), "sub"); |
| 12932 | } |
| 12933 | case SystemZ::BI__builtin_s390_vfisb: |
| 12934 | case SystemZ::BI__builtin_s390_vfidb: { |
| 12935 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12936 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12937 | // Constant-fold the M4 and M5 mask arguments. |
| 12938 | llvm::APSInt M4, M5; |
| 12939 | bool IsConstM4 = E->getArg(1)->isIntegerConstantExpr(M4, getContext()); |
| 12940 | bool IsConstM5 = E->getArg(2)->isIntegerConstantExpr(M5, getContext()); |
| 12941 | assert(IsConstM4 && IsConstM5 && "Constant arg isn't actually constant?")((IsConstM4 && IsConstM5 && "Constant arg isn't actually constant?" ) ? static_cast<void> (0) : __assert_fail ("IsConstM4 && IsConstM5 && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12941, __PRETTY_FUNCTION__)); |
| 12942 | (void)IsConstM4; (void)IsConstM5; |
| 12943 | // Check whether this instance can be represented via a LLVM standard |
| 12944 | // intrinsic. We only support some combinations of M4 and M5. |
| 12945 | Intrinsic::ID ID = Intrinsic::not_intrinsic; |
| 12946 | switch (M4.getZExtValue()) { |
| 12947 | default: break; |
| 12948 | case 0: // IEEE-inexact exception allowed |
| 12949 | switch (M5.getZExtValue()) { |
| 12950 | default: break; |
| 12951 | case 0: ID = Intrinsic::rint; break; |
| 12952 | } |
| 12953 | break; |
| 12954 | case 4: // IEEE-inexact exception suppressed |
| 12955 | switch (M5.getZExtValue()) { |
| 12956 | default: break; |
| 12957 | case 0: ID = Intrinsic::nearbyint; break; |
| 12958 | case 1: ID = Intrinsic::round; break; |
| 12959 | case 5: ID = Intrinsic::trunc; break; |
| 12960 | case 6: ID = Intrinsic::ceil; break; |
| 12961 | case 7: ID = Intrinsic::floor; break; |
| 12962 | } |
| 12963 | break; |
| 12964 | } |
| 12965 | if (ID != Intrinsic::not_intrinsic) { |
| 12966 | Function *F = CGM.getIntrinsic(ID, ResultType); |
| 12967 | return Builder.CreateCall(F, X); |
| 12968 | } |
| 12969 | switch (BuiltinID) { |
| 12970 | case SystemZ::BI__builtin_s390_vfisb: ID = Intrinsic::s390_vfisb; break; |
| 12971 | case SystemZ::BI__builtin_s390_vfidb: ID = Intrinsic::s390_vfidb; break; |
| 12972 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12972); |
| 12973 | } |
| 12974 | Function *F = CGM.getIntrinsic(ID); |
| 12975 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); |
| 12976 | Value *M5Value = llvm::ConstantInt::get(getLLVMContext(), M5); |
| 12977 | return Builder.CreateCall(F, {X, M4Value, M5Value}); |
| 12978 | } |
| 12979 | case SystemZ::BI__builtin_s390_vfmaxsb: |
| 12980 | case SystemZ::BI__builtin_s390_vfmaxdb: { |
| 12981 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 12982 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 12983 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 12984 | // Constant-fold the M4 mask argument. |
| 12985 | llvm::APSInt M4; |
| 12986 | bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext()); |
| 12987 | assert(IsConstM4 && "Constant arg isn't actually constant?")((IsConstM4 && "Constant arg isn't actually constant?" ) ? static_cast<void> (0) : __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 12987, __PRETTY_FUNCTION__)); |
| 12988 | (void)IsConstM4; |
| 12989 | // Check whether this instance can be represented via a LLVM standard |
| 12990 | // intrinsic. We only support some values of M4. |
| 12991 | Intrinsic::ID ID = Intrinsic::not_intrinsic; |
| 12992 | switch (M4.getZExtValue()) { |
| 12993 | default: break; |
| 12994 | case 4: ID = Intrinsic::maxnum; break; |
| 12995 | } |
| 12996 | if (ID != Intrinsic::not_intrinsic) { |
| 12997 | Function *F = CGM.getIntrinsic(ID, ResultType); |
| 12998 | return Builder.CreateCall(F, {X, Y}); |
| 12999 | } |
| 13000 | switch (BuiltinID) { |
| 13001 | case SystemZ::BI__builtin_s390_vfmaxsb: ID = Intrinsic::s390_vfmaxsb; break; |
| 13002 | case SystemZ::BI__builtin_s390_vfmaxdb: ID = Intrinsic::s390_vfmaxdb; break; |
| 13003 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13003); |
| 13004 | } |
| 13005 | Function *F = CGM.getIntrinsic(ID); |
| 13006 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); |
| 13007 | return Builder.CreateCall(F, {X, Y, M4Value}); |
| 13008 | } |
| 13009 | case SystemZ::BI__builtin_s390_vfminsb: |
| 13010 | case SystemZ::BI__builtin_s390_vfmindb: { |
| 13011 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 13012 | Value *X = EmitScalarExpr(E->getArg(0)); |
| 13013 | Value *Y = EmitScalarExpr(E->getArg(1)); |
| 13014 | // Constant-fold the M4 mask argument. |
| 13015 | llvm::APSInt M4; |
| 13016 | bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext()); |
| 13017 | assert(IsConstM4 && "Constant arg isn't actually constant?")((IsConstM4 && "Constant arg isn't actually constant?" ) ? static_cast<void> (0) : __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\"" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13017, __PRETTY_FUNCTION__)); |
| 13018 | (void)IsConstM4; |
| 13019 | // Check whether this instance can be represented via a LLVM standard |
| 13020 | // intrinsic. We only support some values of M4. |
| 13021 | Intrinsic::ID ID = Intrinsic::not_intrinsic; |
| 13022 | switch (M4.getZExtValue()) { |
| 13023 | default: break; |
| 13024 | case 4: ID = Intrinsic::minnum; break; |
| 13025 | } |
| 13026 | if (ID != Intrinsic::not_intrinsic) { |
| 13027 | Function *F = CGM.getIntrinsic(ID, ResultType); |
| 13028 | return Builder.CreateCall(F, {X, Y}); |
| 13029 | } |
| 13030 | switch (BuiltinID) { |
| 13031 | case SystemZ::BI__builtin_s390_vfminsb: ID = Intrinsic::s390_vfminsb; break; |
| 13032 | case SystemZ::BI__builtin_s390_vfmindb: ID = Intrinsic::s390_vfmindb; break; |
| 13033 | default: llvm_unreachable("Unknown BuiltinID")::llvm::llvm_unreachable_internal("Unknown BuiltinID", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13033); |
| 13034 | } |
| 13035 | Function *F = CGM.getIntrinsic(ID); |
| 13036 | Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4); |
| 13037 | return Builder.CreateCall(F, {X, Y, M4Value}); |
| 13038 | } |
| 13039 | |
| 13040 | // Vector intrinsics that output the post-instruction CC value. |
| 13041 | |
| 13042 | #define INTRINSIC_WITH_CC(NAME) \ |
| 13043 | case SystemZ::BI__builtin_##NAME: \ |
| 13044 | return EmitSystemZIntrinsicWithCC(*this, Intrinsic::NAME, E) |
| 13045 | |
| 13046 | INTRINSIC_WITH_CC(s390_vpkshs); |
| 13047 | INTRINSIC_WITH_CC(s390_vpksfs); |
| 13048 | INTRINSIC_WITH_CC(s390_vpksgs); |
| 13049 | |
| 13050 | INTRINSIC_WITH_CC(s390_vpklshs); |
| 13051 | INTRINSIC_WITH_CC(s390_vpklsfs); |
| 13052 | INTRINSIC_WITH_CC(s390_vpklsgs); |
| 13053 | |
| 13054 | INTRINSIC_WITH_CC(s390_vceqbs); |
| 13055 | INTRINSIC_WITH_CC(s390_vceqhs); |
| 13056 | INTRINSIC_WITH_CC(s390_vceqfs); |
| 13057 | INTRINSIC_WITH_CC(s390_vceqgs); |
| 13058 | |
| 13059 | INTRINSIC_WITH_CC(s390_vchbs); |
| 13060 | INTRINSIC_WITH_CC(s390_vchhs); |
| 13061 | INTRINSIC_WITH_CC(s390_vchfs); |
| 13062 | INTRINSIC_WITH_CC(s390_vchgs); |
| 13063 | |
| 13064 | INTRINSIC_WITH_CC(s390_vchlbs); |
| 13065 | INTRINSIC_WITH_CC(s390_vchlhs); |
| 13066 | INTRINSIC_WITH_CC(s390_vchlfs); |
| 13067 | INTRINSIC_WITH_CC(s390_vchlgs); |
| 13068 | |
| 13069 | INTRINSIC_WITH_CC(s390_vfaebs); |
| 13070 | INTRINSIC_WITH_CC(s390_vfaehs); |
| 13071 | INTRINSIC_WITH_CC(s390_vfaefs); |
| 13072 | |
| 13073 | INTRINSIC_WITH_CC(s390_vfaezbs); |
| 13074 | INTRINSIC_WITH_CC(s390_vfaezhs); |
| 13075 | INTRINSIC_WITH_CC(s390_vfaezfs); |
| 13076 | |
| 13077 | INTRINSIC_WITH_CC(s390_vfeebs); |
| 13078 | INTRINSIC_WITH_CC(s390_vfeehs); |
| 13079 | INTRINSIC_WITH_CC(s390_vfeefs); |
| 13080 | |
| 13081 | INTRINSIC_WITH_CC(s390_vfeezbs); |
| 13082 | INTRINSIC_WITH_CC(s390_vfeezhs); |
| 13083 | INTRINSIC_WITH_CC(s390_vfeezfs); |
| 13084 | |
| 13085 | INTRINSIC_WITH_CC(s390_vfenebs); |
| 13086 | INTRINSIC_WITH_CC(s390_vfenehs); |
| 13087 | INTRINSIC_WITH_CC(s390_vfenefs); |
| 13088 | |
| 13089 | INTRINSIC_WITH_CC(s390_vfenezbs); |
| 13090 | INTRINSIC_WITH_CC(s390_vfenezhs); |
| 13091 | INTRINSIC_WITH_CC(s390_vfenezfs); |
| 13092 | |
| 13093 | INTRINSIC_WITH_CC(s390_vistrbs); |
| 13094 | INTRINSIC_WITH_CC(s390_vistrhs); |
| 13095 | INTRINSIC_WITH_CC(s390_vistrfs); |
| 13096 | |
| 13097 | INTRINSIC_WITH_CC(s390_vstrcbs); |
| 13098 | INTRINSIC_WITH_CC(s390_vstrchs); |
| 13099 | INTRINSIC_WITH_CC(s390_vstrcfs); |
| 13100 | |
| 13101 | INTRINSIC_WITH_CC(s390_vstrczbs); |
| 13102 | INTRINSIC_WITH_CC(s390_vstrczhs); |
| 13103 | INTRINSIC_WITH_CC(s390_vstrczfs); |
| 13104 | |
| 13105 | INTRINSIC_WITH_CC(s390_vfcesbs); |
| 13106 | INTRINSIC_WITH_CC(s390_vfcedbs); |
| 13107 | INTRINSIC_WITH_CC(s390_vfchsbs); |
| 13108 | INTRINSIC_WITH_CC(s390_vfchdbs); |
| 13109 | INTRINSIC_WITH_CC(s390_vfchesbs); |
| 13110 | INTRINSIC_WITH_CC(s390_vfchedbs); |
