| File: | build/source/clang/lib/CodeGen/CGBuiltin.cpp |
| Warning: | line 15964, column 5 Undefined or garbage value returned to caller |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This contains code to emit Builtin calls as LLVM code. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #include "ABIInfo.h" |
| 14 | #include "CGCUDARuntime.h" |
| 15 | #include "CGCXXABI.h" |
| 16 | #include "CGObjCRuntime.h" |
| 17 | #include "CGOpenCLRuntime.h" |
| 18 | #include "CGRecordLayout.h" |
| 19 | #include "CodeGenFunction.h" |
| 20 | #include "CodeGenModule.h" |
| 21 | #include "ConstantEmitter.h" |
| 22 | #include "PatternInit.h" |
| 23 | #include "TargetInfo.h" |
| 24 | #include "clang/AST/ASTContext.h" |
| 25 | #include "clang/AST/Attr.h" |
| 26 | #include "clang/AST/Decl.h" |
| 27 | #include "clang/AST/OSLog.h" |
| 28 | #include "clang/Basic/TargetBuiltins.h" |
| 29 | #include "clang/Basic/TargetInfo.h" |
| 30 | #include "clang/CodeGen/CGFunctionInfo.h" |
| 31 | #include "clang/Frontend/FrontendDiagnostic.h" |
| 32 | #include "llvm/ADT/APFloat.h" |
| 33 | #include "llvm/ADT/APInt.h" |
| 34 | #include "llvm/ADT/SmallPtrSet.h" |
| 35 | #include "llvm/ADT/StringExtras.h" |
| 36 | #include "llvm/Analysis/ValueTracking.h" |
| 37 | #include "llvm/IR/DataLayout.h" |
| 38 | #include "llvm/IR/InlineAsm.h" |
| 39 | #include "llvm/IR/Intrinsics.h" |
| 40 | #include "llvm/IR/IntrinsicsAArch64.h" |
| 41 | #include "llvm/IR/IntrinsicsAMDGPU.h" |
| 42 | #include "llvm/IR/IntrinsicsARM.h" |
| 43 | #include "llvm/IR/IntrinsicsBPF.h" |
| 44 | #include "llvm/IR/IntrinsicsHexagon.h" |
| 45 | #include "llvm/IR/IntrinsicsLoongArch.h" |
| 46 | #include "llvm/IR/IntrinsicsNVPTX.h" |
| 47 | #include "llvm/IR/IntrinsicsPowerPC.h" |
| 48 | #include "llvm/IR/IntrinsicsR600.h" |
| 49 | #include "llvm/IR/IntrinsicsRISCV.h" |
| 50 | #include "llvm/IR/IntrinsicsS390.h" |
| 51 | #include "llvm/IR/IntrinsicsVE.h" |
| 52 | #include "llvm/IR/IntrinsicsWebAssembly.h" |
| 53 | #include "llvm/IR/IntrinsicsX86.h" |
| 54 | #include "llvm/IR/MDBuilder.h" |
| 55 | #include "llvm/IR/MatrixBuilder.h" |
| 56 | #include "llvm/Support/ConvertUTF.h" |
| 57 | #include "llvm/Support/ScopedPrinter.h" |
| 58 | #include "llvm/TargetParser/AArch64TargetParser.h" |
| 59 | #include "llvm/TargetParser/X86TargetParser.h" |
| 60 | #include <optional> |
| 61 | #include <sstream> |
| 62 | |
| 63 | using namespace clang; |
| 64 | using namespace CodeGen; |
| 65 | using namespace llvm; |
| 66 | |
| 67 | static void initializeAlloca(CodeGenFunction &CGF, AllocaInst *AI, Value *Size, |
| 68 | Align AlignmentInBytes) { |
| 69 | ConstantInt *Byte; |
| 70 | switch (CGF.getLangOpts().getTrivialAutoVarInit()) { |
| 71 | case LangOptions::TrivialAutoVarInitKind::Uninitialized: |
| 72 | // Nothing to initialize. |
| 73 | return; |
| 74 | case LangOptions::TrivialAutoVarInitKind::Zero: |
| 75 | Byte = CGF.Builder.getInt8(0x00); |
| 76 | break; |
| 77 | case LangOptions::TrivialAutoVarInitKind::Pattern: { |
| 78 | llvm::Type *Int8 = llvm::IntegerType::getInt8Ty(CGF.CGM.getLLVMContext()); |
| 79 | Byte = llvm::dyn_cast<llvm::ConstantInt>( |
| 80 | initializationPatternFor(CGF.CGM, Int8)); |
| 81 | break; |
| 82 | } |
| 83 | } |
| 84 | if (CGF.CGM.stopAutoInit()) |
| 85 | return; |
| 86 | auto *I = CGF.Builder.CreateMemSet(AI, Byte, Size, AlignmentInBytes); |
| 87 | I->addAnnotationMetadata("auto-init"); |
| 88 | } |
| 89 | |
| 90 | /// getBuiltinLibFunction - Given a builtin id for a function like |
| 91 | /// "__builtin_fabsf", return a Function* for "fabsf". |
| 92 | llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD, |
| 93 | unsigned BuiltinID) { |
| 94 | assert(Context.BuiltinInfo.isLibFunction(BuiltinID))(static_cast <bool> (Context.BuiltinInfo.isLibFunction( BuiltinID)) ? void (0) : __assert_fail ("Context.BuiltinInfo.isLibFunction(BuiltinID)" , "clang/lib/CodeGen/CGBuiltin.cpp", 94, __extension__ __PRETTY_FUNCTION__ )); |
| 95 | |
| 96 | // Get the name, skip over the __builtin_ prefix (if necessary). |
| 97 | StringRef Name; |
| 98 | GlobalDecl D(FD); |
| 99 | |
| 100 | // TODO: This list should be expanded or refactored after all GCC-compatible |
| 101 | // std libcall builtins are implemented. |
| 102 | static SmallDenseMap<unsigned, StringRef, 8> F128Builtins{ |
| 103 | {Builtin::BI__builtin_printf, "__printfieee128"}, |
| 104 | {Builtin::BI__builtin_vsnprintf, "__vsnprintfieee128"}, |
| 105 | {Builtin::BI__builtin_vsprintf, "__vsprintfieee128"}, |
| 106 | {Builtin::BI__builtin_sprintf, "__sprintfieee128"}, |
| 107 | {Builtin::BI__builtin_snprintf, "__snprintfieee128"}, |
| 108 | {Builtin::BI__builtin_fprintf, "__fprintfieee128"}, |
| 109 | {Builtin::BI__builtin_nexttowardf128, "__nexttowardieee128"}, |
| 110 | }; |
| 111 | |
| 112 | // The AIX library functions frexpl, ldexpl, and modfl are for 128-bit |
| 113 | // IBM 'long double' (i.e. __ibm128). Map to the 'double' versions |
| 114 | // if it is 64-bit 'long double' mode. |
| 115 | static SmallDenseMap<unsigned, StringRef, 4> AIXLongDouble64Builtins{ |
| 116 | {Builtin::BI__builtin_frexpl, "frexp"}, |
| 117 | {Builtin::BI__builtin_ldexpl, "ldexp"}, |
| 118 | {Builtin::BI__builtin_modfl, "modf"}, |
| 119 | }; |
| 120 | |
| 121 | // If the builtin has been declared explicitly with an assembler label, |
| 122 | // use the mangled name. This differs from the plain label on platforms |
| 123 | // that prefix labels. |
| 124 | if (FD->hasAttr<AsmLabelAttr>()) |
| 125 | Name = getMangledName(D); |
| 126 | else { |
| 127 | // TODO: This mutation should also be applied to other targets other than |
| 128 | // PPC, after backend supports IEEE 128-bit style libcalls. |
| 129 | if (getTriple().isPPC64() && |
| 130 | &getTarget().getLongDoubleFormat() == &llvm::APFloat::IEEEquad() && |
| 131 | F128Builtins.find(BuiltinID) != F128Builtins.end()) |
| 132 | Name = F128Builtins[BuiltinID]; |
| 133 | else if (getTriple().isOSAIX() && |
| 134 | &getTarget().getLongDoubleFormat() == |
| 135 | &llvm::APFloat::IEEEdouble() && |
| 136 | AIXLongDouble64Builtins.find(BuiltinID) != |
| 137 | AIXLongDouble64Builtins.end()) |
| 138 | Name = AIXLongDouble64Builtins[BuiltinID]; |
| 139 | else |
| 140 | Name = Context.BuiltinInfo.getName(BuiltinID).substr(10); |
| 141 | } |
| 142 | |
| 143 | llvm::FunctionType *Ty = |
| 144 | cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType())); |
| 145 | |
| 146 | return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false); |
| 147 | } |
| 148 | |
| 149 | /// Emit the conversions required to turn the given value into an |
| 150 | /// integer of the given size. |
| 151 | static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V, |
| 152 | QualType T, llvm::IntegerType *IntType) { |
| 153 | V = CGF.EmitToMemory(V, T); |
| 154 | |
| 155 | if (V->getType()->isPointerTy()) |
| 156 | return CGF.Builder.CreatePtrToInt(V, IntType); |
| 157 | |
| 158 | assert(V->getType() == IntType)(static_cast <bool> (V->getType() == IntType) ? void (0) : __assert_fail ("V->getType() == IntType", "clang/lib/CodeGen/CGBuiltin.cpp" , 158, __extension__ __PRETTY_FUNCTION__)); |
| 159 | return V; |
| 160 | } |
| 161 | |
| 162 | static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V, |
| 163 | QualType T, llvm::Type *ResultType) { |
| 164 | V = CGF.EmitFromMemory(V, T); |
| 165 | |
| 166 | if (ResultType->isPointerTy()) |
| 167 | return CGF.Builder.CreateIntToPtr(V, ResultType); |
| 168 | |
| 169 | assert(V->getType() == ResultType)(static_cast <bool> (V->getType() == ResultType) ? void (0) : __assert_fail ("V->getType() == ResultType", "clang/lib/CodeGen/CGBuiltin.cpp" , 169, __extension__ __PRETTY_FUNCTION__)); |
| 170 | return V; |
| 171 | } |
| 172 | |
| 173 | /// Utility to insert an atomic instruction based on Intrinsic::ID |
| 174 | /// and the expression node. |
| 175 | static Value *MakeBinaryAtomicValue( |
| 176 | CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E, |
| 177 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 178 | |
| 179 | QualType T = E->getType(); |
| 180 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 180, __extension__ __PRETTY_FUNCTION__ )); |
| 181 | assert(CGF.getContext().hasSameUnqualifiedType(T,(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 182, __extension__ __PRETTY_FUNCTION__ )) |
| 182 | E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 182, __extension__ __PRETTY_FUNCTION__ )); |
| 183 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 183, __extension__ __PRETTY_FUNCTION__ )); |
| 184 | |
| 185 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 186 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 187 | |
| 188 | llvm::IntegerType *IntType = |
| 189 | llvm::IntegerType::get(CGF.getLLVMContext(), |
| 190 | CGF.getContext().getTypeSize(T)); |
| 191 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 192 | |
| 193 | llvm::Value *Args[2]; |
| 194 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 195 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 196 | llvm::Type *ValueType = Args[1]->getType(); |
| 197 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 198 | |
| 199 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
| 200 | Kind, Args[0], Args[1], Ordering); |
| 201 | return EmitFromInt(CGF, Result, T, ValueType); |
| 202 | } |
| 203 | |
| 204 | static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) { |
| 205 | Value *Val = CGF.EmitScalarExpr(E->getArg(0)); |
| 206 | Value *Address = CGF.EmitScalarExpr(E->getArg(1)); |
| 207 | |
| 208 | // Convert the type of the pointer to a pointer to the stored type. |
| 209 | Val = CGF.EmitToMemory(Val, E->getArg(0)->getType()); |
| 210 | unsigned SrcAddrSpace = Address->getType()->getPointerAddressSpace(); |
| 211 | Value *BC = CGF.Builder.CreateBitCast( |
| 212 | Address, llvm::PointerType::get(Val->getType(), SrcAddrSpace), "cast"); |
| 213 | LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType()); |
| 214 | LV.setNontemporal(true); |
| 215 | CGF.EmitStoreOfScalar(Val, LV, false); |
| 216 | return nullptr; |
| 217 | } |
| 218 | |
| 219 | static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) { |
| 220 | Value *Address = CGF.EmitScalarExpr(E->getArg(0)); |
| 221 | |
| 222 | LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType()); |
| 223 | LV.setNontemporal(true); |
| 224 | return CGF.EmitLoadOfScalar(LV, E->getExprLoc()); |
| 225 | } |
| 226 | |
| 227 | static void CheckAtomicAlignment(CodeGenFunction &CGF, const CallExpr *E) { |
| 228 | ASTContext &Ctx = CGF.getContext(); |
| 229 | Address Ptr = CGF.EmitPointerWithAlignment(E->getArg(0)); |
| 230 | unsigned Bytes = Ptr.getElementType()->isPointerTy() |
| 231 | ? Ctx.getTypeSizeInChars(Ctx.VoidPtrTy).getQuantity() |
| 232 | : Ptr.getElementType()->getScalarSizeInBits() / 8; |
| 233 | unsigned Align = Ptr.getAlignment().getQuantity(); |
| 234 | if (Align % Bytes != 0) { |
| 235 | DiagnosticsEngine &Diags = CGF.CGM.getDiags(); |
| 236 | Diags.Report(E->getBeginLoc(), diag::warn_sync_op_misaligned); |
| 237 | } |
| 238 | } |
| 239 | |
| 240 | static RValue EmitBinaryAtomic(CodeGenFunction &CGF, |
| 241 | llvm::AtomicRMWInst::BinOp Kind, |
| 242 | const CallExpr *E) { |
| 243 | CheckAtomicAlignment(CGF, E); |
| 244 | return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E)); |
| 245 | } |
| 246 | |
| 247 | /// Utility to insert an atomic instruction based Intrinsic::ID and |
| 248 | /// the expression node, where the return value is the result of the |
| 249 | /// operation. |
| 250 | static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF, |
| 251 | llvm::AtomicRMWInst::BinOp Kind, |
| 252 | const CallExpr *E, |
| 253 | Instruction::BinaryOps Op, |
| 254 | bool Invert = false) { |
| 255 | CheckAtomicAlignment(CGF, E); |
| 256 | QualType T = E->getType(); |
| 257 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 257, __extension__ __PRETTY_FUNCTION__ )); |
| 258 | assert(CGF.getContext().hasSameUnqualifiedType(T,(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 259, __extension__ __PRETTY_FUNCTION__ )) |
| 259 | E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(0)->getType()->getPointeeType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 259, __extension__ __PRETTY_FUNCTION__ )); |
| 260 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (T, E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 260, __extension__ __PRETTY_FUNCTION__ )); |
| 261 | |
| 262 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 263 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 264 | |
| 265 | llvm::IntegerType *IntType = |
| 266 | llvm::IntegerType::get(CGF.getLLVMContext(), |
| 267 | CGF.getContext().getTypeSize(T)); |
| 268 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 269 | |
| 270 | llvm::Value *Args[2]; |
| 271 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 272 | llvm::Type *ValueType = Args[1]->getType(); |
| 273 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 274 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 275 | |
| 276 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
| 277 | Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent); |
| 278 | Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]); |
| 279 | if (Invert) |
| 280 | Result = |
| 281 | CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result, |
| 282 | llvm::ConstantInt::getAllOnesValue(IntType)); |
| 283 | Result = EmitFromInt(CGF, Result, T, ValueType); |
| 284 | return RValue::get(Result); |
| 285 | } |
| 286 | |
| 287 | /// Utility to insert an atomic cmpxchg instruction. |
| 288 | /// |
| 289 | /// @param CGF The current codegen function. |
| 290 | /// @param E Builtin call expression to convert to cmpxchg. |
| 291 | /// arg0 - address to operate on |
| 292 | /// arg1 - value to compare with |
| 293 | /// arg2 - new value |
| 294 | /// @param ReturnBool Specifies whether to return success flag of |
| 295 | /// cmpxchg result or the old value. |
| 296 | /// |
| 297 | /// @returns result of cmpxchg, according to ReturnBool |
| 298 | /// |
| 299 | /// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics |
| 300 | /// invoke the function EmitAtomicCmpXchgForMSIntrin. |
| 301 | static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E, |
| 302 | bool ReturnBool) { |
| 303 | CheckAtomicAlignment(CGF, E); |
| 304 | QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType(); |
| 305 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 306 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 307 | |
| 308 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
| 309 | CGF.getLLVMContext(), CGF.getContext().getTypeSize(T)); |
| 310 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 311 | |
| 312 | Value *Args[3]; |
| 313 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 314 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 315 | llvm::Type *ValueType = Args[1]->getType(); |
| 316 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 317 | Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType); |
| 318 | |
| 319 | Value *Pair = CGF.Builder.CreateAtomicCmpXchg( |
| 320 | Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent, |
| 321 | llvm::AtomicOrdering::SequentiallyConsistent); |
| 322 | if (ReturnBool) |
| 323 | // Extract boolean success flag and zext it to int. |
| 324 | return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1), |
| 325 | CGF.ConvertType(E->getType())); |
| 326 | else |
| 327 | // Extract old value and emit it using the same type as compare value. |
| 328 | return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T, |
| 329 | ValueType); |
| 330 | } |
| 331 | |
| 332 | /// This function should be invoked to emit atomic cmpxchg for Microsoft's |
| 333 | /// _InterlockedCompareExchange* intrinsics which have the following signature: |
| 334 | /// T _InterlockedCompareExchange(T volatile *Destination, |
| 335 | /// T Exchange, |
| 336 | /// T Comparand); |
| 337 | /// |
| 338 | /// Whereas the llvm 'cmpxchg' instruction has the following syntax: |
| 339 | /// cmpxchg *Destination, Comparand, Exchange. |
| 340 | /// So we need to swap Comparand and Exchange when invoking |
| 341 | /// CreateAtomicCmpXchg. That is the reason we could not use the above utility |
| 342 | /// function MakeAtomicCmpXchgValue since it expects the arguments to be |
| 343 | /// already swapped. |
| 344 | |
| 345 | static |
| 346 | Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E, |
| 347 | AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) { |
| 348 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 348, __extension__ __PRETTY_FUNCTION__ )); |
| 349 | assert(CGF.getContext().hasSameUnqualifiedType((static_cast <bool> (CGF.getContext().hasSameUnqualifiedType ( E->getType(), E->getArg(0)->getType()->getPointeeType ())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 350, __extension__ __PRETTY_FUNCTION__ )) |
| 350 | E->getType(), E->getArg(0)->getType()->getPointeeType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType ( E->getType(), E->getArg(0)->getType()->getPointeeType ())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType( E->getType(), E->getArg(0)->getType()->getPointeeType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 350, __extension__ __PRETTY_FUNCTION__ )); |
| 351 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 352, __extension__ __PRETTY_FUNCTION__ )) |
| 352 | E->getArg(1)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(1)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(1)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 352, __extension__ __PRETTY_FUNCTION__ )); |
| 353 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(2)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 354, __extension__ __PRETTY_FUNCTION__ )) |
| 354 | E->getArg(2)->getType()))(static_cast <bool> (CGF.getContext().hasSameUnqualifiedType (E->getType(), E->getArg(2)->getType())) ? void (0) : __assert_fail ("CGF.getContext().hasSameUnqualifiedType(E->getType(), E->getArg(2)->getType())" , "clang/lib/CodeGen/CGBuiltin.cpp", 354, __extension__ __PRETTY_FUNCTION__ )); |
| 355 | |
| 356 | auto *Destination = CGF.EmitScalarExpr(E->getArg(0)); |
| 357 | auto *Comparand = CGF.EmitScalarExpr(E->getArg(2)); |
| 358 | auto *Exchange = CGF.EmitScalarExpr(E->getArg(1)); |
| 359 | |
| 360 | // For Release ordering, the failure ordering should be Monotonic. |
| 361 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ? |
| 362 | AtomicOrdering::Monotonic : |
| 363 | SuccessOrdering; |
| 364 | |
| 365 | // The atomic instruction is marked volatile for consistency with MSVC. This |
| 366 | // blocks the few atomics optimizations that LLVM has. If we want to optimize |
| 367 | // _Interlocked* operations in the future, we will have to remove the volatile |
| 368 | // marker. |
| 369 | auto *Result = CGF.Builder.CreateAtomicCmpXchg( |
| 370 | Destination, Comparand, Exchange, |
| 371 | SuccessOrdering, FailureOrdering); |
| 372 | Result->setVolatile(true); |
| 373 | return CGF.Builder.CreateExtractValue(Result, 0); |
| 374 | } |
| 375 | |
| 376 | // 64-bit Microsoft platforms support 128 bit cmpxchg operations. They are |
| 377 | // prototyped like this: |
| 378 | // |
| 379 | // unsigned char _InterlockedCompareExchange128...( |
| 380 | // __int64 volatile * _Destination, |
| 381 | // __int64 _ExchangeHigh, |
| 382 | // __int64 _ExchangeLow, |
| 383 | // __int64 * _ComparandResult); |
| 384 | static Value *EmitAtomicCmpXchg128ForMSIntrin(CodeGenFunction &CGF, |
| 385 | const CallExpr *E, |
| 386 | AtomicOrdering SuccessOrdering) { |
| 387 | assert(E->getNumArgs() == 4)(static_cast <bool> (E->getNumArgs() == 4) ? void (0 ) : __assert_fail ("E->getNumArgs() == 4", "clang/lib/CodeGen/CGBuiltin.cpp" , 387, __extension__ __PRETTY_FUNCTION__)); |
| 388 | llvm::Value *Destination = CGF.EmitScalarExpr(E->getArg(0)); |
| 389 | llvm::Value *ExchangeHigh = CGF.EmitScalarExpr(E->getArg(1)); |
| 390 | llvm::Value *ExchangeLow = CGF.EmitScalarExpr(E->getArg(2)); |
| 391 | llvm::Value *ComparandPtr = CGF.EmitScalarExpr(E->getArg(3)); |
| 392 | |
| 393 | assert(Destination->getType()->isPointerTy())(static_cast <bool> (Destination->getType()->isPointerTy ()) ? void (0) : __assert_fail ("Destination->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 393, __extension__ __PRETTY_FUNCTION__ )); |
| 394 | assert(!ExchangeHigh->getType()->isPointerTy())(static_cast <bool> (!ExchangeHigh->getType()->isPointerTy ()) ? void (0) : __assert_fail ("!ExchangeHigh->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 394, __extension__ __PRETTY_FUNCTION__ )); |
| 395 | assert(!ExchangeLow->getType()->isPointerTy())(static_cast <bool> (!ExchangeLow->getType()->isPointerTy ()) ? void (0) : __assert_fail ("!ExchangeLow->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 395, __extension__ __PRETTY_FUNCTION__ )); |
| 396 | assert(ComparandPtr->getType()->isPointerTy())(static_cast <bool> (ComparandPtr->getType()->isPointerTy ()) ? void (0) : __assert_fail ("ComparandPtr->getType()->isPointerTy()" , "clang/lib/CodeGen/CGBuiltin.cpp", 396, __extension__ __PRETTY_FUNCTION__ )); |
| 397 | |
| 398 | // For Release ordering, the failure ordering should be Monotonic. |
| 399 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release |
| 400 | ? AtomicOrdering::Monotonic |
| 401 | : SuccessOrdering; |
| 402 | |
| 403 | // Convert to i128 pointers and values. |
| 404 | llvm::Type *Int128Ty = llvm::IntegerType::get(CGF.getLLVMContext(), 128); |
| 405 | llvm::Type *Int128PtrTy = Int128Ty->getPointerTo(); |
| 406 | Destination = CGF.Builder.CreateBitCast(Destination, Int128PtrTy); |
| 407 | Address ComparandResult(CGF.Builder.CreateBitCast(ComparandPtr, Int128PtrTy), |
| 408 | Int128Ty, CGF.getContext().toCharUnitsFromBits(128)); |
| 409 | |
| 410 | // (((i128)hi) << 64) | ((i128)lo) |
| 411 | ExchangeHigh = CGF.Builder.CreateZExt(ExchangeHigh, Int128Ty); |
| 412 | ExchangeLow = CGF.Builder.CreateZExt(ExchangeLow, Int128Ty); |
| 413 | ExchangeHigh = |
| 414 | CGF.Builder.CreateShl(ExchangeHigh, llvm::ConstantInt::get(Int128Ty, 64)); |
| 415 | llvm::Value *Exchange = CGF.Builder.CreateOr(ExchangeHigh, ExchangeLow); |
| 416 | |
| 417 | // Load the comparand for the instruction. |
| 418 | llvm::Value *Comparand = CGF.Builder.CreateLoad(ComparandResult); |
| 419 | |
| 420 | auto *CXI = CGF.Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
| 421 | SuccessOrdering, FailureOrdering); |
| 422 | |
| 423 | // The atomic instruction is marked volatile for consistency with MSVC. This |
| 424 | // blocks the few atomics optimizations that LLVM has. If we want to optimize |
| 425 | // _Interlocked* operations in the future, we will have to remove the volatile |
| 426 | // marker. |
| 427 | CXI->setVolatile(true); |
| 428 | |
| 429 | // Store the result as an outparameter. |
| 430 | CGF.Builder.CreateStore(CGF.Builder.CreateExtractValue(CXI, 0), |
| 431 | ComparandResult); |
| 432 | |
| 433 | // Get the success boolean and zero extend it to i8. |
| 434 | Value *Success = CGF.Builder.CreateExtractValue(CXI, 1); |
| 435 | return CGF.Builder.CreateZExt(Success, CGF.Int8Ty); |
| 436 | } |
| 437 | |
| 438 | static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E, |
| 439 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 440 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 440, __extension__ __PRETTY_FUNCTION__ )); |
| 441 | |
| 442 | auto *IntTy = CGF.ConvertType(E->getType()); |
| 443 | auto *Result = CGF.Builder.CreateAtomicRMW( |
| 444 | AtomicRMWInst::Add, |
| 445 | CGF.EmitScalarExpr(E->getArg(0)), |
| 446 | ConstantInt::get(IntTy, 1), |
| 447 | Ordering); |
| 448 | return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1)); |
| 449 | } |
| 450 | |
| 451 | static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E, |
| 452 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 453 | assert(E->getArg(0)->getType()->isPointerType())(static_cast <bool> (E->getArg(0)->getType()-> isPointerType()) ? void (0) : __assert_fail ("E->getArg(0)->getType()->isPointerType()" , "clang/lib/CodeGen/CGBuiltin.cpp", 453, __extension__ __PRETTY_FUNCTION__ )); |
| 454 | |
| 455 | auto *IntTy = CGF.ConvertType(E->getType()); |
| 456 | auto *Result = CGF.Builder.CreateAtomicRMW( |
| 457 | AtomicRMWInst::Sub, |
| 458 | CGF.EmitScalarExpr(E->getArg(0)), |
| 459 | ConstantInt::get(IntTy, 1), |
| 460 | Ordering); |
| 461 | return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1)); |
| 462 | } |
| 463 | |
| 464 | // Build a plain volatile load. |
| 465 | static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) { |
| 466 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
| 467 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 468 | CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy); |
| 469 | llvm::Type *ITy = |
| 470 | llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8); |
| 471 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 472 | llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(ITy, Ptr, LoadSize); |
| 473 | Load->setVolatile(true); |
| 474 | return Load; |
| 475 | } |
| 476 | |
| 477 | // Build a plain volatile store. |
| 478 | static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) { |
| 479 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
| 480 | Value *Value = CGF.EmitScalarExpr(E->getArg(1)); |
| 481 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 482 | CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy); |
| 483 | llvm::Type *ITy = |
| 484 | llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8); |
| 485 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 486 | llvm::StoreInst *Store = |
| 487 | CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize); |
| 488 | Store->setVolatile(true); |
| 489 | return Store; |
| 490 | } |
| 491 | |
| 492 | // Emit a simple mangled intrinsic that has 1 argument and a return type |
| 493 | // matching the argument type. Depending on mode, this may be a constrained |
| 494 | // floating-point intrinsic. |
| 495 | static Value *emitUnaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 496 | const CallExpr *E, unsigned IntrinsicID, |
| 497 | unsigned ConstrainedIntrinsicID) { |
| 498 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 499 | |
| 500 | if (CGF.Builder.getIsFPConstrained()) { |
| 501 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 502 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
| 503 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0 }); |
| 504 | } else { |
| 505 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 506 | return CGF.Builder.CreateCall(F, Src0); |
| 507 | } |
| 508 | } |
| 509 | |
| 510 | // Emit an intrinsic that has 2 operands of the same type as its result. |
| 511 | // Depending on mode, this may be a constrained floating-point intrinsic. |
| 512 | static Value *emitBinaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 513 | const CallExpr *E, unsigned IntrinsicID, |
| 514 | unsigned ConstrainedIntrinsicID) { |
| 515 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 516 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 517 | |
| 518 | if (CGF.Builder.getIsFPConstrained()) { |
| 519 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 520 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
| 521 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1 }); |
| 522 | } else { |
| 523 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 524 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | // Emit an intrinsic that has 3 operands of the same type as its result. |
| 529 | // Depending on mode, this may be a constrained floating-point intrinsic. |
| 530 | static Value *emitTernaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 531 | const CallExpr *E, unsigned IntrinsicID, |
| 532 | unsigned ConstrainedIntrinsicID) { |
| 533 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 534 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 535 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); |
| 536 | |
| 537 | if (CGF.Builder.getIsFPConstrained()) { |
| 538 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 539 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
| 540 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1, Src2 }); |
| 541 | } else { |
| 542 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 543 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); |
| 544 | } |
| 545 | } |
| 546 | |
| 547 | // Emit an intrinsic where all operands are of the same type as the result. |
| 548 | // Depending on mode, this may be a constrained floating-point intrinsic. |
| 549 | static Value *emitCallMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 550 | unsigned IntrinsicID, |
| 551 | unsigned ConstrainedIntrinsicID, |
| 552 | llvm::Type *Ty, |
| 553 | ArrayRef<Value *> Args) { |
| 554 | Function *F; |
| 555 | if (CGF.Builder.getIsFPConstrained()) |
| 556 | F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Ty); |
| 557 | else |
| 558 | F = CGF.CGM.getIntrinsic(IntrinsicID, Ty); |
| 559 | |
| 560 | if (CGF.Builder.getIsFPConstrained()) |
| 561 | return CGF.Builder.CreateConstrainedFPCall(F, Args); |
| 562 | else |
| 563 | return CGF.Builder.CreateCall(F, Args); |
| 564 | } |
| 565 | |
| 566 | // Emit a simple mangled intrinsic that has 1 argument and a return type |
| 567 | // matching the argument type. |
| 568 | static Value *emitUnaryBuiltin(CodeGenFunction &CGF, const CallExpr *E, |
| 569 | unsigned IntrinsicID, |
| 570 | llvm::StringRef Name = "") { |
| 571 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 572 | |
| 573 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 574 | return CGF.Builder.CreateCall(F, Src0, Name); |
| 575 | } |
| 576 | |
| 577 | // Emit an intrinsic that has 2 operands of the same type as its result. |
| 578 | static Value *emitBinaryBuiltin(CodeGenFunction &CGF, |
| 579 | const CallExpr *E, |
| 580 | unsigned IntrinsicID) { |
| 581 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 582 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 583 | |
| 584 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 585 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); |
| 586 | } |
| 587 | |
| 588 | // Emit an intrinsic that has 3 operands of the same type as its result. |
| 589 | static Value *emitTernaryBuiltin(CodeGenFunction &CGF, |
| 590 | const CallExpr *E, |
| 591 | unsigned IntrinsicID) { |
| 592 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 593 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 594 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); |
| 595 | |
| 596 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 597 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); |
| 598 | } |
| 599 | |
| 600 | // Emit an intrinsic that has 1 float or double operand, and 1 integer. |
| 601 | static Value *emitFPIntBuiltin(CodeGenFunction &CGF, |
| 602 | const CallExpr *E, |
| 603 | unsigned IntrinsicID) { |
| 604 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 605 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 606 | |
| 607 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 608 | return CGF.Builder.CreateCall(F, {Src0, Src1}); |
| 609 | } |
| 610 | |
| 611 | // Emit an intrinsic that has overloaded integer result and fp operand. |
| 612 | static Value * |
| 613 | emitMaybeConstrainedFPToIntRoundBuiltin(CodeGenFunction &CGF, const CallExpr *E, |
| 614 | unsigned IntrinsicID, |
| 615 | unsigned ConstrainedIntrinsicID) { |
| 616 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); |
| 617 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 618 | |
| 619 | if (CGF.Builder.getIsFPConstrained()) { |
| 620 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 621 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, |
| 622 | {ResultType, Src0->getType()}); |
| 623 | return CGF.Builder.CreateConstrainedFPCall(F, {Src0}); |
| 624 | } else { |
| 625 | Function *F = |
| 626 | CGF.CGM.getIntrinsic(IntrinsicID, {ResultType, Src0->getType()}); |
| 627 | return CGF.Builder.CreateCall(F, Src0); |
| 628 | } |
| 629 | } |
| 630 | |
| 631 | /// EmitFAbs - Emit a call to @llvm.fabs(). |
| 632 | static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) { |
| 633 | Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType()); |
| 634 | llvm::CallInst *Call = CGF.Builder.CreateCall(F, V); |
| 635 | Call->setDoesNotAccessMemory(); |
| 636 | return Call; |
| 637 | } |
| 638 | |
| 639 | /// Emit the computation of the sign bit for a floating point value. Returns |
| 640 | /// the i1 sign bit value. |
| 641 | static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) { |
| 642 | LLVMContext &C = CGF.CGM.getLLVMContext(); |
| 643 | |
| 644 | llvm::Type *Ty = V->getType(); |
| 645 | int Width = Ty->getPrimitiveSizeInBits(); |
| 646 | llvm::Type *IntTy = llvm::IntegerType::get(C, Width); |
| 647 | V = CGF.Builder.CreateBitCast(V, IntTy); |
| 648 | if (Ty->isPPC_FP128Ty()) { |
| 649 | // We want the sign bit of the higher-order double. The bitcast we just |
| 650 | // did works as if the double-double was stored to memory and then |
| 651 | // read as an i128. The "store" will put the higher-order double in the |
| 652 | // lower address in both little- and big-Endian modes, but the "load" |
| 653 | // will treat those bits as a different part of the i128: the low bits in |
| 654 | // little-Endian, the high bits in big-Endian. Therefore, on big-Endian |
| 655 | // we need to shift the high bits down to the low before truncating. |
| 656 | Width >>= 1; |
| 657 | if (CGF.getTarget().isBigEndian()) { |
| 658 | Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width); |
| 659 | V = CGF.Builder.CreateLShr(V, ShiftCst); |
| 660 | } |
| 661 | // We are truncating value in order to extract the higher-order |
| 662 | // double, which we will be using to extract the sign from. |
| 663 | IntTy = llvm::IntegerType::get(C, Width); |
| 664 | V = CGF.Builder.CreateTrunc(V, IntTy); |
| 665 | } |
| 666 | Value *Zero = llvm::Constant::getNullValue(IntTy); |
| 667 | return CGF.Builder.CreateICmpSLT(V, Zero); |
| 668 | } |
| 669 | |
| 670 | static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD, |
| 671 | const CallExpr *E, llvm::Constant *calleeValue) { |
| 672 | CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD)); |
| 673 | return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot()); |
| 674 | } |
| 675 | |
| 676 | /// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.* |
| 677 | /// depending on IntrinsicID. |
| 678 | /// |
| 679 | /// \arg CGF The current codegen function. |
| 680 | /// \arg IntrinsicID The ID for the Intrinsic we wish to generate. |
| 681 | /// \arg X The first argument to the llvm.*.with.overflow.*. |
| 682 | /// \arg Y The second argument to the llvm.*.with.overflow.*. |
| 683 | /// \arg Carry The carry returned by the llvm.*.with.overflow.*. |
| 684 | /// \returns The result (i.e. sum/product) returned by the intrinsic. |
| 685 | static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF, |
| 686 | const llvm::Intrinsic::ID IntrinsicID, |
| 687 | llvm::Value *X, llvm::Value *Y, |
| 688 | llvm::Value *&Carry) { |
| 689 | // Make sure we have integers of the same width. |
| 690 | assert(X->getType() == Y->getType() &&(static_cast <bool> (X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? void (0) : __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 692, __extension__ __PRETTY_FUNCTION__ )) |
| 691 | "Arguments must be the same type. (Did you forget to make sure both "(static_cast <bool> (X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? void (0) : __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 692, __extension__ __PRETTY_FUNCTION__ )) |
| 692 | "arguments have the same integer width?)")(static_cast <bool> (X->getType() == Y->getType() && "Arguments must be the same type. (Did you forget to make sure both " "arguments have the same integer width?)") ? void (0) : __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 692, __extension__ __PRETTY_FUNCTION__ )); |
| 693 | |
| 694 | Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType()); |
| 695 | llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y}); |
| 696 | Carry = CGF.Builder.CreateExtractValue(Tmp, 1); |
| 697 | return CGF.Builder.CreateExtractValue(Tmp, 0); |
| 698 | } |
