| File: | src/gnu/usr.bin/clang/libclangCodeGen/../../../llvm/clang/lib/CodeGen/CGBuiltin.cpp |
| Warning: | line 15278, column 24 Called C++ object pointer is null |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This contains code to emit Builtin calls as LLVM code. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #include "CGCUDARuntime.h" |
| 14 | #include "CGCXXABI.h" |
| 15 | #include "CGObjCRuntime.h" |
| 16 | #include "CGOpenCLRuntime.h" |
| 17 | #include "CGRecordLayout.h" |
| 18 | #include "CodeGenFunction.h" |
| 19 | #include "CodeGenModule.h" |
| 20 | #include "ConstantEmitter.h" |
| 21 | #include "PatternInit.h" |
| 22 | #include "TargetInfo.h" |
| 23 | #include "clang/AST/ASTContext.h" |
| 24 | #include "clang/AST/Attr.h" |
| 25 | #include "clang/AST/Decl.h" |
| 26 | #include "clang/AST/OSLog.h" |
| 27 | #include "clang/Basic/TargetBuiltins.h" |
| 28 | #include "clang/Basic/TargetInfo.h" |
| 29 | #include "clang/CodeGen/CGFunctionInfo.h" |
| 30 | #include "llvm/ADT/APFloat.h" |
| 31 | #include "llvm/ADT/APInt.h" |
| 32 | #include "llvm/ADT/SmallPtrSet.h" |
| 33 | #include "llvm/ADT/StringExtras.h" |
| 34 | #include "llvm/Analysis/ValueTracking.h" |
| 35 | #include "llvm/IR/DataLayout.h" |
| 36 | #include "llvm/IR/InlineAsm.h" |
| 37 | #include "llvm/IR/Intrinsics.h" |
| 38 | #include "llvm/IR/IntrinsicsAArch64.h" |
| 39 | #include "llvm/IR/IntrinsicsAMDGPU.h" |
| 40 | #include "llvm/IR/IntrinsicsARM.h" |
| 41 | #include "llvm/IR/IntrinsicsBPF.h" |
| 42 | #include "llvm/IR/IntrinsicsHexagon.h" |
| 43 | #include "llvm/IR/IntrinsicsNVPTX.h" |
| 44 | #include "llvm/IR/IntrinsicsPowerPC.h" |
| 45 | #include "llvm/IR/IntrinsicsR600.h" |
| 46 | #include "llvm/IR/IntrinsicsRISCV.h" |
| 47 | #include "llvm/IR/IntrinsicsS390.h" |
| 48 | #include "llvm/IR/IntrinsicsWebAssembly.h" |
| 49 | #include "llvm/IR/IntrinsicsX86.h" |
| 50 | #include "llvm/IR/MDBuilder.h" |
| 51 | #include "llvm/IR/MatrixBuilder.h" |
| 52 | #include "llvm/Support/ConvertUTF.h" |
| 53 | #include "llvm/Support/ScopedPrinter.h" |
| 54 | #include "llvm/Support/X86TargetParser.h" |
| 55 | #include <sstream> |
| 56 | |
| 57 | using namespace clang; |
| 58 | using namespace CodeGen; |
| 59 | using namespace llvm; |
| 60 | |
| 61 | static |
| 62 | int64_t clamp(int64_t Value, int64_t Low, int64_t High) { |
| 63 | return std::min(High, std::max(Low, Value)); |
| 64 | } |
| 65 | |
| 66 | static void initializeAlloca(CodeGenFunction &CGF, AllocaInst *AI, Value *Size, |
| 67 | Align AlignmentInBytes) { |
| 68 | ConstantInt *Byte; |
| 69 | switch (CGF.getLangOpts().getTrivialAutoVarInit()) { |
| 70 | case LangOptions::TrivialAutoVarInitKind::Uninitialized: |
| 71 | // Nothing to initialize. |
| 72 | return; |
| 73 | case LangOptions::TrivialAutoVarInitKind::Zero: |
| 74 | Byte = CGF.Builder.getInt8(0x00); |
| 75 | break; |
| 76 | case LangOptions::TrivialAutoVarInitKind::Pattern: { |
| 77 | llvm::Type *Int8 = llvm::IntegerType::getInt8Ty(CGF.CGM.getLLVMContext()); |
| 78 | Byte = llvm::dyn_cast<llvm::ConstantInt>( |
| 79 | initializationPatternFor(CGF.CGM, Int8)); |
| 80 | break; |
| 81 | } |
| 82 | } |
| 83 | if (CGF.CGM.stopAutoInit()) |
| 84 | return; |
| 85 | auto *I = CGF.Builder.CreateMemSet(AI, Byte, Size, AlignmentInBytes); |
| 86 | I->addAnnotationMetadata("auto-init"); |
| 87 | } |
| 88 | |
| 89 | /// getBuiltinLibFunction - Given a builtin id for a function like |
| 90 | /// "__builtin_fabsf", return a Function* for "fabsf". |
| 91 | llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD, |
| 92 | unsigned BuiltinID) { |
| 93 | assert(Context.BuiltinInfo.isLibFunction(BuiltinID))((void)0); |
| 94 | |
| 95 | // Get the name, skip over the __builtin_ prefix (if necessary). |
| 96 | StringRef Name; |
| 97 | GlobalDecl D(FD); |
| 98 | |
| 99 | // If the builtin has been declared explicitly with an assembler label, |
| 100 | // use the mangled name. This differs from the plain label on platforms |
| 101 | // that prefix labels. |
| 102 | if (FD->hasAttr<AsmLabelAttr>()) |
| 103 | Name = getMangledName(D); |
| 104 | else |
| 105 | Name = Context.BuiltinInfo.getName(BuiltinID) + 10; |
| 106 | |
| 107 | llvm::FunctionType *Ty = |
| 108 | cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType())); |
| 109 | |
| 110 | return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false); |
| 111 | } |
| 112 | |
| 113 | /// Emit the conversions required to turn the given value into an |
| 114 | /// integer of the given size. |
| 115 | static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V, |
| 116 | QualType T, llvm::IntegerType *IntType) { |
| 117 | V = CGF.EmitToMemory(V, T); |
| 118 | |
| 119 | if (V->getType()->isPointerTy()) |
| 120 | return CGF.Builder.CreatePtrToInt(V, IntType); |
| 121 | |
| 122 | assert(V->getType() == IntType)((void)0); |
| 123 | return V; |
| 124 | } |
| 125 | |
| 126 | static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V, |
| 127 | QualType T, llvm::Type *ResultType) { |
| 128 | V = CGF.EmitFromMemory(V, T); |
| 129 | |
| 130 | if (ResultType->isPointerTy()) |
| 131 | return CGF.Builder.CreateIntToPtr(V, ResultType); |
| 132 | |
| 133 | assert(V->getType() == ResultType)((void)0); |
| 134 | return V; |
| 135 | } |
| 136 | |
| 137 | /// Utility to insert an atomic instruction based on Intrinsic::ID |
| 138 | /// and the expression node. |
| 139 | static Value *MakeBinaryAtomicValue( |
| 140 | CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E, |
| 141 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 142 | QualType T = E->getType(); |
| 143 | assert(E->getArg(0)->getType()->isPointerType())((void)0); |
| 144 | assert(CGF.getContext().hasSameUnqualifiedType(T,((void)0) |
| 145 | E->getArg(0)->getType()->getPointeeType()))((void)0); |
| 146 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))((void)0); |
| 147 | |
| 148 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 149 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 150 | |
| 151 | llvm::IntegerType *IntType = |
| 152 | llvm::IntegerType::get(CGF.getLLVMContext(), |
| 153 | CGF.getContext().getTypeSize(T)); |
| 154 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 155 | |
| 156 | llvm::Value *Args[2]; |
| 157 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 158 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 159 | llvm::Type *ValueType = Args[1]->getType(); |
| 160 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 161 | |
| 162 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
| 163 | Kind, Args[0], Args[1], Ordering); |
| 164 | return EmitFromInt(CGF, Result, T, ValueType); |
| 165 | } |
| 166 | |
| 167 | static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) { |
| 168 | Value *Val = CGF.EmitScalarExpr(E->getArg(0)); |
| 169 | Value *Address = CGF.EmitScalarExpr(E->getArg(1)); |
| 170 | |
| 171 | // Convert the type of the pointer to a pointer to the stored type. |
| 172 | Val = CGF.EmitToMemory(Val, E->getArg(0)->getType()); |
| 173 | Value *BC = CGF.Builder.CreateBitCast( |
| 174 | Address, llvm::PointerType::getUnqual(Val->getType()), "cast"); |
| 175 | LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType()); |
| 176 | LV.setNontemporal(true); |
| 177 | CGF.EmitStoreOfScalar(Val, LV, false); |
| 178 | return nullptr; |
| 179 | } |
| 180 | |
| 181 | static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) { |
| 182 | Value *Address = CGF.EmitScalarExpr(E->getArg(0)); |
| 183 | |
| 184 | LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType()); |
| 185 | LV.setNontemporal(true); |
| 186 | return CGF.EmitLoadOfScalar(LV, E->getExprLoc()); |
| 187 | } |
| 188 | |
| 189 | static RValue EmitBinaryAtomic(CodeGenFunction &CGF, |
| 190 | llvm::AtomicRMWInst::BinOp Kind, |
| 191 | const CallExpr *E) { |
| 192 | return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E)); |
| 193 | } |
| 194 | |
| 195 | /// Utility to insert an atomic instruction based Intrinsic::ID and |
| 196 | /// the expression node, where the return value is the result of the |
| 197 | /// operation. |
| 198 | static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF, |
| 199 | llvm::AtomicRMWInst::BinOp Kind, |
| 200 | const CallExpr *E, |
| 201 | Instruction::BinaryOps Op, |
| 202 | bool Invert = false) { |
| 203 | QualType T = E->getType(); |
| 204 | assert(E->getArg(0)->getType()->isPointerType())((void)0); |
| 205 | assert(CGF.getContext().hasSameUnqualifiedType(T,((void)0) |
| 206 | E->getArg(0)->getType()->getPointeeType()))((void)0); |
| 207 | assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()))((void)0); |
| 208 | |
| 209 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 210 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 211 | |
| 212 | llvm::IntegerType *IntType = |
| 213 | llvm::IntegerType::get(CGF.getLLVMContext(), |
| 214 | CGF.getContext().getTypeSize(T)); |
| 215 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 216 | |
| 217 | llvm::Value *Args[2]; |
| 218 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 219 | llvm::Type *ValueType = Args[1]->getType(); |
| 220 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 221 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 222 | |
| 223 | llvm::Value *Result = CGF.Builder.CreateAtomicRMW( |
| 224 | Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent); |
| 225 | Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]); |
| 226 | if (Invert) |
| 227 | Result = |
| 228 | CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result, |
| 229 | llvm::ConstantInt::getAllOnesValue(IntType)); |
| 230 | Result = EmitFromInt(CGF, Result, T, ValueType); |
| 231 | return RValue::get(Result); |
| 232 | } |
| 233 | |
| 234 | /// Utility to insert an atomic cmpxchg instruction. |
| 235 | /// |
| 236 | /// @param CGF The current codegen function. |
| 237 | /// @param E Builtin call expression to convert to cmpxchg. |
| 238 | /// arg0 - address to operate on |
| 239 | /// arg1 - value to compare with |
| 240 | /// arg2 - new value |
| 241 | /// @param ReturnBool Specifies whether to return success flag of |
| 242 | /// cmpxchg result or the old value. |
| 243 | /// |
| 244 | /// @returns result of cmpxchg, according to ReturnBool |
| 245 | /// |
| 246 | /// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics |
| 247 | /// invoke the function EmitAtomicCmpXchgForMSIntrin. |
| 248 | static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E, |
| 249 | bool ReturnBool) { |
| 250 | QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType(); |
| 251 | llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0)); |
| 252 | unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace(); |
| 253 | |
| 254 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
| 255 | CGF.getLLVMContext(), CGF.getContext().getTypeSize(T)); |
| 256 | llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace); |
| 257 | |
| 258 | Value *Args[3]; |
| 259 | Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType); |
| 260 | Args[1] = CGF.EmitScalarExpr(E->getArg(1)); |