| 13111 | |
| 13112 | INTRINSIC_WITH_CC(s390_vftcisb); |
| 13113 | INTRINSIC_WITH_CC(s390_vftcidb); |
| 13114 | |
| 13115 | #undef INTRINSIC_WITH_CC |
| 13116 | |
| 13117 | default: |
| 13118 | return nullptr; |
| 13119 | } |
| 13120 | } |
| 13121 | |
| 13122 | namespace { |
| 13123 | // Helper classes for mapping MMA builtins to particular LLVM intrinsic variant. |
| 13124 | struct NVPTXMmaLdstInfo { |
| 13125 | unsigned NumResults; // Number of elements to load/store |
| 13126 | // Intrinsic IDs for row/col variants. 0 if particular layout is unsupported. |
| 13127 | unsigned IID_col; |
| 13128 | unsigned IID_row; |
| 13129 | }; |
| 13130 | |
| 13131 | #define MMA_INTR(geom_op_type, layout) \ |
| 13132 | Intrinsic::nvvm_wmma_##geom_op_type##_##layout##_stride |
| 13133 | #define MMA_LDST(n, geom_op_type) \ |
| 13134 | { n, MMA_INTR(geom_op_type, col), MMA_INTR(geom_op_type, row) } |
| 13135 | |
| 13136 | static NVPTXMmaLdstInfo getNVPTXMmaLdstInfo(unsigned BuiltinID) { |
| 13137 | switch (BuiltinID) { |
| 13138 | // FP MMA loads |
| 13139 | case NVPTX::BI__hmma_m16n16k16_ld_a: |
| 13140 | return MMA_LDST(8, m16n16k16_load_a_f16); |
| 13141 | case NVPTX::BI__hmma_m16n16k16_ld_b: |
| 13142 | return MMA_LDST(8, m16n16k16_load_b_f16); |
| 13143 | case NVPTX::BI__hmma_m16n16k16_ld_c_f16: |
| 13144 | return MMA_LDST(4, m16n16k16_load_c_f16); |
| 13145 | case NVPTX::BI__hmma_m16n16k16_ld_c_f32: |
| 13146 | return MMA_LDST(8, m16n16k16_load_c_f32); |
| 13147 | case NVPTX::BI__hmma_m32n8k16_ld_a: |
| 13148 | return MMA_LDST(8, m32n8k16_load_a_f16); |
| 13149 | case NVPTX::BI__hmma_m32n8k16_ld_b: |
| 13150 | return MMA_LDST(8, m32n8k16_load_b_f16); |
| 13151 | case NVPTX::BI__hmma_m32n8k16_ld_c_f16: |
| 13152 | return MMA_LDST(4, m32n8k16_load_c_f16); |
| 13153 | case NVPTX::BI__hmma_m32n8k16_ld_c_f32: |
| 13154 | return MMA_LDST(8, m32n8k16_load_c_f32); |
| 13155 | case NVPTX::BI__hmma_m8n32k16_ld_a: |
| 13156 | return MMA_LDST(8, m8n32k16_load_a_f16); |
| 13157 | case NVPTX::BI__hmma_m8n32k16_ld_b: |
| 13158 | return MMA_LDST(8, m8n32k16_load_b_f16); |
| 13159 | case NVPTX::BI__hmma_m8n32k16_ld_c_f16: |
| 13160 | return MMA_LDST(4, m8n32k16_load_c_f16); |
| 13161 | case NVPTX::BI__hmma_m8n32k16_ld_c_f32: |
| 13162 | return MMA_LDST(8, m8n32k16_load_c_f32); |
| 13163 | |
| 13164 | // Integer MMA loads |
| 13165 | case NVPTX::BI__imma_m16n16k16_ld_a_s8: |
| 13166 | return MMA_LDST(2, m16n16k16_load_a_s8); |
| 13167 | case NVPTX::BI__imma_m16n16k16_ld_a_u8: |
| 13168 | return MMA_LDST(2, m16n16k16_load_a_u8); |
| 13169 | case NVPTX::BI__imma_m16n16k16_ld_b_s8: |
| 13170 | return MMA_LDST(2, m16n16k16_load_b_s8); |
| 13171 | case NVPTX::BI__imma_m16n16k16_ld_b_u8: |
| 13172 | return MMA_LDST(2, m16n16k16_load_b_u8); |
| 13173 | case NVPTX::BI__imma_m16n16k16_ld_c: |
| 13174 | return MMA_LDST(8, m16n16k16_load_c_s32); |
| 13175 | case NVPTX::BI__imma_m32n8k16_ld_a_s8: |
| 13176 | return MMA_LDST(4, m32n8k16_load_a_s8); |
| 13177 | case NVPTX::BI__imma_m32n8k16_ld_a_u8: |
| 13178 | return MMA_LDST(4, m32n8k16_load_a_u8); |
| 13179 | case NVPTX::BI__imma_m32n8k16_ld_b_s8: |
| 13180 | return MMA_LDST(1, m32n8k16_load_b_s8); |
| 13181 | case NVPTX::BI__imma_m32n8k16_ld_b_u8: |
| 13182 | return MMA_LDST(1, m32n8k16_load_b_u8); |
| 13183 | case NVPTX::BI__imma_m32n8k16_ld_c: |
| 13184 | return MMA_LDST(8, m32n8k16_load_c_s32); |
| 13185 | case NVPTX::BI__imma_m8n32k16_ld_a_s8: |
| 13186 | return MMA_LDST(1, m8n32k16_load_a_s8); |
| 13187 | case NVPTX::BI__imma_m8n32k16_ld_a_u8: |
| 13188 | return MMA_LDST(1, m8n32k16_load_a_u8); |
| 13189 | case NVPTX::BI__imma_m8n32k16_ld_b_s8: |
| 13190 | return MMA_LDST(4, m8n32k16_load_b_s8); |
| 13191 | case NVPTX::BI__imma_m8n32k16_ld_b_u8: |
| 13192 | return MMA_LDST(4, m8n32k16_load_b_u8); |
| 13193 | case NVPTX::BI__imma_m8n32k16_ld_c: |
| 13194 | return MMA_LDST(8, m8n32k16_load_c_s32); |
| 13195 | |
| 13196 | // Sub-integer MMA loads. |
| 13197 | // Only row/col layout is supported by A/B fragments. |
| 13198 | case NVPTX::BI__imma_m8n8k32_ld_a_s4: |
| 13199 | return {1, 0, MMA_INTR(m8n8k32_load_a_s4, row)}; |
| 13200 | case NVPTX::BI__imma_m8n8k32_ld_a_u4: |
| 13201 | return {1, 0, MMA_INTR(m8n8k32_load_a_u4, row)}; |
| 13202 | case NVPTX::BI__imma_m8n8k32_ld_b_s4: |
| 13203 | return {1, MMA_INTR(m8n8k32_load_b_s4, col), 0}; |
| 13204 | case NVPTX::BI__imma_m8n8k32_ld_b_u4: |
| 13205 | return {1, MMA_INTR(m8n8k32_load_b_u4, col), 0}; |
| 13206 | case NVPTX::BI__imma_m8n8k32_ld_c: |
| 13207 | return MMA_LDST(2, m8n8k32_load_c_s32); |
| 13208 | case NVPTX::BI__bmma_m8n8k128_ld_a_b1: |
| 13209 | return {1, 0, MMA_INTR(m8n8k128_load_a_b1, row)}; |
| 13210 | case NVPTX::BI__bmma_m8n8k128_ld_b_b1: |
| 13211 | return {1, MMA_INTR(m8n8k128_load_b_b1, col), 0}; |
| 13212 | case NVPTX::BI__bmma_m8n8k128_ld_c: |
| 13213 | return MMA_LDST(2, m8n8k128_load_c_s32); |
| 13214 | |
| 13215 | // NOTE: We need to follow inconsitent naming scheme used by NVCC. Unlike |
| 13216 | // PTX and LLVM IR where stores always use fragment D, NVCC builtins always |
| 13217 | // use fragment C for both loads and stores. |
| 13218 | // FP MMA stores. |
| 13219 | case NVPTX::BI__hmma_m16n16k16_st_c_f16: |
| 13220 | return MMA_LDST(4, m16n16k16_store_d_f16); |
| 13221 | case NVPTX::BI__hmma_m16n16k16_st_c_f32: |
| 13222 | return MMA_LDST(8, m16n16k16_store_d_f32); |
| 13223 | case NVPTX::BI__hmma_m32n8k16_st_c_f16: |
| 13224 | return MMA_LDST(4, m32n8k16_store_d_f16); |
| 13225 | case NVPTX::BI__hmma_m32n8k16_st_c_f32: |
| 13226 | return MMA_LDST(8, m32n8k16_store_d_f32); |
| 13227 | case NVPTX::BI__hmma_m8n32k16_st_c_f16: |
| 13228 | return MMA_LDST(4, m8n32k16_store_d_f16); |
| 13229 | case NVPTX::BI__hmma_m8n32k16_st_c_f32: |
| 13230 | return MMA_LDST(8, m8n32k16_store_d_f32); |
| 13231 | |
| 13232 | // Integer and sub-integer MMA stores. |
| 13233 | // Another naming quirk. Unlike other MMA builtins that use PTX types in the |
| 13234 | // name, integer loads/stores use LLVM's i32. |
| 13235 | case NVPTX::BI__imma_m16n16k16_st_c_i32: |
| 13236 | return MMA_LDST(8, m16n16k16_store_d_s32); |
| 13237 | case NVPTX::BI__imma_m32n8k16_st_c_i32: |
| 13238 | return MMA_LDST(8, m32n8k16_store_d_s32); |
| 13239 | case NVPTX::BI__imma_m8n32k16_st_c_i32: |
| 13240 | return MMA_LDST(8, m8n32k16_store_d_s32); |
| 13241 | case NVPTX::BI__imma_m8n8k32_st_c_i32: |
| 13242 | return MMA_LDST(2, m8n8k32_store_d_s32); |
| 13243 | case NVPTX::BI__bmma_m8n8k128_st_c_i32: |
| 13244 | return MMA_LDST(2, m8n8k128_store_d_s32); |
| 13245 | |
| 13246 | default: |
| 13247 | llvm_unreachable("Unknown MMA builtin")::llvm::llvm_unreachable_internal("Unknown MMA builtin", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13247); |
| 13248 | } |
| 13249 | } |
| 13250 | #undef MMA_LDST |
| 13251 | #undef MMA_INTR |
| 13252 | |
| 13253 | |
| 13254 | struct NVPTXMmaInfo { |
| 13255 | unsigned NumEltsA; |
| 13256 | unsigned NumEltsB; |
| 13257 | unsigned NumEltsC; |
| 13258 | unsigned NumEltsD; |
| 13259 | std::array<unsigned, 8> Variants; |
| 13260 | |
| 13261 | unsigned getMMAIntrinsic(int Layout, bool Satf) { |
| 13262 | unsigned Index = Layout * 2 + Satf; |
| 13263 | if (Index >= Variants.size()) |
| 13264 | return 0; |
| 13265 | return Variants[Index]; |
| 13266 | } |
| 13267 | }; |
| 13268 | |
| 13269 | // Returns an intrinsic that matches Layout and Satf for valid combinations of |
| 13270 | // Layout and Satf, 0 otherwise. |
| 13271 | static NVPTXMmaInfo getNVPTXMmaInfo(unsigned BuiltinID) { |
| 13272 | // clang-format off |
| 13273 | #define MMA_VARIANTS(geom, type) {{ \ |
| 13274 | Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type, \ |
| 13275 | Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type##_satfinite, \ |
| 13276 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ |
| 13277 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \ |
| 13278 | Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type, \ |
| 13279 | Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type##_satfinite, \ |
| 13280 | Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type, \ |
| 13281 | Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type##_satfinite \ |
| 13282 | }} |
| 13283 | // Sub-integer MMA only supports row.col layout. |
| 13284 | #define MMA_VARIANTS_I4(geom, type) {{ \ |
| 13285 | 0, \ |
| 13286 | 0, \ |
| 13287 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ |
| 13288 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \ |
| 13289 | 0, \ |
| 13290 | 0, \ |
| 13291 | 0, \ |
| 13292 | 0 \ |
| 13293 | }} |
| 13294 | // b1 MMA does not support .satfinite. |
| 13295 | #define MMA_VARIANTS_B1(geom, type) {{ \ |
| 13296 | 0, \ |
| 13297 | 0, \ |
| 13298 | Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type, \ |