| 699 | |
| 700 | static Value *emitRangedBuiltin(CodeGenFunction &CGF, |
| 701 | unsigned IntrinsicID, |
| 702 | int low, int high) { |
| 703 | llvm::MDBuilder MDHelper(CGF.getLLVMContext()); |
| 704 | llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high)); |
| 705 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {}); |
| 706 | llvm::Instruction *Call = CGF.Builder.CreateCall(F); |
| 707 | Call->setMetadata(llvm::LLVMContext::MD_range, RNode); |
| 708 | Call->setMetadata(llvm::LLVMContext::MD_noundef, |
| 709 | llvm::MDNode::get(CGF.getLLVMContext(), std::nullopt)); |
| 710 | return Call; |
| 711 | } |
| 712 | |
| 713 | namespace { |
| 714 | struct WidthAndSignedness { |
| 715 | unsigned Width; |
| 716 | bool Signed; |
| 717 | }; |
| 718 | } |
| 719 | |
| 720 | static WidthAndSignedness |
| 721 | getIntegerWidthAndSignedness(const clang::ASTContext &context, |
| 722 | const clang::QualType Type) { |
| 723 | assert(Type->isIntegerType() && "Given type is not an integer.")(static_cast <bool> (Type->isIntegerType() && "Given type is not an integer.") ? void (0) : __assert_fail ( "Type->isIntegerType() && \"Given type is not an integer.\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 723, __extension__ __PRETTY_FUNCTION__ )); |
| 724 | unsigned Width = Type->isBooleanType() ? 1 |
| 725 | : Type->isBitIntType() ? context.getIntWidth(Type) |
| 726 | : context.getTypeInfo(Type).Width; |
| 727 | bool Signed = Type->isSignedIntegerType(); |
| 728 | return {Width, Signed}; |
| 729 | } |
| 730 | |
| 731 | // Given one or more integer types, this function produces an integer type that |
| 732 | // encompasses them: any value in one of the given types could be expressed in |
| 733 | // the encompassing type. |
| 734 | static struct WidthAndSignedness |
| 735 | EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) { |
| 736 | assert(Types.size() > 0 && "Empty list of types.")(static_cast <bool> (Types.size() > 0 && "Empty list of types." ) ? void (0) : __assert_fail ("Types.size() > 0 && \"Empty list of types.\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 736, __extension__ __PRETTY_FUNCTION__ )); |
| 737 | |
| 738 | // If any of the given types is signed, we must return a signed type. |
| 739 | bool Signed = false; |
| 740 | for (const auto &Type : Types) { |
| 741 | Signed |= Type.Signed; |
| 742 | } |
| 743 | |
| 744 | // The encompassing type must have a width greater than or equal to the width |
| 745 | // of the specified types. Additionally, if the encompassing type is signed, |
| 746 | // its width must be strictly greater than the width of any unsigned types |
| 747 | // given. |
| 748 | unsigned Width = 0; |
| 749 | for (const auto &Type : Types) { |
| 750 | unsigned MinWidth = Type.Width + (Signed && !Type.Signed); |
| 751 | if (Width < MinWidth) { |
| 752 | Width = MinWidth; |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | return {Width, Signed}; |
| 757 | } |
| 758 | |
| 759 | Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) { |
| 760 | llvm::Type *DestType = Int8PtrTy; |
| 761 | if (ArgValue->getType() != DestType) |
| 762 | ArgValue = |
| 763 | Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data()); |
| 764 | |
| 765 | Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend; |
| 766 | return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue); |
| 767 | } |
| 768 | |
| 769 | /// Checks if using the result of __builtin_object_size(p, @p From) in place of |
| 770 | /// __builtin_object_size(p, @p To) is correct |
| 771 | static bool areBOSTypesCompatible(int From, int To) { |
| 772 | // Note: Our __builtin_object_size implementation currently treats Type=0 and |
| 773 | // Type=2 identically. Encoding this implementation detail here may make |
| 774 | // improving __builtin_object_size difficult in the future, so it's omitted. |
| 775 | return From == To || (From == 0 && To == 1) || (From == 3 && To == 2); |
| 776 | } |
| 777 | |
| 778 | static llvm::Value * |
| 779 | getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) { |
| 780 | return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true); |
| 781 | } |
| 782 | |
| 783 | llvm::Value * |
| 784 | CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type, |
| 785 | llvm::IntegerType *ResType, |
| 786 | llvm::Value *EmittedE, |
| 787 | bool IsDynamic) { |
| 788 | uint64_t ObjectSize; |
| 789 | if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type)) |
| 790 | return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic); |
| 791 | return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true); |
| 792 | } |
| 793 | |
| 794 | /// Returns a Value corresponding to the size of the given expression. |
| 795 | /// This Value may be either of the following: |
| 796 | /// - A llvm::Argument (if E is a param with the pass_object_size attribute on |
| 797 | /// it) |
| 798 | /// - A call to the @llvm.objectsize intrinsic |
| 799 | /// |
| 800 | /// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null |
| 801 | /// and we wouldn't otherwise try to reference a pass_object_size parameter, |
| 802 | /// we'll call @llvm.objectsize on EmittedE, rather than emitting E. |
| 803 | llvm::Value * |
| 804 | CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type, |
| 805 | llvm::IntegerType *ResType, |
| 806 | llvm::Value *EmittedE, bool IsDynamic) { |
| 807 | // We need to reference an argument if the pointer is a parameter with the |
| 808 | // pass_object_size attribute. |
| 809 | if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) { |
| 810 | auto *Param = dyn_cast<ParmVarDecl>(D->getDecl()); |
| 811 | auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>(); |
| 812 | if (Param != nullptr && PS != nullptr && |
| 813 | areBOSTypesCompatible(PS->getType(), Type)) { |
| 814 | auto Iter = SizeArguments.find(Param); |
| 815 | assert(Iter != SizeArguments.end())(static_cast <bool> (Iter != SizeArguments.end()) ? void (0) : __assert_fail ("Iter != SizeArguments.end()", "clang/lib/CodeGen/CGBuiltin.cpp" , 815, __extension__ __PRETTY_FUNCTION__)); |
| 816 | |
| 817 | const ImplicitParamDecl *D = Iter->second; |
| 818 | auto DIter = LocalDeclMap.find(D); |
| 819 | assert(DIter != LocalDeclMap.end())(static_cast <bool> (DIter != LocalDeclMap.end()) ? void (0) : __assert_fail ("DIter != LocalDeclMap.end()", "clang/lib/CodeGen/CGBuiltin.cpp" , 819, __extension__ __PRETTY_FUNCTION__)); |
| 820 | |
| 821 | return EmitLoadOfScalar(DIter->second, /*Volatile=*/false, |
| 822 | getContext().getSizeType(), E->getBeginLoc()); |
| 823 | } |
| 824 | } |
| 825 | |
| 826 | // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't |
| 827 | // evaluate E for side-effects. In either case, we shouldn't lower to |
| 828 | // @llvm.objectsize. |
| 829 | if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext()))) |
| 830 | return getDefaultBuiltinObjectSizeResult(Type, ResType); |
| 831 | |
| 832 | Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E); |
| 833 | assert(Ptr->getType()->isPointerTy() &&(static_cast <bool> (Ptr->getType()->isPointerTy( ) && "Non-pointer passed to __builtin_object_size?") ? void (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 834, __extension__ __PRETTY_FUNCTION__ )) |
| 834 | "Non-pointer passed to __builtin_object_size?")(static_cast <bool> (Ptr->getType()->isPointerTy( ) && "Non-pointer passed to __builtin_object_size?") ? void (0) : __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 834, __extension__ __PRETTY_FUNCTION__ )); |
| 835 | |
| 836 | Function *F = |
| 837 | CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()}); |
| 838 | |
| 839 | // LLVM only supports 0 and 2, make sure that we pass along that as a boolean. |
| 840 | Value *Min = Builder.getInt1((Type & 2) != 0); |
| 841 | // For GCC compatibility, __builtin_object_size treat NULL as unknown size. |
| 842 | Value *NullIsUnknown = Builder.getTrue(); |
| 843 | Value *Dynamic = Builder.getInt1(IsDynamic); |
| 844 | return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic}); |
| 845 | } |
| 846 | |
| 847 | namespace { |
| 848 | /// A struct to generically describe a bit test intrinsic. |
| 849 | struct BitTest { |
| 850 | enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set }; |
| 851 | enum InterlockingKind : uint8_t { |
| 852 | Unlocked, |
| 853 | Sequential, |
| 854 | Acquire, |
| 855 | Release, |
| 856 | NoFence |
| 857 | }; |
| 858 | |
| 859 | ActionKind Action; |
| 860 | InterlockingKind Interlocking; |
| 861 | bool Is64Bit; |
| 862 | |
| 863 | static BitTest decodeBitTestBuiltin(unsigned BuiltinID); |
| 864 | }; |
| 865 | } // namespace |
| 866 | |
| 867 | BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) { |
| 868 | switch (BuiltinID) { |
| 869 | // Main portable variants. |
| 870 | case Builtin::BI_bittest: |
| 871 | return {TestOnly, Unlocked, false}; |
| 872 | case Builtin::BI_bittestandcomplement: |
| 873 | return {Complement, Unlocked, false}; |
| 874 | case Builtin::BI_bittestandreset: |
| 875 | return {Reset, Unlocked, false}; |
| 876 | case Builtin::BI_bittestandset: |
| 877 | return {Set, Unlocked, false}; |
| 878 | case Builtin::BI_interlockedbittestandreset: |
| 879 | return {Reset, Sequential, false}; |
| 880 | case Builtin::BI_interlockedbittestandset: |
| 881 | return {Set, Sequential, false}; |
| 882 | |
| 883 | // X86-specific 64-bit variants. |
| 884 | case Builtin::BI_bittest64: |
| 885 | return {TestOnly, Unlocked, true}; |
| 886 | case Builtin::BI_bittestandcomplement64: |
| 887 | return {Complement, Unlocked, true}; |
| 888 | case Builtin::BI_bittestandreset64: |
| 889 | return {Reset, Unlocked, true}; |
| 890 | case Builtin::BI_bittestandset64: |
| 891 | return {Set, Unlocked, true}; |
| 892 | case Builtin::BI_interlockedbittestandreset64: |
| 893 | return {Reset, Sequential, true}; |
| 894 | case Builtin::BI_interlockedbittestandset64: |
| 895 | return {Set, Sequential, true}; |
| 896 | |
| 897 | // ARM/AArch64-specific ordering variants. |
| 898 | case Builtin::BI_interlockedbittestandset_acq: |
| 899 | return {Set, Acquire, false}; |
| 900 | case Builtin::BI_interlockedbittestandset_rel: |
| 901 | return {Set, Release, false}; |
| 902 | case Builtin::BI_interlockedbittestandset_nf: |
| 903 | return {Set, NoFence, false}; |
| 904 | case Builtin::BI_interlockedbittestandreset_acq: |
| 905 | return {Reset, Acquire, false}; |
| 906 | case Builtin::BI_interlockedbittestandreset_rel: |
| 907 | return {Reset, Release, false}; |
| 908 | case Builtin::BI_interlockedbittestandreset_nf: |
| 909 | return {Reset, NoFence, false}; |
| 910 | } |
| 911 | llvm_unreachable("expected only bittest intrinsics")::llvm::llvm_unreachable_internal("expected only bittest intrinsics" , "clang/lib/CodeGen/CGBuiltin.cpp", 911); |
| 912 | } |
| 913 | |
| 914 | static char bitActionToX86BTCode(BitTest::ActionKind A) { |
| 915 | switch (A) { |
| 916 | case BitTest::TestOnly: return '\0'; |
| 917 | case BitTest::Complement: return 'c'; |
| 918 | case BitTest::Reset: return 'r'; |
| 919 | case BitTest::Set: return 's'; |
| 920 | } |
| 921 | llvm_unreachable("invalid action")::llvm::llvm_unreachable_internal("invalid action", "clang/lib/CodeGen/CGBuiltin.cpp" , 921); |
| 922 | } |
| 923 | |
| 924 | static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF, |
| 925 | BitTest BT, |
| 926 | const CallExpr *E, Value *BitBase, |
| 927 | Value *BitPos) { |
| 928 | char Action = bitActionToX86BTCode(BT.Action); |
| 929 | char SizeSuffix = BT.Is64Bit ? 'q' : 'l'; |
| 930 | |
| 931 | // Build the assembly. |
| 932 | SmallString<64> Asm; |
| 933 | raw_svector_ostream AsmOS(Asm); |
| 934 | if (BT.Interlocking != BitTest::Unlocked) |
| 935 | AsmOS << "lock "; |
| 936 | AsmOS << "bt"; |
| 937 | if (Action) |
| 938 | AsmOS << Action; |
| 939 | AsmOS << SizeSuffix << " $2, ($1)"; |
| 940 | |
| 941 | // Build the constraints. FIXME: We should support immediates when possible. |
| 942 | std::string Constraints = "={@ccc},r,r,~{cc},~{memory}"; |
| 943 | std::string MachineClobbers = CGF.getTarget().getClobbers(); |
| 944 | if (!MachineClobbers.empty()) { |
| 945 | Constraints += ','; |
| 946 | Constraints += MachineClobbers; |
| 947 | } |
| 948 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
| 949 | CGF.getLLVMContext(), |
| 950 | CGF.getContext().getTypeSize(E->getArg(1)->getType())); |
| 951 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
| 952 | llvm::FunctionType *FTy = |
| 953 | llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false); |
| 954 | |
| 955 | llvm::InlineAsm *IA = |
| 956 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
| 957 | return CGF.Builder.CreateCall(IA, {BitBase, BitPos}); |
| 958 | } |
| 959 | |
| 960 | static llvm::AtomicOrdering |
| 961 | getBitTestAtomicOrdering(BitTest::InterlockingKind I) { |
| 962 | switch (I) { |
| 963 | case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic; |
| 964 | case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent; |
| 965 | case BitTest::Acquire: return llvm::AtomicOrdering::Acquire; |
| 966 | case BitTest::Release: return llvm::AtomicOrdering::Release; |
| 967 | case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic; |
| 968 | } |
| 969 | llvm_unreachable("invalid interlocking")::llvm::llvm_unreachable_internal("invalid interlocking", "clang/lib/CodeGen/CGBuiltin.cpp" , 969); |
| 970 | } |
| 971 | |
| 972 | /// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of |
| 973 | /// bits and a bit position and read and optionally modify the bit at that |
| 974 | /// position. The position index can be arbitrarily large, i.e. it can be larger |
| 975 | /// than 31 or 63, so we need an indexed load in the general case. |
| 976 | static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF, |
| 977 | unsigned BuiltinID, |
| 978 | const CallExpr *E) { |
| 979 | Value *BitBase = CGF.EmitScalarExpr(E->getArg(0)); |
| 980 | Value *BitPos = CGF.EmitScalarExpr(E->getArg(1)); |
| 981 | |
| 982 | BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID); |
| 983 | |
| 984 | // X86 has special BT, BTC, BTR, and BTS instructions that handle the array |
| 985 | // indexing operation internally. Use them if possible. |
| 986 | if (CGF.getTarget().getTriple().isX86()) |
| 987 | return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos); |
| 988 | |
| 989 | // Otherwise, use generic code to load one byte and test the bit. Use all but |
| 990 | // the bottom three bits as the array index, and the bottom three bits to form |
| 991 | // a mask. |
| 992 | // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0; |
| 993 | Value *ByteIndex = CGF.Builder.CreateAShr( |
| 994 | BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx"); |
| 995 | Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy); |
| 996 | Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8, |
| 997 | ByteIndex, "bittest.byteaddr"), |
| 998 | CGF.Int8Ty, CharUnits::One()); |
| 999 | Value *PosLow = |
| 1000 | CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty), |
| 1001 | llvm::ConstantInt::get(CGF.Int8Ty, 0x7)); |
| 1002 | |
| 1003 | // The updating instructions will need a mask. |
| 1004 | Value *Mask = nullptr; |
| 1005 | if (BT.Action != BitTest::TestOnly) { |
| 1006 | Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow, |
| 1007 | "bittest.mask"); |
| 1008 | } |
| 1009 | |
| 1010 | // Check the action and ordering of the interlocked intrinsics. |
| 1011 | llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking); |
| 1012 | |
| 1013 | Value *OldByte = nullptr; |
| 1014 | if (Ordering != llvm::AtomicOrdering::NotAtomic) { |
| 1015 | // Emit a combined atomicrmw load/store operation for the interlocked |
| 1016 | // intrinsics. |
| 1017 | llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or; |
| 1018 | if (BT.Action == BitTest::Reset) { |
| 1019 | Mask = CGF.Builder.CreateNot(Mask); |
| 1020 | RMWOp = llvm::AtomicRMWInst::And; |
| 1021 | } |
| 1022 | OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask, |
| 1023 | Ordering); |
| 1024 | } else { |
| 1025 | // Emit a plain load for the non-interlocked intrinsics. |
| 1026 | OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte"); |
| 1027 | Value *NewByte = nullptr; |
| 1028 | switch (BT.Action) { |
| 1029 | case BitTest::TestOnly: |
| 1030 | // Don't store anything. |
| 1031 | break; |
| 1032 | case BitTest::Complement: |
| 1033 | NewByte = CGF.Builder.CreateXor(OldByte, Mask); |
| 1034 | break; |
| 1035 | case BitTest::Reset: |
| 1036 | NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask)); |
| 1037 | break; |
| 1038 | case BitTest::Set: |
| 1039 | NewByte = CGF.Builder.CreateOr(OldByte, Mask); |
| 1040 | break; |
| 1041 | } |
| 1042 | if (NewByte) |
| 1043 | CGF.Builder.CreateStore(NewByte, ByteAddr); |
| 1044 | } |
| 1045 | |
| 1046 | // However we loaded the old byte, either by plain load or atomicrmw, shift |
| 1047 | // the bit into the low position and mask it to 0 or 1. |
| 1048 | Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr"); |
| 1049 | return CGF.Builder.CreateAnd( |
| 1050 | ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res"); |
| 1051 | } |
| 1052 | |
| 1053 | static llvm::Value *emitPPCLoadReserveIntrinsic(CodeGenFunction &CGF, |
| 1054 | unsigned BuiltinID, |
| 1055 | const CallExpr *E) { |
| 1056 | Value *Addr = CGF.EmitScalarExpr(E->getArg(0)); |
| 1057 | |
| 1058 | SmallString<64> Asm; |
| 1059 | raw_svector_ostream AsmOS(Asm); |
| 1060 | llvm::IntegerType *RetType = CGF.Int32Ty; |
| 1061 | |
| 1062 | switch (BuiltinID) { |
| 1063 | case clang::PPC::BI__builtin_ppc_ldarx: |
| 1064 | AsmOS << "ldarx "; |
| 1065 | RetType = CGF.Int64Ty; |
| 1066 | break; |
| 1067 | case clang::PPC::BI__builtin_ppc_lwarx: |
| 1068 | AsmOS << "lwarx "; |
| 1069 | RetType = CGF.Int32Ty; |
| 1070 | break; |
| 1071 | case clang::PPC::BI__builtin_ppc_lharx: |
| 1072 | AsmOS << "lharx "; |
| 1073 | RetType = CGF.Int16Ty; |
| 1074 | break; |
| 1075 | case clang::PPC::BI__builtin_ppc_lbarx: |
| 1076 | AsmOS << "lbarx "; |
| 1077 | RetType = CGF.Int8Ty; |
| 1078 | break; |
| 1079 | default: |
| 1080 | llvm_unreachable("Expected only PowerPC load reserve intrinsics")::llvm::llvm_unreachable_internal("Expected only PowerPC load reserve intrinsics" , "clang/lib/CodeGen/CGBuiltin.cpp", 1080); |
| 1081 | } |
| 1082 | |
| 1083 | AsmOS << "$0, ${1:y}"; |
| 1084 | |
| 1085 | std::string Constraints = "=r,*Z,~{memory}"; |
| 1086 | std::string MachineClobbers = CGF.getTarget().getClobbers(); |
| 1087 | if (!MachineClobbers.empty()) { |
| 1088 | Constraints += ','; |
| 1089 | Constraints += MachineClobbers; |
| 1090 | } |
| 1091 | |
| 1092 | llvm::Type *IntPtrType = RetType->getPointerTo(); |
| 1093 | llvm::FunctionType *FTy = |
| 1094 | llvm::FunctionType::get(RetType, {IntPtrType}, false); |
| 1095 | |
| 1096 | llvm::InlineAsm *IA = |
| 1097 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
| 1098 | llvm::CallInst *CI = CGF.Builder.CreateCall(IA, {Addr}); |
| 1099 | CI->addParamAttr( |
| 1100 | 0, Attribute::get(CGF.getLLVMContext(), Attribute::ElementType, RetType)); |
| 1101 | return CI; |
| 1102 | } |
| 1103 | |
| 1104 | namespace { |
| 1105 | enum class MSVCSetJmpKind { |
| 1106 | _setjmpex, |
| 1107 | _setjmp3, |
| 1108 | _setjmp |
| 1109 | }; |
| 1110 | } |
| 1111 | |
| 1112 | /// MSVC handles setjmp a bit differently on different platforms. On every |
| 1113 | /// architecture except 32-bit x86, the frame address is passed. On x86, extra |
| 1114 | /// parameters can be passed as variadic arguments, but we always pass none. |
| 1115 | static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind, |
| 1116 | const CallExpr *E) { |
| 1117 | llvm::Value *Arg1 = nullptr; |
| 1118 | llvm::Type *Arg1Ty = nullptr; |
| 1119 | StringRef Name; |
| 1120 | bool IsVarArg = false; |
| 1121 | if (SJKind == MSVCSetJmpKind::_setjmp3) { |
| 1122 | Name = "_setjmp3"; |
| 1123 | Arg1Ty = CGF.Int32Ty; |
| 1124 | Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0); |
| 1125 | IsVarArg = true; |
| 1126 | } else { |
| 1127 | Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex"; |
| 1128 | Arg1Ty = CGF.Int8PtrTy; |
| 1129 | if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) { |
| 1130 | Arg1 = CGF.Builder.CreateCall( |
| 1131 | CGF.CGM.getIntrinsic(Intrinsic::sponentry, CGF.AllocaInt8PtrTy)); |
| 1132 | } else |
| 1133 | Arg1 = CGF.Builder.CreateCall( |
| 1134 | CGF.CGM.getIntrinsic(Intrinsic::frameaddress, CGF.AllocaInt8PtrTy), |
| 1135 | llvm::ConstantInt::get(CGF.Int32Ty, 0)); |
| 1136 | } |
| 1137 | |
| 1138 | // Mark the call site and declaration with ReturnsTwice. |
| 1139 | llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty}; |
| 1140 | llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get( |
| 1141 | CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex, |
| 1142 | llvm::Attribute::ReturnsTwice); |
| 1143 | llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction( |
| 1144 | llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name, |
| 1145 | ReturnsTwiceAttr, /*Local=*/true); |
| 1146 | |
| 1147 | llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast( |
| 1148 | CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy); |
| 1149 | llvm::Value *Args[] = {Buf, Arg1}; |
| 1150 | llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args); |
| 1151 | CB->setAttributes(ReturnsTwiceAttr); |
| 1152 | return RValue::get(CB); |
| 1153 | } |
| 1154 | |
| 1155 | // Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code, |
| 1156 | // we handle them here. |
| 1157 | enum class CodeGenFunction::MSVCIntrin { |
| 1158 | _BitScanForward, |
| 1159 | _BitScanReverse, |
| 1160 | _InterlockedAnd, |
| 1161 | _InterlockedDecrement, |
| 1162 | _InterlockedExchange, |
| 1163 | _InterlockedExchangeAdd, |
| 1164 | _InterlockedExchangeSub, |
| 1165 | _InterlockedIncrement, |
| 1166 | _InterlockedOr, |
| 1167 | _InterlockedXor, |
| 1168 | _InterlockedExchangeAdd_acq, |
| 1169 | _InterlockedExchangeAdd_rel, |
| 1170 | _InterlockedExchangeAdd_nf, |
| 1171 | _InterlockedExchange_acq, |
| 1172 | _InterlockedExchange_rel, |
| 1173 | _InterlockedExchange_nf, |
| 1174 | _InterlockedCompareExchange_acq, |
| 1175 | _InterlockedCompareExchange_rel, |
| 1176 | _InterlockedCompareExchange_nf, |
| 1177 | _InterlockedCompareExchange128, |
| 1178 | _InterlockedCompareExchange128_acq, |
| 1179 | _InterlockedCompareExchange128_rel, |
| 1180 | _InterlockedCompareExchange128_nf, |
| 1181 | _InterlockedOr_acq, |
| 1182 | _InterlockedOr_rel, |
| 1183 | _InterlockedOr_nf, |
| 1184 | _InterlockedXor_acq, |
| 1185 | _InterlockedXor_rel, |
| 1186 | _InterlockedXor_nf, |
| 1187 | _InterlockedAnd_acq, |
| 1188 | _InterlockedAnd_rel, |
| 1189 | _InterlockedAnd_nf, |
| 1190 | _InterlockedIncrement_acq, |
| 1191 | _InterlockedIncrement_rel, |
| 1192 | _InterlockedIncrement_nf, |
| 1193 | _InterlockedDecrement_acq, |
| 1194 | _InterlockedDecrement_rel, |
| 1195 | _InterlockedDecrement_nf, |
| 1196 | __fastfail, |
| 1197 | }; |
| 1198 | |
| 1199 | static std::optional<CodeGenFunction::MSVCIntrin> |
| 1200 | translateArmToMsvcIntrin(unsigned BuiltinID) { |
| 1201 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
| 1202 | switch (BuiltinID) { |
| 1203 | default: |
| 1204 | return std::nullopt; |
| 1205 | case clang::ARM::BI_BitScanForward: |
| 1206 | case clang::ARM::BI_BitScanForward64: |
| 1207 | return MSVCIntrin::_BitScanForward; |
| 1208 | case clang::ARM::BI_BitScanReverse: |
| 1209 | case clang::ARM::BI_BitScanReverse64: |
| 1210 | return MSVCIntrin::_BitScanReverse; |
| 1211 | case clang::ARM::BI_InterlockedAnd64: |
| 1212 | return MSVCIntrin::_InterlockedAnd; |
| 1213 | case clang::ARM::BI_InterlockedExchange64: |
| 1214 | return MSVCIntrin::_InterlockedExchange; |
| 1215 | case clang::ARM::BI_InterlockedExchangeAdd64: |
| 1216 | return MSVCIntrin::_InterlockedExchangeAdd; |
| 1217 | case clang::ARM::BI_InterlockedExchangeSub64: |
| 1218 | return MSVCIntrin::_InterlockedExchangeSub; |
| 1219 | case clang::ARM::BI_InterlockedOr64: |
| 1220 | return MSVCIntrin::_InterlockedOr; |
| 1221 | case clang::ARM::BI_InterlockedXor64: |
| 1222 | return MSVCIntrin::_InterlockedXor; |
| 1223 | case clang::ARM::BI_InterlockedDecrement64: |
| 1224 | return MSVCIntrin::_InterlockedDecrement; |
| 1225 | case clang::ARM::BI_InterlockedIncrement64: |
| 1226 | return MSVCIntrin::_InterlockedIncrement; |
| 1227 | case clang::ARM::BI_InterlockedExchangeAdd8_acq: |
| 1228 | case clang::ARM::BI_InterlockedExchangeAdd16_acq: |
| 1229 | case clang::ARM::BI_InterlockedExchangeAdd_acq: |
| 1230 | case clang::ARM::BI_InterlockedExchangeAdd64_acq: |
| 1231 | return MSVCIntrin::_InterlockedExchangeAdd_acq; |
| 1232 | case clang::ARM::BI_InterlockedExchangeAdd8_rel: |
| 1233 | case clang::ARM::BI_InterlockedExchangeAdd16_rel: |
| 1234 | case clang::ARM::BI_InterlockedExchangeAdd_rel: |
| 1235 | case clang::ARM::BI_InterlockedExchangeAdd64_rel: |
| 1236 | return MSVCIntrin::_InterlockedExchangeAdd_rel; |
| 1237 | case clang::ARM::BI_InterlockedExchangeAdd8_nf: |
| 1238 | case clang::ARM::BI_InterlockedExchangeAdd16_nf: |
| 1239 | case clang::ARM::BI_InterlockedExchangeAdd_nf: |
| 1240 | case clang::ARM::BI_InterlockedExchangeAdd64_nf: |
| 1241 | return MSVCIntrin::_InterlockedExchangeAdd_nf; |
| 1242 | case clang::ARM::BI_InterlockedExchange8_acq: |
| 1243 | case clang::ARM::BI_InterlockedExchange16_acq: |
| 1244 | case clang::ARM::BI_InterlockedExchange_acq: |
| 1245 | case clang::ARM::BI_InterlockedExchange64_acq: |
| 1246 | return MSVCIntrin::_InterlockedExchange_acq; |
| 1247 | case clang::ARM::BI_InterlockedExchange8_rel: |
| 1248 | case clang::ARM::BI_InterlockedExchange16_rel: |
| 1249 | case clang::ARM::BI_InterlockedExchange_rel: |
| 1250 | case clang::ARM::BI_InterlockedExchange64_rel: |
| 1251 | return MSVCIntrin::_InterlockedExchange_rel; |
| 1252 | case clang::ARM::BI_InterlockedExchange8_nf: |
| 1253 | case clang::ARM::BI_InterlockedExchange16_nf: |
| 1254 | case clang::ARM::BI_InterlockedExchange_nf: |
| 1255 | case clang::ARM::BI_InterlockedExchange64_nf: |
| 1256 | return MSVCIntrin::_InterlockedExchange_nf; |
| 1257 | case clang::ARM::BI_InterlockedCompareExchange8_acq: |
| 1258 | case clang::ARM::BI_InterlockedCompareExchange16_acq: |
| 1259 | case clang::ARM::BI_InterlockedCompareExchange_acq: |
| 1260 | case clang::ARM::BI_InterlockedCompareExchange64_acq: |
| 1261 | return MSVCIntrin::_InterlockedCompareExchange_acq; |
| 1262 | case clang::ARM::BI_InterlockedCompareExchange8_rel: |
| 1263 | case clang::ARM::BI_InterlockedCompareExchange16_rel: |
| 1264 | case clang::ARM::BI_InterlockedCompareExchange_rel: |
| 1265 | case clang::ARM::BI_InterlockedCompareExchange64_rel: |
| 1266 | return MSVCIntrin::_InterlockedCompareExchange_rel; |
| 1267 | case clang::ARM::BI_InterlockedCompareExchange8_nf: |
| 1268 | case clang::ARM::BI_InterlockedCompareExchange16_nf: |
| 1269 | case clang::ARM::BI_InterlockedCompareExchange_nf: |
| 1270 | case clang::ARM::BI_InterlockedCompareExchange64_nf: |
| 1271 | return MSVCIntrin::_InterlockedCompareExchange_nf; |
| 1272 | case clang::ARM::BI_InterlockedOr8_acq: |
| 1273 | case clang::ARM::BI_InterlockedOr16_acq: |
| 1274 | case clang::ARM::BI_InterlockedOr_acq: |
| 1275 | case clang::ARM::BI_InterlockedOr64_acq: |
| 1276 | return MSVCIntrin::_InterlockedOr_acq; |
| 1277 | case clang::ARM::BI_InterlockedOr8_rel: |
| 1278 | case clang::ARM::BI_InterlockedOr16_rel: |
| 1279 | case clang::ARM::BI_InterlockedOr_rel: |
| 1280 | case clang::ARM::BI_InterlockedOr64_rel: |
| 1281 | return MSVCIntrin::_InterlockedOr_rel; |
| 1282 | case clang::ARM::BI_InterlockedOr8_nf: |
| 1283 | case clang::ARM::BI_InterlockedOr16_nf: |
| 1284 | case clang::ARM::BI_InterlockedOr_nf: |
| 1285 | case clang::ARM::BI_InterlockedOr64_nf: |
| 1286 | return MSVCIntrin::_InterlockedOr_nf; |
| 1287 | case clang::ARM::BI_InterlockedXor8_acq: |
| 1288 | case clang::ARM::BI_InterlockedXor16_acq: |
| 1289 | case clang::ARM::BI_InterlockedXor_acq: |
| 1290 | case clang::ARM::BI_InterlockedXor64_acq: |
| 1291 | return MSVCIntrin::_InterlockedXor_acq; |
| 1292 | case clang::ARM::BI_InterlockedXor8_rel: |
| 1293 | case clang::ARM::BI_InterlockedXor16_rel: |
| 1294 | case clang::ARM::BI_InterlockedXor_rel: |
| 1295 | case clang::ARM::BI_InterlockedXor64_rel: |
| 1296 | return MSVCIntrin::_InterlockedXor_rel; |
| 1297 | case clang::ARM::BI_InterlockedXor8_nf: |
| 1298 | case clang::ARM::BI_InterlockedXor16_nf: |
| 1299 | case clang::ARM::BI_InterlockedXor_nf: |
| 1300 | case clang::ARM::BI_InterlockedXor64_nf: |
| 1301 | return MSVCIntrin::_InterlockedXor_nf; |
| 1302 | case clang::ARM::BI_InterlockedAnd8_acq: |
| 1303 | case clang::ARM::BI_InterlockedAnd16_acq: |
| 1304 | case clang::ARM::BI_InterlockedAnd_acq: |
| 1305 | case clang::ARM::BI_InterlockedAnd64_acq: |
| 1306 | return MSVCIntrin::_InterlockedAnd_acq; |
| 1307 | case clang::ARM::BI_InterlockedAnd8_rel: |
| 1308 | case clang::ARM::BI_InterlockedAnd16_rel: |
| 1309 | case clang::ARM::BI_InterlockedAnd_rel: |
| 1310 | case clang::ARM::BI_InterlockedAnd64_rel: |
| 1311 | return MSVCIntrin::_InterlockedAnd_rel; |
| 1312 | case clang::ARM::BI_InterlockedAnd8_nf: |
| 1313 | case clang::ARM::BI_InterlockedAnd16_nf: |
| 1314 | case clang::ARM::BI_InterlockedAnd_nf: |
| 1315 | case clang::ARM::BI_InterlockedAnd64_nf: |
| 1316 | return MSVCIntrin::_InterlockedAnd_nf; |
| 1317 | case clang::ARM::BI_InterlockedIncrement16_acq: |
| 1318 | case clang::ARM::BI_InterlockedIncrement_acq: |
| 1319 | case clang::ARM::BI_InterlockedIncrement64_acq: |
| 1320 | return MSVCIntrin::_InterlockedIncrement_acq; |
| 1321 | case clang::ARM::BI_InterlockedIncrement16_rel: |
| 1322 | case clang::ARM::BI_InterlockedIncrement_rel: |
| 1323 | case clang::ARM::BI_InterlockedIncrement64_rel: |
| 1324 | return MSVCIntrin::_InterlockedIncrement_rel; |
| 1325 | case clang::ARM::BI_InterlockedIncrement16_nf: |
| 1326 | case clang::ARM::BI_InterlockedIncrement_nf: |
| 1327 | case clang::ARM::BI_InterlockedIncrement64_nf: |
| 1328 | return MSVCIntrin::_InterlockedIncrement_nf; |
| 1329 | case clang::ARM::BI_InterlockedDecrement16_acq: |
| 1330 | case clang::ARM::BI_InterlockedDecrement_acq: |
| 1331 | case clang::ARM::BI_InterlockedDecrement64_acq: |
| 1332 | return MSVCIntrin::_InterlockedDecrement_acq; |
| 1333 | case clang::ARM::BI_InterlockedDecrement16_rel: |
| 1334 | case clang::ARM::BI_InterlockedDecrement_rel: |
| 1335 | case clang::ARM::BI_InterlockedDecrement64_rel: |
| 1336 | return MSVCIntrin::_InterlockedDecrement_rel; |
| 1337 | case clang::ARM::BI_InterlockedDecrement16_nf: |
| 1338 | case clang::ARM::BI_InterlockedDecrement_nf: |
| 1339 | case clang::ARM::BI_InterlockedDecrement64_nf: |
| 1340 | return MSVCIntrin::_InterlockedDecrement_nf; |
| 1341 | } |
| 1342 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "clang/lib/CodeGen/CGBuiltin.cpp" , 1342); |
| 1343 | } |
| 1344 | |
| 1345 | static std::optional<CodeGenFunction::MSVCIntrin> |
| 1346 | translateAarch64ToMsvcIntrin(unsigned BuiltinID) { |
| 1347 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
| 1348 | switch (BuiltinID) { |
| 1349 | default: |
| 1350 | return std::nullopt; |
| 1351 | case clang::AArch64::BI_BitScanForward: |
| 1352 | case clang::AArch64::BI_BitScanForward64: |
| 1353 | return MSVCIntrin::_BitScanForward; |
| 1354 | case clang::AArch64::BI_BitScanReverse: |
| 1355 | case clang::AArch64::BI_BitScanReverse64: |
| 1356 | return MSVCIntrin::_BitScanReverse; |
| 1357 | case clang::AArch64::BI_InterlockedAnd64: |
| 1358 | return MSVCIntrin::_InterlockedAnd; |
| 1359 | case clang::AArch64::BI_InterlockedExchange64: |
| 1360 | return MSVCIntrin::_InterlockedExchange; |
| 1361 | case clang::AArch64::BI_InterlockedExchangeAdd64: |
| 1362 | return MSVCIntrin::_InterlockedExchangeAdd; |
| 1363 | case clang::AArch64::BI_InterlockedExchangeSub64: |
| 1364 | return MSVCIntrin::_InterlockedExchangeSub; |
| 1365 | case clang::AArch64::BI_InterlockedOr64: |
| 1366 | return MSVCIntrin::_InterlockedOr; |
| 1367 | case clang::AArch64::BI_InterlockedXor64: |
| 1368 | return MSVCIntrin::_InterlockedXor; |
| 1369 | case clang::AArch64::BI_InterlockedDecrement64: |
| 1370 | return MSVCIntrin::_InterlockedDecrement; |
| 1371 | case clang::AArch64::BI_InterlockedIncrement64: |
| 1372 | return MSVCIntrin::_InterlockedIncrement; |
| 1373 | case clang::AArch64::BI_InterlockedExchangeAdd8_acq: |
| 1374 | case clang::AArch64::BI_InterlockedExchangeAdd16_acq: |
| 1375 | case clang::AArch64::BI_InterlockedExchangeAdd_acq: |
| 1376 | case clang::AArch64::BI_InterlockedExchangeAdd64_acq: |
| 1377 | return MSVCIntrin::_InterlockedExchangeAdd_acq; |
| 1378 | case clang::AArch64::BI_InterlockedExchangeAdd8_rel: |
| 1379 | case clang::AArch64::BI_InterlockedExchangeAdd16_rel: |
| 1380 | case clang::AArch64::BI_InterlockedExchangeAdd_rel: |
| 1381 | case clang::AArch64::BI_InterlockedExchangeAdd64_rel: |
| 1382 | return MSVCIntrin::_InterlockedExchangeAdd_rel; |
| 1383 | case clang::AArch64::BI_InterlockedExchangeAdd8_nf: |
| 1384 | case clang::AArch64::BI_InterlockedExchangeAdd16_nf: |
| 1385 | case clang::AArch64::BI_InterlockedExchangeAdd_nf: |
| 1386 | case clang::AArch64::BI_InterlockedExchangeAdd64_nf: |
| 1387 | return MSVCIntrin::_InterlockedExchangeAdd_nf; |
| 1388 | case clang::AArch64::BI_InterlockedExchange8_acq: |
| 1389 | case clang::AArch64::BI_InterlockedExchange16_acq: |
| 1390 | case clang::AArch64::BI_InterlockedExchange_acq: |
| 1391 | case clang::AArch64::BI_InterlockedExchange64_acq: |
| 1392 | return MSVCIntrin::_InterlockedExchange_acq; |
| 1393 | case clang::AArch64::BI_InterlockedExchange8_rel: |
| 1394 | case clang::AArch64::BI_InterlockedExchange16_rel: |
| 1395 | case clang::AArch64::BI_InterlockedExchange_rel: |
| 1396 | case clang::AArch64::BI_InterlockedExchange64_rel: |
| 1397 | return MSVCIntrin::_InterlockedExchange_rel; |
| 1398 | case clang::AArch64::BI_InterlockedExchange8_nf: |
| 1399 | case clang::AArch64::BI_InterlockedExchange16_nf: |
| 1400 | case clang::AArch64::BI_InterlockedExchange_nf: |
| 1401 | case clang::AArch64::BI_InterlockedExchange64_nf: |
| 1402 | return MSVCIntrin::_InterlockedExchange_nf; |
| 1403 | case clang::AArch64::BI_InterlockedCompareExchange8_acq: |
| 1404 | case clang::AArch64::BI_InterlockedCompareExchange16_acq: |
| 1405 | case clang::AArch64::BI_InterlockedCompareExchange_acq: |
| 1406 | case clang::AArch64::BI_InterlockedCompareExchange64_acq: |
| 1407 | return MSVCIntrin::_InterlockedCompareExchange_acq; |
| 1408 | case clang::AArch64::BI_InterlockedCompareExchange8_rel: |
| 1409 | case clang::AArch64::BI_InterlockedCompareExchange16_rel: |
| 1410 | case clang::AArch64::BI_InterlockedCompareExchange_rel: |
| 1411 | case clang::AArch64::BI_InterlockedCompareExchange64_rel: |
| 1412 | return MSVCIntrin::_InterlockedCompareExchange_rel; |
| 1413 | case clang::AArch64::BI_InterlockedCompareExchange8_nf: |
| 1414 | case clang::AArch64::BI_InterlockedCompareExchange16_nf: |
| 1415 | case clang::AArch64::BI_InterlockedCompareExchange_nf: |
| 1416 | case clang::AArch64::BI_InterlockedCompareExchange64_nf: |
| 1417 | return MSVCIntrin::_InterlockedCompareExchange_nf; |
| 1418 | case clang::AArch64::BI_InterlockedCompareExchange128: |
| 1419 | return MSVCIntrin::_InterlockedCompareExchange128; |
| 1420 | case clang::AArch64::BI_InterlockedCompareExchange128_acq: |
| 1421 | return MSVCIntrin::_InterlockedCompareExchange128_acq; |
| 1422 | case clang::AArch64::BI_InterlockedCompareExchange128_nf: |
| 1423 | return MSVCIntrin::_InterlockedCompareExchange128_nf; |
| 1424 | case clang::AArch64::BI_InterlockedCompareExchange128_rel: |
| 1425 | return MSVCIntrin::_InterlockedCompareExchange128_rel; |
| 1426 | case clang::AArch64::BI_InterlockedOr8_acq: |
| 1427 | case clang::AArch64::BI_InterlockedOr16_acq: |
| 1428 | case clang::AArch64::BI_InterlockedOr_acq: |
| 1429 | case clang::AArch64::BI_InterlockedOr64_acq: |
| 1430 | return MSVCIntrin::_InterlockedOr_acq; |
| 1431 | case clang::AArch64::BI_InterlockedOr8_rel: |
| 1432 | case clang::AArch64::BI_InterlockedOr16_rel: |
| 1433 | case clang::AArch64::BI_InterlockedOr_rel: |
| 1434 | case clang::AArch64::BI_InterlockedOr64_rel: |
| 1435 | return MSVCIntrin::_InterlockedOr_rel; |
| 1436 | case clang::AArch64::BI_InterlockedOr8_nf: |
| 1437 | case clang::AArch64::BI_InterlockedOr16_nf: |