| 261 | llvm::Type *ValueType = Args[1]->getType(); |
| 262 | Args[1] = EmitToInt(CGF, Args[1], T, IntType); |
| 263 | Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType); |
| 264 | |
| 265 | Value *Pair = CGF.Builder.CreateAtomicCmpXchg( |
| 266 | Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent, |
| 267 | llvm::AtomicOrdering::SequentiallyConsistent); |
| 268 | if (ReturnBool) |
| 269 | // Extract boolean success flag and zext it to int. |
| 270 | return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1), |
| 271 | CGF.ConvertType(E->getType())); |
| 272 | else |
| 273 | // Extract old value and emit it using the same type as compare value. |
| 274 | return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T, |
| 275 | ValueType); |
| 276 | } |
| 277 | |
| 278 | /// This function should be invoked to emit atomic cmpxchg for Microsoft's |
| 279 | /// _InterlockedCompareExchange* intrinsics which have the following signature: |
| 280 | /// T _InterlockedCompareExchange(T volatile *Destination, |
| 281 | /// T Exchange, |
| 282 | /// T Comparand); |
| 283 | /// |
| 284 | /// Whereas the llvm 'cmpxchg' instruction has the following syntax: |
| 285 | /// cmpxchg *Destination, Comparand, Exchange. |
| 286 | /// So we need to swap Comparand and Exchange when invoking |
| 287 | /// CreateAtomicCmpXchg. That is the reason we could not use the above utility |
| 288 | /// function MakeAtomicCmpXchgValue since it expects the arguments to be |
| 289 | /// already swapped. |
| 290 | |
| 291 | static |
| 292 | Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E, |
| 293 | AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) { |
| 294 | assert(E->getArg(0)->getType()->isPointerType())((void)0); |
| 295 | assert(CGF.getContext().hasSameUnqualifiedType(((void)0) |
| 296 | E->getType(), E->getArg(0)->getType()->getPointeeType()))((void)0); |
| 297 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),((void)0) |
| 298 | E->getArg(1)->getType()))((void)0); |
| 299 | assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),((void)0) |
| 300 | E->getArg(2)->getType()))((void)0); |
| 301 | |
| 302 | auto *Destination = CGF.EmitScalarExpr(E->getArg(0)); |
| 303 | auto *Comparand = CGF.EmitScalarExpr(E->getArg(2)); |
| 304 | auto *Exchange = CGF.EmitScalarExpr(E->getArg(1)); |
| 305 | |
| 306 | // For Release ordering, the failure ordering should be Monotonic. |
| 307 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ? |
| 308 | AtomicOrdering::Monotonic : |
| 309 | SuccessOrdering; |
| 310 | |
| 311 | // The atomic instruction is marked volatile for consistency with MSVC. This |
| 312 | // blocks the few atomics optimizations that LLVM has. If we want to optimize |
| 313 | // _Interlocked* operations in the future, we will have to remove the volatile |
| 314 | // marker. |
| 315 | auto *Result = CGF.Builder.CreateAtomicCmpXchg( |
| 316 | Destination, Comparand, Exchange, |
| 317 | SuccessOrdering, FailureOrdering); |
| 318 | Result->setVolatile(true); |
| 319 | return CGF.Builder.CreateExtractValue(Result, 0); |
| 320 | } |
| 321 | |
| 322 | // 64-bit Microsoft platforms support 128 bit cmpxchg operations. They are |
| 323 | // prototyped like this: |
| 324 | // |
| 325 | // unsigned char _InterlockedCompareExchange128...( |
| 326 | // __int64 volatile * _Destination, |
| 327 | // __int64 _ExchangeHigh, |
| 328 | // __int64 _ExchangeLow, |
| 329 | // __int64 * _ComparandResult); |
| 330 | static Value *EmitAtomicCmpXchg128ForMSIntrin(CodeGenFunction &CGF, |
| 331 | const CallExpr *E, |
| 332 | AtomicOrdering SuccessOrdering) { |
| 333 | assert(E->getNumArgs() == 4)((void)0); |
| 334 | llvm::Value *Destination = CGF.EmitScalarExpr(E->getArg(0)); |
| 335 | llvm::Value *ExchangeHigh = CGF.EmitScalarExpr(E->getArg(1)); |
| 336 | llvm::Value *ExchangeLow = CGF.EmitScalarExpr(E->getArg(2)); |
| 337 | llvm::Value *ComparandPtr = CGF.EmitScalarExpr(E->getArg(3)); |
| 338 | |
| 339 | assert(Destination->getType()->isPointerTy())((void)0); |
| 340 | assert(!ExchangeHigh->getType()->isPointerTy())((void)0); |
| 341 | assert(!ExchangeLow->getType()->isPointerTy())((void)0); |
| 342 | assert(ComparandPtr->getType()->isPointerTy())((void)0); |
| 343 | |
| 344 | // For Release ordering, the failure ordering should be Monotonic. |
| 345 | auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release |
| 346 | ? AtomicOrdering::Monotonic |
| 347 | : SuccessOrdering; |
| 348 | |
| 349 | // Convert to i128 pointers and values. |
| 350 | llvm::Type *Int128Ty = llvm::IntegerType::get(CGF.getLLVMContext(), 128); |
| 351 | llvm::Type *Int128PtrTy = Int128Ty->getPointerTo(); |
| 352 | Destination = CGF.Builder.CreateBitCast(Destination, Int128PtrTy); |
| 353 | Address ComparandResult(CGF.Builder.CreateBitCast(ComparandPtr, Int128PtrTy), |
| 354 | CGF.getContext().toCharUnitsFromBits(128)); |
| 355 | |
| 356 | // (((i128)hi) << 64) | ((i128)lo) |
| 357 | ExchangeHigh = CGF.Builder.CreateZExt(ExchangeHigh, Int128Ty); |
| 358 | ExchangeLow = CGF.Builder.CreateZExt(ExchangeLow, Int128Ty); |
| 359 | ExchangeHigh = |
| 360 | CGF.Builder.CreateShl(ExchangeHigh, llvm::ConstantInt::get(Int128Ty, 64)); |
| 361 | llvm::Value *Exchange = CGF.Builder.CreateOr(ExchangeHigh, ExchangeLow); |
| 362 | |
| 363 | // Load the comparand for the instruction. |
| 364 | llvm::Value *Comparand = CGF.Builder.CreateLoad(ComparandResult); |
| 365 | |
| 366 | auto *CXI = CGF.Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
| 367 | SuccessOrdering, FailureOrdering); |
| 368 | |
| 369 | // The atomic instruction is marked volatile for consistency with MSVC. This |
| 370 | // blocks the few atomics optimizations that LLVM has. If we want to optimize |
| 371 | // _Interlocked* operations in the future, we will have to remove the volatile |
| 372 | // marker. |
| 373 | CXI->setVolatile(true); |
| 374 | |
| 375 | // Store the result as an outparameter. |
| 376 | CGF.Builder.CreateStore(CGF.Builder.CreateExtractValue(CXI, 0), |
| 377 | ComparandResult); |
| 378 | |
| 379 | // Get the success boolean and zero extend it to i8. |
| 380 | Value *Success = CGF.Builder.CreateExtractValue(CXI, 1); |
| 381 | return CGF.Builder.CreateZExt(Success, CGF.Int8Ty); |
| 382 | } |
| 383 | |
| 384 | static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E, |
| 385 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 386 | assert(E->getArg(0)->getType()->isPointerType())((void)0); |
| 387 | |
| 388 | auto *IntTy = CGF.ConvertType(E->getType()); |
| 389 | auto *Result = CGF.Builder.CreateAtomicRMW( |
| 390 | AtomicRMWInst::Add, |
| 391 | CGF.EmitScalarExpr(E->getArg(0)), |
| 392 | ConstantInt::get(IntTy, 1), |
| 393 | Ordering); |
| 394 | return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1)); |
| 395 | } |
| 396 | |
| 397 | static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E, |
| 398 | AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) { |
| 399 | assert(E->getArg(0)->getType()->isPointerType())((void)0); |
| 400 | |
| 401 | auto *IntTy = CGF.ConvertType(E->getType()); |
| 402 | auto *Result = CGF.Builder.CreateAtomicRMW( |
| 403 | AtomicRMWInst::Sub, |
| 404 | CGF.EmitScalarExpr(E->getArg(0)), |
| 405 | ConstantInt::get(IntTy, 1), |
| 406 | Ordering); |
| 407 | return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1)); |
| 408 | } |
| 409 | |
| 410 | // Build a plain volatile load. |
| 411 | static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) { |
| 412 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
| 413 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 414 | CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy); |
| 415 | llvm::Type *ITy = |
| 416 | llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8); |
| 417 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 418 | llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(ITy, Ptr, LoadSize); |
| 419 | Load->setVolatile(true); |
| 420 | return Load; |
| 421 | } |
| 422 | |
| 423 | // Build a plain volatile store. |
| 424 | static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) { |
| 425 | Value *Ptr = CGF.EmitScalarExpr(E->getArg(0)); |
| 426 | Value *Value = CGF.EmitScalarExpr(E->getArg(1)); |
| 427 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 428 | CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy); |
| 429 | llvm::Type *ITy = |
| 430 | llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8); |
| 431 | Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 432 | llvm::StoreInst *Store = |
| 433 | CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize); |
| 434 | Store->setVolatile(true); |
| 435 | return Store; |
| 436 | } |
| 437 | |
| 438 | // Emit a simple mangled intrinsic that has 1 argument and a return type |
| 439 | // matching the argument type. Depending on mode, this may be a constrained |
| 440 | // floating-point intrinsic. |
| 441 | static Value *emitUnaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 442 | const CallExpr *E, unsigned IntrinsicID, |
| 443 | unsigned ConstrainedIntrinsicID) { |
| 444 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 445 | |
| 446 | if (CGF.Builder.getIsFPConstrained()) { |
| 447 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 448 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
| 449 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0 }); |
| 450 | } else { |
| 451 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 452 | return CGF.Builder.CreateCall(F, Src0); |
| 453 | } |
| 454 | } |
| 455 | |
| 456 | // Emit an intrinsic that has 2 operands of the same type as its result. |
| 457 | // Depending on mode, this may be a constrained floating-point intrinsic. |
| 458 | static Value *emitBinaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 459 | const CallExpr *E, unsigned IntrinsicID, |
| 460 | unsigned ConstrainedIntrinsicID) { |
| 461 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 462 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 463 | |
| 464 | if (CGF.Builder.getIsFPConstrained()) { |
| 465 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 466 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
| 467 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1 }); |
| 468 | } else { |
| 469 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 470 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); |
| 471 | } |
| 472 | } |
| 473 | |
| 474 | // Emit an intrinsic that has 3 operands of the same type as its result. |
| 475 | // Depending on mode, this may be a constrained floating-point intrinsic. |
| 476 | static Value *emitTernaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 477 | const CallExpr *E, unsigned IntrinsicID, |
| 478 | unsigned ConstrainedIntrinsicID) { |
| 479 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 480 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 481 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); |
| 482 | |
| 483 | if (CGF.Builder.getIsFPConstrained()) { |
| 484 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 485 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); |