| 13299 | 0, \ |
| 13300 | 0, \ |
| 13301 | 0, \ |
| 13302 | 0, \ |
| 13303 | 0 \ |
| 13304 | }} |
| 13305 | // clang-format on |
| 13306 | switch (BuiltinID) { |
| 13307 | // FP MMA |
| 13308 | // Note that 'type' argument of MMA_VARIANT uses D_C notation, while |
| 13309 | // NumEltsN of return value are ordered as A,B,C,D. |
| 13310 | case NVPTX::BI__hmma_m16n16k16_mma_f16f16: |
| 13311 | return {8, 8, 4, 4, MMA_VARIANTS(m16n16k16, f16_f16)}; |
| 13312 | case NVPTX::BI__hmma_m16n16k16_mma_f32f16: |
| 13313 | return {8, 8, 4, 8, MMA_VARIANTS(m16n16k16, f32_f16)}; |
| 13314 | case NVPTX::BI__hmma_m16n16k16_mma_f16f32: |
| 13315 | return {8, 8, 8, 4, MMA_VARIANTS(m16n16k16, f16_f32)}; |
| 13316 | case NVPTX::BI__hmma_m16n16k16_mma_f32f32: |
| 13317 | return {8, 8, 8, 8, MMA_VARIANTS(m16n16k16, f32_f32)}; |
| 13318 | case NVPTX::BI__hmma_m32n8k16_mma_f16f16: |
| 13319 | return {8, 8, 4, 4, MMA_VARIANTS(m32n8k16, f16_f16)}; |
| 13320 | case NVPTX::BI__hmma_m32n8k16_mma_f32f16: |
| 13321 | return {8, 8, 4, 8, MMA_VARIANTS(m32n8k16, f32_f16)}; |
| 13322 | case NVPTX::BI__hmma_m32n8k16_mma_f16f32: |
| 13323 | return {8, 8, 8, 4, MMA_VARIANTS(m32n8k16, f16_f32)}; |
| 13324 | case NVPTX::BI__hmma_m32n8k16_mma_f32f32: |
| 13325 | return {8, 8, 8, 8, MMA_VARIANTS(m32n8k16, f32_f32)}; |
| 13326 | case NVPTX::BI__hmma_m8n32k16_mma_f16f16: |
| 13327 | return {8, 8, 4, 4, MMA_VARIANTS(m8n32k16, f16_f16)}; |
| 13328 | case NVPTX::BI__hmma_m8n32k16_mma_f32f16: |
| 13329 | return {8, 8, 4, 8, MMA_VARIANTS(m8n32k16, f32_f16)}; |
| 13330 | case NVPTX::BI__hmma_m8n32k16_mma_f16f32: |
| 13331 | return {8, 8, 8, 4, MMA_VARIANTS(m8n32k16, f16_f32)}; |
| 13332 | case NVPTX::BI__hmma_m8n32k16_mma_f32f32: |
| 13333 | return {8, 8, 8, 8, MMA_VARIANTS(m8n32k16, f32_f32)}; |
| 13334 | |
| 13335 | // Integer MMA |
| 13336 | case NVPTX::BI__imma_m16n16k16_mma_s8: |
| 13337 | return {2, 2, 8, 8, MMA_VARIANTS(m16n16k16, s8)}; |
| 13338 | case NVPTX::BI__imma_m16n16k16_mma_u8: |
| 13339 | return {2, 2, 8, 8, MMA_VARIANTS(m16n16k16, u8)}; |
| 13340 | case NVPTX::BI__imma_m32n8k16_mma_s8: |
| 13341 | return {4, 1, 8, 8, MMA_VARIANTS(m32n8k16, s8)}; |
| 13342 | case NVPTX::BI__imma_m32n8k16_mma_u8: |
| 13343 | return {4, 1, 8, 8, MMA_VARIANTS(m32n8k16, u8)}; |
| 13344 | case NVPTX::BI__imma_m8n32k16_mma_s8: |
| 13345 | return {1, 4, 8, 8, MMA_VARIANTS(m8n32k16, s8)}; |
| 13346 | case NVPTX::BI__imma_m8n32k16_mma_u8: |
| 13347 | return {1, 4, 8, 8, MMA_VARIANTS(m8n32k16, u8)}; |
| 13348 | |
| 13349 | // Sub-integer MMA |
| 13350 | case NVPTX::BI__imma_m8n8k32_mma_s4: |
| 13351 | return {1, 1, 2, 2, MMA_VARIANTS_I4(m8n8k32, s4)}; |
| 13352 | case NVPTX::BI__imma_m8n8k32_mma_u4: |
| 13353 | return {1, 1, 2, 2, MMA_VARIANTS_I4(m8n8k32, u4)}; |
| 13354 | case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1: |
| 13355 | return {1, 1, 2, 2, MMA_VARIANTS_B1(m8n8k128, b1)}; |
| 13356 | default: |
| 13357 | llvm_unreachable("Unexpected builtin ID.")::llvm::llvm_unreachable_internal("Unexpected builtin ID.", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13357); |
| 13358 | } |
| 13359 | #undef MMA_VARIANTS |
| 13360 | #undef MMA_VARIANTS_I4 |
| 13361 | #undef MMA_VARIANTS_B1 |
| 13362 | } |
| 13363 | |
| 13364 | } // namespace |
| 13365 | |
| 13366 | Value * |
| 13367 | CodeGenFunction::EmitNVPTXBuiltinExpr(unsigned BuiltinID, const CallExpr *E) { |
| 13368 | auto MakeLdg = [&](unsigned IntrinsicID) { |
| 13369 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13370 | clang::CharUnits Align = |
| 13371 | getNaturalPointeeTypeAlignment(E->getArg(0)->getType()); |
| 13372 | return Builder.CreateCall( |
| 13373 | CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(), |
| 13374 | Ptr->getType()}), |
| 13375 | {Ptr, ConstantInt::get(Builder.getInt32Ty(), Align.getQuantity())}); |
| 13376 | }; |
| 13377 | auto MakeScopedAtomic = [&](unsigned IntrinsicID) { |
| 13378 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13379 | return Builder.CreateCall( |
| 13380 | CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(), |
| 13381 | Ptr->getType()}), |
| 13382 | {Ptr, EmitScalarExpr(E->getArg(1))}); |
| 13383 | }; |
| 13384 | switch (BuiltinID) { |
| 13385 | case NVPTX::BI__nvvm_atom_add_gen_i: |
| 13386 | case NVPTX::BI__nvvm_atom_add_gen_l: |
| 13387 | case NVPTX::BI__nvvm_atom_add_gen_ll: |
| 13388 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Add, E); |
| 13389 | |
| 13390 | case NVPTX::BI__nvvm_atom_sub_gen_i: |
| 13391 | case NVPTX::BI__nvvm_atom_sub_gen_l: |
| 13392 | case NVPTX::BI__nvvm_atom_sub_gen_ll: |
| 13393 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Sub, E); |
| 13394 | |
| 13395 | case NVPTX::BI__nvvm_atom_and_gen_i: |
| 13396 | case NVPTX::BI__nvvm_atom_and_gen_l: |
| 13397 | case NVPTX::BI__nvvm_atom_and_gen_ll: |
| 13398 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::And, E); |
| 13399 | |
| 13400 | case NVPTX::BI__nvvm_atom_or_gen_i: |
| 13401 | case NVPTX::BI__nvvm_atom_or_gen_l: |
| 13402 | case NVPTX::BI__nvvm_atom_or_gen_ll: |
| 13403 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Or, E); |
| 13404 | |
| 13405 | case NVPTX::BI__nvvm_atom_xor_gen_i: |
| 13406 | case NVPTX::BI__nvvm_atom_xor_gen_l: |
| 13407 | case NVPTX::BI__nvvm_atom_xor_gen_ll: |
| 13408 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xor, E); |
| 13409 | |
| 13410 | case NVPTX::BI__nvvm_atom_xchg_gen_i: |
| 13411 | case NVPTX::BI__nvvm_atom_xchg_gen_l: |
| 13412 | case NVPTX::BI__nvvm_atom_xchg_gen_ll: |
| 13413 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xchg, E); |
| 13414 | |
| 13415 | case NVPTX::BI__nvvm_atom_max_gen_i: |
| 13416 | case NVPTX::BI__nvvm_atom_max_gen_l: |
| 13417 | case NVPTX::BI__nvvm_atom_max_gen_ll: |
| 13418 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E); |
| 13419 | |
| 13420 | case NVPTX::BI__nvvm_atom_max_gen_ui: |
| 13421 | case NVPTX::BI__nvvm_atom_max_gen_ul: |
| 13422 | case NVPTX::BI__nvvm_atom_max_gen_ull: |
| 13423 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMax, E); |
| 13424 | |
| 13425 | case NVPTX::BI__nvvm_atom_min_gen_i: |
| 13426 | case NVPTX::BI__nvvm_atom_min_gen_l: |
| 13427 | case NVPTX::BI__nvvm_atom_min_gen_ll: |
| 13428 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E); |
| 13429 | |
| 13430 | case NVPTX::BI__nvvm_atom_min_gen_ui: |
| 13431 | case NVPTX::BI__nvvm_atom_min_gen_ul: |
| 13432 | case NVPTX::BI__nvvm_atom_min_gen_ull: |
| 13433 | return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMin, E); |
| 13434 | |
| 13435 | case NVPTX::BI__nvvm_atom_cas_gen_i: |
| 13436 | case NVPTX::BI__nvvm_atom_cas_gen_l: |
| 13437 | case NVPTX::BI__nvvm_atom_cas_gen_ll: |
| 13438 | // __nvvm_atom_cas_gen_* should return the old value rather than the |
| 13439 | // success flag. |
| 13440 | return MakeAtomicCmpXchgValue(*this, E, /*ReturnBool=*/false); |
| 13441 | |
| 13442 | case NVPTX::BI__nvvm_atom_add_gen_f: { |
| 13443 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13444 | Value *Val = EmitScalarExpr(E->getArg(1)); |
| 13445 | // atomicrmw only deals with integer arguments so we need to use |
| 13446 | // LLVM's nvvm_atomic_load_add_f32 intrinsic for that. |
| 13447 | Function *FnALAF32 = |
| 13448 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_add_f32, Ptr->getType()); |
| 13449 | return Builder.CreateCall(FnALAF32, {Ptr, Val}); |
| 13450 | } |
| 13451 | |
| 13452 | case NVPTX::BI__nvvm_atom_add_gen_d: { |
| 13453 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13454 | Value *Val = EmitScalarExpr(E->getArg(1)); |
| 13455 | // atomicrmw only deals with integer arguments, so we need to use |
| 13456 | // LLVM's nvvm_atomic_load_add_f64 intrinsic. |
| 13457 | Function *FnALAF64 = |
| 13458 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_add_f64, Ptr->getType()); |
| 13459 | return Builder.CreateCall(FnALAF64, {Ptr, Val}); |
| 13460 | } |
| 13461 | |
| 13462 | case NVPTX::BI__nvvm_atom_inc_gen_ui: { |
| 13463 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13464 | Value *Val = EmitScalarExpr(E->getArg(1)); |
| 13465 | Function *FnALI32 = |
| 13466 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_inc_32, Ptr->getType()); |
| 13467 | return Builder.CreateCall(FnALI32, {Ptr, Val}); |
| 13468 | } |
| 13469 | |
| 13470 | case NVPTX::BI__nvvm_atom_dec_gen_ui: { |
| 13471 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13472 | Value *Val = EmitScalarExpr(E->getArg(1)); |
| 13473 | Function *FnALD32 = |
| 13474 | CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_dec_32, Ptr->getType()); |
| 13475 | return Builder.CreateCall(FnALD32, {Ptr, Val}); |
| 13476 | } |
| 13477 | |
| 13478 | case NVPTX::BI__nvvm_ldg_c: |
| 13479 | case NVPTX::BI__nvvm_ldg_c2: |
| 13480 | case NVPTX::BI__nvvm_ldg_c4: |