| 1438 | case clang::AArch64::BI_InterlockedOr_nf: |
| 1439 | case clang::AArch64::BI_InterlockedOr64_nf: |
| 1440 | return MSVCIntrin::_InterlockedOr_nf; |
| 1441 | case clang::AArch64::BI_InterlockedXor8_acq: |
| 1442 | case clang::AArch64::BI_InterlockedXor16_acq: |
| 1443 | case clang::AArch64::BI_InterlockedXor_acq: |
| 1444 | case clang::AArch64::BI_InterlockedXor64_acq: |
| 1445 | return MSVCIntrin::_InterlockedXor_acq; |
| 1446 | case clang::AArch64::BI_InterlockedXor8_rel: |
| 1447 | case clang::AArch64::BI_InterlockedXor16_rel: |
| 1448 | case clang::AArch64::BI_InterlockedXor_rel: |
| 1449 | case clang::AArch64::BI_InterlockedXor64_rel: |
| 1450 | return MSVCIntrin::_InterlockedXor_rel; |
| 1451 | case clang::AArch64::BI_InterlockedXor8_nf: |
| 1452 | case clang::AArch64::BI_InterlockedXor16_nf: |
| 1453 | case clang::AArch64::BI_InterlockedXor_nf: |
| 1454 | case clang::AArch64::BI_InterlockedXor64_nf: |
| 1455 | return MSVCIntrin::_InterlockedXor_nf; |
| 1456 | case clang::AArch64::BI_InterlockedAnd8_acq: |
| 1457 | case clang::AArch64::BI_InterlockedAnd16_acq: |
| 1458 | case clang::AArch64::BI_InterlockedAnd_acq: |
| 1459 | case clang::AArch64::BI_InterlockedAnd64_acq: |
| 1460 | return MSVCIntrin::_InterlockedAnd_acq; |
| 1461 | case clang::AArch64::BI_InterlockedAnd8_rel: |
| 1462 | case clang::AArch64::BI_InterlockedAnd16_rel: |
| 1463 | case clang::AArch64::BI_InterlockedAnd_rel: |
| 1464 | case clang::AArch64::BI_InterlockedAnd64_rel: |
| 1465 | return MSVCIntrin::_InterlockedAnd_rel; |
| 1466 | case clang::AArch64::BI_InterlockedAnd8_nf: |
| 1467 | case clang::AArch64::BI_InterlockedAnd16_nf: |
| 1468 | case clang::AArch64::BI_InterlockedAnd_nf: |
| 1469 | case clang::AArch64::BI_InterlockedAnd64_nf: |
| 1470 | return MSVCIntrin::_InterlockedAnd_nf; |
| 1471 | case clang::AArch64::BI_InterlockedIncrement16_acq: |
| 1472 | case clang::AArch64::BI_InterlockedIncrement_acq: |
| 1473 | case clang::AArch64::BI_InterlockedIncrement64_acq: |
| 1474 | return MSVCIntrin::_InterlockedIncrement_acq; |
| 1475 | case clang::AArch64::BI_InterlockedIncrement16_rel: |
| 1476 | case clang::AArch64::BI_InterlockedIncrement_rel: |
| 1477 | case clang::AArch64::BI_InterlockedIncrement64_rel: |
| 1478 | return MSVCIntrin::_InterlockedIncrement_rel; |
| 1479 | case clang::AArch64::BI_InterlockedIncrement16_nf: |
| 1480 | case clang::AArch64::BI_InterlockedIncrement_nf: |
| 1481 | case clang::AArch64::BI_InterlockedIncrement64_nf: |
| 1482 | return MSVCIntrin::_InterlockedIncrement_nf; |
| 1483 | case clang::AArch64::BI_InterlockedDecrement16_acq: |
| 1484 | case clang::AArch64::BI_InterlockedDecrement_acq: |
| 1485 | case clang::AArch64::BI_InterlockedDecrement64_acq: |
| 1486 | return MSVCIntrin::_InterlockedDecrement_acq; |
| 1487 | case clang::AArch64::BI_InterlockedDecrement16_rel: |
| 1488 | case clang::AArch64::BI_InterlockedDecrement_rel: |
| 1489 | case clang::AArch64::BI_InterlockedDecrement64_rel: |
| 1490 | return MSVCIntrin::_InterlockedDecrement_rel; |
| 1491 | case clang::AArch64::BI_InterlockedDecrement16_nf: |
| 1492 | case clang::AArch64::BI_InterlockedDecrement_nf: |
| 1493 | case clang::AArch64::BI_InterlockedDecrement64_nf: |
| 1494 | return MSVCIntrin::_InterlockedDecrement_nf; |
| 1495 | } |
| 1496 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "clang/lib/CodeGen/CGBuiltin.cpp" , 1496); |
| 1497 | } |
| 1498 | |
| 1499 | static std::optional<CodeGenFunction::MSVCIntrin> |
| 1500 | translateX86ToMsvcIntrin(unsigned BuiltinID) { |
| 1501 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
| 1502 | switch (BuiltinID) { |
| 1503 | default: |
| 1504 | return std::nullopt; |
| 1505 | case clang::X86::BI_BitScanForward: |
| 1506 | case clang::X86::BI_BitScanForward64: |
| 1507 | return MSVCIntrin::_BitScanForward; |
| 1508 | case clang::X86::BI_BitScanReverse: |
| 1509 | case clang::X86::BI_BitScanReverse64: |
| 1510 | return MSVCIntrin::_BitScanReverse; |
| 1511 | case clang::X86::BI_InterlockedAnd64: |
| 1512 | return MSVCIntrin::_InterlockedAnd; |
| 1513 | case clang::X86::BI_InterlockedCompareExchange128: |
| 1514 | return MSVCIntrin::_InterlockedCompareExchange128; |
| 1515 | case clang::X86::BI_InterlockedExchange64: |
| 1516 | return MSVCIntrin::_InterlockedExchange; |
| 1517 | case clang::X86::BI_InterlockedExchangeAdd64: |
| 1518 | return MSVCIntrin::_InterlockedExchangeAdd; |
| 1519 | case clang::X86::BI_InterlockedExchangeSub64: |
| 1520 | return MSVCIntrin::_InterlockedExchangeSub; |
| 1521 | case clang::X86::BI_InterlockedOr64: |
| 1522 | return MSVCIntrin::_InterlockedOr; |
| 1523 | case clang::X86::BI_InterlockedXor64: |
| 1524 | return MSVCIntrin::_InterlockedXor; |
| 1525 | case clang::X86::BI_InterlockedDecrement64: |
| 1526 | return MSVCIntrin::_InterlockedDecrement; |
| 1527 | case clang::X86::BI_InterlockedIncrement64: |
| 1528 | return MSVCIntrin::_InterlockedIncrement; |
| 1529 | } |
| 1530 | llvm_unreachable("must return from switch")::llvm::llvm_unreachable_internal("must return from switch", "clang/lib/CodeGen/CGBuiltin.cpp" , 1530); |
| 1531 | } |
| 1532 | |
| 1533 | // Emit an MSVC intrinsic. Assumes that arguments have *not* been evaluated. |
| 1534 | Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID, |
| 1535 | const CallExpr *E) { |
| 1536 | switch (BuiltinID) { |
| 1537 | case MSVCIntrin::_BitScanForward: |
| 1538 | case MSVCIntrin::_BitScanReverse: { |
| 1539 | Address IndexAddress(EmitPointerWithAlignment(E->getArg(0))); |
| 1540 | Value *ArgValue = EmitScalarExpr(E->getArg(1)); |
| 1541 | |
| 1542 | llvm::Type *ArgType = ArgValue->getType(); |
| 1543 | llvm::Type *IndexType = IndexAddress.getElementType(); |
| 1544 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1545 | |
| 1546 | Value *ArgZero = llvm::Constant::getNullValue(ArgType); |
| 1547 | Value *ResZero = llvm::Constant::getNullValue(ResultType); |
| 1548 | Value *ResOne = llvm::ConstantInt::get(ResultType, 1); |
| 1549 | |
| 1550 | BasicBlock *Begin = Builder.GetInsertBlock(); |
| 1551 | BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn); |
| 1552 | Builder.SetInsertPoint(End); |
| 1553 | PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result"); |
| 1554 | |
| 1555 | Builder.SetInsertPoint(Begin); |
| 1556 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero); |
| 1557 | BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn); |
| 1558 | Builder.CreateCondBr(IsZero, End, NotZero); |
| 1559 | Result->addIncoming(ResZero, Begin); |
| 1560 | |
| 1561 | Builder.SetInsertPoint(NotZero); |
| 1562 | |
| 1563 | if (BuiltinID == MSVCIntrin::_BitScanForward) { |
| 1564 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 1565 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
| 1566 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
| 1567 | Builder.CreateStore(ZeroCount, IndexAddress, false); |
| 1568 | } else { |
| 1569 | unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth(); |
| 1570 | Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1); |
| 1571 | |
| 1572 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 1573 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
| 1574 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
| 1575 | Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount); |
| 1576 | Builder.CreateStore(Index, IndexAddress, false); |
| 1577 | } |
| 1578 | Builder.CreateBr(End); |
| 1579 | Result->addIncoming(ResOne, NotZero); |
| 1580 | |
| 1581 | Builder.SetInsertPoint(End); |
| 1582 | return Result; |
| 1583 | } |
| 1584 | case MSVCIntrin::_InterlockedAnd: |
| 1585 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E); |
| 1586 | case MSVCIntrin::_InterlockedExchange: |
| 1587 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E); |
| 1588 | case MSVCIntrin::_InterlockedExchangeAdd: |
| 1589 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E); |
| 1590 | case MSVCIntrin::_InterlockedExchangeSub: |
| 1591 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E); |
| 1592 | case MSVCIntrin::_InterlockedOr: |
| 1593 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E); |
| 1594 | case MSVCIntrin::_InterlockedXor: |
| 1595 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E); |
| 1596 | case MSVCIntrin::_InterlockedExchangeAdd_acq: |
| 1597 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 1598 | AtomicOrdering::Acquire); |
| 1599 | case MSVCIntrin::_InterlockedExchangeAdd_rel: |
| 1600 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 1601 | AtomicOrdering::Release); |
| 1602 | case MSVCIntrin::_InterlockedExchangeAdd_nf: |
| 1603 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 1604 | AtomicOrdering::Monotonic); |
| 1605 | case MSVCIntrin::_InterlockedExchange_acq: |
| 1606 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 1607 | AtomicOrdering::Acquire); |
| 1608 | case MSVCIntrin::_InterlockedExchange_rel: |
| 1609 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 1610 | AtomicOrdering::Release); |
| 1611 | case MSVCIntrin::_InterlockedExchange_nf: |
| 1612 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 1613 | AtomicOrdering::Monotonic); |
| 1614 | case MSVCIntrin::_InterlockedCompareExchange_acq: |
| 1615 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire); |
| 1616 | case MSVCIntrin::_InterlockedCompareExchange_rel: |
| 1617 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release); |
| 1618 | case MSVCIntrin::_InterlockedCompareExchange_nf: |
| 1619 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic); |
| 1620 | case MSVCIntrin::_InterlockedCompareExchange128: |
| 1621 | return EmitAtomicCmpXchg128ForMSIntrin( |
| 1622 | *this, E, AtomicOrdering::SequentiallyConsistent); |
| 1623 | case MSVCIntrin::_InterlockedCompareExchange128_acq: |
| 1624 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Acquire); |
| 1625 | case MSVCIntrin::_InterlockedCompareExchange128_rel: |
| 1626 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Release); |
| 1627 | case MSVCIntrin::_InterlockedCompareExchange128_nf: |
| 1628 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Monotonic); |
| 1629 | case MSVCIntrin::_InterlockedOr_acq: |
| 1630 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1631 | AtomicOrdering::Acquire); |
| 1632 | case MSVCIntrin::_InterlockedOr_rel: |
| 1633 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1634 | AtomicOrdering::Release); |
| 1635 | case MSVCIntrin::_InterlockedOr_nf: |
| 1636 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1637 | AtomicOrdering::Monotonic); |
| 1638 | case MSVCIntrin::_InterlockedXor_acq: |
| 1639 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1640 | AtomicOrdering::Acquire); |
| 1641 | case MSVCIntrin::_InterlockedXor_rel: |
| 1642 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1643 | AtomicOrdering::Release); |
| 1644 | case MSVCIntrin::_InterlockedXor_nf: |
| 1645 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1646 | AtomicOrdering::Monotonic); |
| 1647 | case MSVCIntrin::_InterlockedAnd_acq: |
| 1648 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1649 | AtomicOrdering::Acquire); |
| 1650 | case MSVCIntrin::_InterlockedAnd_rel: |
| 1651 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1652 | AtomicOrdering::Release); |
| 1653 | case MSVCIntrin::_InterlockedAnd_nf: |
| 1654 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1655 | AtomicOrdering::Monotonic); |
| 1656 | case MSVCIntrin::_InterlockedIncrement_acq: |
| 1657 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire); |
| 1658 | case MSVCIntrin::_InterlockedIncrement_rel: |
| 1659 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release); |
| 1660 | case MSVCIntrin::_InterlockedIncrement_nf: |
| 1661 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic); |
| 1662 | case MSVCIntrin::_InterlockedDecrement_acq: |
| 1663 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire); |
| 1664 | case MSVCIntrin::_InterlockedDecrement_rel: |
| 1665 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release); |
| 1666 | case MSVCIntrin::_InterlockedDecrement_nf: |
| 1667 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic); |
| 1668 | |
| 1669 | case MSVCIntrin::_InterlockedDecrement: |
| 1670 | return EmitAtomicDecrementValue(*this, E); |
| 1671 | case MSVCIntrin::_InterlockedIncrement: |
| 1672 | return EmitAtomicIncrementValue(*this, E); |
| 1673 | |
| 1674 | case MSVCIntrin::__fastfail: { |
| 1675 | // Request immediate process termination from the kernel. The instruction |
| 1676 | // sequences to do this are documented on MSDN: |
| 1677 | // https://msdn.microsoft.com/en-us/library/dn774154.aspx |
| 1678 | llvm::Triple::ArchType ISA = getTarget().getTriple().getArch(); |
| 1679 | StringRef Asm, Constraints; |
| 1680 | switch (ISA) { |
| 1681 | default: |
| 1682 | ErrorUnsupported(E, "__fastfail call for this architecture"); |
| 1683 | break; |
| 1684 | case llvm::Triple::x86: |
| 1685 | case llvm::Triple::x86_64: |
| 1686 | Asm = "int $$0x29"; |
| 1687 | Constraints = "{cx}"; |
| 1688 | break; |
| 1689 | case llvm::Triple::thumb: |
| 1690 | Asm = "udf #251"; |
| 1691 | Constraints = "{r0}"; |
| 1692 | break; |
| 1693 | case llvm::Triple::aarch64: |
| 1694 | Asm = "brk #0xF003"; |
| 1695 | Constraints = "{w0}"; |
| 1696 | } |
| 1697 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false); |
| 1698 | llvm::InlineAsm *IA = |
| 1699 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
| 1700 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( |
| 1701 | getLLVMContext(), llvm::AttributeList::FunctionIndex, |
| 1702 | llvm::Attribute::NoReturn); |
| 1703 | llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0))); |
| 1704 | CI->setAttributes(NoReturnAttr); |
| 1705 | return CI; |
| 1706 | } |
| 1707 | } |
| 1708 | llvm_unreachable("Incorrect MSVC intrinsic!")::llvm::llvm_unreachable_internal("Incorrect MSVC intrinsic!" , "clang/lib/CodeGen/CGBuiltin.cpp", 1708); |
| 1709 | } |
| 1710 | |
| 1711 | namespace { |
| 1712 | // ARC cleanup for __builtin_os_log_format |
| 1713 | struct CallObjCArcUse final : EHScopeStack::Cleanup { |
| 1714 | CallObjCArcUse(llvm::Value *object) : object(object) {} |
| 1715 | llvm::Value *object; |
| 1716 | |
| 1717 | void Emit(CodeGenFunction &CGF, Flags flags) override { |
| 1718 | CGF.EmitARCIntrinsicUse(object); |
| 1719 | } |
| 1720 | }; |
| 1721 | } |
| 1722 | |
| 1723 | Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E, |
| 1724 | BuiltinCheckKind Kind) { |
| 1725 | assert((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero)(static_cast <bool> ((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind" ) ? void (0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1726, __extension__ __PRETTY_FUNCTION__ )) |
| 1726 | && "Unsupported builtin check kind")(static_cast <bool> ((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && "Unsupported builtin check kind" ) ? void (0) : __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1726, __extension__ __PRETTY_FUNCTION__ )); |
| 1727 | |
| 1728 | Value *ArgValue = EmitScalarExpr(E); |
| 1729 | if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef()) |
| 1730 | return ArgValue; |
| 1731 | |
| 1732 | SanitizerScope SanScope(this); |
| 1733 | Value *Cond = Builder.CreateICmpNE( |
| 1734 | ArgValue, llvm::Constant::getNullValue(ArgValue->getType())); |
| 1735 | EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin), |
| 1736 | SanitizerHandler::InvalidBuiltin, |
| 1737 | {EmitCheckSourceLocation(E->getExprLoc()), |
| 1738 | llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)}, |
| 1739 | std::nullopt); |
| 1740 | return ArgValue; |
| 1741 | } |
| 1742 | |
| 1743 | /// Get the argument type for arguments to os_log_helper. |
| 1744 | static CanQualType getOSLogArgType(ASTContext &C, int Size) { |
| 1745 | QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false); |
| 1746 | return C.getCanonicalType(UnsignedTy); |
| 1747 | } |
| 1748 | |
| 1749 | llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction( |
| 1750 | const analyze_os_log::OSLogBufferLayout &Layout, |
| 1751 | CharUnits BufferAlignment) { |
| 1752 | ASTContext &Ctx = getContext(); |
| 1753 | |
| 1754 | llvm::SmallString<64> Name; |
| 1755 | { |
| 1756 | raw_svector_ostream OS(Name); |
| 1757 | OS << "__os_log_helper"; |
| 1758 | OS << "_" << BufferAlignment.getQuantity(); |
| 1759 | OS << "_" << int(Layout.getSummaryByte()); |
| 1760 | OS << "_" << int(Layout.getNumArgsByte()); |
| 1761 | for (const auto &Item : Layout.Items) |
| 1762 | OS << "_" << int(Item.getSizeByte()) << "_" |
| 1763 | << int(Item.getDescriptorByte()); |
| 1764 | } |
| 1765 | |
| 1766 | if (llvm::Function *F = CGM.getModule().getFunction(Name)) |
| 1767 | return F; |
| 1768 | |
| 1769 | llvm::SmallVector<QualType, 4> ArgTys; |
| 1770 | FunctionArgList Args; |
| 1771 | Args.push_back(ImplicitParamDecl::Create( |
| 1772 | Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"), Ctx.VoidPtrTy, |
| 1773 | ImplicitParamDecl::Other)); |
| 1774 | ArgTys.emplace_back(Ctx.VoidPtrTy); |
| 1775 | |
| 1776 | for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) { |
| 1777 | char Size = Layout.Items[I].getSizeByte(); |
| 1778 | if (!Size) |
| 1779 | continue; |
| 1780 | |
| 1781 | QualType ArgTy = getOSLogArgType(Ctx, Size); |
| 1782 | Args.push_back(ImplicitParamDecl::Create( |
| 1783 | Ctx, nullptr, SourceLocation(), |
| 1784 | &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy, |
| 1785 | ImplicitParamDecl::Other)); |
| 1786 | ArgTys.emplace_back(ArgTy); |
| 1787 | } |
| 1788 | |
| 1789 | QualType ReturnTy = Ctx.VoidTy; |
| 1790 | |
| 1791 | // The helper function has linkonce_odr linkage to enable the linker to merge |
| 1792 | // identical functions. To ensure the merging always happens, 'noinline' is |
| 1793 | // attached to the function when compiling with -Oz. |
| 1794 | const CGFunctionInfo &FI = |
| 1795 | CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args); |
| 1796 | llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI); |
| 1797 | llvm::Function *Fn = llvm::Function::Create( |
| 1798 | FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule()); |
| 1799 | Fn->setVisibility(llvm::GlobalValue::HiddenVisibility); |
| 1800 | CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn, /*IsThunk=*/false); |
| 1801 | CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn); |
| 1802 | Fn->setDoesNotThrow(); |
| 1803 | |
| 1804 | // Attach 'noinline' at -Oz. |
| 1805 | if (CGM.getCodeGenOpts().OptimizeSize == 2) |
| 1806 | Fn->addFnAttr(llvm::Attribute::NoInline); |
| 1807 | |
| 1808 | auto NL = ApplyDebugLocation::CreateEmpty(*this); |
| 1809 | StartFunction(GlobalDecl(), ReturnTy, Fn, FI, Args); |
| 1810 | |
| 1811 | // Create a scope with an artificial location for the body of this function. |
| 1812 | auto AL = ApplyDebugLocation::CreateArtificial(*this); |
| 1813 | |
| 1814 | CharUnits Offset; |
| 1815 | Address BufAddr = |
| 1816 | Address(Builder.CreateLoad(GetAddrOfLocalVar(Args[0]), "buf"), Int8Ty, |
| 1817 | BufferAlignment); |
| 1818 | Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()), |
| 1819 | Builder.CreateConstByteGEP(BufAddr, Offset++, "summary")); |
| 1820 | Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()), |
| 1821 | Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs")); |
| 1822 | |
| 1823 | unsigned I = 1; |
| 1824 | for (const auto &Item : Layout.Items) { |
| 1825 | Builder.CreateStore( |
| 1826 | Builder.getInt8(Item.getDescriptorByte()), |
| 1827 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor")); |
| 1828 | Builder.CreateStore( |
| 1829 | Builder.getInt8(Item.getSizeByte()), |
| 1830 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize")); |
| 1831 | |
| 1832 | CharUnits Size = Item.size(); |
| 1833 | if (!Size.getQuantity()) |
| 1834 | continue; |
| 1835 | |
| 1836 | Address Arg = GetAddrOfLocalVar(Args[I]); |
| 1837 | Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData"); |
| 1838 | Addr = |
| 1839 | Builder.CreateElementBitCast(Addr, Arg.getElementType(), "argDataCast"); |
| 1840 | Builder.CreateStore(Builder.CreateLoad(Arg), Addr); |
| 1841 | Offset += Size; |
| 1842 | ++I; |
| 1843 | } |
| 1844 | |
| 1845 | FinishFunction(); |
| 1846 | |
| 1847 | return Fn; |
| 1848 | } |
| 1849 | |
| 1850 | RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) { |
| 1851 | assert(E.getNumArgs() >= 2 &&(static_cast <bool> (E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? void (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1852, __extension__ __PRETTY_FUNCTION__ )) |
| 1852 | "__builtin_os_log_format takes at least 2 arguments")(static_cast <bool> (E.getNumArgs() >= 2 && "__builtin_os_log_format takes at least 2 arguments" ) ? void (0) : __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1852, __extension__ __PRETTY_FUNCTION__ )); |
| 1853 | ASTContext &Ctx = getContext(); |
| 1854 | analyze_os_log::OSLogBufferLayout Layout; |
| 1855 | analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout); |
| 1856 | Address BufAddr = EmitPointerWithAlignment(E.getArg(0)); |
| 1857 | llvm::SmallVector<llvm::Value *, 4> RetainableOperands; |
| 1858 | |
| 1859 | // Ignore argument 1, the format string. It is not currently used. |
| 1860 | CallArgList Args; |
| 1861 | Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy); |
| 1862 | |
| 1863 | for (const auto &Item : Layout.Items) { |
| 1864 | int Size = Item.getSizeByte(); |
| 1865 | if (!Size) |
| 1866 | continue; |
| 1867 | |
| 1868 | llvm::Value *ArgVal; |
| 1869 | |
| 1870 | if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) { |
| 1871 | uint64_t Val = 0; |
| 1872 | for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I) |
| 1873 | Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8; |
| 1874 | ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val)); |
| 1875 | } else if (const Expr *TheExpr = Item.getExpr()) { |
| 1876 | ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false); |
| 1877 | |
| 1878 | // If a temporary object that requires destruction after the full |
| 1879 | // expression is passed, push a lifetime-extended cleanup to extend its |
| 1880 | // lifetime to the end of the enclosing block scope. |
| 1881 | auto LifetimeExtendObject = [&](const Expr *E) { |
| 1882 | E = E->IgnoreParenCasts(); |
| 1883 | // Extend lifetimes of objects returned by function calls and message |
| 1884 | // sends. |
| 1885 | |
| 1886 | // FIXME: We should do this in other cases in which temporaries are |
| 1887 | // created including arguments of non-ARC types (e.g., C++ |
| 1888 | // temporaries). |
| 1889 | if (isa<CallExpr>(E) || isa<ObjCMessageExpr>(E)) |
| 1890 | return true; |
| 1891 | return false; |
| 1892 | }; |
| 1893 | |
| 1894 | if (TheExpr->getType()->isObjCRetainableType() && |
| 1895 | getLangOpts().ObjCAutoRefCount && LifetimeExtendObject(TheExpr)) { |
| 1896 | assert(getEvaluationKind(TheExpr->getType()) == TEK_Scalar &&(static_cast <bool> (getEvaluationKind(TheExpr->getType ()) == TEK_Scalar && "Only scalar can be a ObjC retainable type" ) ? void (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1897, __extension__ __PRETTY_FUNCTION__ )) |
| 1897 | "Only scalar can be a ObjC retainable type")(static_cast <bool> (getEvaluationKind(TheExpr->getType ()) == TEK_Scalar && "Only scalar can be a ObjC retainable type" ) ? void (0) : __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1897, __extension__ __PRETTY_FUNCTION__ )); |
| 1898 | if (!isa<Constant>(ArgVal)) { |
| 1899 | CleanupKind Cleanup = getARCCleanupKind(); |
| 1900 | QualType Ty = TheExpr->getType(); |
| 1901 | Address Alloca = Address::invalid(); |
| 1902 | Address Addr = CreateMemTemp(Ty, "os.log.arg", &Alloca); |
| 1903 | ArgVal = EmitARCRetain(Ty, ArgVal); |
| 1904 | Builder.CreateStore(ArgVal, Addr); |
| 1905 | pushLifetimeExtendedDestroy(Cleanup, Alloca, Ty, |
| 1906 | CodeGenFunction::destroyARCStrongPrecise, |
| 1907 | Cleanup & EHCleanup); |
| 1908 | |
| 1909 | // Push a clang.arc.use call to ensure ARC optimizer knows that the |
| 1910 | // argument has to be alive. |
| 1911 | if (CGM.getCodeGenOpts().OptimizationLevel != 0) |
| 1912 | pushCleanupAfterFullExpr<CallObjCArcUse>(Cleanup, ArgVal); |
| 1913 | } |
| 1914 | } |
| 1915 | } else { |
| 1916 | ArgVal = Builder.getInt32(Item.getConstValue().getQuantity()); |
| 1917 | } |
| 1918 | |
| 1919 | unsigned ArgValSize = |
| 1920 | CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType()); |
| 1921 | llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(), |
| 1922 | ArgValSize); |
| 1923 | ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy); |
| 1924 | CanQualType ArgTy = getOSLogArgType(Ctx, Size); |
| 1925 | // If ArgVal has type x86_fp80, zero-extend ArgVal. |
| 1926 | ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy)); |
| 1927 | Args.add(RValue::get(ArgVal), ArgTy); |
| 1928 | } |
| 1929 | |
| 1930 | const CGFunctionInfo &FI = |
| 1931 | CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args); |
| 1932 | llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction( |
| 1933 | Layout, BufAddr.getAlignment()); |
| 1934 | EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args); |
| 1935 | return RValue::get(BufAddr.getPointer()); |
| 1936 | } |
| 1937 | |
| 1938 | static bool isSpecialUnsignedMultiplySignedResult( |
| 1939 | unsigned BuiltinID, WidthAndSignedness Op1Info, WidthAndSignedness Op2Info, |
| 1940 | WidthAndSignedness ResultInfo) { |
| 1941 | return BuiltinID == Builtin::BI__builtin_mul_overflow && |
| 1942 | Op1Info.Width == Op2Info.Width && Op2Info.Width == ResultInfo.Width && |
| 1943 | !Op1Info.Signed && !Op2Info.Signed && ResultInfo.Signed; |
| 1944 | } |
| 1945 | |
| 1946 | static RValue EmitCheckedUnsignedMultiplySignedResult( |
| 1947 | CodeGenFunction &CGF, const clang::Expr *Op1, WidthAndSignedness Op1Info, |
| 1948 | const clang::Expr *Op2, WidthAndSignedness Op2Info, |
| 1949 | const clang::Expr *ResultArg, QualType ResultQTy, |
| 1950 | WidthAndSignedness ResultInfo) { |
| 1951 | assert(isSpecialUnsignedMultiplySignedResult((static_cast <bool> (isSpecialUnsignedMultiplySignedResult ( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo ) && "Cannot specialize this multiply") ? void (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1953, __extension__ __PRETTY_FUNCTION__ )) |
| 1952 | Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&(static_cast <bool> (isSpecialUnsignedMultiplySignedResult ( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo ) && "Cannot specialize this multiply") ? void (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1953, __extension__ __PRETTY_FUNCTION__ )) |
| 1953 | "Cannot specialize this multiply")(static_cast <bool> (isSpecialUnsignedMultiplySignedResult ( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo ) && "Cannot specialize this multiply") ? void (0) : __assert_fail ("isSpecialUnsignedMultiplySignedResult( Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Cannot specialize this multiply\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1953, __extension__ __PRETTY_FUNCTION__ )); |
| 1954 | |
| 1955 | llvm::Value *V1 = CGF.EmitScalarExpr(Op1); |
| 1956 | llvm::Value *V2 = CGF.EmitScalarExpr(Op2); |
| 1957 | |
| 1958 | llvm::Value *HasOverflow; |
| 1959 | llvm::Value *Result = EmitOverflowIntrinsic( |
| 1960 | CGF, llvm::Intrinsic::umul_with_overflow, V1, V2, HasOverflow); |
| 1961 | |
| 1962 | // The intrinsic call will detect overflow when the value is > UINT_MAX, |
| 1963 | // however, since the original builtin had a signed result, we need to report |
| 1964 | // an overflow when the result is greater than INT_MAX. |
| 1965 | auto IntMax = llvm::APInt::getSignedMaxValue(ResultInfo.Width); |
| 1966 | llvm::Value *IntMaxValue = llvm::ConstantInt::get(Result->getType(), IntMax); |
| 1967 | |
| 1968 | llvm::Value *IntMaxOverflow = CGF.Builder.CreateICmpUGT(Result, IntMaxValue); |
| 1969 | HasOverflow = CGF.Builder.CreateOr(HasOverflow, IntMaxOverflow); |
| 1970 | |
| 1971 | bool isVolatile = |
| 1972 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 1973 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); |
| 1974 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, |
| 1975 | isVolatile); |
| 1976 | return RValue::get(HasOverflow); |
| 1977 | } |
| 1978 | |
| 1979 | /// Determine if a binop is a checked mixed-sign multiply we can specialize. |
| 1980 | static bool isSpecialMixedSignMultiply(unsigned BuiltinID, |
| 1981 | WidthAndSignedness Op1Info, |
| 1982 | WidthAndSignedness Op2Info, |
| 1983 | WidthAndSignedness ResultInfo) { |
| 1984 | return BuiltinID == Builtin::BI__builtin_mul_overflow && |
| 1985 | std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width && |
| 1986 | Op1Info.Signed != Op2Info.Signed; |
| 1987 | } |
| 1988 | |
| 1989 | /// Emit a checked mixed-sign multiply. This is a cheaper specialization of |
| 1990 | /// the generic checked-binop irgen. |
| 1991 | static RValue |
| 1992 | EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1, |
| 1993 | WidthAndSignedness Op1Info, const clang::Expr *Op2, |
| 1994 | WidthAndSignedness Op2Info, |
| 1995 | const clang::Expr *ResultArg, QualType ResultQTy, |
| 1996 | WidthAndSignedness ResultInfo) { |
| 1997 | assert(isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info,(static_cast <bool> (isSpecialMixedSignMultiply(Builtin ::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize") ? void ( 0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1999, __extension__ __PRETTY_FUNCTION__ )) |
| 1998 | Op2Info, ResultInfo) &&(static_cast <bool> (isSpecialMixedSignMultiply(Builtin ::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize") ? void ( 0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1999, __extension__ __PRETTY_FUNCTION__ )) |
| 1999 | "Not a mixed-sign multipliction we can specialize")(static_cast <bool> (isSpecialMixedSignMultiply(Builtin ::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && "Not a mixed-sign multipliction we can specialize") ? void ( 0) : __assert_fail ("isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 1999, __extension__ __PRETTY_FUNCTION__ )); |
| 2000 | |
| 2001 | // Emit the signed and unsigned operands. |
| 2002 | const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2; |
| 2003 | const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1; |
| 2004 | llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp); |
| 2005 | llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp); |
| 2006 | unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width; |
| 2007 | unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width; |
| 2008 | |
| 2009 | // One of the operands may be smaller than the other. If so, [s|z]ext it. |
| 2010 | if (SignedOpWidth < UnsignedOpWidth) |
| 2011 | Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext"); |
| 2012 | if (UnsignedOpWidth < SignedOpWidth) |
| 2013 | Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext"); |
| 2014 | |
| 2015 | llvm::Type *OpTy = Signed->getType(); |
| 2016 | llvm::Value *Zero = llvm::Constant::getNullValue(OpTy); |
| 2017 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); |
| 2018 | llvm::Type *ResTy = ResultPtr.getElementType(); |
| 2019 | unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width); |
| 2020 | |
| 2021 | // Take the absolute value of the signed operand. |
| 2022 | llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero); |
| 2023 | llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed); |
| 2024 | llvm::Value *AbsSigned = |
| 2025 | CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed); |
| 2026 | |
| 2027 | // Perform a checked unsigned multiplication. |
| 2028 | llvm::Value *UnsignedOverflow; |
| 2029 | llvm::Value *UnsignedResult = |
| 2030 | EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned, |
| 2031 | Unsigned, UnsignedOverflow); |
| 2032 | |
| 2033 | llvm::Value *Overflow, *Result; |
| 2034 | if (ResultInfo.Signed) { |
| 2035 | // Signed overflow occurs if the result is greater than INT_MAX or lesser |
| 2036 | // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative). |
| 2037 | auto IntMax = |
| 2038 | llvm::APInt::getSignedMaxValue(ResultInfo.Width).zext(OpWidth); |
| 2039 | llvm::Value *MaxResult = |
| 2040 | CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax), |
| 2041 | CGF.Builder.CreateZExt(IsNegative, OpTy)); |
| 2042 | llvm::Value *SignedOverflow = |
| 2043 | CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult); |
| 2044 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow); |
| 2045 | |
| 2046 | // Prepare the signed result (possibly by negating it). |
| 2047 | llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult); |
| 2048 | llvm::Value *SignedResult = |
| 2049 | CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult); |
| 2050 | Result = CGF.Builder.CreateTrunc(SignedResult, ResTy); |
| 2051 | } else { |
| 2052 | // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX. |
| 2053 | llvm::Value *Underflow = CGF.Builder.CreateAnd( |
| 2054 | IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult)); |
| 2055 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow); |
| 2056 | if (ResultInfo.Width < OpWidth) { |
| 2057 | auto IntMax = |
| 2058 | llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth); |
| 2059 | llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT( |
| 2060 | UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax)); |
| 2061 | Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow); |
| 2062 | } |
| 2063 | |
| 2064 | // Negate the product if it would be negative in infinite precision. |
| 2065 | Result = CGF.Builder.CreateSelect( |
| 2066 | IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult); |
| 2067 | |
| 2068 | Result = CGF.Builder.CreateTrunc(Result, ResTy); |
| 2069 | } |
| 2070 | assert(Overflow && Result && "Missing overflow or result")(static_cast <bool> (Overflow && Result && "Missing overflow or result") ? void (0) : __assert_fail ("Overflow && Result && \"Missing overflow or result\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2070, __extension__ __PRETTY_FUNCTION__ )); |
| 2071 | |
| 2072 | bool isVolatile = |
| 2073 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 2074 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, |
| 2075 | isVolatile); |
| 2076 | return RValue::get(Overflow); |
| 2077 | } |
| 2078 | |
| 2079 | static bool |
| 2080 | TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty, |
| 2081 | llvm::SmallPtrSetImpl<const Decl *> &Seen) { |
| 2082 | if (const auto *Arr = Ctx.getAsArrayType(Ty)) |
| 2083 | Ty = Ctx.getBaseElementType(Arr); |
| 2084 | |
| 2085 | const auto *Record = Ty->getAsCXXRecordDecl(); |
| 2086 | if (!Record) |
| 2087 | return false; |
| 2088 | |
| 2089 | // We've already checked this type, or are in the process of checking it. |
| 2090 | if (!Seen.insert(Record).second) |
| 2091 | return false; |
| 2092 | |
| 2093 | assert(Record->hasDefinition() &&(static_cast <bool> (Record->hasDefinition() && "Incomplete types should already be diagnosed") ? void (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2094, __extension__ __PRETTY_FUNCTION__ )) |
| 2094 | "Incomplete types should already be diagnosed")(static_cast <bool> (Record->hasDefinition() && "Incomplete types should already be diagnosed") ? void (0) : __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2094, __extension__ __PRETTY_FUNCTION__ )); |
| 2095 | |
| 2096 | if (Record->isDynamicClass()) |
| 2097 | return true; |
| 2098 | |
| 2099 | for (FieldDecl *F : Record->fields()) { |
| 2100 | if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen)) |
| 2101 | return true; |
| 2102 | } |
| 2103 | return false; |
| 2104 | } |
| 2105 | |
| 2106 | /// Determine if the specified type requires laundering by checking if it is a |
| 2107 | /// dynamic class type or contains a subobject which is a dynamic class type. |
| 2108 | static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) { |
| 2109 | if (!CGM.getCodeGenOpts().StrictVTablePointers) |
| 2110 | return false; |
| 2111 | llvm::SmallPtrSet<const Decl *, 16> Seen; |
| 2112 | return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen); |
| 2113 | } |
| 2114 | |
| 2115 | RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) { |
| 2116 | llvm::Value *Src = EmitScalarExpr(E->getArg(0)); |