| 486 | return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1, Src2 }); |
| 487 | } else { |
| 488 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 489 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); |
| 490 | } |
| 491 | } |
| 492 | |
| 493 | // Emit an intrinsic where all operands are of the same type as the result. |
| 494 | // Depending on mode, this may be a constrained floating-point intrinsic. |
| 495 | static Value *emitCallMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, |
| 496 | unsigned IntrinsicID, |
| 497 | unsigned ConstrainedIntrinsicID, |
| 498 | llvm::Type *Ty, |
| 499 | ArrayRef<Value *> Args) { |
| 500 | Function *F; |
| 501 | if (CGF.Builder.getIsFPConstrained()) |
| 502 | F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Ty); |
| 503 | else |
| 504 | F = CGF.CGM.getIntrinsic(IntrinsicID, Ty); |
| 505 | |
| 506 | if (CGF.Builder.getIsFPConstrained()) |
| 507 | return CGF.Builder.CreateConstrainedFPCall(F, Args); |
| 508 | else |
| 509 | return CGF.Builder.CreateCall(F, Args); |
| 510 | } |
| 511 | |
| 512 | // Emit a simple mangled intrinsic that has 1 argument and a return type |
| 513 | // matching the argument type. |
| 514 | static Value *emitUnaryBuiltin(CodeGenFunction &CGF, |
| 515 | const CallExpr *E, |
| 516 | unsigned IntrinsicID) { |
| 517 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 518 | |
| 519 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 520 | return CGF.Builder.CreateCall(F, Src0); |
| 521 | } |
| 522 | |
| 523 | // Emit an intrinsic that has 2 operands of the same type as its result. |
| 524 | static Value *emitBinaryBuiltin(CodeGenFunction &CGF, |
| 525 | const CallExpr *E, |
| 526 | unsigned IntrinsicID) { |
| 527 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 528 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 529 | |
| 530 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 531 | return CGF.Builder.CreateCall(F, { Src0, Src1 }); |
| 532 | } |
| 533 | |
| 534 | // Emit an intrinsic that has 3 operands of the same type as its result. |
| 535 | static Value *emitTernaryBuiltin(CodeGenFunction &CGF, |
| 536 | const CallExpr *E, |
| 537 | unsigned IntrinsicID) { |
| 538 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 539 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 540 | llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2)); |
| 541 | |
| 542 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 543 | return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 }); |
| 544 | } |
| 545 | |
| 546 | // Emit an intrinsic that has 1 float or double operand, and 1 integer. |
| 547 | static Value *emitFPIntBuiltin(CodeGenFunction &CGF, |
| 548 | const CallExpr *E, |
| 549 | unsigned IntrinsicID) { |
| 550 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 551 | llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); |
| 552 | |
| 553 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); |
| 554 | return CGF.Builder.CreateCall(F, {Src0, Src1}); |
| 555 | } |
| 556 | |
| 557 | // Emit an intrinsic that has overloaded integer result and fp operand. |
| 558 | static Value * |
| 559 | emitMaybeConstrainedFPToIntRoundBuiltin(CodeGenFunction &CGF, const CallExpr *E, |
| 560 | unsigned IntrinsicID, |
| 561 | unsigned ConstrainedIntrinsicID) { |
| 562 | llvm::Type *ResultType = CGF.ConvertType(E->getType()); |
| 563 | llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); |
| 564 | |
| 565 | if (CGF.Builder.getIsFPConstrained()) { |
| 566 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); |
| 567 | Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, |
| 568 | {ResultType, Src0->getType()}); |
| 569 | return CGF.Builder.CreateConstrainedFPCall(F, {Src0}); |
| 570 | } else { |
| 571 | Function *F = |
| 572 | CGF.CGM.getIntrinsic(IntrinsicID, {ResultType, Src0->getType()}); |
| 573 | return CGF.Builder.CreateCall(F, Src0); |
| 574 | } |
| 575 | } |
| 576 | |
| 577 | /// EmitFAbs - Emit a call to @llvm.fabs(). |
| 578 | static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) { |
| 579 | Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType()); |
| 580 | llvm::CallInst *Call = CGF.Builder.CreateCall(F, V); |
| 581 | Call->setDoesNotAccessMemory(); |
| 582 | return Call; |
| 583 | } |
| 584 | |
| 585 | /// Emit the computation of the sign bit for a floating point value. Returns |
| 586 | /// the i1 sign bit value. |
| 587 | static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) { |
| 588 | LLVMContext &C = CGF.CGM.getLLVMContext(); |
| 589 | |
| 590 | llvm::Type *Ty = V->getType(); |
| 591 | int Width = Ty->getPrimitiveSizeInBits(); |
| 592 | llvm::Type *IntTy = llvm::IntegerType::get(C, Width); |
| 593 | V = CGF.Builder.CreateBitCast(V, IntTy); |
| 594 | if (Ty->isPPC_FP128Ty()) { |
| 595 | // We want the sign bit of the higher-order double. The bitcast we just |
| 596 | // did works as if the double-double was stored to memory and then |
| 597 | // read as an i128. The "store" will put the higher-order double in the |
| 598 | // lower address in both little- and big-Endian modes, but the "load" |
| 599 | // will treat those bits as a different part of the i128: the low bits in |
| 600 | // little-Endian, the high bits in big-Endian. Therefore, on big-Endian |
| 601 | // we need to shift the high bits down to the low before truncating. |
| 602 | Width >>= 1; |
| 603 | if (CGF.getTarget().isBigEndian()) { |
| 604 | Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width); |
| 605 | V = CGF.Builder.CreateLShr(V, ShiftCst); |
| 606 | } |
| 607 | // We are truncating value in order to extract the higher-order |
| 608 | // double, which we will be using to extract the sign from. |
| 609 | IntTy = llvm::IntegerType::get(C, Width); |
| 610 | V = CGF.Builder.CreateTrunc(V, IntTy); |
| 611 | } |
| 612 | Value *Zero = llvm::Constant::getNullValue(IntTy); |
| 613 | return CGF.Builder.CreateICmpSLT(V, Zero); |
| 614 | } |
| 615 | |
| 616 | static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD, |
| 617 | const CallExpr *E, llvm::Constant *calleeValue) { |
| 618 | CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD)); |
| 619 | return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot()); |
| 620 | } |
| 621 | |
| 622 | /// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.* |
| 623 | /// depending on IntrinsicID. |
| 624 | /// |
| 625 | /// \arg CGF The current codegen function. |
| 626 | /// \arg IntrinsicID The ID for the Intrinsic we wish to generate. |
| 627 | /// \arg X The first argument to the llvm.*.with.overflow.*. |
| 628 | /// \arg Y The second argument to the llvm.*.with.overflow.*. |
| 629 | /// \arg Carry The carry returned by the llvm.*.with.overflow.*. |
| 630 | /// \returns The result (i.e. sum/product) returned by the intrinsic. |
| 631 | static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF, |
| 632 | const llvm::Intrinsic::ID IntrinsicID, |
| 633 | llvm::Value *X, llvm::Value *Y, |
| 634 | llvm::Value *&Carry) { |
| 635 | // Make sure we have integers of the same width. |
| 636 | assert(X->getType() == Y->getType() &&((void)0) |
| 637 | "Arguments must be the same type. (Did you forget to make sure both "((void)0) |
| 638 | "arguments have the same integer width?)")((void)0); |
| 639 | |
| 640 | Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType()); |
| 641 | llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y}); |
| 642 | Carry = CGF.Builder.CreateExtractValue(Tmp, 1); |
| 643 | return CGF.Builder.CreateExtractValue(Tmp, 0); |
| 644 | } |
| 645 | |
| 646 | static Value *emitRangedBuiltin(CodeGenFunction &CGF, |
| 647 | unsigned IntrinsicID, |
| 648 | int low, int high) { |
| 649 | llvm::MDBuilder MDHelper(CGF.getLLVMContext()); |
| 650 | llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high)); |
| 651 | Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {}); |
| 652 | llvm::Instruction *Call = CGF.Builder.CreateCall(F); |
| 653 | Call->setMetadata(llvm::LLVMContext::MD_range, RNode); |
| 654 | return Call; |
| 655 | } |
| 656 | |
| 657 | namespace { |
| 658 | struct WidthAndSignedness { |
| 659 | unsigned Width; |
| 660 | bool Signed; |
| 661 | }; |
| 662 | } |
| 663 | |
| 664 | static WidthAndSignedness |
| 665 | getIntegerWidthAndSignedness(const clang::ASTContext &context, |
| 666 | const clang::QualType Type) { |
| 667 | assert(Type->isIntegerType() && "Given type is not an integer.")((void)0); |
| 668 | unsigned Width = Type->isBooleanType() ? 1 |
| 669 | : Type->isExtIntType() ? context.getIntWidth(Type) |
| 670 | : context.getTypeInfo(Type).Width; |
| 671 | bool Signed = Type->isSignedIntegerType(); |
| 672 | return {Width, Signed}; |
| 673 | } |
| 674 | |
| 675 | // Given one or more integer types, this function produces an integer type that |
| 676 | // encompasses them: any value in one of the given types could be expressed in |
| 677 | // the encompassing type. |
| 678 | static struct WidthAndSignedness |
| 679 | EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) { |
| 680 | assert(Types.size() > 0 && "Empty list of types.")((void)0); |
| 681 | |
| 682 | // If any of the given types is signed, we must return a signed type. |
| 683 | bool Signed = false; |
| 684 | for (const auto &Type : Types) { |
| 685 | Signed |= Type.Signed; |
| 686 | } |
| 687 | |
| 688 | // The encompassing type must have a width greater than or equal to the width |
| 689 | // of the specified types. Additionally, if the encompassing type is signed, |
| 690 | // its width must be strictly greater than the width of any unsigned types |
| 691 | // given. |
| 692 | unsigned Width = 0; |
| 693 | for (const auto &Type : Types) { |
| 694 | unsigned MinWidth = Type.Width + (Signed && !Type.Signed); |
| 695 | if (Width < MinWidth) { |
| 696 | Width = MinWidth; |
| 697 | } |
| 698 | } |
| 699 | |
| 700 | return {Width, Signed}; |
| 701 | } |
| 702 | |
| 703 | Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) { |
| 704 | llvm::Type *DestType = Int8PtrTy; |
| 705 | if (ArgValue->getType() != DestType) |
| 706 | ArgValue = |
| 707 | Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data()); |
| 708 | |
| 709 | Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend; |
| 710 | return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue); |
| 711 | } |
| 712 | |
| 713 | /// Checks if using the result of __builtin_object_size(p, @p From) in place of |
| 714 | /// __builtin_object_size(p, @p To) is correct |
| 715 | static bool areBOSTypesCompatible(int From, int To) { |
| 716 | // Note: Our __builtin_object_size implementation currently treats Type=0 and |
| 717 | // Type=2 identically. Encoding this implementation detail here may make |
| 718 | // improving __builtin_object_size difficult in the future, so it's omitted. |
| 719 | return From == To || (From == 0 && To == 1) || (From == 3 && To == 2); |
| 720 | } |
| 721 | |
| 722 | static llvm::Value * |
| 723 | getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) { |
| 724 | return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true); |