| 13481 | case NVPTX::BI__nvvm_ldg_s: |
| 13482 | case NVPTX::BI__nvvm_ldg_s2: |
| 13483 | case NVPTX::BI__nvvm_ldg_s4: |
| 13484 | case NVPTX::BI__nvvm_ldg_i: |
| 13485 | case NVPTX::BI__nvvm_ldg_i2: |
| 13486 | case NVPTX::BI__nvvm_ldg_i4: |
| 13487 | case NVPTX::BI__nvvm_ldg_l: |
| 13488 | case NVPTX::BI__nvvm_ldg_ll: |
| 13489 | case NVPTX::BI__nvvm_ldg_ll2: |
| 13490 | case NVPTX::BI__nvvm_ldg_uc: |
| 13491 | case NVPTX::BI__nvvm_ldg_uc2: |
| 13492 | case NVPTX::BI__nvvm_ldg_uc4: |
| 13493 | case NVPTX::BI__nvvm_ldg_us: |
| 13494 | case NVPTX::BI__nvvm_ldg_us2: |
| 13495 | case NVPTX::BI__nvvm_ldg_us4: |
| 13496 | case NVPTX::BI__nvvm_ldg_ui: |
| 13497 | case NVPTX::BI__nvvm_ldg_ui2: |
| 13498 | case NVPTX::BI__nvvm_ldg_ui4: |
| 13499 | case NVPTX::BI__nvvm_ldg_ul: |
| 13500 | case NVPTX::BI__nvvm_ldg_ull: |
| 13501 | case NVPTX::BI__nvvm_ldg_ull2: |
| 13502 | // PTX Interoperability section 2.2: "For a vector with an even number of |
| 13503 | // elements, its alignment is set to number of elements times the alignment |
| 13504 | // of its member: n*alignof(t)." |
| 13505 | return MakeLdg(Intrinsic::nvvm_ldg_global_i); |
| 13506 | case NVPTX::BI__nvvm_ldg_f: |
| 13507 | case NVPTX::BI__nvvm_ldg_f2: |
| 13508 | case NVPTX::BI__nvvm_ldg_f4: |
| 13509 | case NVPTX::BI__nvvm_ldg_d: |
| 13510 | case NVPTX::BI__nvvm_ldg_d2: |
| 13511 | return MakeLdg(Intrinsic::nvvm_ldg_global_f); |
| 13512 | |
| 13513 | case NVPTX::BI__nvvm_atom_cta_add_gen_i: |
| 13514 | case NVPTX::BI__nvvm_atom_cta_add_gen_l: |
| 13515 | case NVPTX::BI__nvvm_atom_cta_add_gen_ll: |
| 13516 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_cta); |
| 13517 | case NVPTX::BI__nvvm_atom_sys_add_gen_i: |
| 13518 | case NVPTX::BI__nvvm_atom_sys_add_gen_l: |
| 13519 | case NVPTX::BI__nvvm_atom_sys_add_gen_ll: |
| 13520 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_sys); |
| 13521 | case NVPTX::BI__nvvm_atom_cta_add_gen_f: |
| 13522 | case NVPTX::BI__nvvm_atom_cta_add_gen_d: |
| 13523 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_cta); |
| 13524 | case NVPTX::BI__nvvm_atom_sys_add_gen_f: |
| 13525 | case NVPTX::BI__nvvm_atom_sys_add_gen_d: |
| 13526 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_sys); |
| 13527 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_i: |
| 13528 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_l: |
| 13529 | case NVPTX::BI__nvvm_atom_cta_xchg_gen_ll: |
| 13530 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_cta); |
| 13531 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_i: |
| 13532 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_l: |
| 13533 | case NVPTX::BI__nvvm_atom_sys_xchg_gen_ll: |
| 13534 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_sys); |
| 13535 | case NVPTX::BI__nvvm_atom_cta_max_gen_i: |
| 13536 | case NVPTX::BI__nvvm_atom_cta_max_gen_ui: |
| 13537 | case NVPTX::BI__nvvm_atom_cta_max_gen_l: |
| 13538 | case NVPTX::BI__nvvm_atom_cta_max_gen_ul: |
| 13539 | case NVPTX::BI__nvvm_atom_cta_max_gen_ll: |
| 13540 | case NVPTX::BI__nvvm_atom_cta_max_gen_ull: |
| 13541 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_cta); |
| 13542 | case NVPTX::BI__nvvm_atom_sys_max_gen_i: |
| 13543 | case NVPTX::BI__nvvm_atom_sys_max_gen_ui: |
| 13544 | case NVPTX::BI__nvvm_atom_sys_max_gen_l: |
| 13545 | case NVPTX::BI__nvvm_atom_sys_max_gen_ul: |
| 13546 | case NVPTX::BI__nvvm_atom_sys_max_gen_ll: |
| 13547 | case NVPTX::BI__nvvm_atom_sys_max_gen_ull: |
| 13548 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_sys); |
| 13549 | case NVPTX::BI__nvvm_atom_cta_min_gen_i: |
| 13550 | case NVPTX::BI__nvvm_atom_cta_min_gen_ui: |
| 13551 | case NVPTX::BI__nvvm_atom_cta_min_gen_l: |
| 13552 | case NVPTX::BI__nvvm_atom_cta_min_gen_ul: |
| 13553 | case NVPTX::BI__nvvm_atom_cta_min_gen_ll: |
| 13554 | case NVPTX::BI__nvvm_atom_cta_min_gen_ull: |
| 13555 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_cta); |
| 13556 | case NVPTX::BI__nvvm_atom_sys_min_gen_i: |
| 13557 | case NVPTX::BI__nvvm_atom_sys_min_gen_ui: |
| 13558 | case NVPTX::BI__nvvm_atom_sys_min_gen_l: |
| 13559 | case NVPTX::BI__nvvm_atom_sys_min_gen_ul: |
| 13560 | case NVPTX::BI__nvvm_atom_sys_min_gen_ll: |
| 13561 | case NVPTX::BI__nvvm_atom_sys_min_gen_ull: |
| 13562 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_sys); |
| 13563 | case NVPTX::BI__nvvm_atom_cta_inc_gen_ui: |
| 13564 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_cta); |
| 13565 | case NVPTX::BI__nvvm_atom_cta_dec_gen_ui: |
| 13566 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_cta); |
| 13567 | case NVPTX::BI__nvvm_atom_sys_inc_gen_ui: |
| 13568 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_sys); |
| 13569 | case NVPTX::BI__nvvm_atom_sys_dec_gen_ui: |
| 13570 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_sys); |
| 13571 | case NVPTX::BI__nvvm_atom_cta_and_gen_i: |
| 13572 | case NVPTX::BI__nvvm_atom_cta_and_gen_l: |
| 13573 | case NVPTX::BI__nvvm_atom_cta_and_gen_ll: |
| 13574 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_cta); |
| 13575 | case NVPTX::BI__nvvm_atom_sys_and_gen_i: |
| 13576 | case NVPTX::BI__nvvm_atom_sys_and_gen_l: |
| 13577 | case NVPTX::BI__nvvm_atom_sys_and_gen_ll: |
| 13578 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_sys); |
| 13579 | case NVPTX::BI__nvvm_atom_cta_or_gen_i: |
| 13580 | case NVPTX::BI__nvvm_atom_cta_or_gen_l: |
| 13581 | case NVPTX::BI__nvvm_atom_cta_or_gen_ll: |
| 13582 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_cta); |
| 13583 | case NVPTX::BI__nvvm_atom_sys_or_gen_i: |
| 13584 | case NVPTX::BI__nvvm_atom_sys_or_gen_l: |
| 13585 | case NVPTX::BI__nvvm_atom_sys_or_gen_ll: |
| 13586 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_sys); |
| 13587 | case NVPTX::BI__nvvm_atom_cta_xor_gen_i: |
| 13588 | case NVPTX::BI__nvvm_atom_cta_xor_gen_l: |
| 13589 | case NVPTX::BI__nvvm_atom_cta_xor_gen_ll: |
| 13590 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_cta); |
| 13591 | case NVPTX::BI__nvvm_atom_sys_xor_gen_i: |
| 13592 | case NVPTX::BI__nvvm_atom_sys_xor_gen_l: |
| 13593 | case NVPTX::BI__nvvm_atom_sys_xor_gen_ll: |
| 13594 | return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_sys); |
| 13595 | case NVPTX::BI__nvvm_atom_cta_cas_gen_i: |
| 13596 | case NVPTX::BI__nvvm_atom_cta_cas_gen_l: |
| 13597 | case NVPTX::BI__nvvm_atom_cta_cas_gen_ll: { |
| 13598 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13599 | return Builder.CreateCall( |
| 13600 | CGM.getIntrinsic( |
| 13601 | Intrinsic::nvvm_atomic_cas_gen_i_cta, |
| 13602 | {Ptr->getType()->getPointerElementType(), Ptr->getType()}), |
| 13603 | {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))}); |
| 13604 | } |
| 13605 | case NVPTX::BI__nvvm_atom_sys_cas_gen_i: |
| 13606 | case NVPTX::BI__nvvm_atom_sys_cas_gen_l: |
| 13607 | case NVPTX::BI__nvvm_atom_sys_cas_gen_ll: { |
| 13608 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 13609 | return Builder.CreateCall( |
| 13610 | CGM.getIntrinsic( |
| 13611 | Intrinsic::nvvm_atomic_cas_gen_i_sys, |
| 13612 | {Ptr->getType()->getPointerElementType(), Ptr->getType()}), |
| 13613 | {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))}); |
| 13614 | } |
| 13615 | case NVPTX::BI__nvvm_match_all_sync_i32p: |
| 13616 | case NVPTX::BI__nvvm_match_all_sync_i64p: { |
| 13617 | Value *Mask = EmitScalarExpr(E->getArg(0)); |
| 13618 | Value *Val = EmitScalarExpr(E->getArg(1)); |
| 13619 | Address PredOutPtr = EmitPointerWithAlignment(E->getArg(2)); |
| 13620 | Value *ResultPair = Builder.CreateCall( |
| 13621 | CGM.getIntrinsic(BuiltinID == NVPTX::BI__nvvm_match_all_sync_i32p |
| 13622 | ? Intrinsic::nvvm_match_all_sync_i32p |
| 13623 | : Intrinsic::nvvm_match_all_sync_i64p), |
| 13624 | {Mask, Val}); |
| 13625 | Value *Pred = Builder.CreateZExt(Builder.CreateExtractValue(ResultPair, 1), |
| 13626 | PredOutPtr.getElementType()); |
| 13627 | Builder.CreateStore(Pred, PredOutPtr); |
| 13628 | return Builder.CreateExtractValue(ResultPair, 0); |
| 13629 | } |
| 13630 | |
| 13631 | // FP MMA loads |
| 13632 | case NVPTX::BI__hmma_m16n16k16_ld_a: |
| 13633 | case NVPTX::BI__hmma_m16n16k16_ld_b: |
| 13634 | case NVPTX::BI__hmma_m16n16k16_ld_c_f16: |
| 13635 | case NVPTX::BI__hmma_m16n16k16_ld_c_f32: |
| 13636 | case NVPTX::BI__hmma_m32n8k16_ld_a: |
| 13637 | case NVPTX::BI__hmma_m32n8k16_ld_b: |
| 13638 | case NVPTX::BI__hmma_m32n8k16_ld_c_f16: |
| 13639 | case NVPTX::BI__hmma_m32n8k16_ld_c_f32: |
| 13640 | case NVPTX::BI__hmma_m8n32k16_ld_a: |
| 13641 | case NVPTX::BI__hmma_m8n32k16_ld_b: |
| 13642 | case NVPTX::BI__hmma_m8n32k16_ld_c_f16: |
| 13643 | case NVPTX::BI__hmma_m8n32k16_ld_c_f32: |
| 13644 | // Integer MMA loads. |
| 13645 | case NVPTX::BI__imma_m16n16k16_ld_a_s8: |
| 13646 | case NVPTX::BI__imma_m16n16k16_ld_a_u8: |