| 2117 | llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1)); |
| 2118 | |
| 2119 | // The builtin's shift arg may have a different type than the source arg and |
| 2120 | // result, but the LLVM intrinsic uses the same type for all values. |
| 2121 | llvm::Type *Ty = Src->getType(); |
| 2122 | ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false); |
| 2123 | |
| 2124 | // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same. |
| 2125 | unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl; |
| 2126 | Function *F = CGM.getIntrinsic(IID, Ty); |
| 2127 | return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt })); |
| 2128 | } |
| 2129 | |
| 2130 | // Map math builtins for long-double to f128 version. |
| 2131 | static unsigned mutateLongDoubleBuiltin(unsigned BuiltinID) { |
| 2132 | switch (BuiltinID) { |
| 2133 | #define MUTATE_LDBL(func) \ |
| 2134 | case Builtin::BI__builtin_##func##l: \ |
| 2135 | return Builtin::BI__builtin_##func##f128; |
| 2136 | MUTATE_LDBL(sqrt) |
| 2137 | MUTATE_LDBL(cbrt) |
| 2138 | MUTATE_LDBL(fabs) |
| 2139 | MUTATE_LDBL(log) |
| 2140 | MUTATE_LDBL(log2) |
| 2141 | MUTATE_LDBL(log10) |
| 2142 | MUTATE_LDBL(log1p) |
| 2143 | MUTATE_LDBL(logb) |
| 2144 | MUTATE_LDBL(exp) |
| 2145 | MUTATE_LDBL(exp2) |
| 2146 | MUTATE_LDBL(expm1) |
| 2147 | MUTATE_LDBL(fdim) |
| 2148 | MUTATE_LDBL(hypot) |
| 2149 | MUTATE_LDBL(ilogb) |
| 2150 | MUTATE_LDBL(pow) |
| 2151 | MUTATE_LDBL(fmin) |
| 2152 | MUTATE_LDBL(fmax) |
| 2153 | MUTATE_LDBL(ceil) |
| 2154 | MUTATE_LDBL(trunc) |
| 2155 | MUTATE_LDBL(rint) |
| 2156 | MUTATE_LDBL(nearbyint) |
| 2157 | MUTATE_LDBL(round) |
| 2158 | MUTATE_LDBL(floor) |
| 2159 | MUTATE_LDBL(lround) |
| 2160 | MUTATE_LDBL(llround) |
| 2161 | MUTATE_LDBL(lrint) |
| 2162 | MUTATE_LDBL(llrint) |
| 2163 | MUTATE_LDBL(fmod) |
| 2164 | MUTATE_LDBL(modf) |
| 2165 | MUTATE_LDBL(nan) |
| 2166 | MUTATE_LDBL(nans) |
| 2167 | MUTATE_LDBL(inf) |
| 2168 | MUTATE_LDBL(fma) |
| 2169 | MUTATE_LDBL(sin) |
| 2170 | MUTATE_LDBL(cos) |
| 2171 | MUTATE_LDBL(tan) |
| 2172 | MUTATE_LDBL(sinh) |
| 2173 | MUTATE_LDBL(cosh) |
| 2174 | MUTATE_LDBL(tanh) |
| 2175 | MUTATE_LDBL(asin) |
| 2176 | MUTATE_LDBL(acos) |
| 2177 | MUTATE_LDBL(atan) |
| 2178 | MUTATE_LDBL(asinh) |
| 2179 | MUTATE_LDBL(acosh) |
| 2180 | MUTATE_LDBL(atanh) |
| 2181 | MUTATE_LDBL(atan2) |
| 2182 | MUTATE_LDBL(erf) |
| 2183 | MUTATE_LDBL(erfc) |
| 2184 | MUTATE_LDBL(ldexp) |
| 2185 | MUTATE_LDBL(frexp) |
| 2186 | MUTATE_LDBL(huge_val) |
| 2187 | MUTATE_LDBL(copysign) |
| 2188 | MUTATE_LDBL(nextafter) |
| 2189 | MUTATE_LDBL(nexttoward) |
| 2190 | MUTATE_LDBL(remainder) |
| 2191 | MUTATE_LDBL(remquo) |
| 2192 | MUTATE_LDBL(scalbln) |
| 2193 | MUTATE_LDBL(scalbn) |
| 2194 | MUTATE_LDBL(tgamma) |
| 2195 | MUTATE_LDBL(lgamma) |
| 2196 | #undef MUTATE_LDBL |
| 2197 | default: |
| 2198 | return BuiltinID; |
| 2199 | } |
| 2200 | } |
| 2201 | |
| 2202 | RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, |
| 2203 | const CallExpr *E, |
| 2204 | ReturnValueSlot ReturnValue) { |
| 2205 | const FunctionDecl *FD = GD.getDecl()->getAsFunction(); |
| 2206 | // See if we can constant fold this builtin. If so, don't emit it at all. |
| 2207 | // TODO: Extend this handling to all builtin calls that we can constant-fold. |
| 2208 | Expr::EvalResult Result; |
| 2209 | if (E->isPRValue() && E->EvaluateAsRValue(Result, CGM.getContext()) && |
| 2210 | !Result.hasSideEffects()) { |
| 2211 | if (Result.Val.isInt()) |
| 2212 | return RValue::get(llvm::ConstantInt::get(getLLVMContext(), |
| 2213 | Result.Val.getInt())); |
| 2214 | if (Result.Val.isFloat()) |
| 2215 | return RValue::get(llvm::ConstantFP::get(getLLVMContext(), |
| 2216 | Result.Val.getFloat())); |
| 2217 | } |
| 2218 | |
| 2219 | // If current long-double semantics is IEEE 128-bit, replace math builtins |
| 2220 | // of long-double with f128 equivalent. |
| 2221 | // TODO: This mutation should also be applied to other targets other than PPC, |
| 2222 | // after backend supports IEEE 128-bit style libcalls. |
| 2223 | if (getTarget().getTriple().isPPC64() && |
| 2224 | &getTarget().getLongDoubleFormat() == &llvm::APFloat::IEEEquad()) |
| 2225 | BuiltinID = mutateLongDoubleBuiltin(BuiltinID); |
| 2226 | |
| 2227 | // If the builtin has been declared explicitly with an assembler label, |
| 2228 | // disable the specialized emitting below. Ideally we should communicate the |
| 2229 | // rename in IR, or at least avoid generating the intrinsic calls that are |
| 2230 | // likely to get lowered to the renamed library functions. |
| 2231 | const unsigned BuiltinIDIfNoAsmLabel = |
| 2232 | FD->hasAttr<AsmLabelAttr>() ? 0 : BuiltinID; |
| 2233 | |
| 2234 | // There are LLVM math intrinsics/instructions corresponding to math library |
| 2235 | // functions except the LLVM op will never set errno while the math library |
| 2236 | // might. Also, math builtins have the same semantics as their math library |
| 2237 | // twins. Thus, we can transform math library and builtin calls to their |
| 2238 | // LLVM counterparts if the call is marked 'const' (known to never set errno). |
| 2239 | // In case FP exceptions are enabled, the experimental versions of the |
| 2240 | // intrinsics model those. |
| 2241 | bool ConstWithoutErrnoAndExceptions = |
| 2242 | getContext().BuiltinInfo.isConstWithoutErrnoAndExceptions(BuiltinID); |
| 2243 | bool ConstWithoutExceptions = |
| 2244 | getContext().BuiltinInfo.isConstWithoutExceptions(BuiltinID); |
| 2245 | if (FD->hasAttr<ConstAttr>() || |
| 2246 | ((ConstWithoutErrnoAndExceptions || ConstWithoutExceptions) && |
| 2247 | (!ConstWithoutErrnoAndExceptions || (!getLangOpts().MathErrno)))) { |
| 2248 | switch (BuiltinIDIfNoAsmLabel) { |
| 2249 | case Builtin::BIceil: |
| 2250 | case Builtin::BIceilf: |
| 2251 | case Builtin::BIceill: |
| 2252 | case Builtin::BI__builtin_ceil: |
| 2253 | case Builtin::BI__builtin_ceilf: |
| 2254 | case Builtin::BI__builtin_ceilf16: |
| 2255 | case Builtin::BI__builtin_ceill: |
| 2256 | case Builtin::BI__builtin_ceilf128: |
| 2257 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2258 | Intrinsic::ceil, |
| 2259 | Intrinsic::experimental_constrained_ceil)); |
| 2260 | |
| 2261 | case Builtin::BIcopysign: |
| 2262 | case Builtin::BIcopysignf: |
| 2263 | case Builtin::BIcopysignl: |
| 2264 | case Builtin::BI__builtin_copysign: |
| 2265 | case Builtin::BI__builtin_copysignf: |
| 2266 | case Builtin::BI__builtin_copysignf16: |
| 2267 | case Builtin::BI__builtin_copysignl: |
| 2268 | case Builtin::BI__builtin_copysignf128: |
| 2269 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign)); |
| 2270 | |
| 2271 | case Builtin::BIcos: |
| 2272 | case Builtin::BIcosf: |
| 2273 | case Builtin::BIcosl: |
| 2274 | case Builtin::BI__builtin_cos: |
| 2275 | case Builtin::BI__builtin_cosf: |
| 2276 | case Builtin::BI__builtin_cosf16: |
| 2277 | case Builtin::BI__builtin_cosl: |
| 2278 | case Builtin::BI__builtin_cosf128: |
| 2279 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2280 | Intrinsic::cos, |
| 2281 | Intrinsic::experimental_constrained_cos)); |
| 2282 | |
| 2283 | case Builtin::BIexp: |
| 2284 | case Builtin::BIexpf: |
| 2285 | case Builtin::BIexpl: |
| 2286 | case Builtin::BI__builtin_exp: |
| 2287 | case Builtin::BI__builtin_expf: |
| 2288 | case Builtin::BI__builtin_expf16: |
| 2289 | case Builtin::BI__builtin_expl: |
| 2290 | case Builtin::BI__builtin_expf128: |
| 2291 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2292 | Intrinsic::exp, |
| 2293 | Intrinsic::experimental_constrained_exp)); |
| 2294 | |
| 2295 | case Builtin::BIexp2: |
| 2296 | case Builtin::BIexp2f: |
| 2297 | case Builtin::BIexp2l: |
| 2298 | case Builtin::BI__builtin_exp2: |
| 2299 | case Builtin::BI__builtin_exp2f: |
| 2300 | case Builtin::BI__builtin_exp2f16: |
| 2301 | case Builtin::BI__builtin_exp2l: |
| 2302 | case Builtin::BI__builtin_exp2f128: |
| 2303 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2304 | Intrinsic::exp2, |
| 2305 | Intrinsic::experimental_constrained_exp2)); |
| 2306 | |
| 2307 | case Builtin::BIfabs: |
| 2308 | case Builtin::BIfabsf: |
| 2309 | case Builtin::BIfabsl: |
| 2310 | case Builtin::BI__builtin_fabs: |
| 2311 | case Builtin::BI__builtin_fabsf: |
| 2312 | case Builtin::BI__builtin_fabsf16: |
| 2313 | case Builtin::BI__builtin_fabsl: |
| 2314 | case Builtin::BI__builtin_fabsf128: |
| 2315 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs)); |
| 2316 | |
| 2317 | case Builtin::BIfloor: |
| 2318 | case Builtin::BIfloorf: |
| 2319 | case Builtin::BIfloorl: |
| 2320 | case Builtin::BI__builtin_floor: |
| 2321 | case Builtin::BI__builtin_floorf: |
| 2322 | case Builtin::BI__builtin_floorf16: |
| 2323 | case Builtin::BI__builtin_floorl: |
| 2324 | case Builtin::BI__builtin_floorf128: |
| 2325 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2326 | Intrinsic::floor, |
| 2327 | Intrinsic::experimental_constrained_floor)); |
| 2328 | |
| 2329 | case Builtin::BIfma: |
| 2330 | case Builtin::BIfmaf: |
| 2331 | case Builtin::BIfmal: |
| 2332 | case Builtin::BI__builtin_fma: |
| 2333 | case Builtin::BI__builtin_fmaf: |
| 2334 | case Builtin::BI__builtin_fmaf16: |
| 2335 | case Builtin::BI__builtin_fmal: |
| 2336 | case Builtin::BI__builtin_fmaf128: |
| 2337 | return RValue::get(emitTernaryMaybeConstrainedFPBuiltin(*this, E, |
| 2338 | Intrinsic::fma, |
| 2339 | Intrinsic::experimental_constrained_fma)); |
| 2340 | |
| 2341 | case Builtin::BIfmax: |
| 2342 | case Builtin::BIfmaxf: |
| 2343 | case Builtin::BIfmaxl: |
| 2344 | case Builtin::BI__builtin_fmax: |
| 2345 | case Builtin::BI__builtin_fmaxf: |
| 2346 | case Builtin::BI__builtin_fmaxf16: |
| 2347 | case Builtin::BI__builtin_fmaxl: |
| 2348 | case Builtin::BI__builtin_fmaxf128: |
| 2349 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
| 2350 | Intrinsic::maxnum, |
| 2351 | Intrinsic::experimental_constrained_maxnum)); |
| 2352 | |
| 2353 | case Builtin::BIfmin: |
| 2354 | case Builtin::BIfminf: |
| 2355 | case Builtin::BIfminl: |
| 2356 | case Builtin::BI__builtin_fmin: |
| 2357 | case Builtin::BI__builtin_fminf: |
| 2358 | case Builtin::BI__builtin_fminf16: |
| 2359 | case Builtin::BI__builtin_fminl: |
| 2360 | case Builtin::BI__builtin_fminf128: |
| 2361 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
| 2362 | Intrinsic::minnum, |
| 2363 | Intrinsic::experimental_constrained_minnum)); |
| 2364 | |
| 2365 | // fmod() is a special-case. It maps to the frem instruction rather than an |
| 2366 | // LLVM intrinsic. |
| 2367 | case Builtin::BIfmod: |
| 2368 | case Builtin::BIfmodf: |
| 2369 | case Builtin::BIfmodl: |
| 2370 | case Builtin::BI__builtin_fmod: |
| 2371 | case Builtin::BI__builtin_fmodf: |
| 2372 | case Builtin::BI__builtin_fmodf16: |
| 2373 | case Builtin::BI__builtin_fmodl: |
| 2374 | case Builtin::BI__builtin_fmodf128: { |
| 2375 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 2376 | Value *Arg1 = EmitScalarExpr(E->getArg(0)); |
| 2377 | Value *Arg2 = EmitScalarExpr(E->getArg(1)); |
| 2378 | return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod")); |
| 2379 | } |
| 2380 | |
| 2381 | case Builtin::BIlog: |
| 2382 | case Builtin::BIlogf: |
| 2383 | case Builtin::BIlogl: |
| 2384 | case Builtin::BI__builtin_log: |
| 2385 | case Builtin::BI__builtin_logf: |
| 2386 | case Builtin::BI__builtin_logf16: |
| 2387 | case Builtin::BI__builtin_logl: |
| 2388 | case Builtin::BI__builtin_logf128: |
| 2389 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2390 | Intrinsic::log, |
| 2391 | Intrinsic::experimental_constrained_log)); |
| 2392 | |
| 2393 | case Builtin::BIlog10: |
| 2394 | case Builtin::BIlog10f: |
| 2395 | case Builtin::BIlog10l: |
| 2396 | case Builtin::BI__builtin_log10: |
| 2397 | case Builtin::BI__builtin_log10f: |
| 2398 | case Builtin::BI__builtin_log10f16: |
| 2399 | case Builtin::BI__builtin_log10l: |
| 2400 | case Builtin::BI__builtin_log10f128: |
| 2401 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2402 | Intrinsic::log10, |
| 2403 | Intrinsic::experimental_constrained_log10)); |
| 2404 | |
| 2405 | case Builtin::BIlog2: |
| 2406 | case Builtin::BIlog2f: |
| 2407 | case Builtin::BIlog2l: |
| 2408 | case Builtin::BI__builtin_log2: |
| 2409 | case Builtin::BI__builtin_log2f: |
| 2410 | case Builtin::BI__builtin_log2f16: |
| 2411 | case Builtin::BI__builtin_log2l: |
| 2412 | case Builtin::BI__builtin_log2f128: |
| 2413 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2414 | Intrinsic::log2, |
| 2415 | Intrinsic::experimental_constrained_log2)); |
| 2416 | |
| 2417 | case Builtin::BInearbyint: |
| 2418 | case Builtin::BInearbyintf: |
| 2419 | case Builtin::BInearbyintl: |
| 2420 | case Builtin::BI__builtin_nearbyint: |
| 2421 | case Builtin::BI__builtin_nearbyintf: |
| 2422 | case Builtin::BI__builtin_nearbyintl: |
| 2423 | case Builtin::BI__builtin_nearbyintf128: |
| 2424 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2425 | Intrinsic::nearbyint, |
| 2426 | Intrinsic::experimental_constrained_nearbyint)); |
| 2427 | |
| 2428 | case Builtin::BIpow: |
| 2429 | case Builtin::BIpowf: |
| 2430 | case Builtin::BIpowl: |
| 2431 | case Builtin::BI__builtin_pow: |
| 2432 | case Builtin::BI__builtin_powf: |
| 2433 | case Builtin::BI__builtin_powf16: |
| 2434 | case Builtin::BI__builtin_powl: |
| 2435 | case Builtin::BI__builtin_powf128: |
| 2436 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
| 2437 | Intrinsic::pow, |
| 2438 | Intrinsic::experimental_constrained_pow)); |
| 2439 | |
| 2440 | case Builtin::BIrint: |
| 2441 | case Builtin::BIrintf: |
| 2442 | case Builtin::BIrintl: |
| 2443 | case Builtin::BI__builtin_rint: |
| 2444 | case Builtin::BI__builtin_rintf: |
| 2445 | case Builtin::BI__builtin_rintf16: |
| 2446 | case Builtin::BI__builtin_rintl: |
| 2447 | case Builtin::BI__builtin_rintf128: |
| 2448 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2449 | Intrinsic::rint, |
| 2450 | Intrinsic::experimental_constrained_rint)); |
| 2451 | |
| 2452 | case Builtin::BIround: |
| 2453 | case Builtin::BIroundf: |
| 2454 | case Builtin::BIroundl: |
| 2455 | case Builtin::BI__builtin_round: |
| 2456 | case Builtin::BI__builtin_roundf: |
| 2457 | case Builtin::BI__builtin_roundf16: |
| 2458 | case Builtin::BI__builtin_roundl: |
| 2459 | case Builtin::BI__builtin_roundf128: |
| 2460 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2461 | Intrinsic::round, |
| 2462 | Intrinsic::experimental_constrained_round)); |
| 2463 | |
| 2464 | case Builtin::BIroundeven: |
| 2465 | case Builtin::BIroundevenf: |
| 2466 | case Builtin::BIroundevenl: |
| 2467 | case Builtin::BI__builtin_roundeven: |
| 2468 | case Builtin::BI__builtin_roundevenf: |
| 2469 | case Builtin::BI__builtin_roundevenf16: |
| 2470 | case Builtin::BI__builtin_roundevenl: |
| 2471 | case Builtin::BI__builtin_roundevenf128: |
| 2472 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2473 | Intrinsic::roundeven, |
| 2474 | Intrinsic::experimental_constrained_roundeven)); |
| 2475 | |
| 2476 | case Builtin::BIsin: |
| 2477 | case Builtin::BIsinf: |
| 2478 | case Builtin::BIsinl: |
| 2479 | case Builtin::BI__builtin_sin: |
| 2480 | case Builtin::BI__builtin_sinf: |
| 2481 | case Builtin::BI__builtin_sinf16: |
| 2482 | case Builtin::BI__builtin_sinl: |
| 2483 | case Builtin::BI__builtin_sinf128: |
| 2484 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2485 | Intrinsic::sin, |
| 2486 | Intrinsic::experimental_constrained_sin)); |
| 2487 | |
| 2488 | case Builtin::BIsqrt: |
| 2489 | case Builtin::BIsqrtf: |
| 2490 | case Builtin::BIsqrtl: |
| 2491 | case Builtin::BI__builtin_sqrt: |
| 2492 | case Builtin::BI__builtin_sqrtf: |
| 2493 | case Builtin::BI__builtin_sqrtf16: |
| 2494 | case Builtin::BI__builtin_sqrtl: |
| 2495 | case Builtin::BI__builtin_sqrtf128: |
| 2496 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2497 | Intrinsic::sqrt, |
| 2498 | Intrinsic::experimental_constrained_sqrt)); |
| 2499 | |
| 2500 | case Builtin::BItrunc: |
| 2501 | case Builtin::BItruncf: |
| 2502 | case Builtin::BItruncl: |
| 2503 | case Builtin::BI__builtin_trunc: |
| 2504 | case Builtin::BI__builtin_truncf: |
| 2505 | case Builtin::BI__builtin_truncf16: |
| 2506 | case Builtin::BI__builtin_truncl: |
| 2507 | case Builtin::BI__builtin_truncf128: |
| 2508 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2509 | Intrinsic::trunc, |
| 2510 | Intrinsic::experimental_constrained_trunc)); |
| 2511 | |
| 2512 | case Builtin::BIlround: |
| 2513 | case Builtin::BIlroundf: |
| 2514 | case Builtin::BIlroundl: |
| 2515 | case Builtin::BI__builtin_lround: |
| 2516 | case Builtin::BI__builtin_lroundf: |
| 2517 | case Builtin::BI__builtin_lroundl: |
| 2518 | case Builtin::BI__builtin_lroundf128: |
| 2519 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2520 | *this, E, Intrinsic::lround, |
| 2521 | Intrinsic::experimental_constrained_lround)); |
| 2522 | |
| 2523 | case Builtin::BIllround: |
| 2524 | case Builtin::BIllroundf: |
| 2525 | case Builtin::BIllroundl: |
| 2526 | case Builtin::BI__builtin_llround: |
| 2527 | case Builtin::BI__builtin_llroundf: |
| 2528 | case Builtin::BI__builtin_llroundl: |
| 2529 | case Builtin::BI__builtin_llroundf128: |
| 2530 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2531 | *this, E, Intrinsic::llround, |
| 2532 | Intrinsic::experimental_constrained_llround)); |
| 2533 | |
| 2534 | case Builtin::BIlrint: |
| 2535 | case Builtin::BIlrintf: |
| 2536 | case Builtin::BIlrintl: |
| 2537 | case Builtin::BI__builtin_lrint: |
| 2538 | case Builtin::BI__builtin_lrintf: |
| 2539 | case Builtin::BI__builtin_lrintl: |
| 2540 | case Builtin::BI__builtin_lrintf128: |
| 2541 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2542 | *this, E, Intrinsic::lrint, |
| 2543 | Intrinsic::experimental_constrained_lrint)); |
| 2544 | |
| 2545 | case Builtin::BIllrint: |
| 2546 | case Builtin::BIllrintf: |
| 2547 | case Builtin::BIllrintl: |
| 2548 | case Builtin::BI__builtin_llrint: |
| 2549 | case Builtin::BI__builtin_llrintf: |
| 2550 | case Builtin::BI__builtin_llrintl: |
| 2551 | case Builtin::BI__builtin_llrintf128: |
| 2552 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2553 | *this, E, Intrinsic::llrint, |
| 2554 | Intrinsic::experimental_constrained_llrint)); |
| 2555 | |
| 2556 | default: |
| 2557 | break; |
| 2558 | } |
| 2559 | } |
| 2560 | |
| 2561 | switch (BuiltinIDIfNoAsmLabel) { |
| 2562 | default: break; |
| 2563 | case Builtin::BI__builtin___CFStringMakeConstantString: |
| 2564 | case Builtin::BI__builtin___NSStringMakeConstantString: |
| 2565 | return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType())); |
| 2566 | case Builtin::BI__builtin_stdarg_start: |
| 2567 | case Builtin::BI__builtin_va_start: |
| 2568 | case Builtin::BI__va_start: |
| 2569 | case Builtin::BI__builtin_va_end: |
| 2570 | EmitVAStartEnd(BuiltinID == Builtin::BI__va_start |
| 2571 | ? EmitScalarExpr(E->getArg(0)) |
| 2572 | : EmitVAListRef(E->getArg(0)).getPointer(), |
| 2573 | BuiltinID != Builtin::BI__builtin_va_end); |
| 2574 | return RValue::get(nullptr); |
| 2575 | case Builtin::BI__builtin_va_copy: { |
| 2576 | Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer(); |
| 2577 | Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer(); |
| 2578 | |
| 2579 | llvm::Type *Type = Int8PtrTy; |
| 2580 | |
| 2581 | DstPtr = Builder.CreateBitCast(DstPtr, Type); |
| 2582 | SrcPtr = Builder.CreateBitCast(SrcPtr, Type); |
| 2583 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy), {DstPtr, SrcPtr}); |
| 2584 | return RValue::get(nullptr); |
| 2585 | } |
| 2586 | case Builtin::BI__builtin_abs: |
| 2587 | case Builtin::BI__builtin_labs: |
| 2588 | case Builtin::BI__builtin_llabs: { |
| 2589 | // X < 0 ? -X : X |
| 2590 | // The negation has 'nsw' because abs of INT_MIN is undefined. |
| 2591 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2592 | Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg"); |
| 2593 | Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType()); |
| 2594 | Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond"); |
| 2595 | Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs"); |
| 2596 | return RValue::get(Result); |
| 2597 | } |
| 2598 | case Builtin::BI__builtin_complex: { |
| 2599 | Value *Real = EmitScalarExpr(E->getArg(0)); |
| 2600 | Value *Imag = EmitScalarExpr(E->getArg(1)); |
| 2601 | return RValue::getComplex({Real, Imag}); |
| 2602 | } |
| 2603 | case Builtin::BI__builtin_conj: |
| 2604 | case Builtin::BI__builtin_conjf: |
| 2605 | case Builtin::BI__builtin_conjl: |
| 2606 | case Builtin::BIconj: |
| 2607 | case Builtin::BIconjf: |
| 2608 | case Builtin::BIconjl: { |
| 2609 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2610 | Value *Real = ComplexVal.first; |
| 2611 | Value *Imag = ComplexVal.second; |
| 2612 | Imag = Builder.CreateFNeg(Imag, "neg"); |
| 2613 | return RValue::getComplex(std::make_pair(Real, Imag)); |
| 2614 | } |
| 2615 | case Builtin::BI__builtin_creal: |
| 2616 | case Builtin::BI__builtin_crealf: |
| 2617 | case Builtin::BI__builtin_creall: |
| 2618 | case Builtin::BIcreal: |
| 2619 | case Builtin::BIcrealf: |
| 2620 | case Builtin::BIcreall: { |
| 2621 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2622 | return RValue::get(ComplexVal.first); |
| 2623 | } |
| 2624 | |
| 2625 | case Builtin::BI__builtin_preserve_access_index: { |
| 2626 | // Only enabled preserved access index region when debuginfo |
| 2627 | // is available as debuginfo is needed to preserve user-level |
| 2628 | // access pattern. |
| 2629 | if (!getDebugInfo()) { |
| 2630 | CGM.Error(E->getExprLoc(), "using builtin_preserve_access_index() without -g"); |
| 2631 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
| 2632 | } |
| 2633 | |
| 2634 | // Nested builtin_preserve_access_index() not supported |
| 2635 | if (IsInPreservedAIRegion) { |
| 2636 | CGM.Error(E->getExprLoc(), "nested builtin_preserve_access_index() not supported"); |
| 2637 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
| 2638 | } |
| 2639 | |
| 2640 | IsInPreservedAIRegion = true; |
| 2641 | Value *Res = EmitScalarExpr(E->getArg(0)); |
| 2642 | IsInPreservedAIRegion = false; |
| 2643 | return RValue::get(Res); |
| 2644 | } |
| 2645 | |
| 2646 | case Builtin::BI__builtin_cimag: |
| 2647 | case Builtin::BI__builtin_cimagf: |
| 2648 | case Builtin::BI__builtin_cimagl: |
| 2649 | case Builtin::BIcimag: |
| 2650 | case Builtin::BIcimagf: |
| 2651 | case Builtin::BIcimagl: { |
| 2652 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2653 | return RValue::get(ComplexVal.second); |
| 2654 | } |
| 2655 | |
| 2656 | case Builtin::BI__builtin_clrsb: |
| 2657 | case Builtin::BI__builtin_clrsbl: |
| 2658 | case Builtin::BI__builtin_clrsbll: { |
| 2659 | // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or |
| 2660 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2661 | |
| 2662 | llvm::Type *ArgType = ArgValue->getType(); |
| 2663 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 2664 | |
| 2665 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2666 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
| 2667 | Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg"); |
| 2668 | Value *Inverse = Builder.CreateNot(ArgValue, "not"); |
| 2669 | Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue); |
| 2670 | Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()}); |
| 2671 | Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1)); |
| 2672 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2673 | "cast"); |
| 2674 | return RValue::get(Result); |
| 2675 | } |
| 2676 | case Builtin::BI__builtin_ctzs: |
| 2677 | case Builtin::BI__builtin_ctz: |
| 2678 | case Builtin::BI__builtin_ctzl: |
| 2679 | case Builtin::BI__builtin_ctzll: { |
| 2680 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero); |
| 2681 | |
| 2682 | llvm::Type *ArgType = ArgValue->getType(); |
| 2683 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 2684 | |
| 2685 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2686 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
| 2687 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
| 2688 | if (Result->getType() != ResultType) |
| 2689 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2690 | "cast"); |
| 2691 | return RValue::get(Result); |
| 2692 | } |
| 2693 | case Builtin::BI__builtin_clzs: |
| 2694 | case Builtin::BI__builtin_clz: |
| 2695 | case Builtin::BI__builtin_clzl: |
| 2696 | case Builtin::BI__builtin_clzll: { |
| 2697 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero); |
| 2698 | |
| 2699 | llvm::Type *ArgType = ArgValue->getType(); |
| 2700 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 2701 | |
| 2702 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2703 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
| 2704 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
| 2705 | if (Result->getType() != ResultType) |
| 2706 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2707 | "cast"); |
| 2708 | return RValue::get(Result); |
| 2709 | } |
| 2710 | case Builtin::BI__builtin_ffs: |
| 2711 | case Builtin::BI__builtin_ffsl: |
| 2712 | case Builtin::BI__builtin_ffsll: { |
| 2713 | // ffs(x) -> x ? cttz(x) + 1 : 0 |
| 2714 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2715 | |
| 2716 | llvm::Type *ArgType = ArgValue->getType(); |
| 2717 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 2718 | |
| 2719 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2720 | Value *Tmp = |
| 2721 | Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}), |
| 2722 | llvm::ConstantInt::get(ArgType, 1)); |
| 2723 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
| 2724 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero"); |
| 2725 | Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs"); |
| 2726 | if (Result->getType() != ResultType) |
| 2727 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2728 | "cast"); |
| 2729 | return RValue::get(Result); |
| 2730 | } |
| 2731 | case Builtin::BI__builtin_parity: |
| 2732 | case Builtin::BI__builtin_parityl: |
| 2733 | case Builtin::BI__builtin_parityll: { |
| 2734 | // parity(x) -> ctpop(x) & 1 |
| 2735 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2736 | |
| 2737 | llvm::Type *ArgType = ArgValue->getType(); |
| 2738 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
| 2739 | |
| 2740 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2741 | Value *Tmp = Builder.CreateCall(F, ArgValue); |
| 2742 | Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1)); |
| 2743 | if (Result->getType() != ResultType) |
| 2744 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2745 | "cast"); |
| 2746 | return RValue::get(Result); |
| 2747 | } |
| 2748 | case Builtin::BI__lzcnt16: |
| 2749 | case Builtin::BI__lzcnt: |
| 2750 | case Builtin::BI__lzcnt64: { |
| 2751 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2752 | |
| 2753 | llvm::Type *ArgType = ArgValue->getType(); |
| 2754 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 2755 | |
| 2756 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2757 | Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()}); |
| 2758 | if (Result->getType() != ResultType) |
| 2759 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2760 | "cast"); |
| 2761 | return RValue::get(Result); |
| 2762 | } |
| 2763 | case Builtin::BI__popcnt16: |
| 2764 | case Builtin::BI__popcnt: |
| 2765 | case Builtin::BI__popcnt64: |
| 2766 | case Builtin::BI__builtin_popcount: |
| 2767 | case Builtin::BI__builtin_popcountl: |
| 2768 | case Builtin::BI__builtin_popcountll: { |
| 2769 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2770 | |
| 2771 | llvm::Type *ArgType = ArgValue->getType(); |
| 2772 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
| 2773 | |
| 2774 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2775 | Value *Result = Builder.CreateCall(F, ArgValue); |
| 2776 | if (Result->getType() != ResultType) |
| 2777 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2778 | "cast"); |
| 2779 | return RValue::get(Result); |
| 2780 | } |
| 2781 | case Builtin::BI__builtin_unpredictable: { |
| 2782 | // Always return the argument of __builtin_unpredictable. LLVM does not |
| 2783 | // handle this builtin. Metadata for this builtin should be added directly |
| 2784 | // to instructions such as branches or switches that use it. |
| 2785 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
| 2786 | } |
| 2787 | case Builtin::BI__builtin_expect: { |
| 2788 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2789 | llvm::Type *ArgType = ArgValue->getType(); |
| 2790 | |
| 2791 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); |
| 2792 | // Don't generate llvm.expect on -O0 as the backend won't use it for |
| 2793 | // anything. |
| 2794 | // Note, we still IRGen ExpectedValue because it could have side-effects. |
| 2795 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
| 2796 | return RValue::get(ArgValue); |
| 2797 | |
| 2798 | Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType); |
| 2799 | Value *Result = |
| 2800 | Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval"); |
| 2801 | return RValue::get(Result); |
| 2802 | } |
| 2803 | case Builtin::BI__builtin_expect_with_probability: { |
| 2804 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2805 | llvm::Type *ArgType = ArgValue->getType(); |
| 2806 | |
| 2807 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); |
| 2808 | llvm::APFloat Probability(0.0); |
| 2809 | const Expr *ProbArg = E->getArg(2); |
| 2810 | bool EvalSucceed = ProbArg->EvaluateAsFloat(Probability, CGM.getContext()); |
| 2811 | assert(EvalSucceed && "probability should be able to evaluate as float")(static_cast <bool> (EvalSucceed && "probability should be able to evaluate as float" ) ? void (0) : __assert_fail ("EvalSucceed && \"probability should be able to evaluate as float\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 2811, __extension__ __PRETTY_FUNCTION__ )); |
| 2812 | (void)EvalSucceed; |
| 2813 | bool LoseInfo = false; |
| 2814 | Probability.convert(llvm::APFloat::IEEEdouble(), |
| 2815 | llvm::RoundingMode::Dynamic, &LoseInfo); |
| 2816 | llvm::Type *Ty = ConvertType(ProbArg->getType()); |
| 2817 | Constant *Confidence = ConstantFP::get(Ty, Probability); |
| 2818 | // Don't generate llvm.expect.with.probability on -O0 as the backend |
| 2819 | // won't use it for anything. |
| 2820 | // Note, we still IRGen ExpectedValue because it could have side-effects. |
| 2821 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
| 2822 | return RValue::get(ArgValue); |
| 2823 | |
| 2824 | Function *FnExpect = |
| 2825 | CGM.getIntrinsic(Intrinsic::expect_with_probability, ArgType); |
| 2826 | Value *Result = Builder.CreateCall( |
| 2827 | FnExpect, {ArgValue, ExpectedValue, Confidence}, "expval"); |
| 2828 | return RValue::get(Result); |
| 2829 | } |
| 2830 | case Builtin::BI__builtin_assume_aligned: { |
| 2831 | const Expr *Ptr = E->getArg(0); |
| 2832 | Value *PtrValue = EmitScalarExpr(Ptr); |
| 2833 | if (PtrValue->getType() != VoidPtrTy) |
| 2834 | PtrValue = EmitCastToVoidPtr(PtrValue); |
| 2835 | Value *OffsetValue = |
| 2836 | (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr; |
| 2837 | |
| 2838 | Value *AlignmentValue = EmitScalarExpr(E->getArg(1)); |
| 2839 | ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue); |
| 2840 | if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment)) |
| 2841 | AlignmentCI = ConstantInt::get(AlignmentCI->getType(), |
| 2842 | llvm::Value::MaximumAlignment); |
| 2843 | |
| 2844 | emitAlignmentAssumption(PtrValue, Ptr, |
| 2845 | /*The expr loc is sufficient.*/ SourceLocation(), |
| 2846 | AlignmentCI, OffsetValue); |
| 2847 | return RValue::get(PtrValue); |
| 2848 | } |
| 2849 | case Builtin::BI__assume: |
| 2850 | case Builtin::BI__builtin_assume: { |
| 2851 | if (E->getArg(0)->HasSideEffects(getContext())) |
| 2852 | return RValue::get(nullptr); |
| 2853 | |
| 2854 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2855 | Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume); |
| 2856 | Builder.CreateCall(FnAssume, ArgValue); |
| 2857 | return RValue::get(nullptr); |
| 2858 | } |
| 2859 | case Builtin::BI__builtin_assume_separate_storage: { |
| 2860 | const Expr *Arg0 = E->getArg(0); |
| 2861 | const Expr *Arg1 = E->getArg(1); |
| 2862 | |
| 2863 | Value *Value0 = EmitScalarExpr(Arg0); |
| 2864 | Value *Value1 = EmitScalarExpr(Arg1); |
| 2865 | |
| 2866 | Value *Values[] = {Value0, Value1}; |
| 2867 | OperandBundleDefT<Value *> OBD("separate_storage", Values); |
| 2868 | Builder.CreateAssumption(ConstantInt::getTrue(getLLVMContext()), {OBD}); |
| 2869 | return RValue::get(nullptr); |
| 2870 | } |
| 2871 | case Builtin::BI__arithmetic_fence: { |
| 2872 | // Create the builtin call if FastMath is selected, and the target |
| 2873 | // supports the builtin, otherwise just return the argument. |
| 2874 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 2875 | llvm::FastMathFlags FMF = Builder.getFastMathFlags(); |
| 2876 | bool isArithmeticFenceEnabled = |
| 2877 | FMF.allowReassoc() && |
| 2878 | getContext().getTargetInfo().checkArithmeticFenceSupported(); |
| 2879 | QualType ArgType = E->getArg(0)->getType(); |
| 2880 | if (ArgType->isComplexType()) { |
| 2881 | if (isArithmeticFenceEnabled) { |
| 2882 | QualType ElementType = ArgType->castAs<ComplexType>()->getElementType(); |
| 2883 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2884 | Value *Real = Builder.CreateArithmeticFence(ComplexVal.first, |
| 2885 | ConvertType(ElementType)); |
| 2886 | Value *Imag = Builder.CreateArithmeticFence(ComplexVal.second, |
| 2887 | ConvertType(ElementType)); |
| 2888 | return RValue::getComplex(std::make_pair(Real, Imag)); |
| 2889 | } |
| 2890 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2891 | Value *Real = ComplexVal.first; |
| 2892 | Value *Imag = ComplexVal.second; |
| 2893 | return RValue::getComplex(std::make_pair(Real, Imag)); |
| 2894 | } |
| 2895 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2896 | if (isArithmeticFenceEnabled) |
| 2897 | return RValue::get( |
| 2898 | Builder.CreateArithmeticFence(ArgValue, ConvertType(ArgType))); |
| 2899 | return RValue::get(ArgValue); |
| 2900 | } |
| 2901 | case Builtin::BI__builtin_bswap16: |
| 2902 | case Builtin::BI__builtin_bswap32: |
| 2903 | case Builtin::BI__builtin_bswap64: |
| 2904 | case Builtin::BI_byteswap_ushort: |
| 2905 | case Builtin::BI_byteswap_ulong: |
| 2906 | case Builtin::BI_byteswap_uint64: { |
| 2907 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap)); |
| 2908 | } |
| 2909 | case Builtin::BI__builtin_bitreverse8: |
| 2910 | case Builtin::BI__builtin_bitreverse16: |
| 2911 | case Builtin::BI__builtin_bitreverse32: |
| 2912 | case Builtin::BI__builtin_bitreverse64: { |
| 2913 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse)); |
| 2914 | } |
| 2915 | case Builtin::BI__builtin_rotateleft8: |
| 2916 | case Builtin::BI__builtin_rotateleft16: |
| 2917 | case Builtin::BI__builtin_rotateleft32: |
| 2918 | case Builtin::BI__builtin_rotateleft64: |
| 2919 | case Builtin::BI_rotl8: // Microsoft variants of rotate left |
| 2920 | case Builtin::BI_rotl16: |
| 2921 | case Builtin::BI_rotl: |
| 2922 | case Builtin::BI_lrotl: |
| 2923 | case Builtin::BI_rotl64: |
| 2924 | return emitRotate(E, false); |
| 2925 | |
| 2926 | case Builtin::BI__builtin_rotateright8: |
| 2927 | case Builtin::BI__builtin_rotateright16: |
| 2928 | case Builtin::BI__builtin_rotateright32: |
| 2929 | case Builtin::BI__builtin_rotateright64: |
| 2930 | case Builtin::BI_rotr8: // Microsoft variants of rotate right |
| 2931 | case Builtin::BI_rotr16: |
| 2932 | case Builtin::BI_rotr: |
| 2933 | case Builtin::BI_lrotr: |
| 2934 | case Builtin::BI_rotr64: |
| 2935 | return emitRotate(E, true); |
| 2936 | |
| 2937 | case Builtin::BI__builtin_constant_p: { |
| 2938 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2939 | |
| 2940 | const Expr *Arg = E->getArg(0); |
| 2941 | QualType ArgType = Arg->getType(); |
| 2942 | // FIXME: The allowance for Obj-C pointers and block pointers is historical |
| 2943 | // and likely a mistake. |
| 2944 | if (!ArgType->isIntegralOrEnumerationType() && !ArgType->isFloatingType() && |
| 2945 | !ArgType->isObjCObjectPointerType() && !ArgType->isBlockPointerType()) |