| 725 | } |
| 726 | |
| 727 | llvm::Value * |
| 728 | CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type, |
| 729 | llvm::IntegerType *ResType, |
| 730 | llvm::Value *EmittedE, |
| 731 | bool IsDynamic) { |
| 732 | uint64_t ObjectSize; |
| 733 | if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type)) |
| 734 | return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic); |
| 735 | return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true); |
| 736 | } |
| 737 | |
| 738 | /// Returns a Value corresponding to the size of the given expression. |
| 739 | /// This Value may be either of the following: |
| 740 | /// - A llvm::Argument (if E is a param with the pass_object_size attribute on |
| 741 | /// it) |
| 742 | /// - A call to the @llvm.objectsize intrinsic |
| 743 | /// |
| 744 | /// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null |
| 745 | /// and we wouldn't otherwise try to reference a pass_object_size parameter, |
| 746 | /// we'll call @llvm.objectsize on EmittedE, rather than emitting E. |
| 747 | llvm::Value * |
| 748 | CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type, |
| 749 | llvm::IntegerType *ResType, |
| 750 | llvm::Value *EmittedE, bool IsDynamic) { |
| 751 | // We need to reference an argument if the pointer is a parameter with the |
| 752 | // pass_object_size attribute. |
| 753 | if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) { |
| 754 | auto *Param = dyn_cast<ParmVarDecl>(D->getDecl()); |
| 755 | auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>(); |
| 756 | if (Param != nullptr && PS != nullptr && |
| 757 | areBOSTypesCompatible(PS->getType(), Type)) { |
| 758 | auto Iter = SizeArguments.find(Param); |
| 759 | assert(Iter != SizeArguments.end())((void)0); |
| 760 | |
| 761 | const ImplicitParamDecl *D = Iter->second; |
| 762 | auto DIter = LocalDeclMap.find(D); |
| 763 | assert(DIter != LocalDeclMap.end())((void)0); |
| 764 | |
| 765 | return EmitLoadOfScalar(DIter->second, /*Volatile=*/false, |
| 766 | getContext().getSizeType(), E->getBeginLoc()); |
| 767 | } |
| 768 | } |
| 769 | |
| 770 | // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't |
| 771 | // evaluate E for side-effects. In either case, we shouldn't lower to |
| 772 | // @llvm.objectsize. |
| 773 | if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext()))) |
| 774 | return getDefaultBuiltinObjectSizeResult(Type, ResType); |
| 775 | |
| 776 | Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E); |
| 777 | assert(Ptr->getType()->isPointerTy() &&((void)0) |
| 778 | "Non-pointer passed to __builtin_object_size?")((void)0); |
| 779 | |
| 780 | Function *F = |
| 781 | CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()}); |
| 782 | |
| 783 | // LLVM only supports 0 and 2, make sure that we pass along that as a boolean. |
| 784 | Value *Min = Builder.getInt1((Type & 2) != 0); |
| 785 | // For GCC compatibility, __builtin_object_size treat NULL as unknown size. |
| 786 | Value *NullIsUnknown = Builder.getTrue(); |
| 787 | Value *Dynamic = Builder.getInt1(IsDynamic); |
| 788 | return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic}); |
| 789 | } |
| 790 | |
| 791 | namespace { |
| 792 | /// A struct to generically describe a bit test intrinsic. |
| 793 | struct BitTest { |
| 794 | enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set }; |
| 795 | enum InterlockingKind : uint8_t { |
| 796 | Unlocked, |
| 797 | Sequential, |
| 798 | Acquire, |
| 799 | Release, |
| 800 | NoFence |
| 801 | }; |
| 802 | |
| 803 | ActionKind Action; |
| 804 | InterlockingKind Interlocking; |
| 805 | bool Is64Bit; |
| 806 | |
| 807 | static BitTest decodeBitTestBuiltin(unsigned BuiltinID); |
| 808 | }; |
| 809 | } // namespace |
| 810 | |
| 811 | BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) { |
| 812 | switch (BuiltinID) { |
| 813 | // Main portable variants. |
| 814 | case Builtin::BI_bittest: |
| 815 | return {TestOnly, Unlocked, false}; |
| 816 | case Builtin::BI_bittestandcomplement: |
| 817 | return {Complement, Unlocked, false}; |
| 818 | case Builtin::BI_bittestandreset: |
| 819 | return {Reset, Unlocked, false}; |
| 820 | case Builtin::BI_bittestandset: |
| 821 | return {Set, Unlocked, false}; |
| 822 | case Builtin::BI_interlockedbittestandreset: |
| 823 | return {Reset, Sequential, false}; |
| 824 | case Builtin::BI_interlockedbittestandset: |
| 825 | return {Set, Sequential, false}; |
| 826 | |
| 827 | // X86-specific 64-bit variants. |
| 828 | case Builtin::BI_bittest64: |
| 829 | return {TestOnly, Unlocked, true}; |
| 830 | case Builtin::BI_bittestandcomplement64: |
| 831 | return {Complement, Unlocked, true}; |
| 832 | case Builtin::BI_bittestandreset64: |
| 833 | return {Reset, Unlocked, true}; |
| 834 | case Builtin::BI_bittestandset64: |
| 835 | return {Set, Unlocked, true}; |
| 836 | case Builtin::BI_interlockedbittestandreset64: |
| 837 | return {Reset, Sequential, true}; |
| 838 | case Builtin::BI_interlockedbittestandset64: |
| 839 | return {Set, Sequential, true}; |
| 840 | |
| 841 | // ARM/AArch64-specific ordering variants. |
| 842 | case Builtin::BI_interlockedbittestandset_acq: |
| 843 | return {Set, Acquire, false}; |
| 844 | case Builtin::BI_interlockedbittestandset_rel: |
| 845 | return {Set, Release, false}; |
| 846 | case Builtin::BI_interlockedbittestandset_nf: |
| 847 | return {Set, NoFence, false}; |
| 848 | case Builtin::BI_interlockedbittestandreset_acq: |
| 849 | return {Reset, Acquire, false}; |
| 850 | case Builtin::BI_interlockedbittestandreset_rel: |
| 851 | return {Reset, Release, false}; |
| 852 | case Builtin::BI_interlockedbittestandreset_nf: |
| 853 | return {Reset, NoFence, false}; |
| 854 | } |
| 855 | llvm_unreachable("expected only bittest intrinsics")__builtin_unreachable(); |
| 856 | } |
| 857 | |
| 858 | static char bitActionToX86BTCode(BitTest::ActionKind A) { |
| 859 | switch (A) { |
| 860 | case BitTest::TestOnly: return '\0'; |
| 861 | case BitTest::Complement: return 'c'; |
| 862 | case BitTest::Reset: return 'r'; |
| 863 | case BitTest::Set: return 's'; |
| 864 | } |
| 865 | llvm_unreachable("invalid action")__builtin_unreachable(); |
| 866 | } |
| 867 | |
| 868 | static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF, |
| 869 | BitTest BT, |
| 870 | const CallExpr *E, Value *BitBase, |
| 871 | Value *BitPos) { |
| 872 | char Action = bitActionToX86BTCode(BT.Action); |
| 873 | char SizeSuffix = BT.Is64Bit ? 'q' : 'l'; |
| 874 | |
| 875 | // Build the assembly. |
| 876 | SmallString<64> Asm; |
| 877 | raw_svector_ostream AsmOS(Asm); |
| 878 | if (BT.Interlocking != BitTest::Unlocked) |
| 879 | AsmOS << "lock "; |
| 880 | AsmOS << "bt"; |
| 881 | if (Action) |
| 882 | AsmOS << Action; |
| 883 | AsmOS << SizeSuffix << " $2, ($1)"; |
| 884 | |
| 885 | // Build the constraints. FIXME: We should support immediates when possible. |
| 886 | std::string Constraints = "={@ccc},r,r,~{cc},~{memory}"; |
| 887 | std::string MachineClobbers = CGF.getTarget().getClobbers(); |
| 888 | if (!MachineClobbers.empty()) { |
| 889 | Constraints += ','; |
| 890 | Constraints += MachineClobbers; |
| 891 | } |
| 892 | llvm::IntegerType *IntType = llvm::IntegerType::get( |
| 893 | CGF.getLLVMContext(), |
| 894 | CGF.getContext().getTypeSize(E->getArg(1)->getType())); |
| 895 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
| 896 | llvm::FunctionType *FTy = |
| 897 | llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false); |
| 898 | |
| 899 | llvm::InlineAsm *IA = |
| 900 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
| 901 | return CGF.Builder.CreateCall(IA, {BitBase, BitPos}); |
| 902 | } |
| 903 | |
| 904 | static llvm::AtomicOrdering |
| 905 | getBitTestAtomicOrdering(BitTest::InterlockingKind I) { |
| 906 | switch (I) { |
| 907 | case BitTest::Unlocked: return llvm::AtomicOrdering::NotAtomic; |
| 908 | case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent; |
| 909 | case BitTest::Acquire: return llvm::AtomicOrdering::Acquire; |
| 910 | case BitTest::Release: return llvm::AtomicOrdering::Release; |
| 911 | case BitTest::NoFence: return llvm::AtomicOrdering::Monotonic; |
| 912 | } |
| 913 | llvm_unreachable("invalid interlocking")__builtin_unreachable(); |
| 914 | } |
| 915 | |
| 916 | /// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of |
| 917 | /// bits and a bit position and read and optionally modify the bit at that |
| 918 | /// position. The position index can be arbitrarily large, i.e. it can be larger |
| 919 | /// than 31 or 63, so we need an indexed load in the general case. |
| 920 | static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF, |
| 921 | unsigned BuiltinID, |
| 922 | const CallExpr *E) { |
| 923 | Value *BitBase = CGF.EmitScalarExpr(E->getArg(0)); |
| 924 | Value *BitPos = CGF.EmitScalarExpr(E->getArg(1)); |
| 925 | |
| 926 | BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID); |
| 927 | |
| 928 | // X86 has special BT, BTC, BTR, and BTS instructions that handle the array |
| 929 | // indexing operation internally. Use them if possible. |
| 930 | if (CGF.getTarget().getTriple().isX86()) |
| 931 | return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos); |
| 932 | |
| 933 | // Otherwise, use generic code to load one byte and test the bit. Use all but |
| 934 | // the bottom three bits as the array index, and the bottom three bits to form |
| 935 | // a mask. |
| 936 | // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0; |
| 937 | Value *ByteIndex = CGF.Builder.CreateAShr( |
| 938 | BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx"); |
| 939 | Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy); |
| 940 | Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8, |
| 941 | ByteIndex, "bittest.byteaddr"), |
| 942 | CharUnits::One()); |
| 943 | Value *PosLow = |
| 944 | CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty), |
| 945 | llvm::ConstantInt::get(CGF.Int8Ty, 0x7)); |
| 946 | |
| 947 | // The updating instructions will need a mask. |
| 948 | Value *Mask = nullptr; |
| 949 | if (BT.Action != BitTest::TestOnly) { |
| 950 | Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow, |
| 951 | "bittest.mask"); |
| 952 | } |
| 953 | |
| 954 | // Check the action and ordering of the interlocked intrinsics. |
| 955 | llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking); |
| 956 | |
| 957 | Value *OldByte = nullptr; |
| 958 | if (Ordering != llvm::AtomicOrdering::NotAtomic) { |
| 959 | // Emit a combined atomicrmw load/store operation for the interlocked |
| 960 | // intrinsics. |
| 961 | llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or; |
| 962 | if (BT.Action == BitTest::Reset) { |
| 963 | Mask = CGF.Builder.CreateNot(Mask); |
| 964 | RMWOp = llvm::AtomicRMWInst::And; |
| 965 | } |
| 966 | OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask, |
| 967 | Ordering); |
| 968 | } else { |
| 969 | // Emit a plain load for the non-interlocked intrinsics. |
| 970 | OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte"); |