| 13647 | case NVPTX::BI__imma_m16n16k16_ld_b_s8: |
| 13648 | case NVPTX::BI__imma_m16n16k16_ld_b_u8: |
| 13649 | case NVPTX::BI__imma_m16n16k16_ld_c: |
| 13650 | case NVPTX::BI__imma_m32n8k16_ld_a_s8: |
| 13651 | case NVPTX::BI__imma_m32n8k16_ld_a_u8: |
| 13652 | case NVPTX::BI__imma_m32n8k16_ld_b_s8: |
| 13653 | case NVPTX::BI__imma_m32n8k16_ld_b_u8: |
| 13654 | case NVPTX::BI__imma_m32n8k16_ld_c: |
| 13655 | case NVPTX::BI__imma_m8n32k16_ld_a_s8: |
| 13656 | case NVPTX::BI__imma_m8n32k16_ld_a_u8: |
| 13657 | case NVPTX::BI__imma_m8n32k16_ld_b_s8: |
| 13658 | case NVPTX::BI__imma_m8n32k16_ld_b_u8: |
| 13659 | case NVPTX::BI__imma_m8n32k16_ld_c: |
| 13660 | // Sub-integer MMA loads. |
| 13661 | case NVPTX::BI__imma_m8n8k32_ld_a_s4: |
| 13662 | case NVPTX::BI__imma_m8n8k32_ld_a_u4: |
| 13663 | case NVPTX::BI__imma_m8n8k32_ld_b_s4: |
| 13664 | case NVPTX::BI__imma_m8n8k32_ld_b_u4: |
| 13665 | case NVPTX::BI__imma_m8n8k32_ld_c: |
| 13666 | case NVPTX::BI__bmma_m8n8k128_ld_a_b1: |
| 13667 | case NVPTX::BI__bmma_m8n8k128_ld_b_b1: |
| 13668 | case NVPTX::BI__bmma_m8n8k128_ld_c: |
| 13669 | { |
| 13670 | Address Dst = EmitPointerWithAlignment(E->getArg(0)); |
| 13671 | Value *Src = EmitScalarExpr(E->getArg(1)); |
| 13672 | Value *Ldm = EmitScalarExpr(E->getArg(2)); |
| 13673 | llvm::APSInt isColMajorArg; |
| 13674 | if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext())) |
| 13675 | return nullptr; |
| 13676 | bool isColMajor = isColMajorArg.getSExtValue(); |
| 13677 | NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID); |
| 13678 | unsigned IID = isColMajor ? II.IID_col : II.IID_row; |
| 13679 | if (IID == 0) |
| 13680 | return nullptr; |
| 13681 | |
| 13682 | Value *Result = |
| 13683 | Builder.CreateCall(CGM.getIntrinsic(IID, Src->getType()), {Src, Ldm}); |
| 13684 | |
| 13685 | // Save returned values. |
| 13686 | assert(II.NumResults)((II.NumResults) ? static_cast<void> (0) : __assert_fail ("II.NumResults", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13686, __PRETTY_FUNCTION__)); |
| 13687 | if (II.NumResults == 1) { |
| 13688 | Builder.CreateAlignedStore(Result, Dst.getPointer(), |
| 13689 | CharUnits::fromQuantity(4)); |
| 13690 | } else { |
| 13691 | for (unsigned i = 0; i < II.NumResults; ++i) { |
| 13692 | Builder.CreateAlignedStore( |
| 13693 | Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), |
| 13694 | Dst.getElementType()), |
| 13695 | Builder.CreateGEP(Dst.getPointer(), |
| 13696 | llvm::ConstantInt::get(IntTy, i)), |
| 13697 | CharUnits::fromQuantity(4)); |
| 13698 | } |
| 13699 | } |
| 13700 | return Result; |
| 13701 | } |
| 13702 | |
| 13703 | case NVPTX::BI__hmma_m16n16k16_st_c_f16: |
| 13704 | case NVPTX::BI__hmma_m16n16k16_st_c_f32: |
| 13705 | case NVPTX::BI__hmma_m32n8k16_st_c_f16: |
| 13706 | case NVPTX::BI__hmma_m32n8k16_st_c_f32: |
| 13707 | case NVPTX::BI__hmma_m8n32k16_st_c_f16: |
| 13708 | case NVPTX::BI__hmma_m8n32k16_st_c_f32: |
| 13709 | case NVPTX::BI__imma_m16n16k16_st_c_i32: |
| 13710 | case NVPTX::BI__imma_m32n8k16_st_c_i32: |
| 13711 | case NVPTX::BI__imma_m8n32k16_st_c_i32: |
| 13712 | case NVPTX::BI__imma_m8n8k32_st_c_i32: |
| 13713 | case NVPTX::BI__bmma_m8n8k128_st_c_i32: { |
| 13714 | Value *Dst = EmitScalarExpr(E->getArg(0)); |
| 13715 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 13716 | Value *Ldm = EmitScalarExpr(E->getArg(2)); |
| 13717 | llvm::APSInt isColMajorArg; |
| 13718 | if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext())) |
| 13719 | return nullptr; |
| 13720 | bool isColMajor = isColMajorArg.getSExtValue(); |
| 13721 | NVPTXMmaLdstInfo II = getNVPTXMmaLdstInfo(BuiltinID); |
| 13722 | unsigned IID = isColMajor ? II.IID_col : II.IID_row; |
| 13723 | if (IID == 0) |
| 13724 | return nullptr; |
| 13725 | Function *Intrinsic = |
| 13726 | CGM.getIntrinsic(IID, Dst->getType()); |
| 13727 | llvm::Type *ParamType = Intrinsic->getFunctionType()->getParamType(1); |
| 13728 | SmallVector<Value *, 10> Values = {Dst}; |
| 13729 | for (unsigned i = 0; i < II.NumResults; ++i) { |
| 13730 | Value *V = Builder.CreateAlignedLoad( |
| 13731 | Builder.CreateGEP(Src.getPointer(), llvm::ConstantInt::get(IntTy, i)), |
| 13732 | CharUnits::fromQuantity(4)); |
| 13733 | Values.push_back(Builder.CreateBitCast(V, ParamType)); |
| 13734 | } |
| 13735 | Values.push_back(Ldm); |
| 13736 | Value *Result = Builder.CreateCall(Intrinsic, Values); |
| 13737 | return Result; |
| 13738 | } |
| 13739 | |
| 13740 | // BI__hmma_m16n16k16_mma_<Dtype><CType>(d, a, b, c, layout, satf) --> |
| 13741 | // Intrinsic::nvvm_wmma_m16n16k16_mma_sync<layout A,B><DType><CType><Satf> |
| 13742 | case NVPTX::BI__hmma_m16n16k16_mma_f16f16: |
| 13743 | case NVPTX::BI__hmma_m16n16k16_mma_f32f16: |
| 13744 | case NVPTX::BI__hmma_m16n16k16_mma_f32f32: |
| 13745 | case NVPTX::BI__hmma_m16n16k16_mma_f16f32: |
| 13746 | case NVPTX::BI__hmma_m32n8k16_mma_f16f16: |
| 13747 | case NVPTX::BI__hmma_m32n8k16_mma_f32f16: |
| 13748 | case NVPTX::BI__hmma_m32n8k16_mma_f32f32: |
| 13749 | case NVPTX::BI__hmma_m32n8k16_mma_f16f32: |
| 13750 | case NVPTX::BI__hmma_m8n32k16_mma_f16f16: |
| 13751 | case NVPTX::BI__hmma_m8n32k16_mma_f32f16: |
| 13752 | case NVPTX::BI__hmma_m8n32k16_mma_f32f32: |
| 13753 | case NVPTX::BI__hmma_m8n32k16_mma_f16f32: |
| 13754 | case NVPTX::BI__imma_m16n16k16_mma_s8: |
| 13755 | case NVPTX::BI__imma_m16n16k16_mma_u8: |
| 13756 | case NVPTX::BI__imma_m32n8k16_mma_s8: |
| 13757 | case NVPTX::BI__imma_m32n8k16_mma_u8: |
| 13758 | case NVPTX::BI__imma_m8n32k16_mma_s8: |
| 13759 | case NVPTX::BI__imma_m8n32k16_mma_u8: |
| 13760 | case NVPTX::BI__imma_m8n8k32_mma_s4: |
| 13761 | case NVPTX::BI__imma_m8n8k32_mma_u4: |
| 13762 | case NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1: { |
| 13763 | Address Dst = EmitPointerWithAlignment(E->getArg(0)); |
| 13764 | Address SrcA = EmitPointerWithAlignment(E->getArg(1)); |
| 13765 | Address SrcB = EmitPointerWithAlignment(E->getArg(2)); |
| 13766 | Address SrcC = EmitPointerWithAlignment(E->getArg(3)); |
| 13767 | llvm::APSInt LayoutArg; |
| 13768 | if (!E->getArg(4)->isIntegerConstantExpr(LayoutArg, getContext())) |
| 13769 | return nullptr; |
| 13770 | int Layout = LayoutArg.getSExtValue(); |
| 13771 | if (Layout < 0 || Layout > 3) |
| 13772 | return nullptr; |
| 13773 | llvm::APSInt SatfArg; |
| 13774 | if (BuiltinID == NVPTX::BI__bmma_m8n8k128_mma_xor_popc_b1) |
| 13775 | SatfArg = 0; // .b1 does not have satf argument. |
| 13776 | else if (!E->getArg(5)->isIntegerConstantExpr(SatfArg, getContext())) |
| 13777 | return nullptr; |
| 13778 | bool Satf = SatfArg.getSExtValue(); |
| 13779 | NVPTXMmaInfo MI = getNVPTXMmaInfo(BuiltinID); |
| 13780 | unsigned IID = MI.getMMAIntrinsic(Layout, Satf); |
| 13781 | if (IID == 0) // Unsupported combination of Layout/Satf. |
| 13782 | return nullptr; |
| 13783 | |
| 13784 | SmallVector<Value *, 24> Values; |
| 13785 | Function *Intrinsic = CGM.getIntrinsic(IID); |
| 13786 | llvm::Type *AType = Intrinsic->getFunctionType()->getParamType(0); |
| 13787 | // Load A |
| 13788 | for (unsigned i = 0; i < MI.NumEltsA; ++i) { |
| 13789 | Value *V = Builder.CreateAlignedLoad( |
| 13790 | Builder.CreateGEP(SrcA.getPointer(), |
| 13791 | llvm::ConstantInt::get(IntTy, i)), |
| 13792 | CharUnits::fromQuantity(4)); |
| 13793 | Values.push_back(Builder.CreateBitCast(V, AType)); |
| 13794 | } |
| 13795 | // Load B |
| 13796 | llvm::Type *BType = Intrinsic->getFunctionType()->getParamType(MI.NumEltsA); |
| 13797 | for (unsigned i = 0; i < MI.NumEltsB; ++i) { |
| 13798 | Value *V = Builder.CreateAlignedLoad( |
| 13799 | Builder.CreateGEP(SrcB.getPointer(), |
| 13800 | llvm::ConstantInt::get(IntTy, i)), |
| 13801 | CharUnits::fromQuantity(4)); |
| 13802 | Values.push_back(Builder.CreateBitCast(V, BType)); |
| 13803 | } |
| 13804 | // Load C |
| 13805 | llvm::Type *CType = |
| 13806 | Intrinsic->getFunctionType()->getParamType(MI.NumEltsA + MI.NumEltsB); |
| 13807 | for (unsigned i = 0; i < MI.NumEltsC; ++i) { |
| 13808 | Value *V = Builder.CreateAlignedLoad( |
| 13809 | Builder.CreateGEP(SrcC.getPointer(), |
| 13810 | llvm::ConstantInt::get(IntTy, i)), |
| 13811 | CharUnits::fromQuantity(4)); |
| 13812 | Values.push_back(Builder.CreateBitCast(V, CType)); |
| 13813 | } |
| 13814 | Value *Result = Builder.CreateCall(Intrinsic, Values); |
| 13815 | llvm::Type *DType = Dst.getElementType(); |
| 13816 | for (unsigned i = 0; i < MI.NumEltsD; ++i) |
| 13817 | Builder.CreateAlignedStore( |
| 13818 | Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), DType), |
| 13819 | Builder.CreateGEP(Dst.getPointer(), llvm::ConstantInt::get(IntTy, i)), |
| 13820 | CharUnits::fromQuantity(4)); |
| 13821 | return Result; |
| 13822 | } |
| 13823 | default: |
| 13824 | return nullptr; |
| 13825 | } |
| 13826 | } |