| 2946 | // Per the GCC documentation, only numeric constants are recognized after |
| 2947 | // inlining. |
| 2948 | return RValue::get(ConstantInt::get(ResultType, 0)); |
| 2949 | |
| 2950 | if (Arg->HasSideEffects(getContext())) |
| 2951 | // The argument is unevaluated, so be conservative if it might have |
| 2952 | // side-effects. |
| 2953 | return RValue::get(ConstantInt::get(ResultType, 0)); |
| 2954 | |
| 2955 | Value *ArgValue = EmitScalarExpr(Arg); |
| 2956 | if (ArgType->isObjCObjectPointerType()) { |
| 2957 | // Convert Objective-C objects to id because we cannot distinguish between |
| 2958 | // LLVM types for Obj-C classes as they are opaque. |
| 2959 | ArgType = CGM.getContext().getObjCIdType(); |
| 2960 | ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType)); |
| 2961 | } |
| 2962 | Function *F = |
| 2963 | CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType)); |
| 2964 | Value *Result = Builder.CreateCall(F, ArgValue); |
| 2965 | if (Result->getType() != ResultType) |
| 2966 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false); |
| 2967 | return RValue::get(Result); |
| 2968 | } |
| 2969 | case Builtin::BI__builtin_dynamic_object_size: |
| 2970 | case Builtin::BI__builtin_object_size: { |
| 2971 | unsigned Type = |
| 2972 | E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
| 2973 | auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType())); |
| 2974 | |
| 2975 | // We pass this builtin onto the optimizer so that it can figure out the |
| 2976 | // object size in more complex cases. |
| 2977 | bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size; |
| 2978 | return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType, |
| 2979 | /*EmittedE=*/nullptr, IsDynamic)); |
| 2980 | } |
| 2981 | case Builtin::BI__builtin_prefetch: { |
| 2982 | Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0)); |
| 2983 | // FIXME: Technically these constants should of type 'int', yes? |
| 2984 | RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) : |
| 2985 | llvm::ConstantInt::get(Int32Ty, 0); |
| 2986 | Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : |
| 2987 | llvm::ConstantInt::get(Int32Ty, 3); |
| 2988 | Value *Data = llvm::ConstantInt::get(Int32Ty, 1); |
| 2989 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); |
| 2990 | Builder.CreateCall(F, {Address, RW, Locality, Data}); |
| 2991 | return RValue::get(nullptr); |
| 2992 | } |
| 2993 | case Builtin::BI__builtin_readcyclecounter: { |
| 2994 | Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter); |
| 2995 | return RValue::get(Builder.CreateCall(F)); |
| 2996 | } |
| 2997 | case Builtin::BI__builtin___clear_cache: { |
| 2998 | Value *Begin = EmitScalarExpr(E->getArg(0)); |
| 2999 | Value *End = EmitScalarExpr(E->getArg(1)); |
| 3000 | Function *F = CGM.getIntrinsic(Intrinsic::clear_cache); |
| 3001 | return RValue::get(Builder.CreateCall(F, {Begin, End})); |
| 3002 | } |
| 3003 | case Builtin::BI__builtin_trap: |
| 3004 | EmitTrapCall(Intrinsic::trap); |
| 3005 | return RValue::get(nullptr); |
| 3006 | case Builtin::BI__debugbreak: |
| 3007 | EmitTrapCall(Intrinsic::debugtrap); |
| 3008 | return RValue::get(nullptr); |
| 3009 | case Builtin::BI__builtin_unreachable: { |
| 3010 | EmitUnreachable(E->getExprLoc()); |
| 3011 | |
| 3012 | // We do need to preserve an insertion point. |
| 3013 | EmitBlock(createBasicBlock("unreachable.cont")); |
| 3014 | |
| 3015 | return RValue::get(nullptr); |
| 3016 | } |
| 3017 | |
| 3018 | case Builtin::BI__builtin_powi: |
| 3019 | case Builtin::BI__builtin_powif: |
| 3020 | case Builtin::BI__builtin_powil: { |
| 3021 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); |
| 3022 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); |
| 3023 | |
| 3024 | if (Builder.getIsFPConstrained()) { |
| 3025 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3026 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_powi, |
| 3027 | Src0->getType()); |
| 3028 | return RValue::get(Builder.CreateConstrainedFPCall(F, { Src0, Src1 })); |
| 3029 | } |
| 3030 | |
| 3031 | Function *F = CGM.getIntrinsic(Intrinsic::powi, |
| 3032 | { Src0->getType(), Src1->getType() }); |
| 3033 | return RValue::get(Builder.CreateCall(F, { Src0, Src1 })); |
| 3034 | } |
| 3035 | case Builtin::BI__builtin_isgreater: |
| 3036 | case Builtin::BI__builtin_isgreaterequal: |
| 3037 | case Builtin::BI__builtin_isless: |
| 3038 | case Builtin::BI__builtin_islessequal: |
| 3039 | case Builtin::BI__builtin_islessgreater: |
| 3040 | case Builtin::BI__builtin_isunordered: { |
| 3041 | // Ordered comparisons: we know the arguments to these are matching scalar |
| 3042 | // floating point values. |
| 3043 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3044 | Value *LHS = EmitScalarExpr(E->getArg(0)); |
| 3045 | Value *RHS = EmitScalarExpr(E->getArg(1)); |
| 3046 | |
| 3047 | switch (BuiltinID) { |
| 3048 | default: llvm_unreachable("Unknown ordered comparison")::llvm::llvm_unreachable_internal("Unknown ordered comparison" , "clang/lib/CodeGen/CGBuiltin.cpp", 3048); |
| 3049 | case Builtin::BI__builtin_isgreater: |
| 3050 | LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp"); |
| 3051 | break; |
| 3052 | case Builtin::BI__builtin_isgreaterequal: |
| 3053 | LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp"); |
| 3054 | break; |
| 3055 | case Builtin::BI__builtin_isless: |
| 3056 | LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp"); |
| 3057 | break; |
| 3058 | case Builtin::BI__builtin_islessequal: |
| 3059 | LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp"); |
| 3060 | break; |
| 3061 | case Builtin::BI__builtin_islessgreater: |
| 3062 | LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp"); |
| 3063 | break; |
| 3064 | case Builtin::BI__builtin_isunordered: |
| 3065 | LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp"); |
| 3066 | break; |
| 3067 | } |
| 3068 | // ZExt bool to int type. |
| 3069 | return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType()))); |
| 3070 | } |
| 3071 | case Builtin::BI__builtin_isnan: { |
| 3072 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3073 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 3074 | llvm::Type *Ty = V->getType(); |
| 3075 | const llvm::fltSemantics &Semantics = Ty->getFltSemantics(); |
| 3076 | if (!Builder.getIsFPConstrained() || |
| 3077 | Builder.getDefaultConstrainedExcept() == fp::ebIgnore || |
| 3078 | !Ty->isIEEE()) { |
| 3079 | V = Builder.CreateFCmpUNO(V, V, "cmp"); |
| 3080 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
| 3081 | } |
| 3082 | |
| 3083 | if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM)) |
| 3084 | return RValue::get(Result); |
| 3085 | |
| 3086 | // NaN has all exp bits set and a non zero significand. Therefore: |
| 3087 | // isnan(V) == ((exp mask - (abs(V) & exp mask)) < 0) |
| 3088 | unsigned bitsize = Ty->getScalarSizeInBits(); |
| 3089 | llvm::IntegerType *IntTy = Builder.getIntNTy(bitsize); |
| 3090 | Value *IntV = Builder.CreateBitCast(V, IntTy); |
| 3091 | APInt AndMask = APInt::getSignedMaxValue(bitsize); |
| 3092 | Value *AbsV = |
| 3093 | Builder.CreateAnd(IntV, llvm::ConstantInt::get(IntTy, AndMask)); |
| 3094 | APInt ExpMask = APFloat::getInf(Semantics).bitcastToAPInt(); |
| 3095 | Value *Sub = |
| 3096 | Builder.CreateSub(llvm::ConstantInt::get(IntTy, ExpMask), AbsV); |
| 3097 | // V = sign bit (Sub) <=> V = (Sub < 0) |
| 3098 | V = Builder.CreateLShr(Sub, llvm::ConstantInt::get(IntTy, bitsize - 1)); |
| 3099 | if (bitsize > 32) |
| 3100 | V = Builder.CreateTrunc(V, ConvertType(E->getType())); |
| 3101 | return RValue::get(V); |
| 3102 | } |
| 3103 | |
| 3104 | case Builtin::BI__builtin_nondeterministic_value: { |
| 3105 | llvm::Type *Ty = ConvertType(E->getArg(0)->getType()); |
| 3106 | |
| 3107 | Value *Result = PoisonValue::get(Ty); |
| 3108 | Result = Builder.CreateFreeze(Result); |
| 3109 | |
| 3110 | return RValue::get(Result); |
| 3111 | } |
| 3112 | |
| 3113 | case Builtin::BI__builtin_elementwise_abs: { |
| 3114 | Value *Result; |
| 3115 | QualType QT = E->getArg(0)->getType(); |
| 3116 | |
| 3117 | if (auto *VecTy = QT->getAs<VectorType>()) |
| 3118 | QT = VecTy->getElementType(); |
| 3119 | if (QT->isIntegerType()) |
| 3120 | Result = Builder.CreateBinaryIntrinsic( |
| 3121 | llvm::Intrinsic::abs, EmitScalarExpr(E->getArg(0)), |
| 3122 | Builder.getFalse(), nullptr, "elt.abs"); |
| 3123 | else |
| 3124 | Result = emitUnaryBuiltin(*this, E, llvm::Intrinsic::fabs, "elt.abs"); |
| 3125 | |
| 3126 | return RValue::get(Result); |
| 3127 | } |
| 3128 | |
| 3129 | case Builtin::BI__builtin_elementwise_ceil: |
| 3130 | return RValue::get( |
| 3131 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::ceil, "elt.ceil")); |
| 3132 | case Builtin::BI__builtin_elementwise_exp: |
| 3133 | return RValue::get( |
| 3134 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::exp, "elt.exp")); |
| 3135 | case Builtin::BI__builtin_elementwise_exp2: |
| 3136 | return RValue::get( |
| 3137 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::exp2, "elt.exp2")); |
| 3138 | case Builtin::BI__builtin_elementwise_log: |
| 3139 | return RValue::get( |
| 3140 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::log, "elt.log")); |
| 3141 | case Builtin::BI__builtin_elementwise_log2: |
| 3142 | return RValue::get( |
| 3143 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::log2, "elt.log2")); |
| 3144 | case Builtin::BI__builtin_elementwise_log10: |
| 3145 | return RValue::get( |
| 3146 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::log10, "elt.log10")); |
| 3147 | case Builtin::BI__builtin_elementwise_cos: |
| 3148 | return RValue::get( |
| 3149 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::cos, "elt.cos")); |
| 3150 | case Builtin::BI__builtin_elementwise_floor: |
| 3151 | return RValue::get( |
| 3152 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::floor, "elt.floor")); |
| 3153 | case Builtin::BI__builtin_elementwise_roundeven: |
| 3154 | return RValue::get(emitUnaryBuiltin(*this, E, llvm::Intrinsic::roundeven, |
| 3155 | "elt.roundeven")); |
| 3156 | case Builtin::BI__builtin_elementwise_sin: |
| 3157 | return RValue::get( |
| 3158 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::sin, "elt.sin")); |
| 3159 | |
| 3160 | case Builtin::BI__builtin_elementwise_trunc: |
| 3161 | return RValue::get( |
| 3162 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::trunc, "elt.trunc")); |
| 3163 | case Builtin::BI__builtin_elementwise_canonicalize: |
| 3164 | return RValue::get( |
| 3165 | emitUnaryBuiltin(*this, E, llvm::Intrinsic::canonicalize, "elt.trunc")); |
| 3166 | case Builtin::BI__builtin_elementwise_copysign: |
| 3167 | return RValue::get(emitBinaryBuiltin(*this, E, llvm::Intrinsic::copysign)); |
| 3168 | case Builtin::BI__builtin_elementwise_fma: |
| 3169 | return RValue::get(emitTernaryBuiltin(*this, E, llvm::Intrinsic::fma)); |
| 3170 | case Builtin::BI__builtin_elementwise_add_sat: |
| 3171 | case Builtin::BI__builtin_elementwise_sub_sat: { |
| 3172 | Value *Op0 = EmitScalarExpr(E->getArg(0)); |
| 3173 | Value *Op1 = EmitScalarExpr(E->getArg(1)); |
| 3174 | Value *Result; |
| 3175 | assert(Op0->getType()->isIntOrIntVectorTy() && "integer type expected")(static_cast <bool> (Op0->getType()->isIntOrIntVectorTy () && "integer type expected") ? void (0) : __assert_fail ("Op0->getType()->isIntOrIntVectorTy() && \"integer type expected\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3175, __extension__ __PRETTY_FUNCTION__ )); |
| 3176 | QualType Ty = E->getArg(0)->getType(); |
| 3177 | if (auto *VecTy = Ty->getAs<VectorType>()) |
| 3178 | Ty = VecTy->getElementType(); |
| 3179 | bool IsSigned = Ty->isSignedIntegerType(); |
| 3180 | unsigned Opc; |
| 3181 | if (BuiltinIDIfNoAsmLabel == Builtin::BI__builtin_elementwise_add_sat) |
| 3182 | Opc = IsSigned ? llvm::Intrinsic::sadd_sat : llvm::Intrinsic::uadd_sat; |
| 3183 | else |
| 3184 | Opc = IsSigned ? llvm::Intrinsic::ssub_sat : llvm::Intrinsic::usub_sat; |
| 3185 | Result = Builder.CreateBinaryIntrinsic(Opc, Op0, Op1, nullptr, "elt.sat"); |
| 3186 | return RValue::get(Result); |
| 3187 | } |
| 3188 | |
| 3189 | case Builtin::BI__builtin_elementwise_max: { |
| 3190 | Value *Op0 = EmitScalarExpr(E->getArg(0)); |
| 3191 | Value *Op1 = EmitScalarExpr(E->getArg(1)); |
| 3192 | Value *Result; |
| 3193 | if (Op0->getType()->isIntOrIntVectorTy()) { |
| 3194 | QualType Ty = E->getArg(0)->getType(); |
| 3195 | if (auto *VecTy = Ty->getAs<VectorType>()) |
| 3196 | Ty = VecTy->getElementType(); |
| 3197 | Result = Builder.CreateBinaryIntrinsic(Ty->isSignedIntegerType() |
| 3198 | ? llvm::Intrinsic::smax |
| 3199 | : llvm::Intrinsic::umax, |
| 3200 | Op0, Op1, nullptr, "elt.max"); |
| 3201 | } else |
| 3202 | Result = Builder.CreateMaxNum(Op0, Op1, "elt.max"); |
| 3203 | return RValue::get(Result); |
| 3204 | } |
| 3205 | case Builtin::BI__builtin_elementwise_min: { |
| 3206 | Value *Op0 = EmitScalarExpr(E->getArg(0)); |
| 3207 | Value *Op1 = EmitScalarExpr(E->getArg(1)); |
| 3208 | Value *Result; |
| 3209 | if (Op0->getType()->isIntOrIntVectorTy()) { |
| 3210 | QualType Ty = E->getArg(0)->getType(); |
| 3211 | if (auto *VecTy = Ty->getAs<VectorType>()) |
| 3212 | Ty = VecTy->getElementType(); |
| 3213 | Result = Builder.CreateBinaryIntrinsic(Ty->isSignedIntegerType() |
| 3214 | ? llvm::Intrinsic::smin |
| 3215 | : llvm::Intrinsic::umin, |
| 3216 | Op0, Op1, nullptr, "elt.min"); |
| 3217 | } else |
| 3218 | Result = Builder.CreateMinNum(Op0, Op1, "elt.min"); |
| 3219 | return RValue::get(Result); |
| 3220 | } |
| 3221 | |
| 3222 | case Builtin::BI__builtin_reduce_max: { |
| 3223 | auto GetIntrinsicID = [](QualType QT) { |
| 3224 | if (auto *VecTy = QT->getAs<VectorType>()) |
| 3225 | QT = VecTy->getElementType(); |
| 3226 | if (QT->isSignedIntegerType()) |
| 3227 | return llvm::Intrinsic::vector_reduce_smax; |
| 3228 | if (QT->isUnsignedIntegerType()) |
| 3229 | return llvm::Intrinsic::vector_reduce_umax; |
| 3230 | assert(QT->isFloatingType() && "must have a float here")(static_cast <bool> (QT->isFloatingType() && "must have a float here") ? void (0) : __assert_fail ("QT->isFloatingType() && \"must have a float here\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3230, __extension__ __PRETTY_FUNCTION__ )); |
| 3231 | return llvm::Intrinsic::vector_reduce_fmax; |
| 3232 | }; |
| 3233 | return RValue::get(emitUnaryBuiltin( |
| 3234 | *this, E, GetIntrinsicID(E->getArg(0)->getType()), "rdx.min")); |
| 3235 | } |
| 3236 | |
| 3237 | case Builtin::BI__builtin_reduce_min: { |
| 3238 | auto GetIntrinsicID = [](QualType QT) { |
| 3239 | if (auto *VecTy = QT->getAs<VectorType>()) |
| 3240 | QT = VecTy->getElementType(); |
| 3241 | if (QT->isSignedIntegerType()) |
| 3242 | return llvm::Intrinsic::vector_reduce_smin; |
| 3243 | if (QT->isUnsignedIntegerType()) |
| 3244 | return llvm::Intrinsic::vector_reduce_umin; |
| 3245 | assert(QT->isFloatingType() && "must have a float here")(static_cast <bool> (QT->isFloatingType() && "must have a float here") ? void (0) : __assert_fail ("QT->isFloatingType() && \"must have a float here\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3245, __extension__ __PRETTY_FUNCTION__ )); |
| 3246 | return llvm::Intrinsic::vector_reduce_fmin; |
| 3247 | }; |
| 3248 | |
| 3249 | return RValue::get(emitUnaryBuiltin( |
| 3250 | *this, E, GetIntrinsicID(E->getArg(0)->getType()), "rdx.min")); |
| 3251 | } |
| 3252 | |
| 3253 | case Builtin::BI__builtin_reduce_add: |
| 3254 | return RValue::get(emitUnaryBuiltin( |
| 3255 | *this, E, llvm::Intrinsic::vector_reduce_add, "rdx.add")); |
| 3256 | case Builtin::BI__builtin_reduce_mul: |
| 3257 | return RValue::get(emitUnaryBuiltin( |
| 3258 | *this, E, llvm::Intrinsic::vector_reduce_mul, "rdx.mul")); |
| 3259 | case Builtin::BI__builtin_reduce_xor: |
| 3260 | return RValue::get(emitUnaryBuiltin( |
| 3261 | *this, E, llvm::Intrinsic::vector_reduce_xor, "rdx.xor")); |
| 3262 | case Builtin::BI__builtin_reduce_or: |
| 3263 | return RValue::get(emitUnaryBuiltin( |
| 3264 | *this, E, llvm::Intrinsic::vector_reduce_or, "rdx.or")); |
| 3265 | case Builtin::BI__builtin_reduce_and: |
| 3266 | return RValue::get(emitUnaryBuiltin( |
| 3267 | *this, E, llvm::Intrinsic::vector_reduce_and, "rdx.and")); |
| 3268 | |
| 3269 | case Builtin::BI__builtin_matrix_transpose: { |
| 3270 | auto *MatrixTy = E->getArg(0)->getType()->castAs<ConstantMatrixType>(); |
| 3271 | Value *MatValue = EmitScalarExpr(E->getArg(0)); |
| 3272 | MatrixBuilder MB(Builder); |
| 3273 | Value *Result = MB.CreateMatrixTranspose(MatValue, MatrixTy->getNumRows(), |
| 3274 | MatrixTy->getNumColumns()); |
| 3275 | return RValue::get(Result); |
| 3276 | } |
| 3277 | |
| 3278 | case Builtin::BI__builtin_matrix_column_major_load: { |
| 3279 | MatrixBuilder MB(Builder); |
| 3280 | // Emit everything that isn't dependent on the first parameter type |
| 3281 | Value *Stride = EmitScalarExpr(E->getArg(3)); |
| 3282 | const auto *ResultTy = E->getType()->getAs<ConstantMatrixType>(); |
| 3283 | auto *PtrTy = E->getArg(0)->getType()->getAs<PointerType>(); |
| 3284 | assert(PtrTy && "arg0 must be of pointer type")(static_cast <bool> (PtrTy && "arg0 must be of pointer type" ) ? void (0) : __assert_fail ("PtrTy && \"arg0 must be of pointer type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3284, __extension__ __PRETTY_FUNCTION__ )); |
| 3285 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); |
| 3286 | |
| 3287 | Address Src = EmitPointerWithAlignment(E->getArg(0)); |
| 3288 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(0)->getType(), |
| 3289 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3290 | Value *Result = MB.CreateColumnMajorLoad( |
| 3291 | Src.getElementType(), Src.getPointer(), |
| 3292 | Align(Src.getAlignment().getQuantity()), Stride, IsVolatile, |
| 3293 | ResultTy->getNumRows(), ResultTy->getNumColumns(), |
| 3294 | "matrix"); |
| 3295 | return RValue::get(Result); |
| 3296 | } |
| 3297 | |
| 3298 | case Builtin::BI__builtin_matrix_column_major_store: { |
| 3299 | MatrixBuilder MB(Builder); |
| 3300 | Value *Matrix = EmitScalarExpr(E->getArg(0)); |
| 3301 | Address Dst = EmitPointerWithAlignment(E->getArg(1)); |
| 3302 | Value *Stride = EmitScalarExpr(E->getArg(2)); |
| 3303 | |
| 3304 | const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>(); |
| 3305 | auto *PtrTy = E->getArg(1)->getType()->getAs<PointerType>(); |
| 3306 | assert(PtrTy && "arg1 must be of pointer type")(static_cast <bool> (PtrTy && "arg1 must be of pointer type" ) ? void (0) : __assert_fail ("PtrTy && \"arg1 must be of pointer type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3306, __extension__ __PRETTY_FUNCTION__ )); |
| 3307 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); |
| 3308 | |
| 3309 | EmitNonNullArgCheck(RValue::get(Dst.getPointer()), E->getArg(1)->getType(), |
| 3310 | E->getArg(1)->getExprLoc(), FD, 0); |
| 3311 | Value *Result = MB.CreateColumnMajorStore( |
| 3312 | Matrix, Dst.getPointer(), Align(Dst.getAlignment().getQuantity()), |
| 3313 | Stride, IsVolatile, MatrixTy->getNumRows(), MatrixTy->getNumColumns()); |
| 3314 | return RValue::get(Result); |
| 3315 | } |
| 3316 | |
| 3317 | case Builtin::BIfinite: |
| 3318 | case Builtin::BI__finite: |
| 3319 | case Builtin::BIfinitef: |
| 3320 | case Builtin::BI__finitef: |
| 3321 | case Builtin::BIfinitel: |
| 3322 | case Builtin::BI__finitel: |
| 3323 | case Builtin::BI__builtin_isinf: |
| 3324 | case Builtin::BI__builtin_isfinite: { |
| 3325 | // isinf(x) --> fabs(x) == infinity |
| 3326 | // isfinite(x) --> fabs(x) != infinity |
| 3327 | // x != NaN via the ordered compare in either case. |
| 3328 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3329 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 3330 | llvm::Type *Ty = V->getType(); |
| 3331 | if (!Builder.getIsFPConstrained() || |
| 3332 | Builder.getDefaultConstrainedExcept() == fp::ebIgnore || |
| 3333 | !Ty->isIEEE()) { |
| 3334 | Value *Fabs = EmitFAbs(*this, V); |
| 3335 | Constant *Infinity = ConstantFP::getInfinity(V->getType()); |
| 3336 | CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf) |
| 3337 | ? CmpInst::FCMP_OEQ |
| 3338 | : CmpInst::FCMP_ONE; |
| 3339 | Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf"); |
| 3340 | return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType()))); |
| 3341 | } |
| 3342 | |
| 3343 | if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM)) |
| 3344 | return RValue::get(Result); |
| 3345 | |
| 3346 | // Inf values have all exp bits set and a zero significand. Therefore: |
| 3347 | // isinf(V) == ((V << 1) == ((exp mask) << 1)) |
| 3348 | // isfinite(V) == ((V << 1) < ((exp mask) << 1)) using unsigned comparison |
| 3349 | unsigned bitsize = Ty->getScalarSizeInBits(); |
| 3350 | llvm::IntegerType *IntTy = Builder.getIntNTy(bitsize); |
| 3351 | Value *IntV = Builder.CreateBitCast(V, IntTy); |
| 3352 | Value *Shl1 = Builder.CreateShl(IntV, 1); |
| 3353 | const llvm::fltSemantics &Semantics = Ty->getFltSemantics(); |
| 3354 | APInt ExpMask = APFloat::getInf(Semantics).bitcastToAPInt(); |
| 3355 | Value *ExpMaskShl1 = llvm::ConstantInt::get(IntTy, ExpMask.shl(1)); |
| 3356 | if (BuiltinID == Builtin::BI__builtin_isinf) |
| 3357 | V = Builder.CreateICmpEQ(Shl1, ExpMaskShl1); |
| 3358 | else |
| 3359 | V = Builder.CreateICmpULT(Shl1, ExpMaskShl1); |
| 3360 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
| 3361 | } |
| 3362 | |
| 3363 | case Builtin::BI__builtin_isinf_sign: { |
| 3364 | // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0 |
| 3365 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3366 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
| 3367 | Value *Arg = EmitScalarExpr(E->getArg(0)); |
| 3368 | Value *AbsArg = EmitFAbs(*this, Arg); |
| 3369 | Value *IsInf = Builder.CreateFCmpOEQ( |
| 3370 | AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf"); |
| 3371 | Value *IsNeg = EmitSignBit(*this, Arg); |
| 3372 | |
| 3373 | llvm::Type *IntTy = ConvertType(E->getType()); |
| 3374 | Value *Zero = Constant::getNullValue(IntTy); |
| 3375 | Value *One = ConstantInt::get(IntTy, 1); |
| 3376 | Value *NegativeOne = ConstantInt::get(IntTy, -1); |
| 3377 | Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One); |
| 3378 | Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero); |
| 3379 | return RValue::get(Result); |
| 3380 | } |
| 3381 | |
| 3382 | case Builtin::BI__builtin_isnormal: { |
| 3383 | // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min |
| 3384 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3385 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
| 3386 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 3387 | Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq"); |
| 3388 | |
| 3389 | Value *Abs = EmitFAbs(*this, V); |
| 3390 | Value *IsLessThanInf = |
| 3391 | Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf"); |
| 3392 | APFloat Smallest = APFloat::getSmallestNormalized( |
| 3393 | getContext().getFloatTypeSemantics(E->getArg(0)->getType())); |
| 3394 | Value *IsNormal = |
| 3395 | Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest), |
| 3396 | "isnormal"); |
| 3397 | V = Builder.CreateAnd(Eq, IsLessThanInf, "and"); |
| 3398 | V = Builder.CreateAnd(V, IsNormal, "and"); |
| 3399 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
| 3400 | } |
| 3401 | |
| 3402 | case Builtin::BI__builtin_flt_rounds: { |
| 3403 | Function *F = CGM.getIntrinsic(Intrinsic::get_rounding); |
| 3404 | |
| 3405 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 3406 | Value *Result = Builder.CreateCall(F); |
| 3407 | if (Result->getType() != ResultType) |
| 3408 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 3409 | "cast"); |
| 3410 | return RValue::get(Result); |
| 3411 | } |
| 3412 | |
| 3413 | case Builtin::BI__builtin_set_flt_rounds: { |
| 3414 | Function *F = CGM.getIntrinsic(Intrinsic::set_rounding); |
| 3415 | |
| 3416 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 3417 | Builder.CreateCall(F, V); |
| 3418 | return RValue::get(nullptr); |
| 3419 | } |
| 3420 | |
| 3421 | case Builtin::BI__builtin_fpclassify: { |
| 3422 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3423 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
| 3424 | Value *V = EmitScalarExpr(E->getArg(5)); |
| 3425 | llvm::Type *Ty = ConvertType(E->getArg(5)->getType()); |
| 3426 | |
| 3427 | // Create Result |
| 3428 | BasicBlock *Begin = Builder.GetInsertBlock(); |
| 3429 | BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn); |
| 3430 | Builder.SetInsertPoint(End); |
| 3431 | PHINode *Result = |
| 3432 | Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4, |
| 3433 | "fpclassify_result"); |
| 3434 | |
| 3435 | // if (V==0) return FP_ZERO |
| 3436 | Builder.SetInsertPoint(Begin); |
| 3437 | Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty), |
| 3438 | "iszero"); |
| 3439 | Value *ZeroLiteral = EmitScalarExpr(E->getArg(4)); |
| 3440 | BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn); |
| 3441 | Builder.CreateCondBr(IsZero, End, NotZero); |
| 3442 | Result->addIncoming(ZeroLiteral, Begin); |
| 3443 | |
| 3444 | // if (V != V) return FP_NAN |
| 3445 | Builder.SetInsertPoint(NotZero); |
| 3446 | Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp"); |
| 3447 | Value *NanLiteral = EmitScalarExpr(E->getArg(0)); |
| 3448 | BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn); |
| 3449 | Builder.CreateCondBr(IsNan, End, NotNan); |
| 3450 | Result->addIncoming(NanLiteral, NotZero); |
| 3451 | |
| 3452 | // if (fabs(V) == infinity) return FP_INFINITY |
| 3453 | Builder.SetInsertPoint(NotNan); |
| 3454 | Value *VAbs = EmitFAbs(*this, V); |
| 3455 | Value *IsInf = |
| 3456 | Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()), |
| 3457 | "isinf"); |
| 3458 | Value *InfLiteral = EmitScalarExpr(E->getArg(1)); |
| 3459 | BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn); |
| 3460 | Builder.CreateCondBr(IsInf, End, NotInf); |
| 3461 | Result->addIncoming(InfLiteral, NotNan); |
| 3462 | |
| 3463 | // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL |
| 3464 | Builder.SetInsertPoint(NotInf); |
| 3465 | APFloat Smallest = APFloat::getSmallestNormalized( |
| 3466 | getContext().getFloatTypeSemantics(E->getArg(5)->getType())); |
| 3467 | Value *IsNormal = |
| 3468 | Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest), |
| 3469 | "isnormal"); |
| 3470 | Value *NormalResult = |
| 3471 | Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)), |
| 3472 | EmitScalarExpr(E->getArg(3))); |
| 3473 | Builder.CreateBr(End); |
| 3474 | Result->addIncoming(NormalResult, NotInf); |
| 3475 | |
| 3476 | // return Result |
| 3477 | Builder.SetInsertPoint(End); |
| 3478 | return RValue::get(Result); |
| 3479 | } |
| 3480 | |
| 3481 | case Builtin::BIalloca: |
| 3482 | case Builtin::BI_alloca: |
| 3483 | case Builtin::BI__builtin_alloca_uninitialized: |
| 3484 | case Builtin::BI__builtin_alloca: { |
| 3485 | Value *Size = EmitScalarExpr(E->getArg(0)); |
| 3486 | const TargetInfo &TI = getContext().getTargetInfo(); |
| 3487 | // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__. |
| 3488 | const Align SuitableAlignmentInBytes = |
| 3489 | CGM.getContext() |
| 3490 | .toCharUnitsFromBits(TI.getSuitableAlign()) |
| 3491 | .getAsAlign(); |
| 3492 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
| 3493 | AI->setAlignment(SuitableAlignmentInBytes); |
| 3494 | if (BuiltinID != Builtin::BI__builtin_alloca_uninitialized) |
| 3495 | initializeAlloca(*this, AI, Size, SuitableAlignmentInBytes); |
| 3496 | return RValue::get(AI); |
| 3497 | } |
| 3498 | |
| 3499 | case Builtin::BI__builtin_alloca_with_align_uninitialized: |
| 3500 | case Builtin::BI__builtin_alloca_with_align: { |
| 3501 | Value *Size = EmitScalarExpr(E->getArg(0)); |
| 3502 | Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1)); |
| 3503 | auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue); |
| 3504 | unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue(); |
| 3505 | const Align AlignmentInBytes = |
| 3506 | CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getAsAlign(); |
| 3507 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
| 3508 | AI->setAlignment(AlignmentInBytes); |
| 3509 | if (BuiltinID != Builtin::BI__builtin_alloca_with_align_uninitialized) |
| 3510 | initializeAlloca(*this, AI, Size, AlignmentInBytes); |
| 3511 | return RValue::get(AI); |
| 3512 | } |
| 3513 | |
| 3514 | case Builtin::BIbzero: |
| 3515 | case Builtin::BI__builtin_bzero: { |
| 3516 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3517 | Value *SizeVal = EmitScalarExpr(E->getArg(1)); |
| 3518 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3519 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3520 | Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false); |
| 3521 | return RValue::get(nullptr); |
| 3522 | } |
| 3523 | case Builtin::BImemcpy: |
| 3524 | case Builtin::BI__builtin_memcpy: |
| 3525 | case Builtin::BImempcpy: |
| 3526 | case Builtin::BI__builtin_mempcpy: { |
| 3527 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3528 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3529 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3530 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3531 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3532 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 3533 | E->getArg(1)->getExprLoc(), FD, 1); |
| 3534 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
| 3535 | if (BuiltinID == Builtin::BImempcpy || |
| 3536 | BuiltinID == Builtin::BI__builtin_mempcpy) |
| 3537 | return RValue::get(Builder.CreateInBoundsGEP(Dest.getElementType(), |
| 3538 | Dest.getPointer(), SizeVal)); |
| 3539 | else |
| 3540 | return RValue::get(Dest.getPointer()); |
| 3541 | } |
| 3542 | |
| 3543 | case Builtin::BI__builtin_memcpy_inline: { |
| 3544 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3545 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3546 | uint64_t Size = |
| 3547 | E->getArg(2)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
| 3548 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3549 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3550 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 3551 | E->getArg(1)->getExprLoc(), FD, 1); |
| 3552 | Builder.CreateMemCpyInline(Dest, Src, Size); |
| 3553 | return RValue::get(nullptr); |
| 3554 | } |
| 3555 | |
| 3556 | case Builtin::BI__builtin_char_memchr: |
| 3557 | BuiltinID = Builtin::BI__builtin_memchr; |
| 3558 | break; |
| 3559 | |
| 3560 | case Builtin::BI__builtin___memcpy_chk: { |
| 3561 | // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2. |
| 3562 | Expr::EvalResult SizeResult, DstSizeResult; |
| 3563 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 3564 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 3565 | break; |
| 3566 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 3567 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 3568 | if (Size.ugt(DstSize)) |
| 3569 | break; |
| 3570 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3571 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3572 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 3573 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
| 3574 | return RValue::get(Dest.getPointer()); |
| 3575 | } |
| 3576 | |
| 3577 | case Builtin::BI__builtin_objc_memmove_collectable: { |
| 3578 | Address DestAddr = EmitPointerWithAlignment(E->getArg(0)); |
| 3579 | Address SrcAddr = EmitPointerWithAlignment(E->getArg(1)); |
| 3580 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3581 | CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, |
| 3582 | DestAddr, SrcAddr, SizeVal); |
| 3583 | return RValue::get(DestAddr.getPointer()); |
| 3584 | } |
| 3585 | |
| 3586 | case Builtin::BI__builtin___memmove_chk: { |
| 3587 | // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2. |
| 3588 | Expr::EvalResult SizeResult, DstSizeResult; |
| 3589 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 3590 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 3591 | break; |
| 3592 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 3593 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 3594 | if (Size.ugt(DstSize)) |
| 3595 | break; |
| 3596 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3597 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3598 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 3599 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
| 3600 | return RValue::get(Dest.getPointer()); |
| 3601 | } |
| 3602 | |
| 3603 | case Builtin::BImemmove: |
| 3604 | case Builtin::BI__builtin_memmove: { |
| 3605 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3606 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3607 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3608 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3609 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3610 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 3611 | E->getArg(1)->getExprLoc(), FD, 1); |
| 3612 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
| 3613 | return RValue::get(Dest.getPointer()); |
| 3614 | } |
| 3615 | case Builtin::BImemset: |
| 3616 | case Builtin::BI__builtin_memset: { |
| 3617 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3618 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
| 3619 | Builder.getInt8Ty()); |
| 3620 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3621 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3622 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3623 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
| 3624 | return RValue::get(Dest.getPointer()); |
| 3625 | } |
| 3626 | case Builtin::BI__builtin_memset_inline: { |
| 3627 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3628 | Value *ByteVal = |
| 3629 | Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), Builder.getInt8Ty()); |
| 3630 | uint64_t Size = |
| 3631 | E->getArg(2)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
| 3632 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3633 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3634 | Builder.CreateMemSetInline(Dest, ByteVal, Size); |
| 3635 | return RValue::get(nullptr); |
| 3636 | } |
| 3637 | case Builtin::BI__builtin___memset_chk: { |
| 3638 | // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2. |
| 3639 | Expr::EvalResult SizeResult, DstSizeResult; |
| 3640 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 3641 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 3642 | break; |
| 3643 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 3644 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 3645 | if (Size.ugt(DstSize)) |
| 3646 | break; |
| 3647 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3648 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
| 3649 | Builder.getInt8Ty()); |
| 3650 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 3651 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
| 3652 | return RValue::get(Dest.getPointer()); |
| 3653 | } |
| 3654 | case Builtin::BI__builtin_wmemchr: { |
| 3655 | // The MSVC runtime library does not provide a definition of wmemchr, so we |
| 3656 | // need an inline implementation. |
| 3657 | if (!getTarget().getTriple().isOSMSVCRT()) |
| 3658 | break; |
| 3659 | |
| 3660 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); |
| 3661 | Value *Str = EmitScalarExpr(E->getArg(0)); |
| 3662 | Value *Chr = EmitScalarExpr(E->getArg(1)); |
| 3663 | Value *Size = EmitScalarExpr(E->getArg(2)); |
| 3664 | |
| 3665 | BasicBlock *Entry = Builder.GetInsertBlock(); |
| 3666 | BasicBlock *CmpEq = createBasicBlock("wmemchr.eq"); |
| 3667 | BasicBlock *Next = createBasicBlock("wmemchr.next"); |
| 3668 | BasicBlock *Exit = createBasicBlock("wmemchr.exit"); |
| 3669 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); |
| 3670 | Builder.CreateCondBr(SizeEq0, Exit, CmpEq); |
| 3671 | |
| 3672 | EmitBlock(CmpEq); |
| 3673 | PHINode *StrPhi = Builder.CreatePHI(Str->getType(), 2); |
| 3674 | StrPhi->addIncoming(Str, Entry); |
| 3675 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); |
| 3676 | SizePhi->addIncoming(Size, Entry); |
| 3677 | CharUnits WCharAlign = |
| 3678 | getContext().getTypeAlignInChars(getContext().WCharTy); |