| 971 | Value *NewByte = nullptr; |
| 972 | switch (BT.Action) { |
| 973 | case BitTest::TestOnly: |
| 974 | // Don't store anything. |
| 975 | break; |
| 976 | case BitTest::Complement: |
| 977 | NewByte = CGF.Builder.CreateXor(OldByte, Mask); |
| 978 | break; |
| 979 | case BitTest::Reset: |
| 980 | NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask)); |
| 981 | break; |
| 982 | case BitTest::Set: |
| 983 | NewByte = CGF.Builder.CreateOr(OldByte, Mask); |
| 984 | break; |
| 985 | } |
| 986 | if (NewByte) |
| 987 | CGF.Builder.CreateStore(NewByte, ByteAddr); |
| 988 | } |
| 989 | |
| 990 | // However we loaded the old byte, either by plain load or atomicrmw, shift |
| 991 | // the bit into the low position and mask it to 0 or 1. |
| 992 | Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr"); |
| 993 | return CGF.Builder.CreateAnd( |
| 994 | ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res"); |
| 995 | } |
| 996 | |
| 997 | static llvm::Value *emitPPCLoadReserveIntrinsic(CodeGenFunction &CGF, |
| 998 | unsigned BuiltinID, |
| 999 | const CallExpr *E) { |
| 1000 | Value *Addr = CGF.EmitScalarExpr(E->getArg(0)); |
| 1001 | |
| 1002 | SmallString<64> Asm; |
| 1003 | raw_svector_ostream AsmOS(Asm); |
| 1004 | llvm::IntegerType *RetType = CGF.Int32Ty; |
| 1005 | |
| 1006 | switch (BuiltinID) { |
| 1007 | case clang::PPC::BI__builtin_ppc_ldarx: |
| 1008 | AsmOS << "ldarx "; |
| 1009 | RetType = CGF.Int64Ty; |
| 1010 | break; |
| 1011 | case clang::PPC::BI__builtin_ppc_lwarx: |
| 1012 | AsmOS << "lwarx "; |
| 1013 | RetType = CGF.Int32Ty; |
| 1014 | break; |
| 1015 | case clang::PPC::BI__builtin_ppc_lharx: |
| 1016 | AsmOS << "lharx "; |
| 1017 | RetType = CGF.Int16Ty; |
| 1018 | break; |
| 1019 | case clang::PPC::BI__builtin_ppc_lbarx: |
| 1020 | AsmOS << "lbarx "; |
| 1021 | RetType = CGF.Int8Ty; |
| 1022 | break; |
| 1023 | default: |
| 1024 | llvm_unreachable("Expected only PowerPC load reserve intrinsics")__builtin_unreachable(); |
| 1025 | } |
| 1026 | |
| 1027 | AsmOS << "$0, ${1:y}"; |
| 1028 | |
| 1029 | std::string Constraints = "=r,*Z,~{memory}"; |
| 1030 | std::string MachineClobbers = CGF.getTarget().getClobbers(); |
| 1031 | if (!MachineClobbers.empty()) { |
| 1032 | Constraints += ','; |
| 1033 | Constraints += MachineClobbers; |
| 1034 | } |
| 1035 | |
| 1036 | llvm::Type *IntPtrType = RetType->getPointerTo(); |
| 1037 | llvm::FunctionType *FTy = |
| 1038 | llvm::FunctionType::get(RetType, {IntPtrType}, false); |
| 1039 | |
| 1040 | llvm::InlineAsm *IA = |
| 1041 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
| 1042 | return CGF.Builder.CreateCall(IA, {Addr}); |
| 1043 | } |
| 1044 | |
| 1045 | namespace { |
| 1046 | enum class MSVCSetJmpKind { |
| 1047 | _setjmpex, |
| 1048 | _setjmp3, |
| 1049 | _setjmp |
| 1050 | }; |
| 1051 | } |
| 1052 | |
| 1053 | /// MSVC handles setjmp a bit differently on different platforms. On every |
| 1054 | /// architecture except 32-bit x86, the frame address is passed. On x86, extra |
| 1055 | /// parameters can be passed as variadic arguments, but we always pass none. |
| 1056 | static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind, |
| 1057 | const CallExpr *E) { |
| 1058 | llvm::Value *Arg1 = nullptr; |
| 1059 | llvm::Type *Arg1Ty = nullptr; |
| 1060 | StringRef Name; |
| 1061 | bool IsVarArg = false; |
| 1062 | if (SJKind == MSVCSetJmpKind::_setjmp3) { |
| 1063 | Name = "_setjmp3"; |
| 1064 | Arg1Ty = CGF.Int32Ty; |
| 1065 | Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0); |
| 1066 | IsVarArg = true; |
| 1067 | } else { |
| 1068 | Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex"; |
| 1069 | Arg1Ty = CGF.Int8PtrTy; |
| 1070 | if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) { |
| 1071 | Arg1 = CGF.Builder.CreateCall( |
| 1072 | CGF.CGM.getIntrinsic(Intrinsic::sponentry, CGF.AllocaInt8PtrTy)); |
| 1073 | } else |
| 1074 | Arg1 = CGF.Builder.CreateCall( |
| 1075 | CGF.CGM.getIntrinsic(Intrinsic::frameaddress, CGF.AllocaInt8PtrTy), |
| 1076 | llvm::ConstantInt::get(CGF.Int32Ty, 0)); |
| 1077 | } |
| 1078 | |
| 1079 | // Mark the call site and declaration with ReturnsTwice. |
| 1080 | llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty}; |
| 1081 | llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get( |
| 1082 | CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex, |
| 1083 | llvm::Attribute::ReturnsTwice); |
| 1084 | llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction( |
| 1085 | llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name, |
| 1086 | ReturnsTwiceAttr, /*Local=*/true); |
| 1087 | |
| 1088 | llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast( |
| 1089 | CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy); |
| 1090 | llvm::Value *Args[] = {Buf, Arg1}; |
| 1091 | llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args); |
| 1092 | CB->setAttributes(ReturnsTwiceAttr); |
| 1093 | return RValue::get(CB); |
| 1094 | } |
| 1095 | |
| 1096 | // Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code, |
| 1097 | // we handle them here. |
| 1098 | enum class CodeGenFunction::MSVCIntrin { |
| 1099 | _BitScanForward, |
| 1100 | _BitScanReverse, |
| 1101 | _InterlockedAnd, |
| 1102 | _InterlockedDecrement, |
| 1103 | _InterlockedExchange, |
| 1104 | _InterlockedExchangeAdd, |
| 1105 | _InterlockedExchangeSub, |
| 1106 | _InterlockedIncrement, |
| 1107 | _InterlockedOr, |
| 1108 | _InterlockedXor, |
| 1109 | _InterlockedExchangeAdd_acq, |
| 1110 | _InterlockedExchangeAdd_rel, |
| 1111 | _InterlockedExchangeAdd_nf, |
| 1112 | _InterlockedExchange_acq, |
| 1113 | _InterlockedExchange_rel, |
| 1114 | _InterlockedExchange_nf, |
| 1115 | _InterlockedCompareExchange_acq, |
| 1116 | _InterlockedCompareExchange_rel, |
| 1117 | _InterlockedCompareExchange_nf, |
| 1118 | _InterlockedCompareExchange128, |
| 1119 | _InterlockedCompareExchange128_acq, |
| 1120 | _InterlockedCompareExchange128_rel, |
| 1121 | _InterlockedCompareExchange128_nf, |
| 1122 | _InterlockedOr_acq, |
| 1123 | _InterlockedOr_rel, |
| 1124 | _InterlockedOr_nf, |
| 1125 | _InterlockedXor_acq, |
| 1126 | _InterlockedXor_rel, |
| 1127 | _InterlockedXor_nf, |
| 1128 | _InterlockedAnd_acq, |
| 1129 | _InterlockedAnd_rel, |
| 1130 | _InterlockedAnd_nf, |
| 1131 | _InterlockedIncrement_acq, |
| 1132 | _InterlockedIncrement_rel, |
| 1133 | _InterlockedIncrement_nf, |
| 1134 | _InterlockedDecrement_acq, |
| 1135 | _InterlockedDecrement_rel, |
| 1136 | _InterlockedDecrement_nf, |
| 1137 | __fastfail, |
| 1138 | }; |
| 1139 | |
| 1140 | static Optional<CodeGenFunction::MSVCIntrin> |
| 1141 | translateArmToMsvcIntrin(unsigned BuiltinID) { |
| 1142 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
| 1143 | switch (BuiltinID) { |
| 1144 | default: |
| 1145 | return None; |
| 1146 | case ARM::BI_BitScanForward: |
| 1147 | case ARM::BI_BitScanForward64: |
| 1148 | return MSVCIntrin::_BitScanForward; |
| 1149 | case ARM::BI_BitScanReverse: |
| 1150 | case ARM::BI_BitScanReverse64: |
| 1151 | return MSVCIntrin::_BitScanReverse; |
| 1152 | case ARM::BI_InterlockedAnd64: |
| 1153 | return MSVCIntrin::_InterlockedAnd; |
| 1154 | case ARM::BI_InterlockedExchange64: |
| 1155 | return MSVCIntrin::_InterlockedExchange; |
| 1156 | case ARM::BI_InterlockedExchangeAdd64: |
| 1157 | return MSVCIntrin::_InterlockedExchangeAdd; |
| 1158 | case ARM::BI_InterlockedExchangeSub64: |
| 1159 | return MSVCIntrin::_InterlockedExchangeSub; |
| 1160 | case ARM::BI_InterlockedOr64: |
| 1161 | return MSVCIntrin::_InterlockedOr; |
| 1162 | case ARM::BI_InterlockedXor64: |
| 1163 | return MSVCIntrin::_InterlockedXor; |
| 1164 | case ARM::BI_InterlockedDecrement64: |
| 1165 | return MSVCIntrin::_InterlockedDecrement; |
| 1166 | case ARM::BI_InterlockedIncrement64: |
| 1167 | return MSVCIntrin::_InterlockedIncrement; |
| 1168 | case ARM::BI_InterlockedExchangeAdd8_acq: |
| 1169 | case ARM::BI_InterlockedExchangeAdd16_acq: |
| 1170 | case ARM::BI_InterlockedExchangeAdd_acq: |
| 1171 | case ARM::BI_InterlockedExchangeAdd64_acq: |
| 1172 | return MSVCIntrin::_InterlockedExchangeAdd_acq; |
| 1173 | case ARM::BI_InterlockedExchangeAdd8_rel: |
| 1174 | case ARM::BI_InterlockedExchangeAdd16_rel: |
| 1175 | case ARM::BI_InterlockedExchangeAdd_rel: |
| 1176 | case ARM::BI_InterlockedExchangeAdd64_rel: |
| 1177 | return MSVCIntrin::_InterlockedExchangeAdd_rel; |
| 1178 | case ARM::BI_InterlockedExchangeAdd8_nf: |
| 1179 | case ARM::BI_InterlockedExchangeAdd16_nf: |
| 1180 | case ARM::BI_InterlockedExchangeAdd_nf: |
| 1181 | case ARM::BI_InterlockedExchangeAdd64_nf: |
| 1182 | return MSVCIntrin::_InterlockedExchangeAdd_nf; |
| 1183 | case ARM::BI_InterlockedExchange8_acq: |
| 1184 | case ARM::BI_InterlockedExchange16_acq: |
| 1185 | case ARM::BI_InterlockedExchange_acq: |
| 1186 | case ARM::BI_InterlockedExchange64_acq: |
| 1187 | return MSVCIntrin::_InterlockedExchange_acq; |
| 1188 | case ARM::BI_InterlockedExchange8_rel: |
| 1189 | case ARM::BI_InterlockedExchange16_rel: |
| 1190 | case ARM::BI_InterlockedExchange_rel: |
| 1191 | case ARM::BI_InterlockedExchange64_rel: |
| 1192 | return MSVCIntrin::_InterlockedExchange_rel; |
| 1193 | case ARM::BI_InterlockedExchange8_nf: |
| 1194 | case ARM::BI_InterlockedExchange16_nf: |
| 1195 | case ARM::BI_InterlockedExchange_nf: |
| 1196 | case ARM::BI_InterlockedExchange64_nf: |
| 1197 | return MSVCIntrin::_InterlockedExchange_nf; |
| 1198 | case ARM::BI_InterlockedCompareExchange8_acq: |
| 1199 | case ARM::BI_InterlockedCompareExchange16_acq: |
| 1200 | case ARM::BI_InterlockedCompareExchange_acq: |
| 1201 | case ARM::BI_InterlockedCompareExchange64_acq: |
| 1202 | return MSVCIntrin::_InterlockedCompareExchange_acq; |
| 1203 | case ARM::BI_InterlockedCompareExchange8_rel: |
| 1204 | case ARM::BI_InterlockedCompareExchange16_rel: |
| 1205 | case ARM::BI_InterlockedCompareExchange_rel: |
| 1206 | case ARM::BI_InterlockedCompareExchange64_rel: |
| 1207 | return MSVCIntrin::_InterlockedCompareExchange_rel; |
| 1208 | case ARM::BI_InterlockedCompareExchange8_nf: |
| 1209 | case ARM::BI_InterlockedCompareExchange16_nf: |
| 1210 | case ARM::BI_InterlockedCompareExchange_nf: |
| 1211 | case ARM::BI_InterlockedCompareExchange64_nf: |
| 1212 | return MSVCIntrin::_InterlockedCompareExchange_nf; |
| 1213 | case ARM::BI_InterlockedOr8_acq: |
| 1214 | case ARM::BI_InterlockedOr16_acq: |
| 1215 | case ARM::BI_InterlockedOr_acq: |
| 1216 | case ARM::BI_InterlockedOr64_acq: |
| 1217 | return MSVCIntrin::_InterlockedOr_acq; |
| 1218 | case ARM::BI_InterlockedOr8_rel: |
| 1219 | case ARM::BI_InterlockedOr16_rel: |
| 1220 | case ARM::BI_InterlockedOr_rel: |
| 1221 | case ARM::BI_InterlockedOr64_rel: |
| 1222 | return MSVCIntrin::_InterlockedOr_rel; |
| 1223 | case ARM::BI_InterlockedOr8_nf: |
| 1224 | case ARM::BI_InterlockedOr16_nf: |
| 1225 | case ARM::BI_InterlockedOr_nf: |
| 1226 | case ARM::BI_InterlockedOr64_nf: |
| 1227 | return MSVCIntrin::_InterlockedOr_nf; |
| 1228 | case ARM::BI_InterlockedXor8_acq: |
| 1229 | case ARM::BI_InterlockedXor16_acq: |
| 1230 | case ARM::BI_InterlockedXor_acq: |