| 13827 | |
| 13828 | Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID, |
| 13829 | const CallExpr *E) { |
| 13830 | switch (BuiltinID) { |
| 13831 | case WebAssembly::BI__builtin_wasm_memory_size: { |
| 13832 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 13833 | Value *I = EmitScalarExpr(E->getArg(0)); |
| 13834 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_size, ResultType); |
| 13835 | return Builder.CreateCall(Callee, I); |
| 13836 | } |
| 13837 | case WebAssembly::BI__builtin_wasm_memory_grow: { |
| 13838 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 13839 | Value *Args[] = { |
| 13840 | EmitScalarExpr(E->getArg(0)), |
| 13841 | EmitScalarExpr(E->getArg(1)) |
| 13842 | }; |
| 13843 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_grow, ResultType); |
| 13844 | return Builder.CreateCall(Callee, Args); |
| 13845 | } |
| 13846 | case WebAssembly::BI__builtin_wasm_memory_init: { |
| 13847 | llvm::APSInt SegConst; |
| 13848 | if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext())) |
| 13849 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13849); |
| 13850 | llvm::APSInt MemConst; |
| 13851 | if (!E->getArg(1)->isIntegerConstantExpr(MemConst, getContext())) |
| 13852 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13852); |
| 13853 | if (!MemConst.isNullValue()) |
| 13854 | ErrorUnsupported(E, "non-zero memory index"); |
| 13855 | Value *Args[] = {llvm::ConstantInt::get(getLLVMContext(), SegConst), |
| 13856 | llvm::ConstantInt::get(getLLVMContext(), MemConst), |
| 13857 | EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)), |
| 13858 | EmitScalarExpr(E->getArg(4))}; |
| 13859 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_init); |
| 13860 | return Builder.CreateCall(Callee, Args); |
| 13861 | } |
| 13862 | case WebAssembly::BI__builtin_wasm_data_drop: { |
| 13863 | llvm::APSInt SegConst; |
| 13864 | if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext())) |
| 13865 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13865); |
| 13866 | Value *Arg = llvm::ConstantInt::get(getLLVMContext(), SegConst); |
| 13867 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_data_drop); |
| 13868 | return Builder.CreateCall(Callee, {Arg}); |
| 13869 | } |
| 13870 | case WebAssembly::BI__builtin_wasm_throw: { |
| 13871 | Value *Tag = EmitScalarExpr(E->getArg(0)); |
| 13872 | Value *Obj = EmitScalarExpr(E->getArg(1)); |
| 13873 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_throw); |
| 13874 | return Builder.CreateCall(Callee, {Tag, Obj}); |
| 13875 | } |
| 13876 | case WebAssembly::BI__builtin_wasm_rethrow_in_catch: { |
| 13877 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_rethrow_in_catch); |
| 13878 | return Builder.CreateCall(Callee); |
| 13879 | } |
| 13880 | case WebAssembly::BI__builtin_wasm_atomic_wait_i32: { |
| 13881 | Value *Addr = EmitScalarExpr(E->getArg(0)); |
| 13882 | Value *Expected = EmitScalarExpr(E->getArg(1)); |
| 13883 | Value *Timeout = EmitScalarExpr(E->getArg(2)); |
| 13884 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_wait_i32); |
| 13885 | return Builder.CreateCall(Callee, {Addr, Expected, Timeout}); |
| 13886 | } |
| 13887 | case WebAssembly::BI__builtin_wasm_atomic_wait_i64: { |
| 13888 | Value *Addr = EmitScalarExpr(E->getArg(0)); |
| 13889 | Value *Expected = EmitScalarExpr(E->getArg(1)); |
| 13890 | Value *Timeout = EmitScalarExpr(E->getArg(2)); |
| 13891 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_wait_i64); |
| 13892 | return Builder.CreateCall(Callee, {Addr, Expected, Timeout}); |
| 13893 | } |
| 13894 | case WebAssembly::BI__builtin_wasm_atomic_notify: { |
| 13895 | Value *Addr = EmitScalarExpr(E->getArg(0)); |
| 13896 | Value *Count = EmitScalarExpr(E->getArg(1)); |
| 13897 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_notify); |
| 13898 | return Builder.CreateCall(Callee, {Addr, Count}); |
| 13899 | } |
| 13900 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f32: |
| 13901 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f64: |
| 13902 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f32: |
| 13903 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f64: |
| 13904 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32x4_f32x4: |
| 13905 | case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64x2_f64x2: { |
| 13906 | Value *Src = EmitScalarExpr(E->getArg(0)); |
| 13907 | llvm::Type *ResT = ConvertType(E->getType()); |
| 13908 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_saturate_signed, |
| 13909 | {ResT, Src->getType()}); |
| 13910 | return Builder.CreateCall(Callee, {Src}); |
| 13911 | } |
| 13912 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f32: |
| 13913 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f64: |
| 13914 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f32: |
| 13915 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f64: |
| 13916 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32x4_f32x4: |
| 13917 | case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64x2_f64x2: { |
| 13918 | Value *Src = EmitScalarExpr(E->getArg(0)); |
| 13919 | llvm::Type *ResT = ConvertType(E->getType()); |
| 13920 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_saturate_unsigned, |
| 13921 | {ResT, Src->getType()}); |
| 13922 | return Builder.CreateCall(Callee, {Src}); |
| 13923 | } |
| 13924 | case WebAssembly::BI__builtin_wasm_min_f32: |
| 13925 | case WebAssembly::BI__builtin_wasm_min_f64: |
| 13926 | case WebAssembly::BI__builtin_wasm_min_f32x4: |
| 13927 | case WebAssembly::BI__builtin_wasm_min_f64x2: { |
| 13928 | Value *LHS = EmitScalarExpr(E->getArg(0)); |
| 13929 | Value *RHS = EmitScalarExpr(E->getArg(1)); |
| 13930 | Function *Callee = CGM.getIntrinsic(Intrinsic::minimum, |
| 13931 | ConvertType(E->getType())); |
| 13932 | return Builder.CreateCall(Callee, {LHS, RHS}); |
| 13933 | } |
| 13934 | case WebAssembly::BI__builtin_wasm_max_f32: |
| 13935 | case WebAssembly::BI__builtin_wasm_max_f64: |
| 13936 | case WebAssembly::BI__builtin_wasm_max_f32x4: |
| 13937 | case WebAssembly::BI__builtin_wasm_max_f64x2: { |
| 13938 | Value *LHS = EmitScalarExpr(E->getArg(0)); |
| 13939 | Value *RHS = EmitScalarExpr(E->getArg(1)); |
| 13940 | Function *Callee = CGM.getIntrinsic(Intrinsic::maximum, |
| 13941 | ConvertType(E->getType())); |
| 13942 | return Builder.CreateCall(Callee, {LHS, RHS}); |
| 13943 | } |
| 13944 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i8x16: |
| 13945 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i8x16: |
| 13946 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i16x8: |
| 13947 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i16x8: |
| 13948 | case WebAssembly::BI__builtin_wasm_extract_lane_i32x4: |
| 13949 | case WebAssembly::BI__builtin_wasm_extract_lane_i64x2: |
| 13950 | case WebAssembly::BI__builtin_wasm_extract_lane_f32x4: |
| 13951 | case WebAssembly::BI__builtin_wasm_extract_lane_f64x2: { |
| 13952 | llvm::APSInt LaneConst; |
| 13953 | if (!E->getArg(1)->isIntegerConstantExpr(LaneConst, getContext())) |
| 13954 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13954); |
| 13955 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 13956 | Value *Lane = llvm::ConstantInt::get(getLLVMContext(), LaneConst); |
| 13957 | Value *Extract = Builder.CreateExtractElement(Vec, Lane); |
| 13958 | switch (BuiltinID) { |
| 13959 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i8x16: |
| 13960 | case WebAssembly::BI__builtin_wasm_extract_lane_s_i16x8: |
| 13961 | return Builder.CreateSExt(Extract, ConvertType(E->getType())); |
| 13962 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i8x16: |
| 13963 | case WebAssembly::BI__builtin_wasm_extract_lane_u_i16x8: |
| 13964 | return Builder.CreateZExt(Extract, ConvertType(E->getType())); |
| 13965 | case WebAssembly::BI__builtin_wasm_extract_lane_i32x4: |
| 13966 | case WebAssembly::BI__builtin_wasm_extract_lane_i64x2: |
| 13967 | case WebAssembly::BI__builtin_wasm_extract_lane_f32x4: |
| 13968 | case WebAssembly::BI__builtin_wasm_extract_lane_f64x2: |
| 13969 | return Extract; |
| 13970 | default: |
| 13971 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13971); |
| 13972 | } |
| 13973 | } |
| 13974 | case WebAssembly::BI__builtin_wasm_replace_lane_i8x16: |
| 13975 | case WebAssembly::BI__builtin_wasm_replace_lane_i16x8: |
| 13976 | case WebAssembly::BI__builtin_wasm_replace_lane_i32x4: |
| 13977 | case WebAssembly::BI__builtin_wasm_replace_lane_i64x2: |
| 13978 | case WebAssembly::BI__builtin_wasm_replace_lane_f32x4: |
| 13979 | case WebAssembly::BI__builtin_wasm_replace_lane_f64x2: { |
| 13980 | llvm::APSInt LaneConst; |
| 13981 | if (!E->getArg(1)->isIntegerConstantExpr(LaneConst, getContext())) |