| 3679 | Value *StrCh = Builder.CreateAlignedLoad(WCharTy, StrPhi, WCharAlign); |
| 3680 | Value *FoundChr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 0); |
| 3681 | Value *StrEqChr = Builder.CreateICmpEQ(StrCh, Chr); |
| 3682 | Builder.CreateCondBr(StrEqChr, Exit, Next); |
| 3683 | |
| 3684 | EmitBlock(Next); |
| 3685 | Value *NextStr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 1); |
| 3686 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); |
| 3687 | Value *NextSizeEq0 = |
| 3688 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); |
| 3689 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpEq); |
| 3690 | StrPhi->addIncoming(NextStr, Next); |
| 3691 | SizePhi->addIncoming(NextSize, Next); |
| 3692 | |
| 3693 | EmitBlock(Exit); |
| 3694 | PHINode *Ret = Builder.CreatePHI(Str->getType(), 3); |
| 3695 | Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Entry); |
| 3696 | Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Next); |
| 3697 | Ret->addIncoming(FoundChr, CmpEq); |
| 3698 | return RValue::get(Ret); |
| 3699 | } |
| 3700 | case Builtin::BI__builtin_wmemcmp: { |
| 3701 | // The MSVC runtime library does not provide a definition of wmemcmp, so we |
| 3702 | // need an inline implementation. |
| 3703 | if (!getTarget().getTriple().isOSMSVCRT()) |
| 3704 | break; |
| 3705 | |
| 3706 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); |
| 3707 | |
| 3708 | Value *Dst = EmitScalarExpr(E->getArg(0)); |
| 3709 | Value *Src = EmitScalarExpr(E->getArg(1)); |
| 3710 | Value *Size = EmitScalarExpr(E->getArg(2)); |
| 3711 | |
| 3712 | BasicBlock *Entry = Builder.GetInsertBlock(); |
| 3713 | BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt"); |
| 3714 | BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt"); |
| 3715 | BasicBlock *Next = createBasicBlock("wmemcmp.next"); |
| 3716 | BasicBlock *Exit = createBasicBlock("wmemcmp.exit"); |
| 3717 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); |
| 3718 | Builder.CreateCondBr(SizeEq0, Exit, CmpGT); |
| 3719 | |
| 3720 | EmitBlock(CmpGT); |
| 3721 | PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2); |
| 3722 | DstPhi->addIncoming(Dst, Entry); |
| 3723 | PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2); |
| 3724 | SrcPhi->addIncoming(Src, Entry); |
| 3725 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); |
| 3726 | SizePhi->addIncoming(Size, Entry); |
| 3727 | CharUnits WCharAlign = |
| 3728 | getContext().getTypeAlignInChars(getContext().WCharTy); |
| 3729 | Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign); |
| 3730 | Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign); |
| 3731 | Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh); |
| 3732 | Builder.CreateCondBr(DstGtSrc, Exit, CmpLT); |
| 3733 | |
| 3734 | EmitBlock(CmpLT); |
| 3735 | Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh); |
| 3736 | Builder.CreateCondBr(DstLtSrc, Exit, Next); |
| 3737 | |
| 3738 | EmitBlock(Next); |
| 3739 | Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1); |
| 3740 | Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1); |
| 3741 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); |
| 3742 | Value *NextSizeEq0 = |
| 3743 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); |
| 3744 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT); |
| 3745 | DstPhi->addIncoming(NextDst, Next); |
| 3746 | SrcPhi->addIncoming(NextSrc, Next); |
| 3747 | SizePhi->addIncoming(NextSize, Next); |
| 3748 | |
| 3749 | EmitBlock(Exit); |
| 3750 | PHINode *Ret = Builder.CreatePHI(IntTy, 4); |
| 3751 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry); |
| 3752 | Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT); |
| 3753 | Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT); |
| 3754 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Next); |
| 3755 | return RValue::get(Ret); |
| 3756 | } |
| 3757 | case Builtin::BI__builtin_dwarf_cfa: { |
| 3758 | // The offset in bytes from the first argument to the CFA. |
| 3759 | // |
| 3760 | // Why on earth is this in the frontend? Is there any reason at |
| 3761 | // all that the backend can't reasonably determine this while |
| 3762 | // lowering llvm.eh.dwarf.cfa()? |
| 3763 | // |
| 3764 | // TODO: If there's a satisfactory reason, add a target hook for |
| 3765 | // this instead of hard-coding 0, which is correct for most targets. |
| 3766 | int32_t Offset = 0; |
| 3767 | |
| 3768 | Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa); |
| 3769 | return RValue::get(Builder.CreateCall(F, |
| 3770 | llvm::ConstantInt::get(Int32Ty, Offset))); |
| 3771 | } |
| 3772 | case Builtin::BI__builtin_return_address: { |
| 3773 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
| 3774 | getContext().UnsignedIntTy); |
| 3775 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
| 3776 | return RValue::get(Builder.CreateCall(F, Depth)); |
| 3777 | } |
| 3778 | case Builtin::BI_ReturnAddress: { |
| 3779 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
| 3780 | return RValue::get(Builder.CreateCall(F, Builder.getInt32(0))); |
| 3781 | } |
| 3782 | case Builtin::BI__builtin_frame_address: { |
| 3783 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
| 3784 | getContext().UnsignedIntTy); |
| 3785 | Function *F = CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy); |
| 3786 | return RValue::get(Builder.CreateCall(F, Depth)); |
| 3787 | } |
| 3788 | case Builtin::BI__builtin_extract_return_addr: { |
| 3789 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 3790 | Value *Result = getTargetHooks().decodeReturnAddress(*this, Address); |
| 3791 | return RValue::get(Result); |
| 3792 | } |
| 3793 | case Builtin::BI__builtin_frob_return_addr: { |
| 3794 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 3795 | Value *Result = getTargetHooks().encodeReturnAddress(*this, Address); |
| 3796 | return RValue::get(Result); |
| 3797 | } |
| 3798 | case Builtin::BI__builtin_dwarf_sp_column: { |
| 3799 | llvm::IntegerType *Ty |
| 3800 | = cast<llvm::IntegerType>(ConvertType(E->getType())); |
| 3801 | int Column = getTargetHooks().getDwarfEHStackPointer(CGM); |
| 3802 | if (Column == -1) { |
| 3803 | CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column"); |
| 3804 | return RValue::get(llvm::UndefValue::get(Ty)); |
| 3805 | } |
| 3806 | return RValue::get(llvm::ConstantInt::get(Ty, Column, true)); |
| 3807 | } |
| 3808 | case Builtin::BI__builtin_init_dwarf_reg_size_table: { |
| 3809 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 3810 | if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address)) |
| 3811 | CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table"); |
| 3812 | return RValue::get(llvm::UndefValue::get(ConvertType(E->getType()))); |
| 3813 | } |
| 3814 | case Builtin::BI__builtin_eh_return: { |
| 3815 | Value *Int = EmitScalarExpr(E->getArg(0)); |
| 3816 | Value *Ptr = EmitScalarExpr(E->getArg(1)); |
| 3817 | |
| 3818 | llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType()); |
| 3819 | assert((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) &&(static_cast <bool> ((IntTy->getBitWidth() == 32 || IntTy ->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? void (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3820, __extension__ __PRETTY_FUNCTION__ )) |
| 3820 | "LLVM's __builtin_eh_return only supports 32- and 64-bit variants")(static_cast <bool> ((IntTy->getBitWidth() == 32 || IntTy ->getBitWidth() == 64) && "LLVM's __builtin_eh_return only supports 32- and 64-bit variants" ) ? void (0) : __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 3820, __extension__ __PRETTY_FUNCTION__ )); |
| 3821 | Function *F = |
| 3822 | CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32 |
| 3823 | : Intrinsic::eh_return_i64); |
| 3824 | Builder.CreateCall(F, {Int, Ptr}); |
| 3825 | Builder.CreateUnreachable(); |
| 3826 | |
| 3827 | // We do need to preserve an insertion point. |
| 3828 | EmitBlock(createBasicBlock("builtin_eh_return.cont")); |
| 3829 | |
| 3830 | return RValue::get(nullptr); |
| 3831 | } |
| 3832 | case Builtin::BI__builtin_unwind_init: { |
| 3833 | Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init); |
| 3834 | Builder.CreateCall(F); |
| 3835 | return RValue::get(nullptr); |
| 3836 | } |
| 3837 | case Builtin::BI__builtin_extend_pointer: { |
| 3838 | // Extends a pointer to the size of an _Unwind_Word, which is |
| 3839 | // uint64_t on all platforms. Generally this gets poked into a |
| 3840 | // register and eventually used as an address, so if the |
| 3841 | // addressing registers are wider than pointers and the platform |
| 3842 | // doesn't implicitly ignore high-order bits when doing |
| 3843 | // addressing, we need to make sure we zext / sext based on |
| 3844 | // the platform's expectations. |
| 3845 | // |
| 3846 | // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html |
| 3847 | |
| 3848 | // Cast the pointer to intptr_t. |
| 3849 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 3850 | Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast"); |
| 3851 | |
| 3852 | // If that's 64 bits, we're done. |
| 3853 | if (IntPtrTy->getBitWidth() == 64) |
| 3854 | return RValue::get(Result); |
| 3855 | |
| 3856 | // Otherwise, ask the codegen data what to do. |
| 3857 | if (getTargetHooks().extendPointerWithSExt()) |
| 3858 | return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext")); |
| 3859 | else |
| 3860 | return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext")); |
| 3861 | } |
| 3862 | case Builtin::BI__builtin_setjmp: { |
| 3863 | // Buffer is a void**. |
| 3864 | Address Buf = EmitPointerWithAlignment(E->getArg(0)); |
| 3865 | |
| 3866 | // Store the frame pointer to the setjmp buffer. |
| 3867 | Value *FrameAddr = Builder.CreateCall( |
| 3868 | CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy), |
| 3869 | ConstantInt::get(Int32Ty, 0)); |
| 3870 | Builder.CreateStore(FrameAddr, Buf); |
| 3871 | |
| 3872 | // Store the stack pointer to the setjmp buffer. |
| 3873 | Value *StackAddr = |
| 3874 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave)); |
| 3875 | Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2); |
| 3876 | Builder.CreateStore(StackAddr, StackSaveSlot); |
| 3877 | |
| 3878 | // Call LLVM's EH setjmp, which is lightweight. |
| 3879 | Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp); |
| 3880 | Buf = Builder.CreateElementBitCast(Buf, Int8Ty); |
| 3881 | return RValue::get(Builder.CreateCall(F, Buf.getPointer())); |
| 3882 | } |
| 3883 | case Builtin::BI__builtin_longjmp: { |
| 3884 | Value *Buf = EmitScalarExpr(E->getArg(0)); |
| 3885 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); |
| 3886 | |
| 3887 | // Call LLVM's EH longjmp, which is lightweight. |
| 3888 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf); |
| 3889 | |
| 3890 | // longjmp doesn't return; mark this as unreachable. |
| 3891 | Builder.CreateUnreachable(); |
| 3892 | |
| 3893 | // We do need to preserve an insertion point. |
| 3894 | EmitBlock(createBasicBlock("longjmp.cont")); |
| 3895 | |
| 3896 | return RValue::get(nullptr); |
| 3897 | } |
| 3898 | case Builtin::BI__builtin_launder: { |
| 3899 | const Expr *Arg = E->getArg(0); |
| 3900 | QualType ArgTy = Arg->getType()->getPointeeType(); |
| 3901 | Value *Ptr = EmitScalarExpr(Arg); |
| 3902 | if (TypeRequiresBuiltinLaunder(CGM, ArgTy)) |
| 3903 | Ptr = Builder.CreateLaunderInvariantGroup(Ptr); |
| 3904 | |
| 3905 | return RValue::get(Ptr); |
| 3906 | } |
| 3907 | case Builtin::BI__sync_fetch_and_add: |
| 3908 | case Builtin::BI__sync_fetch_and_sub: |
| 3909 | case Builtin::BI__sync_fetch_and_or: |
| 3910 | case Builtin::BI__sync_fetch_and_and: |
| 3911 | case Builtin::BI__sync_fetch_and_xor: |
| 3912 | case Builtin::BI__sync_fetch_and_nand: |
| 3913 | case Builtin::BI__sync_add_and_fetch: |
| 3914 | case Builtin::BI__sync_sub_and_fetch: |
| 3915 | case Builtin::BI__sync_and_and_fetch: |
| 3916 | case Builtin::BI__sync_or_and_fetch: |
| 3917 | case Builtin::BI__sync_xor_and_fetch: |
| 3918 | case Builtin::BI__sync_nand_and_fetch: |
| 3919 | case Builtin::BI__sync_val_compare_and_swap: |
| 3920 | case Builtin::BI__sync_bool_compare_and_swap: |
| 3921 | case Builtin::BI__sync_lock_test_and_set: |
| 3922 | case Builtin::BI__sync_lock_release: |
| 3923 | case Builtin::BI__sync_swap: |
| 3924 | llvm_unreachable("Shouldn't make it through sema")::llvm::llvm_unreachable_internal("Shouldn't make it through sema" , "clang/lib/CodeGen/CGBuiltin.cpp", 3924); |
| 3925 | case Builtin::BI__sync_fetch_and_add_1: |
| 3926 | case Builtin::BI__sync_fetch_and_add_2: |
| 3927 | case Builtin::BI__sync_fetch_and_add_4: |
| 3928 | case Builtin::BI__sync_fetch_and_add_8: |
| 3929 | case Builtin::BI__sync_fetch_and_add_16: |
| 3930 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E); |
| 3931 | case Builtin::BI__sync_fetch_and_sub_1: |
| 3932 | case Builtin::BI__sync_fetch_and_sub_2: |
| 3933 | case Builtin::BI__sync_fetch_and_sub_4: |
| 3934 | case Builtin::BI__sync_fetch_and_sub_8: |
| 3935 | case Builtin::BI__sync_fetch_and_sub_16: |
| 3936 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E); |
| 3937 | case Builtin::BI__sync_fetch_and_or_1: |
| 3938 | case Builtin::BI__sync_fetch_and_or_2: |
| 3939 | case Builtin::BI__sync_fetch_and_or_4: |
| 3940 | case Builtin::BI__sync_fetch_and_or_8: |
| 3941 | case Builtin::BI__sync_fetch_and_or_16: |
| 3942 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E); |
| 3943 | case Builtin::BI__sync_fetch_and_and_1: |
| 3944 | case Builtin::BI__sync_fetch_and_and_2: |
| 3945 | case Builtin::BI__sync_fetch_and_and_4: |
| 3946 | case Builtin::BI__sync_fetch_and_and_8: |
| 3947 | case Builtin::BI__sync_fetch_and_and_16: |
| 3948 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E); |
| 3949 | case Builtin::BI__sync_fetch_and_xor_1: |
| 3950 | case Builtin::BI__sync_fetch_and_xor_2: |
| 3951 | case Builtin::BI__sync_fetch_and_xor_4: |
| 3952 | case Builtin::BI__sync_fetch_and_xor_8: |
| 3953 | case Builtin::BI__sync_fetch_and_xor_16: |
| 3954 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E); |
| 3955 | case Builtin::BI__sync_fetch_and_nand_1: |
| 3956 | case Builtin::BI__sync_fetch_and_nand_2: |
| 3957 | case Builtin::BI__sync_fetch_and_nand_4: |
| 3958 | case Builtin::BI__sync_fetch_and_nand_8: |
| 3959 | case Builtin::BI__sync_fetch_and_nand_16: |
| 3960 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E); |
| 3961 | |
| 3962 | // Clang extensions: not overloaded yet. |
| 3963 | case Builtin::BI__sync_fetch_and_min: |
| 3964 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E); |
| 3965 | case Builtin::BI__sync_fetch_and_max: |
| 3966 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E); |
| 3967 | case Builtin::BI__sync_fetch_and_umin: |
| 3968 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E); |
| 3969 | case Builtin::BI__sync_fetch_and_umax: |
| 3970 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E); |
| 3971 | |
| 3972 | case Builtin::BI__sync_add_and_fetch_1: |
| 3973 | case Builtin::BI__sync_add_and_fetch_2: |
| 3974 | case Builtin::BI__sync_add_and_fetch_4: |
| 3975 | case Builtin::BI__sync_add_and_fetch_8: |
| 3976 | case Builtin::BI__sync_add_and_fetch_16: |
| 3977 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E, |
| 3978 | llvm::Instruction::Add); |
| 3979 | case Builtin::BI__sync_sub_and_fetch_1: |
| 3980 | case Builtin::BI__sync_sub_and_fetch_2: |
| 3981 | case Builtin::BI__sync_sub_and_fetch_4: |
| 3982 | case Builtin::BI__sync_sub_and_fetch_8: |
| 3983 | case Builtin::BI__sync_sub_and_fetch_16: |
| 3984 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E, |
| 3985 | llvm::Instruction::Sub); |
| 3986 | case Builtin::BI__sync_and_and_fetch_1: |
| 3987 | case Builtin::BI__sync_and_and_fetch_2: |
| 3988 | case Builtin::BI__sync_and_and_fetch_4: |
| 3989 | case Builtin::BI__sync_and_and_fetch_8: |
| 3990 | case Builtin::BI__sync_and_and_fetch_16: |
| 3991 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E, |
| 3992 | llvm::Instruction::And); |
| 3993 | case Builtin::BI__sync_or_and_fetch_1: |
| 3994 | case Builtin::BI__sync_or_and_fetch_2: |
| 3995 | case Builtin::BI__sync_or_and_fetch_4: |
| 3996 | case Builtin::BI__sync_or_and_fetch_8: |
| 3997 | case Builtin::BI__sync_or_and_fetch_16: |
| 3998 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E, |
| 3999 | llvm::Instruction::Or); |
| 4000 | case Builtin::BI__sync_xor_and_fetch_1: |
| 4001 | case Builtin::BI__sync_xor_and_fetch_2: |
| 4002 | case Builtin::BI__sync_xor_and_fetch_4: |
| 4003 | case Builtin::BI__sync_xor_and_fetch_8: |
| 4004 | case Builtin::BI__sync_xor_and_fetch_16: |
| 4005 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E, |
| 4006 | llvm::Instruction::Xor); |
| 4007 | case Builtin::BI__sync_nand_and_fetch_1: |
| 4008 | case Builtin::BI__sync_nand_and_fetch_2: |
| 4009 | case Builtin::BI__sync_nand_and_fetch_4: |
| 4010 | case Builtin::BI__sync_nand_and_fetch_8: |
| 4011 | case Builtin::BI__sync_nand_and_fetch_16: |
| 4012 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E, |
| 4013 | llvm::Instruction::And, true); |
| 4014 | |
| 4015 | case Builtin::BI__sync_val_compare_and_swap_1: |
| 4016 | case Builtin::BI__sync_val_compare_and_swap_2: |
| 4017 | case Builtin::BI__sync_val_compare_and_swap_4: |
| 4018 | case Builtin::BI__sync_val_compare_and_swap_8: |
| 4019 | case Builtin::BI__sync_val_compare_and_swap_16: |
| 4020 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, false)); |
| 4021 | |
| 4022 | case Builtin::BI__sync_bool_compare_and_swap_1: |
| 4023 | case Builtin::BI__sync_bool_compare_and_swap_2: |
| 4024 | case Builtin::BI__sync_bool_compare_and_swap_4: |
| 4025 | case Builtin::BI__sync_bool_compare_and_swap_8: |
| 4026 | case Builtin::BI__sync_bool_compare_and_swap_16: |
| 4027 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, true)); |
| 4028 | |
| 4029 | case Builtin::BI__sync_swap_1: |
| 4030 | case Builtin::BI__sync_swap_2: |
| 4031 | case Builtin::BI__sync_swap_4: |
| 4032 | case Builtin::BI__sync_swap_8: |
| 4033 | case Builtin::BI__sync_swap_16: |
| 4034 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
| 4035 | |
| 4036 | case Builtin::BI__sync_lock_test_and_set_1: |
| 4037 | case Builtin::BI__sync_lock_test_and_set_2: |
| 4038 | case Builtin::BI__sync_lock_test_and_set_4: |
| 4039 | case Builtin::BI__sync_lock_test_and_set_8: |
| 4040 | case Builtin::BI__sync_lock_test_and_set_16: |
| 4041 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
| 4042 | |
| 4043 | case Builtin::BI__sync_lock_release_1: |
| 4044 | case Builtin::BI__sync_lock_release_2: |
| 4045 | case Builtin::BI__sync_lock_release_4: |
| 4046 | case Builtin::BI__sync_lock_release_8: |
| 4047 | case Builtin::BI__sync_lock_release_16: { |
| 4048 | CheckAtomicAlignment(*this, E); |
| 4049 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 4050 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 4051 | CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy); |
| 4052 | llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(), |
| 4053 | StoreSize.getQuantity() * 8); |
| 4054 | Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 4055 | llvm::StoreInst *Store = |
| 4056 | Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr, |
| 4057 | StoreSize); |
| 4058 | Store->setAtomic(llvm::AtomicOrdering::Release); |
| 4059 | return RValue::get(nullptr); |
| 4060 | } |
| 4061 | |
| 4062 | case Builtin::BI__sync_synchronize: { |
| 4063 | // We assume this is supposed to correspond to a C++0x-style |
| 4064 | // sequentially-consistent fence (i.e. this is only usable for |
| 4065 | // synchronization, not device I/O or anything like that). This intrinsic |
| 4066 | // is really badly designed in the sense that in theory, there isn't |
| 4067 | // any way to safely use it... but in practice, it mostly works |
| 4068 | // to use it with non-atomic loads and stores to get acquire/release |
| 4069 | // semantics. |
| 4070 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent); |
| 4071 | return RValue::get(nullptr); |
| 4072 | } |
| 4073 | |
| 4074 | case Builtin::BI__builtin_nontemporal_load: |
| 4075 | return RValue::get(EmitNontemporalLoad(*this, E)); |
| 4076 | case Builtin::BI__builtin_nontemporal_store: |
| 4077 | return RValue::get(EmitNontemporalStore(*this, E)); |
| 4078 | case Builtin::BI__c11_atomic_is_lock_free: |
| 4079 | case Builtin::BI__atomic_is_lock_free: { |
| 4080 | // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the |
| 4081 | // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since |
| 4082 | // _Atomic(T) is always properly-aligned. |
| 4083 | const char *LibCallName = "__atomic_is_lock_free"; |
| 4084 | CallArgList Args; |
| 4085 | Args.add(RValue::get(EmitScalarExpr(E->getArg(0))), |
| 4086 | getContext().getSizeType()); |
| 4087 | if (BuiltinID == Builtin::BI__atomic_is_lock_free) |
| 4088 | Args.add(RValue::get(EmitScalarExpr(E->getArg(1))), |
| 4089 | getContext().VoidPtrTy); |
| 4090 | else |
| 4091 | Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)), |
| 4092 | getContext().VoidPtrTy); |
| 4093 | const CGFunctionInfo &FuncInfo = |
| 4094 | CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args); |
| 4095 | llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo); |
| 4096 | llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName); |
| 4097 | return EmitCall(FuncInfo, CGCallee::forDirect(Func), |
| 4098 | ReturnValueSlot(), Args); |
| 4099 | } |
| 4100 | |
| 4101 | case Builtin::BI__atomic_test_and_set: { |
| 4102 | // Look at the argument type to determine whether this is a volatile |
| 4103 | // operation. The parameter type is always volatile. |
| 4104 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
| 4105 | bool Volatile = |
| 4106 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
| 4107 | |
| 4108 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 4109 | unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace(); |
| 4110 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); |
| 4111 | Value *NewVal = Builder.getInt8(1); |
| 4112 | Value *Order = EmitScalarExpr(E->getArg(1)); |
| 4113 | if (isa<llvm::ConstantInt>(Order)) { |
| 4114 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 4115 | AtomicRMWInst *Result = nullptr; |
| 4116 | switch (ord) { |
| 4117 | case 0: // memory_order_relaxed |
| 4118 | default: // invalid order |
| 4119 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 4120 | llvm::AtomicOrdering::Monotonic); |
| 4121 | break; |
| 4122 | case 1: // memory_order_consume |
| 4123 | case 2: // memory_order_acquire |
| 4124 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 4125 | llvm::AtomicOrdering::Acquire); |
| 4126 | break; |
| 4127 | case 3: // memory_order_release |
| 4128 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 4129 | llvm::AtomicOrdering::Release); |
| 4130 | break; |
| 4131 | case 4: // memory_order_acq_rel |
| 4132 | |
| 4133 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 4134 | llvm::AtomicOrdering::AcquireRelease); |
| 4135 | break; |
| 4136 | case 5: // memory_order_seq_cst |
| 4137 | Result = Builder.CreateAtomicRMW( |
| 4138 | llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 4139 | llvm::AtomicOrdering::SequentiallyConsistent); |
| 4140 | break; |
| 4141 | } |
| 4142 | Result->setVolatile(Volatile); |
| 4143 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
| 4144 | } |
| 4145 | |
| 4146 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 4147 | |
| 4148 | llvm::BasicBlock *BBs[5] = { |
| 4149 | createBasicBlock("monotonic", CurFn), |
| 4150 | createBasicBlock("acquire", CurFn), |
| 4151 | createBasicBlock("release", CurFn), |
| 4152 | createBasicBlock("acqrel", CurFn), |
| 4153 | createBasicBlock("seqcst", CurFn) |
| 4154 | }; |
| 4155 | llvm::AtomicOrdering Orders[5] = { |
| 4156 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire, |
| 4157 | llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease, |
| 4158 | llvm::AtomicOrdering::SequentiallyConsistent}; |
| 4159 | |
| 4160 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 4161 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
| 4162 | |
| 4163 | Builder.SetInsertPoint(ContBB); |
| 4164 | PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set"); |
| 4165 | |
| 4166 | for (unsigned i = 0; i < 5; ++i) { |
| 4167 | Builder.SetInsertPoint(BBs[i]); |
| 4168 | AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, |
| 4169 | Ptr, NewVal, Orders[i]); |
| 4170 | RMW->setVolatile(Volatile); |
| 4171 | Result->addIncoming(RMW, BBs[i]); |
| 4172 | Builder.CreateBr(ContBB); |
| 4173 | } |
| 4174 | |
| 4175 | SI->addCase(Builder.getInt32(0), BBs[0]); |
| 4176 | SI->addCase(Builder.getInt32(1), BBs[1]); |
| 4177 | SI->addCase(Builder.getInt32(2), BBs[1]); |
| 4178 | SI->addCase(Builder.getInt32(3), BBs[2]); |
| 4179 | SI->addCase(Builder.getInt32(4), BBs[3]); |
| 4180 | SI->addCase(Builder.getInt32(5), BBs[4]); |
| 4181 | |
| 4182 | Builder.SetInsertPoint(ContBB); |
| 4183 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
| 4184 | } |
| 4185 | |
| 4186 | case Builtin::BI__atomic_clear: { |
| 4187 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
| 4188 | bool Volatile = |
| 4189 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
| 4190 | |
| 4191 | Address Ptr = EmitPointerWithAlignment(E->getArg(0)); |
| 4192 | Ptr = Builder.CreateElementBitCast(Ptr, Int8Ty); |
| 4193 | Value *NewVal = Builder.getInt8(0); |
| 4194 | Value *Order = EmitScalarExpr(E->getArg(1)); |
| 4195 | if (isa<llvm::ConstantInt>(Order)) { |
| 4196 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 4197 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
| 4198 | switch (ord) { |
| 4199 | case 0: // memory_order_relaxed |
| 4200 | default: // invalid order |
| 4201 | Store->setOrdering(llvm::AtomicOrdering::Monotonic); |
| 4202 | break; |
| 4203 | case 3: // memory_order_release |
| 4204 | Store->setOrdering(llvm::AtomicOrdering::Release); |
| 4205 | break; |
| 4206 | case 5: // memory_order_seq_cst |
| 4207 | Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent); |
| 4208 | break; |
| 4209 | } |
| 4210 | return RValue::get(nullptr); |
| 4211 | } |
| 4212 | |
| 4213 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 4214 | |
| 4215 | llvm::BasicBlock *BBs[3] = { |
| 4216 | createBasicBlock("monotonic", CurFn), |
| 4217 | createBasicBlock("release", CurFn), |
| 4218 | createBasicBlock("seqcst", CurFn) |
| 4219 | }; |
| 4220 | llvm::AtomicOrdering Orders[3] = { |
| 4221 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release, |
| 4222 | llvm::AtomicOrdering::SequentiallyConsistent}; |
| 4223 | |
| 4224 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 4225 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
| 4226 | |
| 4227 | for (unsigned i = 0; i < 3; ++i) { |
| 4228 | Builder.SetInsertPoint(BBs[i]); |
| 4229 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
| 4230 | Store->setOrdering(Orders[i]); |
| 4231 | Builder.CreateBr(ContBB); |
| 4232 | } |
| 4233 | |
| 4234 | SI->addCase(Builder.getInt32(0), BBs[0]); |
| 4235 | SI->addCase(Builder.getInt32(3), BBs[1]); |
| 4236 | SI->addCase(Builder.getInt32(5), BBs[2]); |
| 4237 | |
| 4238 | Builder.SetInsertPoint(ContBB); |
| 4239 | return RValue::get(nullptr); |
| 4240 | } |
| 4241 | |
| 4242 | case Builtin::BI__atomic_thread_fence: |
| 4243 | case Builtin::BI__atomic_signal_fence: |
| 4244 | case Builtin::BI__c11_atomic_thread_fence: |
| 4245 | case Builtin::BI__c11_atomic_signal_fence: { |
| 4246 | llvm::SyncScope::ID SSID; |
| 4247 | if (BuiltinID == Builtin::BI__atomic_signal_fence || |
| 4248 | BuiltinID == Builtin::BI__c11_atomic_signal_fence) |
| 4249 | SSID = llvm::SyncScope::SingleThread; |
| 4250 | else |
| 4251 | SSID = llvm::SyncScope::System; |
| 4252 | Value *Order = EmitScalarExpr(E->getArg(0)); |
| 4253 | if (isa<llvm::ConstantInt>(Order)) { |
| 4254 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 4255 | switch (ord) { |
| 4256 | case 0: // memory_order_relaxed |
| 4257 | default: // invalid order |
| 4258 | break; |
| 4259 | case 1: // memory_order_consume |
| 4260 | case 2: // memory_order_acquire |
| 4261 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
| 4262 | break; |
| 4263 | case 3: // memory_order_release |
| 4264 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
| 4265 | break; |
| 4266 | case 4: // memory_order_acq_rel |
| 4267 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
| 4268 | break; |
| 4269 | case 5: // memory_order_seq_cst |
| 4270 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
| 4271 | break; |
| 4272 | } |
| 4273 | return RValue::get(nullptr); |
| 4274 | } |
| 4275 | |
| 4276 | llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB; |
| 4277 | AcquireBB = createBasicBlock("acquire", CurFn); |
| 4278 | ReleaseBB = createBasicBlock("release", CurFn); |
| 4279 | AcqRelBB = createBasicBlock("acqrel", CurFn); |
| 4280 | SeqCstBB = createBasicBlock("seqcst", CurFn); |
| 4281 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 4282 | |
| 4283 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 4284 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB); |
| 4285 | |
| 4286 | Builder.SetInsertPoint(AcquireBB); |
| 4287 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
| 4288 | Builder.CreateBr(ContBB); |
| 4289 | SI->addCase(Builder.getInt32(1), AcquireBB); |
| 4290 | SI->addCase(Builder.getInt32(2), AcquireBB); |
| 4291 | |
| 4292 | Builder.SetInsertPoint(ReleaseBB); |
| 4293 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
| 4294 | Builder.CreateBr(ContBB); |
| 4295 | SI->addCase(Builder.getInt32(3), ReleaseBB); |
| 4296 | |
| 4297 | Builder.SetInsertPoint(AcqRelBB); |
| 4298 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
| 4299 | Builder.CreateBr(ContBB); |
| 4300 | SI->addCase(Builder.getInt32(4), AcqRelBB); |
| 4301 | |
| 4302 | Builder.SetInsertPoint(SeqCstBB); |
| 4303 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
| 4304 | Builder.CreateBr(ContBB); |
| 4305 | SI->addCase(Builder.getInt32(5), SeqCstBB); |
| 4306 | |
| 4307 | Builder.SetInsertPoint(ContBB); |
| 4308 | return RValue::get(nullptr); |
| 4309 | } |
| 4310 | |
| 4311 | case Builtin::BI__builtin_signbit: |
| 4312 | case Builtin::BI__builtin_signbitf: |
| 4313 | case Builtin::BI__builtin_signbitl: { |
| 4314 | return RValue::get( |
| 4315 | Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))), |
| 4316 | ConvertType(E->getType()))); |
| 4317 | } |
| 4318 | case Builtin::BI__warn_memset_zero_len: |
| 4319 | return RValue::getIgnored(); |
| 4320 | case Builtin::BI__annotation: { |
| 4321 | // Re-encode each wide string to UTF8 and make an MDString. |
| 4322 | SmallVector<Metadata *, 1> Strings; |
| 4323 | for (const Expr *Arg : E->arguments()) { |
| 4324 | const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts()); |
| 4325 | assert(Str->getCharByteWidth() == 2)(static_cast <bool> (Str->getCharByteWidth() == 2) ? void (0) : __assert_fail ("Str->getCharByteWidth() == 2", "clang/lib/CodeGen/CGBuiltin.cpp", 4325, __extension__ __PRETTY_FUNCTION__ )); |
| 4326 | StringRef WideBytes = Str->getBytes(); |
| 4327 | std::string StrUtf8; |
| 4328 | if (!convertUTF16ToUTF8String( |
| 4329 | ArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) { |
| 4330 | CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument"); |
| 4331 | continue; |
| 4332 | } |
| 4333 | Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8)); |
| 4334 | } |
| 4335 | |
| 4336 | // Build and MDTuple of MDStrings and emit the intrinsic call. |
| 4337 | llvm::Function *F = |
| 4338 | CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {}); |
| 4339 | MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings); |
| 4340 | Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple)); |
| 4341 | return RValue::getIgnored(); |
| 4342 | } |
| 4343 | case Builtin::BI__builtin_annotation: { |
| 4344 | llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0)); |
| 4345 | llvm::Function *F = |
| 4346 | CGM.getIntrinsic(llvm::Intrinsic::annotation, |
| 4347 | {AnnVal->getType(), CGM.ConstGlobalsPtrTy}); |
| 4348 | |
| 4349 | // Get the annotation string, go through casts. Sema requires this to be a |
| 4350 | // non-wide string literal, potentially casted, so the cast<> is safe. |
| 4351 | const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts(); |
| 4352 | StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString(); |
| 4353 | return RValue::get( |
| 4354 | EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc(), nullptr)); |
| 4355 | } |
| 4356 | case Builtin::BI__builtin_addcb: |
| 4357 | case Builtin::BI__builtin_addcs: |
| 4358 | case Builtin::BI__builtin_addc: |
| 4359 | case Builtin::BI__builtin_addcl: |
| 4360 | case Builtin::BI__builtin_addcll: |
| 4361 | case Builtin::BI__builtin_subcb: |
| 4362 | case Builtin::BI__builtin_subcs: |
| 4363 | case Builtin::BI__builtin_subc: |
| 4364 | case Builtin::BI__builtin_subcl: |
| 4365 | case Builtin::BI__builtin_subcll: { |
| 4366 | |
| 4367 | // We translate all of these builtins from expressions of the form: |
| 4368 | // int x = ..., y = ..., carryin = ..., carryout, result; |
| 4369 | // result = __builtin_addc(x, y, carryin, &carryout); |
| 4370 | // |
| 4371 | // to LLVM IR of the form: |
| 4372 | // |
| 4373 | // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y) |
| 4374 | // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0 |
| 4375 | // %carry1 = extractvalue {i32, i1} %tmp1, 1 |
| 4376 | // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1, |
| 4377 | // i32 %carryin) |
| 4378 | // %result = extractvalue {i32, i1} %tmp2, 0 |
| 4379 | // %carry2 = extractvalue {i32, i1} %tmp2, 1 |
| 4380 | // %tmp3 = or i1 %carry1, %carry2 |
| 4381 | // %tmp4 = zext i1 %tmp3 to i32 |
| 4382 | // store i32 %tmp4, i32* %carryout |
| 4383 | |
| 4384 | // Scalarize our inputs. |
| 4385 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
| 4386 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
| 4387 | llvm::Value *Carryin = EmitScalarExpr(E->getArg(2)); |
| 4388 | Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3)); |
| 4389 | |
| 4390 | // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow. |
| 4391 | llvm::Intrinsic::ID IntrinsicId; |
| 4392 | switch (BuiltinID) { |
| 4393 | default: llvm_unreachable("Unknown multiprecision builtin id.")::llvm::llvm_unreachable_internal("Unknown multiprecision builtin id." , "clang/lib/CodeGen/CGBuiltin.cpp", 4393); |
| 4394 | case Builtin::BI__builtin_addcb: |
| 4395 | case Builtin::BI__builtin_addcs: |
| 4396 | case Builtin::BI__builtin_addc: |
| 4397 | case Builtin::BI__builtin_addcl: |
| 4398 | case Builtin::BI__builtin_addcll: |
| 4399 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
| 4400 | break; |
| 4401 | case Builtin::BI__builtin_subcb: |
| 4402 | case Builtin::BI__builtin_subcs: |
| 4403 | case Builtin::BI__builtin_subc: |
| 4404 | case Builtin::BI__builtin_subcl: |
| 4405 | case Builtin::BI__builtin_subcll: |
| 4406 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
| 4407 | break; |
| 4408 | } |
| 4409 | |
| 4410 | // Construct our resulting LLVM IR expression. |
| 4411 | llvm::Value *Carry1; |
| 4412 | llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId, |
| 4413 | X, Y, Carry1); |
| 4414 | llvm::Value *Carry2; |
| 4415 | llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId, |
| 4416 | Sum1, Carryin, Carry2); |
| 4417 | llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2), |
| 4418 | X->getType()); |
| 4419 | Builder.CreateStore(CarryOut, CarryOutPtr); |
| 4420 | return RValue::get(Sum2); |
| 4421 | } |
| 4422 | |
| 4423 | case Builtin::BI__builtin_add_overflow: |
| 4424 | case Builtin::BI__builtin_sub_overflow: |
| 4425 | case Builtin::BI__builtin_mul_overflow: { |
| 4426 | const clang::Expr *LeftArg = E->getArg(0); |
| 4427 | const clang::Expr *RightArg = E->getArg(1); |
| 4428 | const clang::Expr *ResultArg = E->getArg(2); |
| 4429 | |
| 4430 | clang::QualType ResultQTy = |
| 4431 | ResultArg->getType()->castAs<PointerType>()->getPointeeType(); |
| 4432 | |
| 4433 | WidthAndSignedness LeftInfo = |
| 4434 | getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType()); |
| 4435 | WidthAndSignedness RightInfo = |
| 4436 | getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType()); |