| 1231 | case ARM::BI_InterlockedXor64_acq: |
| 1232 | return MSVCIntrin::_InterlockedXor_acq; |
| 1233 | case ARM::BI_InterlockedXor8_rel: |
| 1234 | case ARM::BI_InterlockedXor16_rel: |
| 1235 | case ARM::BI_InterlockedXor_rel: |
| 1236 | case ARM::BI_InterlockedXor64_rel: |
| 1237 | return MSVCIntrin::_InterlockedXor_rel; |
| 1238 | case ARM::BI_InterlockedXor8_nf: |
| 1239 | case ARM::BI_InterlockedXor16_nf: |
| 1240 | case ARM::BI_InterlockedXor_nf: |
| 1241 | case ARM::BI_InterlockedXor64_nf: |
| 1242 | return MSVCIntrin::_InterlockedXor_nf; |
| 1243 | case ARM::BI_InterlockedAnd8_acq: |
| 1244 | case ARM::BI_InterlockedAnd16_acq: |
| 1245 | case ARM::BI_InterlockedAnd_acq: |
| 1246 | case ARM::BI_InterlockedAnd64_acq: |
| 1247 | return MSVCIntrin::_InterlockedAnd_acq; |
| 1248 | case ARM::BI_InterlockedAnd8_rel: |
| 1249 | case ARM::BI_InterlockedAnd16_rel: |
| 1250 | case ARM::BI_InterlockedAnd_rel: |
| 1251 | case ARM::BI_InterlockedAnd64_rel: |
| 1252 | return MSVCIntrin::_InterlockedAnd_rel; |
| 1253 | case ARM::BI_InterlockedAnd8_nf: |
| 1254 | case ARM::BI_InterlockedAnd16_nf: |
| 1255 | case ARM::BI_InterlockedAnd_nf: |
| 1256 | case ARM::BI_InterlockedAnd64_nf: |
| 1257 | return MSVCIntrin::_InterlockedAnd_nf; |
| 1258 | case ARM::BI_InterlockedIncrement16_acq: |
| 1259 | case ARM::BI_InterlockedIncrement_acq: |
| 1260 | case ARM::BI_InterlockedIncrement64_acq: |
| 1261 | return MSVCIntrin::_InterlockedIncrement_acq; |
| 1262 | case ARM::BI_InterlockedIncrement16_rel: |
| 1263 | case ARM::BI_InterlockedIncrement_rel: |
| 1264 | case ARM::BI_InterlockedIncrement64_rel: |
| 1265 | return MSVCIntrin::_InterlockedIncrement_rel; |
| 1266 | case ARM::BI_InterlockedIncrement16_nf: |
| 1267 | case ARM::BI_InterlockedIncrement_nf: |
| 1268 | case ARM::BI_InterlockedIncrement64_nf: |
| 1269 | return MSVCIntrin::_InterlockedIncrement_nf; |
| 1270 | case ARM::BI_InterlockedDecrement16_acq: |
| 1271 | case ARM::BI_InterlockedDecrement_acq: |
| 1272 | case ARM::BI_InterlockedDecrement64_acq: |
| 1273 | return MSVCIntrin::_InterlockedDecrement_acq; |
| 1274 | case ARM::BI_InterlockedDecrement16_rel: |
| 1275 | case ARM::BI_InterlockedDecrement_rel: |
| 1276 | case ARM::BI_InterlockedDecrement64_rel: |
| 1277 | return MSVCIntrin::_InterlockedDecrement_rel; |
| 1278 | case ARM::BI_InterlockedDecrement16_nf: |
| 1279 | case ARM::BI_InterlockedDecrement_nf: |
| 1280 | case ARM::BI_InterlockedDecrement64_nf: |
| 1281 | return MSVCIntrin::_InterlockedDecrement_nf; |
| 1282 | } |
| 1283 | llvm_unreachable("must return from switch")__builtin_unreachable(); |
| 1284 | } |
| 1285 | |
| 1286 | static Optional<CodeGenFunction::MSVCIntrin> |
| 1287 | translateAarch64ToMsvcIntrin(unsigned BuiltinID) { |
| 1288 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
| 1289 | switch (BuiltinID) { |
| 1290 | default: |
| 1291 | return None; |
| 1292 | case AArch64::BI_BitScanForward: |
| 1293 | case AArch64::BI_BitScanForward64: |
| 1294 | return MSVCIntrin::_BitScanForward; |
| 1295 | case AArch64::BI_BitScanReverse: |
| 1296 | case AArch64::BI_BitScanReverse64: |
| 1297 | return MSVCIntrin::_BitScanReverse; |
| 1298 | case AArch64::BI_InterlockedAnd64: |
| 1299 | return MSVCIntrin::_InterlockedAnd; |
| 1300 | case AArch64::BI_InterlockedExchange64: |
| 1301 | return MSVCIntrin::_InterlockedExchange; |
| 1302 | case AArch64::BI_InterlockedExchangeAdd64: |
| 1303 | return MSVCIntrin::_InterlockedExchangeAdd; |
| 1304 | case AArch64::BI_InterlockedExchangeSub64: |
| 1305 | return MSVCIntrin::_InterlockedExchangeSub; |
| 1306 | case AArch64::BI_InterlockedOr64: |
| 1307 | return MSVCIntrin::_InterlockedOr; |
| 1308 | case AArch64::BI_InterlockedXor64: |
| 1309 | return MSVCIntrin::_InterlockedXor; |
| 1310 | case AArch64::BI_InterlockedDecrement64: |
| 1311 | return MSVCIntrin::_InterlockedDecrement; |
| 1312 | case AArch64::BI_InterlockedIncrement64: |
| 1313 | return MSVCIntrin::_InterlockedIncrement; |
| 1314 | case AArch64::BI_InterlockedExchangeAdd8_acq: |
| 1315 | case AArch64::BI_InterlockedExchangeAdd16_acq: |
| 1316 | case AArch64::BI_InterlockedExchangeAdd_acq: |
| 1317 | case AArch64::BI_InterlockedExchangeAdd64_acq: |
| 1318 | return MSVCIntrin::_InterlockedExchangeAdd_acq; |
| 1319 | case AArch64::BI_InterlockedExchangeAdd8_rel: |
| 1320 | case AArch64::BI_InterlockedExchangeAdd16_rel: |
| 1321 | case AArch64::BI_InterlockedExchangeAdd_rel: |
| 1322 | case AArch64::BI_InterlockedExchangeAdd64_rel: |
| 1323 | return MSVCIntrin::_InterlockedExchangeAdd_rel; |
| 1324 | case AArch64::BI_InterlockedExchangeAdd8_nf: |
| 1325 | case AArch64::BI_InterlockedExchangeAdd16_nf: |
| 1326 | case AArch64::BI_InterlockedExchangeAdd_nf: |
| 1327 | case AArch64::BI_InterlockedExchangeAdd64_nf: |
| 1328 | return MSVCIntrin::_InterlockedExchangeAdd_nf; |
| 1329 | case AArch64::BI_InterlockedExchange8_acq: |
| 1330 | case AArch64::BI_InterlockedExchange16_acq: |
| 1331 | case AArch64::BI_InterlockedExchange_acq: |
| 1332 | case AArch64::BI_InterlockedExchange64_acq: |
| 1333 | return MSVCIntrin::_InterlockedExchange_acq; |
| 1334 | case AArch64::BI_InterlockedExchange8_rel: |
| 1335 | case AArch64::BI_InterlockedExchange16_rel: |
| 1336 | case AArch64::BI_InterlockedExchange_rel: |
| 1337 | case AArch64::BI_InterlockedExchange64_rel: |
| 1338 | return MSVCIntrin::_InterlockedExchange_rel; |
| 1339 | case AArch64::BI_InterlockedExchange8_nf: |
| 1340 | case AArch64::BI_InterlockedExchange16_nf: |
| 1341 | case AArch64::BI_InterlockedExchange_nf: |
| 1342 | case AArch64::BI_InterlockedExchange64_nf: |
| 1343 | return MSVCIntrin::_InterlockedExchange_nf; |
| 1344 | case AArch64::BI_InterlockedCompareExchange8_acq: |
| 1345 | case AArch64::BI_InterlockedCompareExchange16_acq: |
| 1346 | case AArch64::BI_InterlockedCompareExchange_acq: |
| 1347 | case AArch64::BI_InterlockedCompareExchange64_acq: |
| 1348 | return MSVCIntrin::_InterlockedCompareExchange_acq; |
| 1349 | case AArch64::BI_InterlockedCompareExchange8_rel: |
| 1350 | case AArch64::BI_InterlockedCompareExchange16_rel: |
| 1351 | case AArch64::BI_InterlockedCompareExchange_rel: |
| 1352 | case AArch64::BI_InterlockedCompareExchange64_rel: |
| 1353 | return MSVCIntrin::_InterlockedCompareExchange_rel; |
| 1354 | case AArch64::BI_InterlockedCompareExchange8_nf: |
| 1355 | case AArch64::BI_InterlockedCompareExchange16_nf: |
| 1356 | case AArch64::BI_InterlockedCompareExchange_nf: |
| 1357 | case AArch64::BI_InterlockedCompareExchange64_nf: |
| 1358 | return MSVCIntrin::_InterlockedCompareExchange_nf; |
| 1359 | case AArch64::BI_InterlockedCompareExchange128: |
| 1360 | return MSVCIntrin::_InterlockedCompareExchange128; |
| 1361 | case AArch64::BI_InterlockedCompareExchange128_acq: |
| 1362 | return MSVCIntrin::_InterlockedCompareExchange128_acq; |
| 1363 | case AArch64::BI_InterlockedCompareExchange128_nf: |
| 1364 | return MSVCIntrin::_InterlockedCompareExchange128_nf; |
| 1365 | case AArch64::BI_InterlockedCompareExchange128_rel: |
| 1366 | return MSVCIntrin::_InterlockedCompareExchange128_rel; |
| 1367 | case AArch64::BI_InterlockedOr8_acq: |
| 1368 | case AArch64::BI_InterlockedOr16_acq: |
| 1369 | case AArch64::BI_InterlockedOr_acq: |
| 1370 | case AArch64::BI_InterlockedOr64_acq: |
| 1371 | return MSVCIntrin::_InterlockedOr_acq; |
| 1372 | case AArch64::BI_InterlockedOr8_rel: |
| 1373 | case AArch64::BI_InterlockedOr16_rel: |
| 1374 | case AArch64::BI_InterlockedOr_rel: |
| 1375 | case AArch64::BI_InterlockedOr64_rel: |
| 1376 | return MSVCIntrin::_InterlockedOr_rel; |
| 1377 | case AArch64::BI_InterlockedOr8_nf: |
| 1378 | case AArch64::BI_InterlockedOr16_nf: |
| 1379 | case AArch64::BI_InterlockedOr_nf: |
| 1380 | case AArch64::BI_InterlockedOr64_nf: |
| 1381 | return MSVCIntrin::_InterlockedOr_nf; |
| 1382 | case AArch64::BI_InterlockedXor8_acq: |
| 1383 | case AArch64::BI_InterlockedXor16_acq: |
| 1384 | case AArch64::BI_InterlockedXor_acq: |
| 1385 | case AArch64::BI_InterlockedXor64_acq: |
| 1386 | return MSVCIntrin::_InterlockedXor_acq; |
| 1387 | case AArch64::BI_InterlockedXor8_rel: |
| 1388 | case AArch64::BI_InterlockedXor16_rel: |
| 1389 | case AArch64::BI_InterlockedXor_rel: |
| 1390 | case AArch64::BI_InterlockedXor64_rel: |
| 1391 | return MSVCIntrin::_InterlockedXor_rel; |
| 1392 | case AArch64::BI_InterlockedXor8_nf: |
| 1393 | case AArch64::BI_InterlockedXor16_nf: |
| 1394 | case AArch64::BI_InterlockedXor_nf: |
| 1395 | case AArch64::BI_InterlockedXor64_nf: |
| 1396 | return MSVCIntrin::_InterlockedXor_nf; |
| 1397 | case AArch64::BI_InterlockedAnd8_acq: |
| 1398 | case AArch64::BI_InterlockedAnd16_acq: |
| 1399 | case AArch64::BI_InterlockedAnd_acq: |
| 1400 | case AArch64::BI_InterlockedAnd64_acq: |
| 1401 | return MSVCIntrin::_InterlockedAnd_acq; |
| 1402 | case AArch64::BI_InterlockedAnd8_rel: |
| 1403 | case AArch64::BI_InterlockedAnd16_rel: |
| 1404 | case AArch64::BI_InterlockedAnd_rel: |
| 1405 | case AArch64::BI_InterlockedAnd64_rel: |
| 1406 | return MSVCIntrin::_InterlockedAnd_rel; |
| 1407 | case AArch64::BI_InterlockedAnd8_nf: |
| 1408 | case AArch64::BI_InterlockedAnd16_nf: |
| 1409 | case AArch64::BI_InterlockedAnd_nf: |
| 1410 | case AArch64::BI_InterlockedAnd64_nf: |
| 1411 | return MSVCIntrin::_InterlockedAnd_nf; |
| 1412 | case AArch64::BI_InterlockedIncrement16_acq: |
| 1413 | case AArch64::BI_InterlockedIncrement_acq: |
| 1414 | case AArch64::BI_InterlockedIncrement64_acq: |
| 1415 | return MSVCIntrin::_InterlockedIncrement_acq; |
| 1416 | case AArch64::BI_InterlockedIncrement16_rel: |
| 1417 | case AArch64::BI_InterlockedIncrement_rel: |
| 1418 | case AArch64::BI_InterlockedIncrement64_rel: |
| 1419 | return MSVCIntrin::_InterlockedIncrement_rel; |
| 1420 | case AArch64::BI_InterlockedIncrement16_nf: |
| 1421 | case AArch64::BI_InterlockedIncrement_nf: |
| 1422 | case AArch64::BI_InterlockedIncrement64_nf: |
| 1423 | return MSVCIntrin::_InterlockedIncrement_nf; |
| 1424 | case AArch64::BI_InterlockedDecrement16_acq: |
| 1425 | case AArch64::BI_InterlockedDecrement_acq: |
| 1426 | case AArch64::BI_InterlockedDecrement64_acq: |
| 1427 | return MSVCIntrin::_InterlockedDecrement_acq; |
| 1428 | case AArch64::BI_InterlockedDecrement16_rel: |
| 1429 | case AArch64::BI_InterlockedDecrement_rel: |
| 1430 | case AArch64::BI_InterlockedDecrement64_rel: |
| 1431 | return MSVCIntrin::_InterlockedDecrement_rel; |
| 1432 | case AArch64::BI_InterlockedDecrement16_nf: |
| 1433 | case AArch64::BI_InterlockedDecrement_nf: |
| 1434 | case AArch64::BI_InterlockedDecrement64_nf: |
| 1435 | return MSVCIntrin::_InterlockedDecrement_nf; |
| 1436 | } |
| 1437 | llvm_unreachable("must return from switch")__builtin_unreachable(); |
| 1438 | } |
| 1439 | |
| 1440 | static Optional<CodeGenFunction::MSVCIntrin> |
| 1441 | translateX86ToMsvcIntrin(unsigned BuiltinID) { |
| 1442 | using MSVCIntrin = CodeGenFunction::MSVCIntrin; |
| 1443 | switch (BuiltinID) { |
| 1444 | default: |
| 1445 | return None; |
| 1446 | case clang::X86::BI_BitScanForward: |
| 1447 | case clang::X86::BI_BitScanForward64: |