| 13982 | llvm_unreachable("Constant arg isn't actually constant?")::llvm::llvm_unreachable_internal("Constant arg isn't actually constant?" , "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13982); |
| 13983 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 13984 | Value *Lane = llvm::ConstantInt::get(getLLVMContext(), LaneConst); |
| 13985 | Value *Val = EmitScalarExpr(E->getArg(2)); |
| 13986 | switch (BuiltinID) { |
| 13987 | case WebAssembly::BI__builtin_wasm_replace_lane_i8x16: |
| 13988 | case WebAssembly::BI__builtin_wasm_replace_lane_i16x8: { |
| 13989 | llvm::Type *ElemType = ConvertType(E->getType())->getVectorElementType(); |
| 13990 | Value *Trunc = Builder.CreateTrunc(Val, ElemType); |
| 13991 | return Builder.CreateInsertElement(Vec, Trunc, Lane); |
| 13992 | } |
| 13993 | case WebAssembly::BI__builtin_wasm_replace_lane_i32x4: |
| 13994 | case WebAssembly::BI__builtin_wasm_replace_lane_i64x2: |
| 13995 | case WebAssembly::BI__builtin_wasm_replace_lane_f32x4: |
| 13996 | case WebAssembly::BI__builtin_wasm_replace_lane_f64x2: |
| 13997 | return Builder.CreateInsertElement(Vec, Val, Lane); |
| 13998 | default: |
| 13999 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 13999); |
| 14000 | } |
| 14001 | } |
| 14002 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i8x16: |
| 14003 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i8x16: |
| 14004 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i16x8: |
| 14005 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i16x8: |
| 14006 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i8x16: |
| 14007 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i8x16: |
| 14008 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i16x8: |
| 14009 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i16x8: { |
| 14010 | unsigned IntNo; |
| 14011 | switch (BuiltinID) { |
| 14012 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i8x16: |
| 14013 | case WebAssembly::BI__builtin_wasm_add_saturate_s_i16x8: |
| 14014 | IntNo = Intrinsic::sadd_sat; |
| 14015 | break; |
| 14016 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i8x16: |
| 14017 | case WebAssembly::BI__builtin_wasm_add_saturate_u_i16x8: |
| 14018 | IntNo = Intrinsic::uadd_sat; |
| 14019 | break; |
| 14020 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i8x16: |
| 14021 | case WebAssembly::BI__builtin_wasm_sub_saturate_s_i16x8: |
| 14022 | IntNo = Intrinsic::wasm_sub_saturate_signed; |
| 14023 | break; |
| 14024 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i8x16: |
| 14025 | case WebAssembly::BI__builtin_wasm_sub_saturate_u_i16x8: |
| 14026 | IntNo = Intrinsic::wasm_sub_saturate_unsigned; |
| 14027 | break; |
| 14028 | default: |
| 14029 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14029); |
| 14030 | } |
| 14031 | Value *LHS = EmitScalarExpr(E->getArg(0)); |
| 14032 | Value *RHS = EmitScalarExpr(E->getArg(1)); |
| 14033 | Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType())); |
| 14034 | return Builder.CreateCall(Callee, {LHS, RHS}); |
| 14035 | } |
| 14036 | case WebAssembly::BI__builtin_wasm_bitselect: { |
| 14037 | Value *V1 = EmitScalarExpr(E->getArg(0)); |
| 14038 | Value *V2 = EmitScalarExpr(E->getArg(1)); |
| 14039 | Value *C = EmitScalarExpr(E->getArg(2)); |
| 14040 | Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_bitselect, |
| 14041 | ConvertType(E->getType())); |
| 14042 | return Builder.CreateCall(Callee, {V1, V2, C}); |
| 14043 | } |
| 14044 | case WebAssembly::BI__builtin_wasm_any_true_i8x16: |
| 14045 | case WebAssembly::BI__builtin_wasm_any_true_i16x8: |
| 14046 | case WebAssembly::BI__builtin_wasm_any_true_i32x4: |
| 14047 | case WebAssembly::BI__builtin_wasm_any_true_i64x2: |
| 14048 | case WebAssembly::BI__builtin_wasm_all_true_i8x16: |
| 14049 | case WebAssembly::BI__builtin_wasm_all_true_i16x8: |
| 14050 | case WebAssembly::BI__builtin_wasm_all_true_i32x4: |
| 14051 | case WebAssembly::BI__builtin_wasm_all_true_i64x2: { |
| 14052 | unsigned IntNo; |
| 14053 | switch (BuiltinID) { |
| 14054 | case WebAssembly::BI__builtin_wasm_any_true_i8x16: |
| 14055 | case WebAssembly::BI__builtin_wasm_any_true_i16x8: |
| 14056 | case WebAssembly::BI__builtin_wasm_any_true_i32x4: |
| 14057 | case WebAssembly::BI__builtin_wasm_any_true_i64x2: |
| 14058 | IntNo = Intrinsic::wasm_anytrue; |
| 14059 | break; |
| 14060 | case WebAssembly::BI__builtin_wasm_all_true_i8x16: |
| 14061 | case WebAssembly::BI__builtin_wasm_all_true_i16x8: |
| 14062 | case WebAssembly::BI__builtin_wasm_all_true_i32x4: |
| 14063 | case WebAssembly::BI__builtin_wasm_all_true_i64x2: |
| 14064 | IntNo = Intrinsic::wasm_alltrue; |
| 14065 | break; |
| 14066 | default: |
| 14067 | llvm_unreachable("unexpected builtin ID")::llvm::llvm_unreachable_internal("unexpected builtin ID", "/build/llvm-toolchain-snapshot-9~svn362543/tools/clang/lib/CodeGen/CGBuiltin.cpp" , 14067); |
| 14068 | } |
| 14069 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 14070 | Function *Callee = CGM.getIntrinsic(IntNo, Vec->getType()); |
| 14071 | return Builder.CreateCall(Callee, {Vec}); |
| 14072 | } |
| 14073 | case WebAssembly::BI__builtin_wasm_abs_f32x4: |
| 14074 | case WebAssembly::BI__builtin_wasm_abs_f64x2: { |
| 14075 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 14076 | Function *Callee = CGM.getIntrinsic(Intrinsic::fabs, Vec->getType()); |
| 14077 | return Builder.CreateCall(Callee, {Vec}); |
| 14078 | } |
| 14079 | case WebAssembly::BI__builtin_wasm_sqrt_f32x4: |
| 14080 | case WebAssembly::BI__builtin_wasm_sqrt_f64x2: { |
| 14081 | Value *Vec = EmitScalarExpr(E->getArg(0)); |
| 14082 | Function *Callee = CGM.getIntrinsic(Intrinsic::sqrt, Vec->getType()); |
| 14083 | return Builder.CreateCall(Callee, {Vec}); |
| 14084 | } |
| 14085 | |
| 14086 | default: |
| 14087 | return nullptr; |
| 14088 | } |
| 14089 | } |
| 14090 | |
| 14091 | Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID, |
| 14092 | const CallExpr *E) { |
| 14093 | SmallVector<llvm::Value *, 4> Ops; |
| 14094 | Intrinsic::ID ID = Intrinsic::not_intrinsic; |
| 14095 | |
| 14096 | auto MakeCircLd = [&](unsigned IntID, bool HasImm) { |
| 14097 | // The base pointer is passed by address, so it needs to be loaded. |
| 14098 | Address BP = EmitPointerWithAlignment(E->getArg(0)); |
| 14099 | BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy), |
| 14100 | BP.getAlignment()); |
| 14101 | llvm::Value *Base = Builder.CreateLoad(BP); |
| 14102 | // Operands are Base, Increment, Modifier, Start. |
| 14103 | if (HasImm) |
| 14104 | Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), |
| 14105 | EmitScalarExpr(E->getArg(3)) }; |
| 14106 | else |
| 14107 | Ops = { Base, EmitScalarExpr(E->getArg(1)), |
| 14108 | EmitScalarExpr(E->getArg(2)) }; |
| 14109 | |
| 14110 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops); |
| 14111 | llvm::Value *NewBase = Builder.CreateExtractValue(Result, 1); |
| 14112 | llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), |
| 14113 | NewBase->getType()->getPointerTo()); |
| 14114 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 14115 | // The intrinsic generates two results. The new value for the base pointer |
| 14116 | // needs to be stored. |
| 14117 | Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment()); |
| 14118 | return Builder.CreateExtractValue(Result, 0); |
| 14119 | }; |
| 14120 | |
| 14121 | auto MakeCircSt = [&](unsigned IntID, bool HasImm) { |
| 14122 | // The base pointer is passed by address, so it needs to be loaded. |
| 14123 | Address BP = EmitPointerWithAlignment(E->getArg(0)); |
| 14124 | BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy), |
| 14125 | BP.getAlignment()); |
| 14126 | llvm::Value *Base = Builder.CreateLoad(BP); |
| 14127 | // Operands are Base, Increment, Modifier, Value, Start. |
| 14128 | if (HasImm) |
| 14129 | Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), |
| 14130 | EmitScalarExpr(E->getArg(3)), EmitScalarExpr(E->getArg(4)) }; |
| 14131 | else |
| 14132 | Ops = { Base, EmitScalarExpr(E->getArg(1)), |
| 14133 | EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)) }; |
| 14134 | |
| 14135 | llvm::Value *NewBase = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops); |
| 14136 | llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), |
| 14137 | NewBase->getType()->getPointerTo()); |
| 14138 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 14139 | // The intrinsic generates one result, which is the new value for the base |
| 14140 | // pointer. It needs to be stored. |
| 14141 | return Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment()); |
| 14142 | }; |
| 14143 | |
| 14144 | // Handle the conversion of bit-reverse load intrinsics to bit code. |
| 14145 | // The intrinsic call after this function only reads from memory and the |