| 4437 | WidthAndSignedness ResultInfo = |
| 4438 | getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy); |
| 4439 | |
| 4440 | // Handle mixed-sign multiplication as a special case, because adding |
| 4441 | // runtime or backend support for our generic irgen would be too expensive. |
| 4442 | if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo)) |
| 4443 | return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg, |
| 4444 | RightInfo, ResultArg, ResultQTy, |
| 4445 | ResultInfo); |
| 4446 | |
| 4447 | if (isSpecialUnsignedMultiplySignedResult(BuiltinID, LeftInfo, RightInfo, |
| 4448 | ResultInfo)) |
| 4449 | return EmitCheckedUnsignedMultiplySignedResult( |
| 4450 | *this, LeftArg, LeftInfo, RightArg, RightInfo, ResultArg, ResultQTy, |
| 4451 | ResultInfo); |
| 4452 | |
| 4453 | WidthAndSignedness EncompassingInfo = |
| 4454 | EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo}); |
| 4455 | |
| 4456 | llvm::Type *EncompassingLLVMTy = |
| 4457 | llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width); |
| 4458 | |
| 4459 | llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy); |
| 4460 | |
| 4461 | llvm::Intrinsic::ID IntrinsicId; |
| 4462 | switch (BuiltinID) { |
| 4463 | default: |
| 4464 | llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "clang/lib/CodeGen/CGBuiltin.cpp", 4464); |
| 4465 | case Builtin::BI__builtin_add_overflow: |
| 4466 | IntrinsicId = EncompassingInfo.Signed |
| 4467 | ? llvm::Intrinsic::sadd_with_overflow |
| 4468 | : llvm::Intrinsic::uadd_with_overflow; |
| 4469 | break; |
| 4470 | case Builtin::BI__builtin_sub_overflow: |
| 4471 | IntrinsicId = EncompassingInfo.Signed |
| 4472 | ? llvm::Intrinsic::ssub_with_overflow |
| 4473 | : llvm::Intrinsic::usub_with_overflow; |
| 4474 | break; |
| 4475 | case Builtin::BI__builtin_mul_overflow: |
| 4476 | IntrinsicId = EncompassingInfo.Signed |
| 4477 | ? llvm::Intrinsic::smul_with_overflow |
| 4478 | : llvm::Intrinsic::umul_with_overflow; |
| 4479 | break; |
| 4480 | } |
| 4481 | |
| 4482 | llvm::Value *Left = EmitScalarExpr(LeftArg); |
| 4483 | llvm::Value *Right = EmitScalarExpr(RightArg); |
| 4484 | Address ResultPtr = EmitPointerWithAlignment(ResultArg); |
| 4485 | |
| 4486 | // Extend each operand to the encompassing type. |
| 4487 | Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed); |
| 4488 | Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed); |
| 4489 | |
| 4490 | // Perform the operation on the extended values. |
| 4491 | llvm::Value *Overflow, *Result; |
| 4492 | Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow); |
| 4493 | |
| 4494 | if (EncompassingInfo.Width > ResultInfo.Width) { |
| 4495 | // The encompassing type is wider than the result type, so we need to |
| 4496 | // truncate it. |
| 4497 | llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy); |
| 4498 | |
| 4499 | // To see if the truncation caused an overflow, we will extend |
| 4500 | // the result and then compare it to the original result. |
| 4501 | llvm::Value *ResultTruncExt = Builder.CreateIntCast( |
| 4502 | ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed); |
| 4503 | llvm::Value *TruncationOverflow = |
| 4504 | Builder.CreateICmpNE(Result, ResultTruncExt); |
| 4505 | |
| 4506 | Overflow = Builder.CreateOr(Overflow, TruncationOverflow); |
| 4507 | Result = ResultTrunc; |
| 4508 | } |
| 4509 | |
| 4510 | // Finally, store the result using the pointer. |
| 4511 | bool isVolatile = |
| 4512 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 4513 | Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile); |
| 4514 | |
| 4515 | return RValue::get(Overflow); |
| 4516 | } |
| 4517 | |
| 4518 | case Builtin::BI__builtin_uadd_overflow: |
| 4519 | case Builtin::BI__builtin_uaddl_overflow: |
| 4520 | case Builtin::BI__builtin_uaddll_overflow: |
| 4521 | case Builtin::BI__builtin_usub_overflow: |
| 4522 | case Builtin::BI__builtin_usubl_overflow: |
| 4523 | case Builtin::BI__builtin_usubll_overflow: |
| 4524 | case Builtin::BI__builtin_umul_overflow: |
| 4525 | case Builtin::BI__builtin_umull_overflow: |
| 4526 | case Builtin::BI__builtin_umulll_overflow: |
| 4527 | case Builtin::BI__builtin_sadd_overflow: |
| 4528 | case Builtin::BI__builtin_saddl_overflow: |
| 4529 | case Builtin::BI__builtin_saddll_overflow: |
| 4530 | case Builtin::BI__builtin_ssub_overflow: |
| 4531 | case Builtin::BI__builtin_ssubl_overflow: |
| 4532 | case Builtin::BI__builtin_ssubll_overflow: |
| 4533 | case Builtin::BI__builtin_smul_overflow: |
| 4534 | case Builtin::BI__builtin_smull_overflow: |
| 4535 | case Builtin::BI__builtin_smulll_overflow: { |
| 4536 | |
| 4537 | // We translate all of these builtins directly to the relevant llvm IR node. |
| 4538 | |
| 4539 | // Scalarize our inputs. |
| 4540 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
| 4541 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
| 4542 | Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2)); |
| 4543 | |
| 4544 | // Decide which of the overflow intrinsics we are lowering to: |
| 4545 | llvm::Intrinsic::ID IntrinsicId; |
| 4546 | switch (BuiltinID) { |
| 4547 | default: llvm_unreachable("Unknown overflow builtin id.")::llvm::llvm_unreachable_internal("Unknown overflow builtin id." , "clang/lib/CodeGen/CGBuiltin.cpp", 4547); |
| 4548 | case Builtin::BI__builtin_uadd_overflow: |
| 4549 | case Builtin::BI__builtin_uaddl_overflow: |
| 4550 | case Builtin::BI__builtin_uaddll_overflow: |
| 4551 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
| 4552 | break; |
| 4553 | case Builtin::BI__builtin_usub_overflow: |
| 4554 | case Builtin::BI__builtin_usubl_overflow: |
| 4555 | case Builtin::BI__builtin_usubll_overflow: |
| 4556 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
| 4557 | break; |
| 4558 | case Builtin::BI__builtin_umul_overflow: |
| 4559 | case Builtin::BI__builtin_umull_overflow: |
| 4560 | case Builtin::BI__builtin_umulll_overflow: |
| 4561 | IntrinsicId = llvm::Intrinsic::umul_with_overflow; |
| 4562 | break; |
| 4563 | case Builtin::BI__builtin_sadd_overflow: |
| 4564 | case Builtin::BI__builtin_saddl_overflow: |
| 4565 | case Builtin::BI__builtin_saddll_overflow: |
| 4566 | IntrinsicId = llvm::Intrinsic::sadd_with_overflow; |
| 4567 | break; |
| 4568 | case Builtin::BI__builtin_ssub_overflow: |
| 4569 | case Builtin::BI__builtin_ssubl_overflow: |
| 4570 | case Builtin::BI__builtin_ssubll_overflow: |
| 4571 | IntrinsicId = llvm::Intrinsic::ssub_with_overflow; |
| 4572 | break; |
| 4573 | case Builtin::BI__builtin_smul_overflow: |
| 4574 | case Builtin::BI__builtin_smull_overflow: |
| 4575 | case Builtin::BI__builtin_smulll_overflow: |
| 4576 | IntrinsicId = llvm::Intrinsic::smul_with_overflow; |
| 4577 | break; |
| 4578 | } |
| 4579 | |
| 4580 | |
| 4581 | llvm::Value *Carry; |
| 4582 | llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry); |
| 4583 | Builder.CreateStore(Sum, SumOutPtr); |
| 4584 | |
| 4585 | return RValue::get(Carry); |
| 4586 | } |
| 4587 | case Builtin::BIaddressof: |
| 4588 | case Builtin::BI__addressof: |
| 4589 | case Builtin::BI__builtin_addressof: |
| 4590 | return RValue::get(EmitLValue(E->getArg(0)).getPointer(*this)); |
| 4591 | case Builtin::BI__builtin_function_start: |
| 4592 | return RValue::get(CGM.GetFunctionStart( |
| 4593 | E->getArg(0)->getAsBuiltinConstantDeclRef(CGM.getContext()))); |
| 4594 | case Builtin::BI__builtin_operator_new: |
| 4595 | return EmitBuiltinNewDeleteCall( |
| 4596 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false); |
| 4597 | case Builtin::BI__builtin_operator_delete: |
| 4598 | EmitBuiltinNewDeleteCall( |
| 4599 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true); |
| 4600 | return RValue::get(nullptr); |
| 4601 | |
| 4602 | case Builtin::BI__builtin_is_aligned: |
| 4603 | return EmitBuiltinIsAligned(E); |
| 4604 | case Builtin::BI__builtin_align_up: |
| 4605 | return EmitBuiltinAlignTo(E, true); |
| 4606 | case Builtin::BI__builtin_align_down: |
| 4607 | return EmitBuiltinAlignTo(E, false); |
| 4608 | |
| 4609 | case Builtin::BI__noop: |
| 4610 | // __noop always evaluates to an integer literal zero. |
| 4611 | return RValue::get(ConstantInt::get(IntTy, 0)); |
| 4612 | case Builtin::BI__builtin_call_with_static_chain: { |
| 4613 | const CallExpr *Call = cast<CallExpr>(E->getArg(0)); |
| 4614 | const Expr *Chain = E->getArg(1); |
| 4615 | return EmitCall(Call->getCallee()->getType(), |
| 4616 | EmitCallee(Call->getCallee()), Call, ReturnValue, |
| 4617 | EmitScalarExpr(Chain)); |
| 4618 | } |
| 4619 | case Builtin::BI_InterlockedExchange8: |
| 4620 | case Builtin::BI_InterlockedExchange16: |
| 4621 | case Builtin::BI_InterlockedExchange: |
| 4622 | case Builtin::BI_InterlockedExchangePointer: |
| 4623 | return RValue::get( |
| 4624 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E)); |
| 4625 | case Builtin::BI_InterlockedCompareExchangePointer: |
| 4626 | case Builtin::BI_InterlockedCompareExchangePointer_nf: { |
| 4627 | llvm::Type *RTy; |
| 4628 | llvm::IntegerType *IntType = |
| 4629 | IntegerType::get(getLLVMContext(), |
| 4630 | getContext().getTypeSize(E->getType())); |
| 4631 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
| 4632 | |
| 4633 | llvm::Value *Destination = |
| 4634 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType); |
| 4635 | |
| 4636 | llvm::Value *Exchange = EmitScalarExpr(E->getArg(1)); |
| 4637 | RTy = Exchange->getType(); |
| 4638 | Exchange = Builder.CreatePtrToInt(Exchange, IntType); |
| 4639 | |
| 4640 | llvm::Value *Comparand = |
| 4641 | Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType); |
| 4642 | |
| 4643 | auto Ordering = |
| 4644 | BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ? |
| 4645 | AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent; |
| 4646 | |
| 4647 | auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
| 4648 | Ordering, Ordering); |
| 4649 | Result->setVolatile(true); |
| 4650 | |
| 4651 | return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result, |
| 4652 | 0), |
| 4653 | RTy)); |
| 4654 | } |
| 4655 | case Builtin::BI_InterlockedCompareExchange8: |
| 4656 | case Builtin::BI_InterlockedCompareExchange16: |
| 4657 | case Builtin::BI_InterlockedCompareExchange: |
| 4658 | case Builtin::BI_InterlockedCompareExchange64: |
| 4659 | return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E)); |
| 4660 | case Builtin::BI_InterlockedIncrement16: |
| 4661 | case Builtin::BI_InterlockedIncrement: |
| 4662 | return RValue::get( |
| 4663 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E)); |
| 4664 | case Builtin::BI_InterlockedDecrement16: |
| 4665 | case Builtin::BI_InterlockedDecrement: |
| 4666 | return RValue::get( |
| 4667 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E)); |
| 4668 | case Builtin::BI_InterlockedAnd8: |
| 4669 | case Builtin::BI_InterlockedAnd16: |
| 4670 | case Builtin::BI_InterlockedAnd: |
| 4671 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E)); |
| 4672 | case Builtin::BI_InterlockedExchangeAdd8: |
| 4673 | case Builtin::BI_InterlockedExchangeAdd16: |
| 4674 | case Builtin::BI_InterlockedExchangeAdd: |
| 4675 | return RValue::get( |
| 4676 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E)); |
| 4677 | case Builtin::BI_InterlockedExchangeSub8: |
| 4678 | case Builtin::BI_InterlockedExchangeSub16: |
| 4679 | case Builtin::BI_InterlockedExchangeSub: |
| 4680 | return RValue::get( |
| 4681 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E)); |
| 4682 | case Builtin::BI_InterlockedOr8: |
| 4683 | case Builtin::BI_InterlockedOr16: |
| 4684 | case Builtin::BI_InterlockedOr: |
| 4685 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E)); |
| 4686 | case Builtin::BI_InterlockedXor8: |
| 4687 | case Builtin::BI_InterlockedXor16: |
| 4688 | case Builtin::BI_InterlockedXor: |
| 4689 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E)); |
| 4690 | |
| 4691 | case Builtin::BI_bittest64: |
| 4692 | case Builtin::BI_bittest: |
| 4693 | case Builtin::BI_bittestandcomplement64: |
| 4694 | case Builtin::BI_bittestandcomplement: |
| 4695 | case Builtin::BI_bittestandreset64: |
| 4696 | case Builtin::BI_bittestandreset: |
| 4697 | case Builtin::BI_bittestandset64: |
| 4698 | case Builtin::BI_bittestandset: |
| 4699 | case Builtin::BI_interlockedbittestandreset: |
| 4700 | case Builtin::BI_interlockedbittestandreset64: |
| 4701 | case Builtin::BI_interlockedbittestandset64: |
| 4702 | case Builtin::BI_interlockedbittestandset: |
| 4703 | case Builtin::BI_interlockedbittestandset_acq: |
| 4704 | case Builtin::BI_interlockedbittestandset_rel: |
| 4705 | case Builtin::BI_interlockedbittestandset_nf: |
| 4706 | case Builtin::BI_interlockedbittestandreset_acq: |
| 4707 | case Builtin::BI_interlockedbittestandreset_rel: |
| 4708 | case Builtin::BI_interlockedbittestandreset_nf: |
| 4709 | return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E)); |
| 4710 | |
| 4711 | // These builtins exist to emit regular volatile loads and stores not |
| 4712 | // affected by the -fms-volatile setting. |
| 4713 | case Builtin::BI__iso_volatile_load8: |
| 4714 | case Builtin::BI__iso_volatile_load16: |
| 4715 | case Builtin::BI__iso_volatile_load32: |
| 4716 | case Builtin::BI__iso_volatile_load64: |
| 4717 | return RValue::get(EmitISOVolatileLoad(*this, E)); |
| 4718 | case Builtin::BI__iso_volatile_store8: |
| 4719 | case Builtin::BI__iso_volatile_store16: |
| 4720 | case Builtin::BI__iso_volatile_store32: |
| 4721 | case Builtin::BI__iso_volatile_store64: |
| 4722 | return RValue::get(EmitISOVolatileStore(*this, E)); |
| 4723 | |
| 4724 | case Builtin::BI__exception_code: |
| 4725 | case Builtin::BI_exception_code: |
| 4726 | return RValue::get(EmitSEHExceptionCode()); |
| 4727 | case Builtin::BI__exception_info: |
| 4728 | case Builtin::BI_exception_info: |
| 4729 | return RValue::get(EmitSEHExceptionInfo()); |
| 4730 | case Builtin::BI__abnormal_termination: |
| 4731 | case Builtin::BI_abnormal_termination: |
| 4732 | return RValue::get(EmitSEHAbnormalTermination()); |
| 4733 | case Builtin::BI_setjmpex: |
| 4734 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && |
| 4735 | E->getArg(0)->getType()->isPointerType()) |
| 4736 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
| 4737 | break; |
| 4738 | case Builtin::BI_setjmp: |
| 4739 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && |
| 4740 | E->getArg(0)->getType()->isPointerType()) { |
| 4741 | if (getTarget().getTriple().getArch() == llvm::Triple::x86) |
| 4742 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E); |
| 4743 | else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64) |
| 4744 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
| 4745 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E); |
| 4746 | } |
| 4747 | break; |
| 4748 | |
| 4749 | // C++ std:: builtins. |
| 4750 | case Builtin::BImove: |
| 4751 | case Builtin::BImove_if_noexcept: |
| 4752 | case Builtin::BIforward: |
| 4753 | case Builtin::BIforward_like: |
| 4754 | case Builtin::BIas_const: |
| 4755 | return RValue::get(EmitLValue(E->getArg(0)).getPointer(*this)); |
| 4756 | case Builtin::BI__GetExceptionInfo: { |
| 4757 | if (llvm::GlobalVariable *GV = |
| 4758 | CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType())) |
| 4759 | return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy)); |
| 4760 | break; |
| 4761 | } |
| 4762 | |
| 4763 | case Builtin::BI__fastfail: |
| 4764 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E)); |
| 4765 | |
| 4766 | case Builtin::BI__builtin_coro_id: |
| 4767 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_id); |
| 4768 | case Builtin::BI__builtin_coro_promise: |
| 4769 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise); |
| 4770 | case Builtin::BI__builtin_coro_resume: |
| 4771 | EmitCoroutineIntrinsic(E, Intrinsic::coro_resume); |
| 4772 | return RValue::get(nullptr); |
| 4773 | case Builtin::BI__builtin_coro_frame: |
| 4774 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame); |
| 4775 | case Builtin::BI__builtin_coro_noop: |
| 4776 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop); |
| 4777 | case Builtin::BI__builtin_coro_free: |
| 4778 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_free); |
| 4779 | case Builtin::BI__builtin_coro_destroy: |
| 4780 | EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy); |
| 4781 | return RValue::get(nullptr); |
| 4782 | case Builtin::BI__builtin_coro_done: |
| 4783 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_done); |
| 4784 | case Builtin::BI__builtin_coro_alloc: |
| 4785 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc); |
| 4786 | case Builtin::BI__builtin_coro_begin: |
| 4787 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin); |
| 4788 | case Builtin::BI__builtin_coro_end: |
| 4789 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_end); |
| 4790 | case Builtin::BI__builtin_coro_suspend: |
| 4791 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend); |
| 4792 | case Builtin::BI__builtin_coro_size: |
| 4793 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_size); |
| 4794 | case Builtin::BI__builtin_coro_align: |
| 4795 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_align); |
| 4796 | |
| 4797 | // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions |
| 4798 | case Builtin::BIread_pipe: |
| 4799 | case Builtin::BIwrite_pipe: { |
| 4800 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 4801 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 4802 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4803 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4804 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4805 | |
| 4806 | // Type of the generic packet parameter. |
| 4807 | unsigned GenericAS = |
| 4808 | getContext().getTargetAddressSpace(LangAS::opencl_generic); |
| 4809 | llvm::Type *I8PTy = llvm::PointerType::get( |
| 4810 | llvm::Type::getInt8Ty(getLLVMContext()), GenericAS); |
| 4811 | |
| 4812 | // Testing which overloaded version we should generate the call for. |
| 4813 | if (2U == E->getNumArgs()) { |
| 4814 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2" |
| 4815 | : "__write_pipe_2"; |
| 4816 | // Creating a generic function type to be able to call with any builtin or |
| 4817 | // user defined type. |
| 4818 | llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty}; |
| 4819 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4820 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4821 | Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy); |
| 4822 | return RValue::get( |
| 4823 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4824 | {Arg0, BCast, PacketSize, PacketAlign})); |
| 4825 | } else { |
| 4826 | assert(4 == E->getNumArgs() &&(static_cast <bool> (4 == E->getNumArgs() && "Illegal number of parameters to pipe function") ? void (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 4827, __extension__ __PRETTY_FUNCTION__ )) |
| 4827 | "Illegal number of parameters to pipe function")(static_cast <bool> (4 == E->getNumArgs() && "Illegal number of parameters to pipe function") ? void (0) : __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 4827, __extension__ __PRETTY_FUNCTION__ )); |
| 4828 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4" |
| 4829 | : "__write_pipe_4"; |
| 4830 | |
| 4831 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy, |
| 4832 | Int32Ty, Int32Ty}; |
| 4833 | Value *Arg2 = EmitScalarExpr(E->getArg(2)), |
| 4834 | *Arg3 = EmitScalarExpr(E->getArg(3)); |
| 4835 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4836 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4837 | Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy); |
| 4838 | // We know the third argument is an integer type, but we may need to cast |
| 4839 | // it to i32. |
| 4840 | if (Arg2->getType() != Int32Ty) |
| 4841 | Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty); |
| 4842 | return RValue::get( |
| 4843 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4844 | {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign})); |
| 4845 | } |
| 4846 | } |
| 4847 | // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write |
| 4848 | // functions |
| 4849 | case Builtin::BIreserve_read_pipe: |
| 4850 | case Builtin::BIreserve_write_pipe: |
| 4851 | case Builtin::BIwork_group_reserve_read_pipe: |
| 4852 | case Builtin::BIwork_group_reserve_write_pipe: |
| 4853 | case Builtin::BIsub_group_reserve_read_pipe: |
| 4854 | case Builtin::BIsub_group_reserve_write_pipe: { |
| 4855 | // Composing the mangled name for the function. |
| 4856 | const char *Name; |
| 4857 | if (BuiltinID == Builtin::BIreserve_read_pipe) |
| 4858 | Name = "__reserve_read_pipe"; |
| 4859 | else if (BuiltinID == Builtin::BIreserve_write_pipe) |
| 4860 | Name = "__reserve_write_pipe"; |
| 4861 | else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe) |
| 4862 | Name = "__work_group_reserve_read_pipe"; |
| 4863 | else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe) |
| 4864 | Name = "__work_group_reserve_write_pipe"; |
| 4865 | else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe) |
| 4866 | Name = "__sub_group_reserve_read_pipe"; |
| 4867 | else |
| 4868 | Name = "__sub_group_reserve_write_pipe"; |
| 4869 | |
| 4870 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 4871 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 4872 | llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy); |
| 4873 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4874 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4875 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4876 | |
| 4877 | // Building the generic function prototype. |
| 4878 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty}; |
| 4879 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4880 | ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4881 | // We know the second argument is an integer type, but we may need to cast |
| 4882 | // it to i32. |
| 4883 | if (Arg1->getType() != Int32Ty) |
| 4884 | Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty); |
| 4885 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4886 | {Arg0, Arg1, PacketSize, PacketAlign})); |
| 4887 | } |
| 4888 | // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write |
| 4889 | // functions |
| 4890 | case Builtin::BIcommit_read_pipe: |
| 4891 | case Builtin::BIcommit_write_pipe: |
| 4892 | case Builtin::BIwork_group_commit_read_pipe: |
| 4893 | case Builtin::BIwork_group_commit_write_pipe: |
| 4894 | case Builtin::BIsub_group_commit_read_pipe: |
| 4895 | case Builtin::BIsub_group_commit_write_pipe: { |
| 4896 | const char *Name; |
| 4897 | if (BuiltinID == Builtin::BIcommit_read_pipe) |
| 4898 | Name = "__commit_read_pipe"; |
| 4899 | else if (BuiltinID == Builtin::BIcommit_write_pipe) |
| 4900 | Name = "__commit_write_pipe"; |
| 4901 | else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe) |
| 4902 | Name = "__work_group_commit_read_pipe"; |
| 4903 | else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe) |
| 4904 | Name = "__work_group_commit_write_pipe"; |
| 4905 | else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe) |
| 4906 | Name = "__sub_group_commit_read_pipe"; |
| 4907 | else |
| 4908 | Name = "__sub_group_commit_write_pipe"; |
| 4909 | |
| 4910 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 4911 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 4912 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4913 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4914 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4915 | |
| 4916 | // Building the generic function prototype. |
| 4917 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty}; |
| 4918 | llvm::FunctionType *FTy = |
| 4919 | llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()), |
| 4920 | llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4921 | |
| 4922 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4923 | {Arg0, Arg1, PacketSize, PacketAlign})); |
| 4924 | } |
| 4925 | // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions |
| 4926 | case Builtin::BIget_pipe_num_packets: |
| 4927 | case Builtin::BIget_pipe_max_packets: { |
| 4928 | const char *BaseName; |
| 4929 | const auto *PipeTy = E->getArg(0)->getType()->castAs<PipeType>(); |
| 4930 | if (BuiltinID == Builtin::BIget_pipe_num_packets) |
| 4931 | BaseName = "__get_pipe_num_packets"; |
| 4932 | else |
| 4933 | BaseName = "__get_pipe_max_packets"; |
| 4934 | std::string Name = std::string(BaseName) + |
| 4935 | std::string(PipeTy->isReadOnly() ? "_ro" : "_wo"); |
| 4936 | |
| 4937 | // Building the generic function prototype. |
| 4938 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); |
| 4939 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4940 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4941 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4942 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty}; |
| 4943 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4944 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4945 | |
| 4946 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4947 | {Arg0, PacketSize, PacketAlign})); |
| 4948 | } |
| 4949 | |
| 4950 | // OpenCL v2.0 s6.13.9 - Address space qualifier functions. |
| 4951 | case Builtin::BIto_global: |
| 4952 | case Builtin::BIto_local: |
| 4953 | case Builtin::BIto_private: { |
| 4954 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
| 4955 | auto NewArgT = llvm::PointerType::get(Int8Ty, |
| 4956 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4957 | auto NewRetT = llvm::PointerType::get(Int8Ty, |
| 4958 | CGM.getContext().getTargetAddressSpace( |
| 4959 | E->getType()->getPointeeType().getAddressSpace())); |
| 4960 | auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false); |
| 4961 | llvm::Value *NewArg; |
| 4962 | if (Arg0->getType()->getPointerAddressSpace() != |
| 4963 | NewArgT->getPointerAddressSpace()) |
| 4964 | NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT); |
| 4965 | else |
| 4966 | NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT); |
| 4967 | auto NewName = std::string("__") + E->getDirectCallee()->getName().str(); |
| 4968 | auto NewCall = |
| 4969 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg}); |
| 4970 | return RValue::get(Builder.CreateBitOrPointerCast(NewCall, |
| 4971 | ConvertType(E->getType()))); |
| 4972 | } |
| 4973 | |
| 4974 | // OpenCL v2.0, s6.13.17 - Enqueue kernel function. |
| 4975 | // It contains four different overload formats specified in Table 6.13.17.1. |
| 4976 | case Builtin::BIenqueue_kernel: { |
| 4977 | StringRef Name; // Generated function call name |
| 4978 | unsigned NumArgs = E->getNumArgs(); |
| 4979 | |
| 4980 | llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy); |
| 4981 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 4982 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4983 | |
| 4984 | llvm::Value *Queue = EmitScalarExpr(E->getArg(0)); |
| 4985 | llvm::Value *Flags = EmitScalarExpr(E->getArg(1)); |
| 4986 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(2)); |
| 4987 | llvm::Value *Range = NDRangeL.getAddress(*this).getPointer(); |
| 4988 | llvm::Type *RangeTy = NDRangeL.getAddress(*this).getType(); |
| 4989 | |
| 4990 | if (NumArgs == 4) { |
| 4991 | // The most basic form of the call with parameters: |
| 4992 | // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void) |
| 4993 | Name = "__enqueue_kernel_basic"; |
| 4994 | llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy, |
| 4995 | GenericVoidPtrTy}; |
| 4996 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4997 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4998 | |
| 4999 | auto Info = |
| 5000 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
| 5001 | llvm::Value *Kernel = |
| 5002 | Builder.CreatePointerCast(Info.KernelHandle, GenericVoidPtrTy); |
| 5003 | llvm::Value *Block = |
| 5004 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 5005 | |
| 5006 | AttrBuilder B(Builder.getContext()); |
| 5007 | B.addByValAttr(NDRangeL.getAddress(*this).getElementType()); |
| 5008 | llvm::AttributeList ByValAttrSet = |
| 5009 | llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B); |
| 5010 | |
| 5011 | auto RTCall = |
| 5012 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet), |
| 5013 | {Queue, Flags, Range, Kernel, Block}); |
| 5014 | RTCall->setAttributes(ByValAttrSet); |
| 5015 | return RValue::get(RTCall); |
| 5016 | } |
| 5017 | assert(NumArgs >= 5 && "Invalid enqueue_kernel signature")(static_cast <bool> (NumArgs >= 5 && "Invalid enqueue_kernel signature" ) ? void (0) : __assert_fail ("NumArgs >= 5 && \"Invalid enqueue_kernel signature\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5017, __extension__ __PRETTY_FUNCTION__ )); |
| 5018 | |
| 5019 | // Create a temporary array to hold the sizes of local pointer arguments |
| 5020 | // for the block. \p First is the position of the first size argument. |
| 5021 | auto CreateArrayForSizeVar = [=](unsigned First) |
| 5022 | -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> { |
| 5023 | llvm::APInt ArraySize(32, NumArgs - First); |
| 5024 | QualType SizeArrayTy = getContext().getConstantArrayType( |
| 5025 | getContext().getSizeType(), ArraySize, nullptr, ArrayType::Normal, |
| 5026 | /*IndexTypeQuals=*/0); |
| 5027 | auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes"); |
| 5028 | llvm::Value *TmpPtr = Tmp.getPointer(); |
| 5029 | llvm::Value *TmpSize = EmitLifetimeStart( |
| 5030 | CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr); |
| 5031 | llvm::Value *ElemPtr; |
| 5032 | // Each of the following arguments specifies the size of the corresponding |
| 5033 | // argument passed to the enqueued block. |
| 5034 | auto *Zero = llvm::ConstantInt::get(IntTy, 0); |
| 5035 | for (unsigned I = First; I < NumArgs; ++I) { |
| 5036 | auto *Index = llvm::ConstantInt::get(IntTy, I - First); |
| 5037 | auto *GEP = Builder.CreateGEP(Tmp.getElementType(), TmpPtr, |
| 5038 | {Zero, Index}); |
| 5039 | if (I == First) |
| 5040 | ElemPtr = GEP; |
| 5041 | auto *V = |
| 5042 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy); |
| 5043 | Builder.CreateAlignedStore( |
| 5044 | V, GEP, CGM.getDataLayout().getPrefTypeAlign(SizeTy)); |
| 5045 | } |
| 5046 | return std::tie(ElemPtr, TmpSize, TmpPtr); |
| 5047 | }; |
| 5048 | |
| 5049 | // Could have events and/or varargs. |
| 5050 | if (E->getArg(3)->getType()->isBlockPointerType()) { |
| 5051 | // No events passed, but has variadic arguments. |
| 5052 | Name = "__enqueue_kernel_varargs"; |
| 5053 | auto Info = |
| 5054 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
| 5055 | llvm::Value *Kernel = |
| 5056 | Builder.CreatePointerCast(Info.KernelHandle, GenericVoidPtrTy); |
| 5057 | auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 5058 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
| 5059 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4); |
| 5060 | |
| 5061 | // Create a vector of the arguments, as well as a constant value to |
| 5062 | // express to the runtime the number of variadic arguments. |
| 5063 | llvm::Value *const Args[] = {Queue, Flags, |
| 5064 | Range, Kernel, |
| 5065 | Block, ConstantInt::get(IntTy, NumArgs - 4), |
| 5066 | ElemPtr}; |
| 5067 | llvm::Type *const ArgTys[] = { |
| 5068 | QueueTy, IntTy, RangeTy, GenericVoidPtrTy, |
| 5069 | GenericVoidPtrTy, IntTy, ElemPtr->getType()}; |
| 5070 | |
| 5071 | llvm::FunctionType *FTy = llvm::FunctionType::get(Int32Ty, ArgTys, false); |
| 5072 | auto Call = RValue::get( |
| 5073 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Args)); |
| 5074 | if (TmpSize) |
| 5075 | EmitLifetimeEnd(TmpSize, TmpPtr); |
| 5076 | return Call; |
| 5077 | } |
| 5078 | // Any calls now have event arguments passed. |
| 5079 | if (NumArgs >= 7) { |
| 5080 | llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy); |
| 5081 | llvm::PointerType *EventPtrTy = EventTy->getPointerTo( |
| 5082 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 5083 | |
| 5084 | llvm::Value *NumEvents = |
| 5085 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty); |
| 5086 | |
| 5087 | // Since SemaOpenCLBuiltinEnqueueKernel allows fifth and sixth arguments |
| 5088 | // to be a null pointer constant (including `0` literal), we can take it |
| 5089 | // into account and emit null pointer directly. |
| 5090 | llvm::Value *EventWaitList = nullptr; |
| 5091 | if (E->getArg(4)->isNullPointerConstant( |
| 5092 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
| 5093 | EventWaitList = llvm::ConstantPointerNull::get(EventPtrTy); |
| 5094 | } else { |
| 5095 | EventWaitList = E->getArg(4)->getType()->isArrayType() |
| 5096 | ? EmitArrayToPointerDecay(E->getArg(4)).getPointer() |
| 5097 | : EmitScalarExpr(E->getArg(4)); |
| 5098 | // Convert to generic address space. |
| 5099 | EventWaitList = Builder.CreatePointerCast(EventWaitList, EventPtrTy); |
| 5100 | } |
| 5101 | llvm::Value *EventRet = nullptr; |
| 5102 | if (E->getArg(5)->isNullPointerConstant( |
| 5103 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
| 5104 | EventRet = llvm::ConstantPointerNull::get(EventPtrTy); |
| 5105 | } else { |
| 5106 | EventRet = |
| 5107 | Builder.CreatePointerCast(EmitScalarExpr(E->getArg(5)), EventPtrTy); |
| 5108 | } |
| 5109 | |
| 5110 | auto Info = |
| 5111 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6)); |
| 5112 | llvm::Value *Kernel = |
| 5113 | Builder.CreatePointerCast(Info.KernelHandle, GenericVoidPtrTy); |
| 5114 | llvm::Value *Block = |
| 5115 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 5116 | |
| 5117 | std::vector<llvm::Type *> ArgTys = { |
| 5118 | QueueTy, Int32Ty, RangeTy, Int32Ty, |
| 5119 | EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy}; |
| 5120 | |
| 5121 | std::vector<llvm::Value *> Args = {Queue, Flags, Range, |
| 5122 | NumEvents, EventWaitList, EventRet, |
| 5123 | Kernel, Block}; |
| 5124 | |
| 5125 | if (NumArgs == 7) { |
| 5126 | // Has events but no variadics. |
| 5127 | Name = "__enqueue_kernel_basic_events"; |
| 5128 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 5129 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 5130 | return RValue::get( |
| 5131 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 5132 | llvm::ArrayRef<llvm::Value *>(Args))); |
| 5133 | } |
| 5134 | // Has event info and variadics |
| 5135 | // Pass the number of variadics to the runtime function too. |
| 5136 | Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7)); |
| 5137 | ArgTys.push_back(Int32Ty); |
| 5138 | Name = "__enqueue_kernel_events_varargs"; |
| 5139 | |
| 5140 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
| 5141 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7); |
| 5142 | Args.push_back(ElemPtr); |
| 5143 | ArgTys.push_back(ElemPtr->getType()); |
| 5144 | |
| 5145 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 5146 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 5147 | auto Call = |
| 5148 | RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 5149 | llvm::ArrayRef<llvm::Value *>(Args))); |
| 5150 | if (TmpSize) |
| 5151 | EmitLifetimeEnd(TmpSize, TmpPtr); |
| 5152 | return Call; |
| 5153 | } |
| 5154 | [[fallthrough]]; |
| 5155 | } |
| 5156 | // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block |
| 5157 | // parameter. |
| 5158 | case Builtin::BIget_kernel_work_group_size: { |
| 5159 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 5160 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 5161 | auto Info = |
| 5162 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
| 5163 | Value *Kernel = |
| 5164 | Builder.CreatePointerCast(Info.KernelHandle, GenericVoidPtrTy); |
| 5165 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 5166 | return RValue::get(EmitRuntimeCall( |
| 5167 | CGM.CreateRuntimeFunction( |
| 5168 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
| 5169 | false), |
| 5170 | "__get_kernel_work_group_size_impl"), |
| 5171 | {Kernel, Arg})); |
| 5172 | } |
| 5173 | case Builtin::BIget_kernel_preferred_work_group_size_multiple: { |
| 5174 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 5175 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 5176 | auto Info = |
| 5177 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