| 1448 | return MSVCIntrin::_BitScanForward; |
| 1449 | case clang::X86::BI_BitScanReverse: |
| 1450 | case clang::X86::BI_BitScanReverse64: |
| 1451 | return MSVCIntrin::_BitScanReverse; |
| 1452 | case clang::X86::BI_InterlockedAnd64: |
| 1453 | return MSVCIntrin::_InterlockedAnd; |
| 1454 | case clang::X86::BI_InterlockedCompareExchange128: |
| 1455 | return MSVCIntrin::_InterlockedCompareExchange128; |
| 1456 | case clang::X86::BI_InterlockedExchange64: |
| 1457 | return MSVCIntrin::_InterlockedExchange; |
| 1458 | case clang::X86::BI_InterlockedExchangeAdd64: |
| 1459 | return MSVCIntrin::_InterlockedExchangeAdd; |
| 1460 | case clang::X86::BI_InterlockedExchangeSub64: |
| 1461 | return MSVCIntrin::_InterlockedExchangeSub; |
| 1462 | case clang::X86::BI_InterlockedOr64: |
| 1463 | return MSVCIntrin::_InterlockedOr; |
| 1464 | case clang::X86::BI_InterlockedXor64: |
| 1465 | return MSVCIntrin::_InterlockedXor; |
| 1466 | case clang::X86::BI_InterlockedDecrement64: |
| 1467 | return MSVCIntrin::_InterlockedDecrement; |
| 1468 | case clang::X86::BI_InterlockedIncrement64: |
| 1469 | return MSVCIntrin::_InterlockedIncrement; |
| 1470 | } |
| 1471 | llvm_unreachable("must return from switch")__builtin_unreachable(); |
| 1472 | } |
| 1473 | |
| 1474 | // Emit an MSVC intrinsic. Assumes that arguments have *not* been evaluated. |
| 1475 | Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID, |
| 1476 | const CallExpr *E) { |
| 1477 | switch (BuiltinID) { |
| 1478 | case MSVCIntrin::_BitScanForward: |
| 1479 | case MSVCIntrin::_BitScanReverse: { |
| 1480 | Address IndexAddress(EmitPointerWithAlignment(E->getArg(0))); |
| 1481 | Value *ArgValue = EmitScalarExpr(E->getArg(1)); |
| 1482 | |
| 1483 | llvm::Type *ArgType = ArgValue->getType(); |
| 1484 | llvm::Type *IndexType = |
| 1485 | IndexAddress.getPointer()->getType()->getPointerElementType(); |
| 1486 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 1487 | |
| 1488 | Value *ArgZero = llvm::Constant::getNullValue(ArgType); |
| 1489 | Value *ResZero = llvm::Constant::getNullValue(ResultType); |
| 1490 | Value *ResOne = llvm::ConstantInt::get(ResultType, 1); |
| 1491 | |
| 1492 | BasicBlock *Begin = Builder.GetInsertBlock(); |
| 1493 | BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn); |
| 1494 | Builder.SetInsertPoint(End); |
| 1495 | PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result"); |
| 1496 | |
| 1497 | Builder.SetInsertPoint(Begin); |
| 1498 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero); |
| 1499 | BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn); |
| 1500 | Builder.CreateCondBr(IsZero, End, NotZero); |
| 1501 | Result->addIncoming(ResZero, Begin); |
| 1502 | |
| 1503 | Builder.SetInsertPoint(NotZero); |
| 1504 | |
| 1505 | if (BuiltinID == MSVCIntrin::_BitScanForward) { |
| 1506 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 1507 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
| 1508 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
| 1509 | Builder.CreateStore(ZeroCount, IndexAddress, false); |
| 1510 | } else { |
| 1511 | unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth(); |
| 1512 | Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1); |
| 1513 | |
| 1514 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 1515 | Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()}); |
| 1516 | ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false); |
| 1517 | Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount); |
| 1518 | Builder.CreateStore(Index, IndexAddress, false); |
| 1519 | } |
| 1520 | Builder.CreateBr(End); |
| 1521 | Result->addIncoming(ResOne, NotZero); |
| 1522 | |
| 1523 | Builder.SetInsertPoint(End); |
| 1524 | return Result; |
| 1525 | } |
| 1526 | case MSVCIntrin::_InterlockedAnd: |
| 1527 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E); |
| 1528 | case MSVCIntrin::_InterlockedExchange: |
| 1529 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E); |
| 1530 | case MSVCIntrin::_InterlockedExchangeAdd: |
| 1531 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E); |
| 1532 | case MSVCIntrin::_InterlockedExchangeSub: |
| 1533 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E); |
| 1534 | case MSVCIntrin::_InterlockedOr: |
| 1535 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E); |
| 1536 | case MSVCIntrin::_InterlockedXor: |
| 1537 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E); |
| 1538 | case MSVCIntrin::_InterlockedExchangeAdd_acq: |
| 1539 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 1540 | AtomicOrdering::Acquire); |
| 1541 | case MSVCIntrin::_InterlockedExchangeAdd_rel: |
| 1542 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 1543 | AtomicOrdering::Release); |
| 1544 | case MSVCIntrin::_InterlockedExchangeAdd_nf: |
| 1545 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E, |
| 1546 | AtomicOrdering::Monotonic); |
| 1547 | case MSVCIntrin::_InterlockedExchange_acq: |
| 1548 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 1549 | AtomicOrdering::Acquire); |
| 1550 | case MSVCIntrin::_InterlockedExchange_rel: |
| 1551 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 1552 | AtomicOrdering::Release); |
| 1553 | case MSVCIntrin::_InterlockedExchange_nf: |
| 1554 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E, |
| 1555 | AtomicOrdering::Monotonic); |
| 1556 | case MSVCIntrin::_InterlockedCompareExchange_acq: |
| 1557 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire); |
| 1558 | case MSVCIntrin::_InterlockedCompareExchange_rel: |
| 1559 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release); |
| 1560 | case MSVCIntrin::_InterlockedCompareExchange_nf: |
| 1561 | return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic); |
| 1562 | case MSVCIntrin::_InterlockedCompareExchange128: |
| 1563 | return EmitAtomicCmpXchg128ForMSIntrin( |
| 1564 | *this, E, AtomicOrdering::SequentiallyConsistent); |
| 1565 | case MSVCIntrin::_InterlockedCompareExchange128_acq: |
| 1566 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Acquire); |
| 1567 | case MSVCIntrin::_InterlockedCompareExchange128_rel: |
| 1568 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Release); |
| 1569 | case MSVCIntrin::_InterlockedCompareExchange128_nf: |
| 1570 | return EmitAtomicCmpXchg128ForMSIntrin(*this, E, AtomicOrdering::Monotonic); |
| 1571 | case MSVCIntrin::_InterlockedOr_acq: |
| 1572 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1573 | AtomicOrdering::Acquire); |
| 1574 | case MSVCIntrin::_InterlockedOr_rel: |
| 1575 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1576 | AtomicOrdering::Release); |
| 1577 | case MSVCIntrin::_InterlockedOr_nf: |
| 1578 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E, |
| 1579 | AtomicOrdering::Monotonic); |
| 1580 | case MSVCIntrin::_InterlockedXor_acq: |
| 1581 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1582 | AtomicOrdering::Acquire); |
| 1583 | case MSVCIntrin::_InterlockedXor_rel: |
| 1584 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1585 | AtomicOrdering::Release); |
| 1586 | case MSVCIntrin::_InterlockedXor_nf: |
| 1587 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E, |
| 1588 | AtomicOrdering::Monotonic); |
| 1589 | case MSVCIntrin::_InterlockedAnd_acq: |
| 1590 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1591 | AtomicOrdering::Acquire); |
| 1592 | case MSVCIntrin::_InterlockedAnd_rel: |
| 1593 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1594 | AtomicOrdering::Release); |
| 1595 | case MSVCIntrin::_InterlockedAnd_nf: |
| 1596 | return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E, |
| 1597 | AtomicOrdering::Monotonic); |
| 1598 | case MSVCIntrin::_InterlockedIncrement_acq: |
| 1599 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire); |
| 1600 | case MSVCIntrin::_InterlockedIncrement_rel: |
| 1601 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release); |
| 1602 | case MSVCIntrin::_InterlockedIncrement_nf: |
| 1603 | return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic); |
| 1604 | case MSVCIntrin::_InterlockedDecrement_acq: |
| 1605 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire); |
| 1606 | case MSVCIntrin::_InterlockedDecrement_rel: |
| 1607 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release); |
| 1608 | case MSVCIntrin::_InterlockedDecrement_nf: |
| 1609 | return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic); |
| 1610 | |
| 1611 | case MSVCIntrin::_InterlockedDecrement: |
| 1612 | return EmitAtomicDecrementValue(*this, E); |
| 1613 | case MSVCIntrin::_InterlockedIncrement: |
| 1614 | return EmitAtomicIncrementValue(*this, E); |
| 1615 | |
| 1616 | case MSVCIntrin::__fastfail: { |
| 1617 | // Request immediate process termination from the kernel. The instruction |
| 1618 | // sequences to do this are documented on MSDN: |
| 1619 | // https://msdn.microsoft.com/en-us/library/dn774154.aspx |
| 1620 | llvm::Triple::ArchType ISA = getTarget().getTriple().getArch(); |
| 1621 | StringRef Asm, Constraints; |
| 1622 | switch (ISA) { |
| 1623 | default: |
| 1624 | ErrorUnsupported(E, "__fastfail call for this architecture"); |
| 1625 | break; |
| 1626 | case llvm::Triple::x86: |
| 1627 | case llvm::Triple::x86_64: |
| 1628 | Asm = "int $$0x29"; |
| 1629 | Constraints = "{cx}"; |
| 1630 | break; |
| 1631 | case llvm::Triple::thumb: |
| 1632 | Asm = "udf #251"; |
| 1633 | Constraints = "{r0}"; |
| 1634 | break; |
| 1635 | case llvm::Triple::aarch64: |
| 1636 | Asm = "brk #0xF003"; |
| 1637 | Constraints = "{w0}"; |
| 1638 | } |
| 1639 | llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false); |
| 1640 | llvm::InlineAsm *IA = |
| 1641 | llvm::InlineAsm::get(FTy, Asm, Constraints, /*hasSideEffects=*/true); |
| 1642 | llvm::AttributeList NoReturnAttr = llvm::AttributeList::get( |
| 1643 | getLLVMContext(), llvm::AttributeList::FunctionIndex, |
| 1644 | llvm::Attribute::NoReturn); |
| 1645 | llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0))); |
| 1646 | CI->setAttributes(NoReturnAttr); |
| 1647 | return CI; |
| 1648 | } |
| 1649 | } |
| 1650 | llvm_unreachable("Incorrect MSVC intrinsic!")__builtin_unreachable(); |
| 1651 | } |
| 1652 | |
| 1653 | namespace { |
| 1654 | // ARC cleanup for __builtin_os_log_format |
| 1655 | struct CallObjCArcUse final : EHScopeStack::Cleanup { |
| 1656 | CallObjCArcUse(llvm::Value *object) : object(object) {} |
| 1657 | llvm::Value *object; |
| 1658 | |
| 1659 | void Emit(CodeGenFunction &CGF, Flags flags) override { |
| 1660 | CGF.EmitARCIntrinsicUse(object); |
| 1661 | } |
| 1662 | }; |
| 1663 | } |
| 1664 | |
| 1665 | Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E, |
| 1666 | BuiltinCheckKind Kind) { |
| 1667 | assert((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero)((void)0) |
| 1668 | && "Unsupported builtin check kind")((void)0); |
| 1669 | |
| 1670 | Value *ArgValue = EmitScalarExpr(E); |
| 1671 | if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef()) |
| 1672 | return ArgValue; |
| 1673 | |
| 1674 | SanitizerScope SanScope(this); |