| 14146 | // write to memory is dealt by the store instruction. |
| 14147 | auto MakeBrevLd = [&](unsigned IntID, llvm::Type *DestTy) { |
| 14148 | // The intrinsic generates one result, which is the new value for the base |
| 14149 | // pointer. It needs to be returned. The result of the load instruction is |
| 14150 | // passed to intrinsic by address, so the value needs to be stored. |
| 14151 | llvm::Value *BaseAddress = |
| 14152 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int8PtrTy); |
| 14153 | |
| 14154 | // Expressions like &(*pt++) will be incremented per evaluation. |
| 14155 | // EmitPointerWithAlignment and EmitScalarExpr evaluates the expression |
| 14156 | // per call. |
| 14157 | Address DestAddr = EmitPointerWithAlignment(E->getArg(1)); |
| 14158 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), Int8PtrTy), |
| 14159 | DestAddr.getAlignment()); |
| 14160 | llvm::Value *DestAddress = DestAddr.getPointer(); |
| 14161 | |
| 14162 | // Operands are Base, Dest, Modifier. |
| 14163 | // The intrinsic format in LLVM IR is defined as |
| 14164 | // { ValueType, i8* } (i8*, i32). |
| 14165 | Ops = {BaseAddress, EmitScalarExpr(E->getArg(2))}; |
| 14166 | |
| 14167 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops); |
| 14168 | // The value needs to be stored as the variable is passed by reference. |
| 14169 | llvm::Value *DestVal = Builder.CreateExtractValue(Result, 0); |
| 14170 | |
| 14171 | // The store needs to be truncated to fit the destination type. |
| 14172 | // While i32 and i64 are natively supported on Hexagon, i8 and i16 needs |
| 14173 | // to be handled with stores of respective destination type. |
| 14174 | DestVal = Builder.CreateTrunc(DestVal, DestTy); |
| 14175 | |
| 14176 | llvm::Value *DestForStore = |
| 14177 | Builder.CreateBitCast(DestAddress, DestVal->getType()->getPointerTo()); |
| 14178 | Builder.CreateAlignedStore(DestVal, DestForStore, DestAddr.getAlignment()); |
| 14179 | // The updated value of the base pointer is returned. |
| 14180 | return Builder.CreateExtractValue(Result, 1); |
| 14181 | }; |
| 14182 | |
| 14183 | switch (BuiltinID) { |
| 14184 | case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry: |
| 14185 | case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry_128B: { |
| 14186 | Address Dest = EmitPointerWithAlignment(E->getArg(2)); |
| 14187 | unsigned Size; |
| 14188 | if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vaddcarry) { |
| 14189 | Size = 512; |
| 14190 | ID = Intrinsic::hexagon_V6_vaddcarry; |
| 14191 | } else { |
| 14192 | Size = 1024; |
| 14193 | ID = Intrinsic::hexagon_V6_vaddcarry_128B; |
| 14194 | } |
| 14195 | Dest = Builder.CreateBitCast(Dest, |
| 14196 | llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0)); |
| 14197 | LoadInst *QLd = Builder.CreateLoad(Dest); |
| 14198 | Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd }; |
| 14199 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); |
| 14200 | llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1); |
| 14201 | llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)), |
| 14202 | Vprd->getType()->getPointerTo(0)); |
| 14203 | Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment()); |
| 14204 | return Builder.CreateExtractValue(Result, 0); |
| 14205 | } |
| 14206 | case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry: |
| 14207 | case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry_128B: { |
| 14208 | Address Dest = EmitPointerWithAlignment(E->getArg(2)); |
| 14209 | unsigned Size; |
| 14210 | if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vsubcarry) { |
| 14211 | Size = 512; |
| 14212 | ID = Intrinsic::hexagon_V6_vsubcarry; |
| 14213 | } else { |
| 14214 | Size = 1024; |
| 14215 | ID = Intrinsic::hexagon_V6_vsubcarry_128B; |
| 14216 | } |
| 14217 | Dest = Builder.CreateBitCast(Dest, |
| 14218 | llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0)); |
| 14219 | LoadInst *QLd = Builder.CreateLoad(Dest); |
| 14220 | Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd }; |
| 14221 | llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops); |
| 14222 | llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1); |
| 14223 | llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)), |
| 14224 | Vprd->getType()->getPointerTo(0)); |
| 14225 | Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment()); |
| 14226 | return Builder.CreateExtractValue(Result, 0); |
| 14227 | } |
| 14228 | case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pci: |
| 14229 | return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pci, /*HasImm*/true); |
| 14230 | case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pci: |
| 14231 | return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pci, /*HasImm*/true); |
| 14232 | case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pci: |
| 14233 | return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pci, /*HasImm*/true); |
| 14234 | case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pci: |
| 14235 | return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pci, /*HasImm*/true); |
| 14236 | case Hexagon::BI__builtin_HEXAGON_L2_loadri_pci: |
| 14237 | return MakeCircLd(Intrinsic::hexagon_L2_loadri_pci, /*HasImm*/true); |
| 14238 | case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pci: |
| 14239 | return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pci, /*HasImm*/true); |
| 14240 | case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pcr: |
| 14241 | return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pcr, /*HasImm*/false); |
| 14242 | case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pcr: |
| 14243 | return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pcr, /*HasImm*/false); |
| 14244 | case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pcr: |
| 14245 | return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pcr, /*HasImm*/false); |
| 14246 | case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pcr: |
| 14247 | return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pcr, /*HasImm*/false); |
| 14248 | case Hexagon::BI__builtin_HEXAGON_L2_loadri_pcr: |
| 14249 | return MakeCircLd(Intrinsic::hexagon_L2_loadri_pcr, /*HasImm*/false); |
| 14250 | case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pcr: |
| 14251 | return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pcr, /*HasImm*/false); |
| 14252 | case Hexagon::BI__builtin_HEXAGON_S2_storerb_pci: |
| 14253 | return MakeCircSt(Intrinsic::hexagon_S2_storerb_pci, /*HasImm*/true); |
| 14254 | case Hexagon::BI__builtin_HEXAGON_S2_storerh_pci: |
| 14255 | return MakeCircSt(Intrinsic::hexagon_S2_storerh_pci, /*HasImm*/true); |
| 14256 | case Hexagon::BI__builtin_HEXAGON_S2_storerf_pci: |
| 14257 | return MakeCircSt(Intrinsic::hexagon_S2_storerf_pci, /*HasImm*/true); |
| 14258 | case Hexagon::BI__builtin_HEXAGON_S2_storeri_pci: |
| 14259 | return MakeCircSt(Intrinsic::hexagon_S2_storeri_pci, /*HasImm*/true); |
| 14260 | case Hexagon::BI__builtin_HEXAGON_S2_storerd_pci: |
| 14261 | return MakeCircSt(Intrinsic::hexagon_S2_storerd_pci, /*HasImm*/true); |
| 14262 | case Hexagon::BI__builtin_HEXAGON_S2_storerb_pcr: |
| 14263 | return MakeCircSt(Intrinsic::hexagon_S2_storerb_pcr, /*HasImm*/false); |
| 14264 | case Hexagon::BI__builtin_HEXAGON_S2_storerh_pcr: |
| 14265 | return MakeCircSt(Intrinsic::hexagon_S2_storerh_pcr, /*HasImm*/false); |
| 14266 | case Hexagon::BI__builtin_HEXAGON_S2_storerf_pcr: |
| 14267 | return MakeCircSt(Intrinsic::hexagon_S2_storerf_pcr, /*HasImm*/false); |
| 14268 | case Hexagon::BI__builtin_HEXAGON_S2_storeri_pcr: |
| 14269 | return MakeCircSt(Intrinsic::hexagon_S2_storeri_pcr, /*HasImm*/false); |
| 14270 | case Hexagon::BI__builtin_HEXAGON_S2_storerd_pcr: |
| 14271 | return MakeCircSt(Intrinsic::hexagon_S2_storerd_pcr, /*HasImm*/false); |
| 14272 | case Hexagon::BI__builtin_brev_ldub: |
| 14273 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrub_pbr, Int8Ty); |
| 14274 | case Hexagon::BI__builtin_brev_ldb: |
| 14275 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrb_pbr, Int8Ty); |
| 14276 | case Hexagon::BI__builtin_brev_lduh: |
| 14277 | return MakeBrevLd(Intrinsic::hexagon_L2_loadruh_pbr, Int16Ty); |
| 14278 | case Hexagon::BI__builtin_brev_ldh: |
| 14279 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrh_pbr, Int16Ty); |
| 14280 | case Hexagon::BI__builtin_brev_ldw: |
| 14281 | return MakeBrevLd(Intrinsic::hexagon_L2_loadri_pbr, Int32Ty); |
| 14282 | case Hexagon::BI__builtin_brev_ldd: |
| 14283 | return MakeBrevLd(Intrinsic::hexagon_L2_loadrd_pbr, Int64Ty); |
| 14284 | default: |
| 14285 | break; |
| 14286 | } // switch |
| 14287 | |
| 14288 | return nullptr; |
| 14289 | } |