| 5178 | Value *Kernel = |
| 5179 | Builder.CreatePointerCast(Info.KernelHandle, GenericVoidPtrTy); |
| 5180 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 5181 | return RValue::get(EmitRuntimeCall( |
| 5182 | CGM.CreateRuntimeFunction( |
| 5183 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
| 5184 | false), |
| 5185 | "__get_kernel_preferred_work_group_size_multiple_impl"), |
| 5186 | {Kernel, Arg})); |
| 5187 | } |
| 5188 | case Builtin::BIget_kernel_max_sub_group_size_for_ndrange: |
| 5189 | case Builtin::BIget_kernel_sub_group_count_for_ndrange: { |
| 5190 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 5191 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 5192 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(0)); |
| 5193 | llvm::Value *NDRange = NDRangeL.getAddress(*this).getPointer(); |
| 5194 | auto Info = |
| 5195 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1)); |
| 5196 | Value *Kernel = |
| 5197 | Builder.CreatePointerCast(Info.KernelHandle, GenericVoidPtrTy); |
| 5198 | Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 5199 | const char *Name = |
| 5200 | BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange |
| 5201 | ? "__get_kernel_max_sub_group_size_for_ndrange_impl" |
| 5202 | : "__get_kernel_sub_group_count_for_ndrange_impl"; |
| 5203 | return RValue::get(EmitRuntimeCall( |
| 5204 | CGM.CreateRuntimeFunction( |
| 5205 | llvm::FunctionType::get( |
| 5206 | IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy}, |
| 5207 | false), |
| 5208 | Name), |
| 5209 | {NDRange, Kernel, Block})); |
| 5210 | } |
| 5211 | |
| 5212 | case Builtin::BI__builtin_store_half: |
| 5213 | case Builtin::BI__builtin_store_halff: { |
| 5214 | Value *Val = EmitScalarExpr(E->getArg(0)); |
| 5215 | Address Address = EmitPointerWithAlignment(E->getArg(1)); |
| 5216 | Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy()); |
| 5217 | Builder.CreateStore(HalfVal, Address); |
| 5218 | return RValue::get(nullptr); |
| 5219 | } |
| 5220 | case Builtin::BI__builtin_load_half: { |
| 5221 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
| 5222 | Value *HalfVal = Builder.CreateLoad(Address); |
| 5223 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy())); |
| 5224 | } |
| 5225 | case Builtin::BI__builtin_load_halff: { |
| 5226 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
| 5227 | Value *HalfVal = Builder.CreateLoad(Address); |
| 5228 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy())); |
| 5229 | } |
| 5230 | case Builtin::BIprintf: |
| 5231 | if (getTarget().getTriple().isNVPTX() || |
| 5232 | getTarget().getTriple().isAMDGCN()) { |
| 5233 | if (getLangOpts().OpenMPIsDevice) |
| 5234 | return EmitOpenMPDevicePrintfCallExpr(E); |
| 5235 | if (getTarget().getTriple().isNVPTX()) |
| 5236 | return EmitNVPTXDevicePrintfCallExpr(E); |
| 5237 | if (getTarget().getTriple().isAMDGCN() && getLangOpts().HIP) |
| 5238 | return EmitAMDGPUDevicePrintfCallExpr(E); |
| 5239 | } |
| 5240 | |
| 5241 | break; |
| 5242 | case Builtin::BI__builtin_canonicalize: |
| 5243 | case Builtin::BI__builtin_canonicalizef: |
| 5244 | case Builtin::BI__builtin_canonicalizef16: |
| 5245 | case Builtin::BI__builtin_canonicalizel: |
| 5246 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize)); |
| 5247 | |
| 5248 | case Builtin::BI__builtin_thread_pointer: { |
| 5249 | if (!getContext().getTargetInfo().isTLSSupported()) |
| 5250 | CGM.ErrorUnsupported(E, "__builtin_thread_pointer"); |
| 5251 | // Fall through - it's already mapped to the intrinsic by ClangBuiltin. |
| 5252 | break; |
| 5253 | } |
| 5254 | case Builtin::BI__builtin_os_log_format: |
| 5255 | return emitBuiltinOSLogFormat(*E); |
| 5256 | |
| 5257 | case Builtin::BI__xray_customevent: { |
| 5258 | if (!ShouldXRayInstrumentFunction()) |
| 5259 | return RValue::getIgnored(); |
| 5260 | |
| 5261 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
| 5262 | XRayInstrKind::Custom)) |
| 5263 | return RValue::getIgnored(); |
| 5264 | |
| 5265 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
| 5266 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents()) |
| 5267 | return RValue::getIgnored(); |
| 5268 | |
| 5269 | Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent); |
| 5270 | auto FTy = F->getFunctionType(); |
| 5271 | auto Arg0 = E->getArg(0); |
| 5272 | auto Arg0Val = EmitScalarExpr(Arg0); |
| 5273 | auto Arg0Ty = Arg0->getType(); |
| 5274 | auto PTy0 = FTy->getParamType(0); |
| 5275 | if (PTy0 != Arg0Val->getType()) { |
| 5276 | if (Arg0Ty->isArrayType()) |
| 5277 | Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer(); |
| 5278 | else |
| 5279 | Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0); |
| 5280 | } |
| 5281 | auto Arg1 = EmitScalarExpr(E->getArg(1)); |
| 5282 | auto PTy1 = FTy->getParamType(1); |
| 5283 | if (PTy1 != Arg1->getType()) |
| 5284 | Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1); |
| 5285 | return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1})); |
| 5286 | } |
| 5287 | |
| 5288 | case Builtin::BI__xray_typedevent: { |
| 5289 | // TODO: There should be a way to always emit events even if the current |
| 5290 | // function is not instrumented. Losing events in a stream can cripple |
| 5291 | // a trace. |
| 5292 | if (!ShouldXRayInstrumentFunction()) |
| 5293 | return RValue::getIgnored(); |
| 5294 | |
| 5295 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
| 5296 | XRayInstrKind::Typed)) |
| 5297 | return RValue::getIgnored(); |
| 5298 | |
| 5299 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
| 5300 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents()) |
| 5301 | return RValue::getIgnored(); |
| 5302 | |
| 5303 | Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent); |
| 5304 | auto FTy = F->getFunctionType(); |
| 5305 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
| 5306 | auto PTy0 = FTy->getParamType(0); |
| 5307 | if (PTy0 != Arg0->getType()) |
| 5308 | Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0); |
| 5309 | auto Arg1 = E->getArg(1); |
| 5310 | auto Arg1Val = EmitScalarExpr(Arg1); |
| 5311 | auto Arg1Ty = Arg1->getType(); |
| 5312 | auto PTy1 = FTy->getParamType(1); |
| 5313 | if (PTy1 != Arg1Val->getType()) { |
| 5314 | if (Arg1Ty->isArrayType()) |
| 5315 | Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer(); |
| 5316 | else |
| 5317 | Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1); |
| 5318 | } |
| 5319 | auto Arg2 = EmitScalarExpr(E->getArg(2)); |
| 5320 | auto PTy2 = FTy->getParamType(2); |
| 5321 | if (PTy2 != Arg2->getType()) |
| 5322 | Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2); |
| 5323 | return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2})); |
| 5324 | } |
| 5325 | |
| 5326 | case Builtin::BI__builtin_ms_va_start: |
| 5327 | case Builtin::BI__builtin_ms_va_end: |
| 5328 | return RValue::get( |
| 5329 | EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(), |
| 5330 | BuiltinID == Builtin::BI__builtin_ms_va_start)); |
| 5331 | |
| 5332 | case Builtin::BI__builtin_ms_va_copy: { |
| 5333 | // Lower this manually. We can't reliably determine whether or not any |
| 5334 | // given va_copy() is for a Win64 va_list from the calling convention |
| 5335 | // alone, because it's legal to do this from a System V ABI function. |
| 5336 | // With opaque pointer types, we won't have enough information in LLVM |
| 5337 | // IR to determine this from the argument types, either. Best to do it |
| 5338 | // now, while we have enough information. |
| 5339 | Address DestAddr = EmitMSVAListRef(E->getArg(0)); |
| 5340 | Address SrcAddr = EmitMSVAListRef(E->getArg(1)); |
| 5341 | |
| 5342 | llvm::Type *BPP = Int8PtrPtrTy; |
| 5343 | |
| 5344 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"), |
| 5345 | Int8PtrTy, DestAddr.getAlignment()); |
| 5346 | SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"), |
| 5347 | Int8PtrTy, SrcAddr.getAlignment()); |
| 5348 | |
| 5349 | Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val"); |
| 5350 | return RValue::get(Builder.CreateStore(ArgPtr, DestAddr)); |
| 5351 | } |
| 5352 | |
| 5353 | case Builtin::BI__builtin_get_device_side_mangled_name: { |
| 5354 | auto Name = CGM.getCUDARuntime().getDeviceSideName( |
| 5355 | cast<DeclRefExpr>(E->getArg(0)->IgnoreImpCasts())->getDecl()); |
| 5356 | auto Str = CGM.GetAddrOfConstantCString(Name, ""); |
| 5357 | llvm::Constant *Zeros[] = {llvm::ConstantInt::get(SizeTy, 0), |
| 5358 | llvm::ConstantInt::get(SizeTy, 0)}; |
| 5359 | auto *Ptr = llvm::ConstantExpr::getGetElementPtr(Str.getElementType(), |
| 5360 | Str.getPointer(), Zeros); |
| 5361 | return RValue::get(Ptr); |
| 5362 | } |
| 5363 | } |
| 5364 | |
| 5365 | // If this is an alias for a lib function (e.g. __builtin_sin), emit |
| 5366 | // the call using the normal call path, but using the unmangled |
| 5367 | // version of the function name. |
| 5368 | if (getContext().BuiltinInfo.isLibFunction(BuiltinID)) |
| 5369 | return emitLibraryCall(*this, FD, E, |
| 5370 | CGM.getBuiltinLibFunction(FD, BuiltinID)); |
| 5371 | |
| 5372 | // If this is a predefined lib function (e.g. malloc), emit the call |
| 5373 | // using exactly the normal call path. |
| 5374 | if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID)) |
| 5375 | return emitLibraryCall(*this, FD, E, |
| 5376 | cast<llvm::Constant>(EmitScalarExpr(E->getCallee()))); |
| 5377 | |
| 5378 | // Check that a call to a target specific builtin has the correct target |
| 5379 | // features. |
| 5380 | // This is down here to avoid non-target specific builtins, however, if |
| 5381 | // generic builtins start to require generic target features then we |
| 5382 | // can move this up to the beginning of the function. |
| 5383 | checkTargetFeatures(E, FD); |
| 5384 | |
| 5385 | if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID)) |
| 5386 | LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth); |
| 5387 | |
| 5388 | // See if we have a target specific intrinsic. |
| 5389 | StringRef Name = getContext().BuiltinInfo.getName(BuiltinID); |
| 5390 | Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic; |
| 5391 | StringRef Prefix = |
| 5392 | llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch()); |
| 5393 | if (!Prefix.empty()) { |
| 5394 | IntrinsicID = Intrinsic::getIntrinsicForClangBuiltin(Prefix.data(), Name); |
| 5395 | // NOTE we don't need to perform a compatibility flag check here since the |
| 5396 | // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the |
| 5397 | // MS builtins via ALL_MS_LANGUAGES and are filtered earlier. |
| 5398 | if (IntrinsicID == Intrinsic::not_intrinsic) |
| 5399 | IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name); |
| 5400 | } |
| 5401 | |
| 5402 | if (IntrinsicID != Intrinsic::not_intrinsic) { |
| 5403 | SmallVector<Value*, 16> Args; |
| 5404 | |
| 5405 | // Find out if any arguments are required to be integer constant |
| 5406 | // expressions. |
| 5407 | unsigned ICEArguments = 0; |
| 5408 | ASTContext::GetBuiltinTypeError Error; |
| 5409 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); |
| 5410 | assert(Error == ASTContext::GE_None && "Should not codegen an error")(static_cast <bool> (Error == ASTContext::GE_None && "Should not codegen an error") ? void (0) : __assert_fail ("Error == ASTContext::GE_None && \"Should not codegen an error\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5410, __extension__ __PRETTY_FUNCTION__ )); |
| 5411 | |
| 5412 | Function *F = CGM.getIntrinsic(IntrinsicID); |
| 5413 | llvm::FunctionType *FTy = F->getFunctionType(); |
| 5414 | |
| 5415 | for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) { |
| 5416 | Value *ArgValue; |
| 5417 | // If this is a normal argument, just emit it as a scalar. |
| 5418 | if ((ICEArguments & (1 << i)) == 0) { |
| 5419 | ArgValue = EmitScalarExpr(E->getArg(i)); |
| 5420 | } else { |
| 5421 | // If this is required to be a constant, constant fold it so that we |
| 5422 | // know that the generated intrinsic gets a ConstantInt. |
| 5423 | ArgValue = llvm::ConstantInt::get( |
| 5424 | getLLVMContext(), |
| 5425 | *E->getArg(i)->getIntegerConstantExpr(getContext())); |
| 5426 | } |
| 5427 | |
| 5428 | // If the intrinsic arg type is different from the builtin arg type |
| 5429 | // we need to do a bit cast. |
| 5430 | llvm::Type *PTy = FTy->getParamType(i); |
| 5431 | if (PTy != ArgValue->getType()) { |
| 5432 | // XXX - vector of pointers? |
| 5433 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) { |
| 5434 | if (PtrTy->getAddressSpace() != |
| 5435 | ArgValue->getType()->getPointerAddressSpace()) { |
| 5436 | ArgValue = Builder.CreateAddrSpaceCast( |
| 5437 | ArgValue, |
| 5438 | ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace())); |
| 5439 | } |
| 5440 | } |
| 5441 | |
| 5442 | assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&(static_cast <bool> (PTy->canLosslesslyBitCastTo(FTy ->getParamType(i)) && "Must be able to losslessly bit cast to param" ) ? void (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5443, __extension__ __PRETTY_FUNCTION__ )) |
| 5443 | "Must be able to losslessly bit cast to param")(static_cast <bool> (PTy->canLosslesslyBitCastTo(FTy ->getParamType(i)) && "Must be able to losslessly bit cast to param" ) ? void (0) : __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5443, __extension__ __PRETTY_FUNCTION__ )); |
| 5444 | // Cast vector type (e.g., v256i32) to x86_amx, this only happen |
| 5445 | // in amx intrinsics. |
| 5446 | if (PTy->isX86_AMXTy()) |
| 5447 | ArgValue = Builder.CreateIntrinsic(Intrinsic::x86_cast_vector_to_tile, |
| 5448 | {ArgValue->getType()}, {ArgValue}); |
| 5449 | else |
| 5450 | ArgValue = Builder.CreateBitCast(ArgValue, PTy); |
| 5451 | } |
| 5452 | |
| 5453 | Args.push_back(ArgValue); |
| 5454 | } |
| 5455 | |
| 5456 | Value *V = Builder.CreateCall(F, Args); |
| 5457 | QualType BuiltinRetType = E->getType(); |
| 5458 | |
| 5459 | llvm::Type *RetTy = VoidTy; |
| 5460 | if (!BuiltinRetType->isVoidType()) |
| 5461 | RetTy = ConvertType(BuiltinRetType); |
| 5462 | |
| 5463 | if (RetTy != V->getType()) { |
| 5464 | // XXX - vector of pointers? |
| 5465 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) { |
| 5466 | if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) { |
| 5467 | V = Builder.CreateAddrSpaceCast( |
| 5468 | V, V->getType()->getPointerTo(PtrTy->getAddressSpace())); |
| 5469 | } |
| 5470 | } |
| 5471 | |
| 5472 | assert(V->getType()->canLosslesslyBitCastTo(RetTy) &&(static_cast <bool> (V->getType()->canLosslesslyBitCastTo (RetTy) && "Must be able to losslessly bit cast result type" ) ? void (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5473, __extension__ __PRETTY_FUNCTION__ )) |
| 5473 | "Must be able to losslessly bit cast result type")(static_cast <bool> (V->getType()->canLosslesslyBitCastTo (RetTy) && "Must be able to losslessly bit cast result type" ) ? void (0) : __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5473, __extension__ __PRETTY_FUNCTION__ )); |
| 5474 | // Cast x86_amx to vector type (e.g., v256i32), this only happen |
| 5475 | // in amx intrinsics. |
| 5476 | if (V->getType()->isX86_AMXTy()) |
| 5477 | V = Builder.CreateIntrinsic(Intrinsic::x86_cast_tile_to_vector, {RetTy}, |
| 5478 | {V}); |
| 5479 | else |
| 5480 | V = Builder.CreateBitCast(V, RetTy); |
| 5481 | } |
| 5482 | |
| 5483 | if (RetTy->isVoidTy()) |
| 5484 | return RValue::get(nullptr); |
| 5485 | |
| 5486 | return RValue::get(V); |
| 5487 | } |
| 5488 | |
| 5489 | // Some target-specific builtins can have aggregate return values, e.g. |
| 5490 | // __builtin_arm_mve_vld2q_u32. So if the result is an aggregate, force |
| 5491 | // ReturnValue to be non-null, so that the target-specific emission code can |
| 5492 | // always just emit into it. |
| 5493 | TypeEvaluationKind EvalKind = getEvaluationKind(E->getType()); |
| 5494 | if (EvalKind == TEK_Aggregate && ReturnValue.isNull()) { |
| 5495 | Address DestPtr = CreateMemTemp(E->getType(), "agg.tmp"); |
| 5496 | ReturnValue = ReturnValueSlot(DestPtr, false); |
| 5497 | } |
| 5498 | |
| 5499 | // Now see if we can emit a target-specific builtin. |
| 5500 | if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E, ReturnValue)) { |
| 5501 | switch (EvalKind) { |
| 5502 | case TEK_Scalar: |
| 5503 | if (V->getType()->isVoidTy()) |
| 5504 | return RValue::get(nullptr); |
| 5505 | return RValue::get(V); |
| 5506 | case TEK_Aggregate: |
| 5507 | return RValue::getAggregate(ReturnValue.getValue(), |
| 5508 | ReturnValue.isVolatile()); |
| 5509 | case TEK_Complex: |
| 5510 | llvm_unreachable("No current target builtin returns complex")::llvm::llvm_unreachable_internal("No current target builtin returns complex" , "clang/lib/CodeGen/CGBuiltin.cpp", 5510); |
| 5511 | } |
| 5512 | llvm_unreachable("Bad evaluation kind in EmitBuiltinExpr")::llvm::llvm_unreachable_internal("Bad evaluation kind in EmitBuiltinExpr" , "clang/lib/CodeGen/CGBuiltin.cpp", 5512); |
| 5513 | } |
| 5514 | |
| 5515 | ErrorUnsupported(E, "builtin function"); |
| 5516 | |
| 5517 | // Unknown builtin, for now just dump it out and return undef. |
| 5518 | return GetUndefRValue(E->getType()); |
| 5519 | } |
| 5520 | |
| 5521 | static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF, |
| 5522 | unsigned BuiltinID, const CallExpr *E, |
| 5523 | ReturnValueSlot ReturnValue, |
| 5524 | llvm::Triple::ArchType Arch) { |
| 5525 | switch (Arch) { |
| 5526 | case llvm::Triple::arm: |
| 5527 | case llvm::Triple::armeb: |
| 5528 | case llvm::Triple::thumb: |
| 5529 | case llvm::Triple::thumbeb: |
| 5530 | return CGF->EmitARMBuiltinExpr(BuiltinID, E, ReturnValue, Arch); |
| 5531 | case llvm::Triple::aarch64: |
| 5532 | case llvm::Triple::aarch64_32: |
| 5533 | case llvm::Triple::aarch64_be: |
| 5534 | return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch); |
| 5535 | case llvm::Triple::bpfeb: |
| 5536 | case llvm::Triple::bpfel: |
| 5537 | return CGF->EmitBPFBuiltinExpr(BuiltinID, E); |
| 5538 | case llvm::Triple::x86: |
| 5539 | case llvm::Triple::x86_64: |
| 5540 | return CGF->EmitX86BuiltinExpr(BuiltinID, E); |
| 5541 | case llvm::Triple::ppc: |
| 5542 | case llvm::Triple::ppcle: |
| 5543 | case llvm::Triple::ppc64: |
| 5544 | case llvm::Triple::ppc64le: |
| 5545 | return CGF->EmitPPCBuiltinExpr(BuiltinID, E); |
| 5546 | case llvm::Triple::r600: |
| 5547 | case llvm::Triple::amdgcn: |
| 5548 | return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E); |
| 5549 | case llvm::Triple::systemz: |
| 5550 | return CGF->EmitSystemZBuiltinExpr(BuiltinID, E); |
| 5551 | case llvm::Triple::nvptx: |
| 5552 | case llvm::Triple::nvptx64: |
| 5553 | return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E); |
| 5554 | case llvm::Triple::wasm32: |
| 5555 | case llvm::Triple::wasm64: |
| 5556 | return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E); |
| 5557 | case llvm::Triple::hexagon: |
| 5558 | return CGF->EmitHexagonBuiltinExpr(BuiltinID, E); |
| 5559 | case llvm::Triple::riscv32: |
| 5560 | case llvm::Triple::riscv64: |
| 5561 | return CGF->EmitRISCVBuiltinExpr(BuiltinID, E, ReturnValue); |
| 5562 | case llvm::Triple::loongarch32: |
| 5563 | case llvm::Triple::loongarch64: |
| 5564 | return CGF->EmitLoongArchBuiltinExpr(BuiltinID, E); |
| 5565 | default: |
| 5566 | return nullptr; |
| 5567 | } |
| 5568 | } |
| 5569 | |
| 5570 | Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID, |
| 5571 | const CallExpr *E, |
| 5572 | ReturnValueSlot ReturnValue) { |
| 5573 | if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) { |
| 5574 | assert(getContext().getAuxTargetInfo() && "Missing aux target info")(static_cast <bool> (getContext().getAuxTargetInfo() && "Missing aux target info") ? void (0) : __assert_fail ("getContext().getAuxTargetInfo() && \"Missing aux target info\"" , "clang/lib/CodeGen/CGBuiltin.cpp", 5574, __extension__ __PRETTY_FUNCTION__ )); |
| 5575 | return EmitTargetArchBuiltinExpr( |
| 5576 | this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E, |
| 5577 | ReturnValue, getContext().getAuxTargetInfo()->getTriple().getArch()); |
| 5578 | } |
| 5579 | |
| 5580 | return EmitTargetArchBuiltinExpr(this, BuiltinID, E, ReturnValue, |
| 5581 | getTarget().getTriple().getArch()); |
| 5582 | } |
| 5583 | |
| 5584 | static llvm::FixedVectorType *GetNeonType(CodeGenFunction *CGF, |
| 5585 | NeonTypeFlags TypeFlags, |
| 5586 | bool HasLegalHalfType = true, |
| 5587 | bool V1Ty = false, |
| 5588 | bool AllowBFloatArgsAndRet = true) { |
| 5589 | int IsQuad = TypeFlags.isQuad(); |
| 5590 | switch (TypeFlags.getEltType()) { |
| 5591 | case NeonTypeFlags::Int8: |
| 5592 | case NeonTypeFlags::Poly8: |
| 5593 | return llvm::FixedVectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad)); |
| 5594 | case NeonTypeFlags::Int16: |
| 5595 | case NeonTypeFlags::Poly16: |
| 5596 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 5597 | case NeonTypeFlags::BFloat16: |
| 5598 | if (AllowBFloatArgsAndRet) |
| 5599 | return llvm::FixedVectorType::get(CGF->BFloatTy, V1Ty ? 1 : (4 << IsQuad)); |
| 5600 | else |
| 5601 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 5602 | case NeonTypeFlags::Float16: |
| 5603 | if (HasLegalHalfType) |
| 5604 | return llvm::FixedVectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad)); |
| 5605 | else |
| 5606 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 5607 | case NeonTypeFlags::Int32: |
| 5608 | return llvm::FixedVectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad)); |
| 5609 | case NeonTypeFlags::Int64: |
| 5610 | case NeonTypeFlags::Poly64: |
| 5611 | return llvm::FixedVectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad)); |
| 5612 | case NeonTypeFlags::Poly128: |
| 5613 | // FIXME: i128 and f128 doesn't get fully support in Clang and llvm. |
| 5614 | // There is a lot of i128 and f128 API missing. |
| 5615 | // so we use v16i8 to represent poly128 and get pattern matched. |
| 5616 | return llvm::FixedVectorType::get(CGF->Int8Ty, 16); |
| 5617 | case NeonTypeFlags::Float32: |
| 5618 | return llvm::FixedVectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad)); |
| 5619 | case NeonTypeFlags::Float64: |
| 5620 | return llvm::FixedVectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad)); |
| 5621 | } |
| 5622 | llvm_unreachable("Unknown vector element type!")::llvm::llvm_unreachable_internal("Unknown vector element type!" , "clang/lib/CodeGen/CGBuiltin.cpp", 5622); |
| 5623 | } |
| 5624 | |
| 5625 | static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF, |
| 5626 | NeonTypeFlags IntTypeFlags) { |
| 5627 | int IsQuad = IntTypeFlags.isQuad(); |
| 5628 | switch (IntTypeFlags.getEltType()) { |
| 5629 | case NeonTypeFlags::Int16: |
| 5630 | return llvm::FixedVectorType::get(CGF->HalfTy, (4 << IsQuad)); |
| 5631 | case NeonTypeFlags::Int32: |
| 5632 | return llvm::FixedVectorType::get(CGF->FloatTy, (2 << IsQuad)); |
| 5633 | case NeonTypeFlags::Int64: |
| 5634 | return llvm::FixedVectorType::get(CGF->DoubleTy, (1 << IsQuad)); |
| 5635 | default: |
| 5636 | llvm_unreachable("Type can't be converted to floating-point!")::llvm::llvm_unreachable_internal("Type can't be converted to floating-point!" , "clang/lib/CodeGen/CGBuiltin.cpp", 5636); |
| 5637 | } |
| 5638 | } |
| 5639 | |
| 5640 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C, |
| 5641 | const ElementCount &Count) { |
| 5642 | Value *SV = llvm::ConstantVector::getSplat(Count, C); |
| 5643 | return Builder.CreateShuffleVector(V, V, SV, "lane"); |
| 5644 | } |
| 5645 | |
| 5646 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) { |
| 5647 | ElementCount EC = cast<llvm::VectorType>(V->getType())->getElementCount(); |
| 5648 | return EmitNeonSplat(V, C, EC); |
| 5649 | } |
| 5650 | |
| 5651 | Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops, |
| 5652 | const char *name, |
| 5653 | unsigned shift, bool rightshift) { |
| 5654 | unsigned j = 0; |
| 5655 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); |
| 5656 | ai != ae; ++ai, ++j) { |
| 5657 | if (F->isConstrainedFPIntrinsic()) |
| 5658 | if (ai->getType()->isMetadataTy()) |
| 5659 | continue; |
| 5660 | if (shift > 0 && shift == j) |
| 5661 | Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift); |
| 5662 | else |
| 5663 | Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name); |
| 5664 | } |
| 5665 | |
| 5666 | if (F->isConstrainedFPIntrinsic()) |
| 5667 | return Builder.CreateConstrainedFPCall(F, Ops, name); |
| 5668 | else |
| 5669 | return Builder.CreateCall(F, Ops, name); |
| 5670 | } |
| 5671 | |
| 5672 | Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty, |
| 5673 | bool neg) { |
| 5674 | int SV = cast<ConstantInt>(V)->getSExtValue(); |
| 5675 | return ConstantInt::get(Ty, neg ? -SV : SV); |
| 5676 | } |
| 5677 | |
| 5678 | // Right-shift a vector by a constant. |
| 5679 | Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift, |
| 5680 | llvm::Type *Ty, bool usgn, |
| 5681 | const char *name) { |
| 5682 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); |
| 5683 | |
| 5684 | int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue(); |
| 5685 | int EltSize = VTy->getScalarSizeInBits(); |
| 5686 | |
| 5687 | Vec = Builder.CreateBitCast(Vec, Ty); |
| 5688 | |
| 5689 | // lshr/ashr are undefined when the shift amount is equal to the vector |
| 5690 | // element size. |
| 5691 | if (ShiftAmt == EltSize) { |
| 5692 | if (usgn) { |
| 5693 | // Right-shifting an unsigned value by its size yields 0. |
| 5694 | return llvm::ConstantAggregateZero::get(VTy); |
| 5695 | } else { |
| 5696 | // Right-shifting a signed value by its size is equivalent |
| 5697 | // to a shift of size-1. |
| 5698 | --ShiftAmt; |
| 5699 | Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt); |
| 5700 | } |
| 5701 | } |
| 5702 | |
| 5703 | Shift = EmitNeonShiftVector(Shift, Ty, false); |
| 5704 | if (usgn) |
| 5705 | return Builder.CreateLShr(Vec, Shift, name); |
| 5706 | else |
| 5707 | return Builder.CreateAShr(Vec, Shift, name); |
| 5708 | } |
| 5709 | |
| 5710 | enum { |
| 5711 | AddRetType = (1 << 0), |
| 5712 | Add1ArgType = (1 << 1), |
| 5713 | Add2ArgTypes = (1 << 2), |
| 5714 | |
| 5715 | VectorizeRetType = (1 << 3), |
| 5716 | VectorizeArgTypes = (1 << 4), |
| 5717 | |
| 5718 | InventFloatType = (1 << 5), |
| 5719 | UnsignedAlts = (1 << 6), |
| 5720 | |
| 5721 | Use64BitVectors = (1 << 7), |
| 5722 | Use128BitVectors = (1 << 8), |
| 5723 | |
| 5724 | Vectorize1ArgType = Add1ArgType | VectorizeArgTypes, |
| 5725 | VectorRet = AddRetType | VectorizeRetType, |
| 5726 | VectorRetGetArgs01 = |
| 5727 | AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes, |
| 5728 | FpCmpzModifiers = |
| 5729 | AddRetType | VectorizeRetType | Add1ArgType | InventFloatType |
| 5730 | }; |
| 5731 | |
| 5732 | namespace { |
| 5733 | struct ARMVectorIntrinsicInfo { |
| 5734 | const char *NameHint; |
| 5735 | unsigned BuiltinID; |
| 5736 | unsigned LLVMIntrinsic; |
| 5737 | unsigned AltLLVMIntrinsic; |
| 5738 | uint64_t TypeModifier; |
| 5739 | |
| 5740 | bool operator<(unsigned RHSBuiltinID) const { |
| 5741 | return BuiltinID < RHSBuiltinID; |
| 5742 | } |
| 5743 | bool operator<(const ARMVectorIntrinsicInfo &TE) const { |
| 5744 | return BuiltinID < TE.BuiltinID; |
| 5745 | } |
| 5746 | }; |
| 5747 | } // end anonymous namespace |
| 5748 | |
| 5749 | #define NEONMAP0(NameBase) \ |
| 5750 | { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 } |
| 5751 | |
| 5752 | #define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \ |
| 5753 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
| 5754 | Intrinsic::LLVMIntrinsic, 0, TypeModifier } |
| 5755 | |
| 5756 | #define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \ |
| 5757 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
| 5758 | Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \ |
| 5759 | TypeModifier } |
| 5760 | |
| 5761 | static const ARMVectorIntrinsicInfo ARMSIMDIntrinsicMap [] = { |
| 5762 | NEONMAP1(__a32_vcvt_bf16_f32, arm_neon_vcvtfp2bf, 0), |
| 5763 | NEONMAP0(splat_lane_v), |
| 5764 | NEONMAP0(splat_laneq_v), |
| 5765 | NEONMAP0(splatq_lane_v), |
| 5766 | NEONMAP0(splatq_laneq_v), |
| 5767 | NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
| 5768 | NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
| 5769 | NEONMAP1(vabs_v, arm_neon_vabs, 0), |
| 5770 | NEONMAP1(vabsq_v, arm_neon_vabs, 0), |
| 5771 | NEONMAP0(vadd_v), |
| 5772 | NEONMAP0(vaddhn_v), |
| 5773 | NEONMAP0(vaddq_v), |
| 5774 | NEONMAP1(vaesdq_u8, arm_neon_aesd, 0), |
| 5775 | NEONMAP1(vaeseq_u8, arm_neon_aese, 0), |
| 5776 | NEONMAP1(vaesimcq_u8, arm_neon_aesimc, 0), |
| 5777 | NEONMAP1(vaesmcq_u8, arm_neon_aesmc, 0), |
| 5778 | NEONMAP1(vbfdot_f32, arm_neon_bfdot, 0), |
| 5779 | NEONMAP1(vbfdotq_f32, arm_neon_bfdot, 0), |
| 5780 | NEONMAP1(vbfmlalbq_f32, arm_neon_bfmlalb, 0), |
| 5781 | NEONMAP1(vbfmlaltq_f32, arm_neon_bfmlalt, 0), |
| 5782 | NEONMAP1(vbfmmlaq_f32, arm_neon_bfmmla, 0), |
| 5783 | NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType), |
| 5784 | NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType), |
| 5785 | NEONMAP1(vcadd_rot270_f16, arm_neon_vcadd_rot270, Add1ArgType), |
| 5786 | NEONMAP1(vcadd_rot270_f32, arm_neon_vcadd_rot270, Add1ArgType), |
| 5787 | NEONMAP1(vcadd_rot90_f16, arm_neon_vcadd_rot90, Add1ArgType), |
| 5788 | NEONMAP1(vcadd_rot90_f32, arm_neon_vcadd_rot90, Add1ArgType), |
| 5789 | NEONMAP1(vcaddq_rot270_f16, arm_neon_vcadd_rot270, Add1ArgType), |
| 5790 | NEONMAP1(vcaddq_rot270_f32, arm_neon_vcadd_rot270, Add1ArgType), |
| 5791 | NEONMAP1(vcaddq_rot270_f64, arm_neon_vcadd_rot270, Add1ArgType), |
| 5792 | NEONMAP1(vcaddq_rot90_f16, arm_neon_vcadd_rot90, Add1ArgType), |
| 5793 | NEONMAP1(vcaddq_rot90_f32, arm_neon_vcadd_rot90, Add1ArgType), |
| 5794 | NEONMAP1(vcaddq_rot90_f64, arm_neon_vcadd_rot90, Add1ArgType), |
| 5795 | NEONMAP1(vcage_v, arm_neon_vacge, 0), |
| 5796 | NEONMAP1(vcageq_v, arm_neon_vacge, 0), |
| 5797 | NEONMAP1(vcagt_v, arm_neon_vacgt, 0), |
| 5798 | NEONMAP1(vcagtq_v, arm_neon_vacgt, 0), |
| 5799 | NEONMAP1(vcale_v, arm_neon_vacge, 0), |
| 5800 | NEONMAP1(vcaleq_v, arm_neon_vacge, 0), |
| 5801 | NEONMAP1(vcalt_v, arm_neon_vacgt, 0), |
| 5802 | NEONMAP1(vcaltq_v, arm_neon_vacgt, 0), |
| 5803 | NEONMAP0(vceqz_v), |
| 5804 | NEONMAP0(vceqzq_v), |
| 5805 | NEONMAP0(vcgez_v), |
| 5806 | NEONMAP0(vcgezq_v), |
| 5807 | NEONMAP0(vcgtz_v), |
| 5808 | NEONMAP0(vcgtzq_v), |
| 5809 | NEONMAP0(vclez_v), |
| 5810 | NEONMAP0(vclezq_v), |
| 5811 | NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType), |
| 5812 | NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType), |
| 5813 | NEONMAP0(vcltz_v), |
| 5814 | NEONMAP0(vcltzq_v), |
| 5815 | NEONMAP1(vclz_v, ctlz, Add1ArgType), |
| 5816 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), |
| 5817 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), |
| 5818 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), |
| 5819 | NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0), |
| 5820 | NEONMAP0(vcvt_f16_s16), |
| 5821 | NEONMAP0(vcvt_f16_u16), |
| 5822 | NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0), |
| 5823 | NEONMAP0(vcvt_f32_v), |
| 5824 | NEONMAP1(vcvt_n_f16_s16, arm_neon_vcvtfxs2fp, 0), |
| 5825 | NEONMAP1(vcvt_n_f16_u16, arm_neon_vcvtfxu2fp, 0), |
| 5826 | NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 5827 | NEONMAP1(vcvt_n_s16_f16, arm_neon_vcvtfp2fxs, 0), |
| 5828 | NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
| 5829 | NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
| 5830 | NEONMAP1(vcvt_n_u16_f16, arm_neon_vcvtfp2fxu, 0), |
| 5831 | NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
| 5832 | NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
| 5833 | NEONMAP0(vcvt_s16_f16), |
| 5834 | NEONMAP0(vcvt_s32_v), |
| 5835 | NEONMAP0(vcvt_s64_v), |
| 5836 | NEONMAP0(vcvt_u16_f16), |
| 5837 | NEONMAP0(vcvt_u32_v), |
| 5838 | NEONMAP0(vcvt_u64_v), |
| 5839 | NEONMAP1(vcvta_s16_f16, arm_neon_vcvtas, 0), |
| 5840 | NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0), |
| 5841 | NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0), |
| 5842 | NEONMAP1(vcvta_u16_f16, arm_neon_vcvtau, 0), |
| 5843 | NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0), |
| 5844 | NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0), |
| 5845 | NEONMAP1(vcvtaq_s16_f16, arm_neon_vcvtas, 0), |
| 5846 | NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0), |
| 5847 | NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0), |
| 5848 | NEONMAP1(vcvtaq_u16_f16, arm_neon_vcvtau, 0), |
| 5849 | NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0), |
| 5850 | NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0), |
| 5851 | NEONMAP1(vcvth_bf16_f32, arm_neon_vcvtbfp2bf, 0), |
| 5852 | NEONMAP1(vcvtm_s16_f16, arm_neon_vcvtms, 0), |
| 5853 | NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0), |
| 5854 | NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0), |
| 5855 | NEONMAP1(vcvtm_u16_f16, arm_neon_vcvtmu, 0), |
| 5856 | NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0), |
| 5857 | NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0), |
| 5858 | NEONMAP1(vcvtmq_s16_f16, arm_neon_vcvtms, 0), |
| 5859 | NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0), |
| 5860 | NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0), |
| 5861 | NEONMAP1(vcvtmq_u16_f16, arm_neon_vcvtmu, 0), |
| 5862 | NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0), |
| 5863 | NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0), |
| 5864 | NEONMAP1(vcvtn_s16_f16, arm_neon_vcvtns, 0), |
| 5865 | NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0), |
| 5866 | NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0), |
| 5867 | NEONMAP1(vcvtn_u16_f16, arm_neon_vcvtnu, 0), |
| 5868 | NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0), |
| 5869 | NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0), |
| 5870 | NEONMAP1(vcvtnq_s16_f16, arm_neon_vcvtns, 0), |
| 5871 | NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0), |
| 5872 | NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0), |
| 5873 | NEONMAP1(vcvtnq_u16_f16, arm_neon_vcvtnu, 0), |
| 5874 | NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0), |
| 5875 | NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0), |
| 5876 | NEONMAP1(vcvtp_s16_f16, arm_neon_vcvtps, 0), |
| 5877 | NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0), |
| 5878 | NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0), |
| 5879 | NEONMAP1(vcvtp_u16_f16, arm_neon_vcvtpu, 0), |
| 5880 | NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0), |
| 5881 | NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0), |
| 5882 | NEONMAP1(vcvtpq_s16_f16, arm_neon_vcvtps, 0), |
| 5883 | NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0), |
| 5884 | NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0), |
| 5885 | NEONMAP1(vcvtpq_u16_f16, arm_neon_vcvtpu, 0), |
| 5886 | NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0), |
| 5887 | NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0), |
| 5888 | NEONMAP0(vcvtq_f16_s16), |
| 5889 | NEONMAP0(vcvtq_f16_u16), |
| 5890 | NEONMAP0(vcvtq_f32_v), |
| 5891 | NEONMAP1(vcvtq_n_f16_s16, arm_neon_vcvtfxs2fp, 0), |
| 5892 | NEONMAP1(vcvtq_n_f16_u16, arm_neon_vcvtfxu2fp, 0), |
| 5893 | NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 5894 | NEONMAP1(vcvtq_n_s16_f16, arm_neon_vcvtfp2fxs, 0), |
| 5895 | NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
| 5896 | NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
| 5897 | NEONMAP1(vcvtq_n_u16_f16, arm_neon_vcvtfp2fxu, 0), |
| 5898 | NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
| 5899 | NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
| 5900 | NEONMAP0(vcvtq_s16_f16), |
| 5901 | NEONMAP0(vcvtq_s32_v), |
| 5902 | NEONMAP0(vcvtq_s64_v), |
| 5903 | NEONMAP0(vcvtq_u16_f16), |
| 5904 | NEONMAP0(vcvtq_u32_v), |
| 5905 | NEONMAP0(vcvtq_u64_v), |
| 5906 | NEONMAP1(vdot_s32, arm_neon_sdot, 0), |
| 5907 | NEONMAP1(vdot_u32, arm_neon_udot, 0), |
| 5908 | NEONMAP1(vdotq_s32, arm_neon_sdot, 0), |
| 5909 | NEONMAP1(vdotq_u32, arm_neon_udot, 0), |
| 5910 | NEONMAP0(vext_v), |
| 5911 | NEONMAP0(vextq_v), |
| 5912 | NEONMAP0(vfma_v), |
| 5913 | NEONMAP0(vfmaq_v), |
| 5914 | NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), |