| 1675 | Value *Cond = Builder.CreateICmpNE( |
| 1676 | ArgValue, llvm::Constant::getNullValue(ArgValue->getType())); |
| 1677 | EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin), |
| 1678 | SanitizerHandler::InvalidBuiltin, |
| 1679 | {EmitCheckSourceLocation(E->getExprLoc()), |
| 1680 | llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)}, |
| 1681 | None); |
| 1682 | return ArgValue; |
| 1683 | } |
| 1684 | |
| 1685 | /// Get the argument type for arguments to os_log_helper. |
| 1686 | static CanQualType getOSLogArgType(ASTContext &C, int Size) { |
| 1687 | QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false); |
| 1688 | return C.getCanonicalType(UnsignedTy); |
| 1689 | } |
| 1690 | |
| 1691 | llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction( |
| 1692 | const analyze_os_log::OSLogBufferLayout &Layout, |
| 1693 | CharUnits BufferAlignment) { |
| 1694 | ASTContext &Ctx = getContext(); |
| 1695 | |
| 1696 | llvm::SmallString<64> Name; |
| 1697 | { |
| 1698 | raw_svector_ostream OS(Name); |
| 1699 | OS << "__os_log_helper"; |
| 1700 | OS << "_" << BufferAlignment.getQuantity(); |
| 1701 | OS << "_" << int(Layout.getSummaryByte()); |
| 1702 | OS << "_" << int(Layout.getNumArgsByte()); |
| 1703 | for (const auto &Item : Layout.Items) |
| 1704 | OS << "_" << int(Item.getSizeByte()) << "_" |
| 1705 | << int(Item.getDescriptorByte()); |
| 1706 | } |
| 1707 | |
| 1708 | if (llvm::Function *F = CGM.getModule().getFunction(Name)) |
| 1709 | return F; |
| 1710 | |
| 1711 | llvm::SmallVector<QualType, 4> ArgTys; |
| 1712 | FunctionArgList Args; |
| 1713 | Args.push_back(ImplicitParamDecl::Create( |
| 1714 | Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"), Ctx.VoidPtrTy, |
| 1715 | ImplicitParamDecl::Other)); |
| 1716 | ArgTys.emplace_back(Ctx.VoidPtrTy); |
| 1717 | |
| 1718 | for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) { |
| 1719 | char Size = Layout.Items[I].getSizeByte(); |
| 1720 | if (!Size) |
| 1721 | continue; |
| 1722 | |
| 1723 | QualType ArgTy = getOSLogArgType(Ctx, Size); |
| 1724 | Args.push_back(ImplicitParamDecl::Create( |
| 1725 | Ctx, nullptr, SourceLocation(), |
| 1726 | &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy, |
| 1727 | ImplicitParamDecl::Other)); |
| 1728 | ArgTys.emplace_back(ArgTy); |
| 1729 | } |
| 1730 | |
| 1731 | QualType ReturnTy = Ctx.VoidTy; |
| 1732 | |
| 1733 | // The helper function has linkonce_odr linkage to enable the linker to merge |
| 1734 | // identical functions. To ensure the merging always happens, 'noinline' is |
| 1735 | // attached to the function when compiling with -Oz. |
| 1736 | const CGFunctionInfo &FI = |
| 1737 | CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args); |
| 1738 | llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI); |
| 1739 | llvm::Function *Fn = llvm::Function::Create( |
| 1740 | FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule()); |
| 1741 | Fn->setVisibility(llvm::GlobalValue::HiddenVisibility); |
| 1742 | CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn, /*IsThunk=*/false); |
| 1743 | CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn); |
| 1744 | Fn->setDoesNotThrow(); |
| 1745 | |
| 1746 | // Attach 'noinline' at -Oz. |
| 1747 | if (CGM.getCodeGenOpts().OptimizeSize == 2) |
| 1748 | Fn->addFnAttr(llvm::Attribute::NoInline); |
| 1749 | |
| 1750 | auto NL = ApplyDebugLocation::CreateEmpty(*this); |
| 1751 | StartFunction(GlobalDecl(), ReturnTy, Fn, FI, Args); |
| 1752 | |
| 1753 | // Create a scope with an artificial location for the body of this function. |
| 1754 | auto AL = ApplyDebugLocation::CreateArtificial(*this); |
| 1755 | |
| 1756 | CharUnits Offset; |
| 1757 | Address BufAddr(Builder.CreateLoad(GetAddrOfLocalVar(Args[0]), "buf"), |
| 1758 | BufferAlignment); |
| 1759 | Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()), |
| 1760 | Builder.CreateConstByteGEP(BufAddr, Offset++, "summary")); |
| 1761 | Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()), |
| 1762 | Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs")); |
| 1763 | |
| 1764 | unsigned I = 1; |
| 1765 | for (const auto &Item : Layout.Items) { |
| 1766 | Builder.CreateStore( |
| 1767 | Builder.getInt8(Item.getDescriptorByte()), |
| 1768 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor")); |
| 1769 | Builder.CreateStore( |
| 1770 | Builder.getInt8(Item.getSizeByte()), |
| 1771 | Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize")); |
| 1772 | |
| 1773 | CharUnits Size = Item.size(); |
| 1774 | if (!Size.getQuantity()) |
| 1775 | continue; |
| 1776 | |
| 1777 | Address Arg = GetAddrOfLocalVar(Args[I]); |
| 1778 | Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData"); |
| 1779 | Addr = Builder.CreateBitCast(Addr, Arg.getPointer()->getType(), |
| 1780 | "argDataCast"); |
| 1781 | Builder.CreateStore(Builder.CreateLoad(Arg), Addr); |
| 1782 | Offset += Size; |
| 1783 | ++I; |
| 1784 | } |
| 1785 | |
| 1786 | FinishFunction(); |
| 1787 | |
| 1788 | return Fn; |
| 1789 | } |
| 1790 | |
| 1791 | RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) { |
| 1792 | assert(E.getNumArgs() >= 2 &&((void)0) |
| 1793 | "__builtin_os_log_format takes at least 2 arguments")((void)0); |
| 1794 | ASTContext &Ctx = getContext(); |
| 1795 | analyze_os_log::OSLogBufferLayout Layout; |
| 1796 | analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout); |
| 1797 | Address BufAddr = EmitPointerWithAlignment(E.getArg(0)); |
| 1798 | llvm::SmallVector<llvm::Value *, 4> RetainableOperands; |
| 1799 | |
| 1800 | // Ignore argument 1, the format string. It is not currently used. |
| 1801 | CallArgList Args; |
| 1802 | Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy); |
| 1803 | |
| 1804 | for (const auto &Item : Layout.Items) { |
| 1805 | int Size = Item.getSizeByte(); |
| 1806 | if (!Size) |
| 1807 | continue; |
| 1808 | |
| 1809 | llvm::Value *ArgVal; |
| 1810 | |
| 1811 | if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) { |
| 1812 | uint64_t Val = 0; |
| 1813 | for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I) |
| 1814 | Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8; |
| 1815 | ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val)); |
| 1816 | } else if (const Expr *TheExpr = Item.getExpr()) { |
| 1817 | ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false); |
| 1818 | |
| 1819 | // If a temporary object that requires destruction after the full |
| 1820 | // expression is passed, push a lifetime-extended cleanup to extend its |
| 1821 | // lifetime to the end of the enclosing block scope. |
| 1822 | auto LifetimeExtendObject = [&](const Expr *E) { |
| 1823 | E = E->IgnoreParenCasts(); |
| 1824 | // Extend lifetimes of objects returned by function calls and message |
| 1825 | // sends. |
| 1826 | |
| 1827 | // FIXME: We should do this in other cases in which temporaries are |
| 1828 | // created including arguments of non-ARC types (e.g., C++ |
| 1829 | // temporaries). |
| 1830 | if (isa<CallExpr>(E) || isa<ObjCMessageExpr>(E)) |
| 1831 | return true; |
| 1832 | return false; |
| 1833 | }; |
| 1834 | |
| 1835 | if (TheExpr->getType()->isObjCRetainableType() && |
| 1836 | getLangOpts().ObjCAutoRefCount && LifetimeExtendObject(TheExpr)) { |
| 1837 | assert(getEvaluationKind(TheExpr->getType()) == TEK_Scalar &&((void)0) |
| 1838 | "Only scalar can be a ObjC retainable type")((void)0); |
| 1839 | if (!isa<Constant>(ArgVal)) { |
| 1840 | CleanupKind Cleanup = getARCCleanupKind(); |
| 1841 | QualType Ty = TheExpr->getType(); |
| 1842 | Address Alloca = Address::invalid(); |
| 1843 | Address Addr = CreateMemTemp(Ty, "os.log.arg", &Alloca); |
| 1844 | ArgVal = EmitARCRetain(Ty, ArgVal); |
| 1845 | Builder.CreateStore(ArgVal, Addr); |
| 1846 | pushLifetimeExtendedDestroy(Cleanup, Alloca, Ty, |
| 1847 | CodeGenFunction::destroyARCStrongPrecise, |
| 1848 | Cleanup & EHCleanup); |
| 1849 | |
| 1850 | // Push a clang.arc.use call to ensure ARC optimizer knows that the |
| 1851 | // argument has to be alive. |
| 1852 | if (CGM.getCodeGenOpts().OptimizationLevel != 0) |
| 1853 | pushCleanupAfterFullExpr<CallObjCArcUse>(Cleanup, ArgVal); |
| 1854 | } |
| 1855 | } |
| 1856 | } else { |
| 1857 | ArgVal = Builder.getInt32(Item.getConstValue().getQuantity()); |
| 1858 | } |
| 1859 | |
| 1860 | unsigned ArgValSize = |
| 1861 | CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType()); |
| 1862 | llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(), |
| 1863 | ArgValSize); |
| 1864 | ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy); |
| 1865 | CanQualType ArgTy = getOSLogArgType(Ctx, Size); |
| 1866 | // If ArgVal has type x86_fp80, zero-extend ArgVal. |
| 1867 | ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy)); |
| 1868 | Args.add(RValue::get(ArgVal), ArgTy); |
| 1869 | } |
| 1870 | |
| 1871 | const CGFunctionInfo &FI = |
| 1872 | CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args); |
| 1873 | llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction( |
| 1874 | Layout, BufAddr.getAlignment()); |
| 1875 | EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args); |
| 1876 | return RValue::get(BufAddr.getPointer()); |
| 1877 | } |
| 1878 | |
| 1879 | static bool isSpecialUnsignedMultiplySignedResult( |
| 1880 | unsigned BuiltinID, WidthAndSignedness Op1Info, WidthAndSignedness Op2Info, |
| 1881 | WidthAndSignedness ResultInfo) { |
| 1882 | return BuiltinID == Builtin::BI__builtin_mul_overflow && |
| 1883 | Op1Info.Width == Op2Info.Width && Op2Info.Width == ResultInfo.Width && |
| 1884 | !Op1Info.Signed && !Op2Info.Signed && ResultInfo.Signed; |
| 1885 | } |
| 1886 | |
| 1887 | static RValue EmitCheckedUnsignedMultiplySignedResult( |
| 1888 | CodeGenFunction &CGF, const clang::Expr *Op1, WidthAndSignedness Op1Info, |
| 1889 | const clang::Expr *Op2, WidthAndSignedness Op2Info, |
| 1890 | const clang::Expr *ResultArg, QualType ResultQTy, |
| 1891 | WidthAndSignedness ResultInfo) { |
| 1892 | assert(isSpecialUnsignedMultiplySignedResult(((void)0) |
| 1893 | Builtin::BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) &&((void)0) |
| 1894 | "Cannot specialize this multiply")((void)0); |
| 1895 | |
| 1896 | llvm::Value *V1 = CGF.EmitScalarExpr(Op1); |
| 1897 | llvm::Value *V2 = CGF.EmitScalarExpr(Op2); |
| 1898 | |
| 1899 | llvm::Value *HasOverflow; |
| 1900 | llvm::Value *Result = EmitOverflowIntrinsic( |
| 1901 | CGF, llvm::Intrinsic::umul_with_overflow, V1, V2, HasOverflow); |
| 1902 | |
| 1903 | // The intrinsic call will detect overflow when the value is > UINT_MAX, |
| 1904 | // however, since the original builtin had a signed result, we need to report |
| 1905 | // an overflow when the result is greater than INT_MAX. |
| 1906 | auto IntMax = llvm::APInt::getSignedMaxValue(ResultInfo.Width); |
| 1907 | llvm::Value *IntMaxValue = llvm::ConstantInt::get(Result->getType(), IntMax); |
| 1908 | |
| 1909 | llvm::Value *IntMaxOverflow = CGF.Builder.CreateICmpUGT(Result, IntMaxValue); |
| 1910 | HasOverflow = CGF.Builder.CreateOr(HasOverflow, IntMaxOverflow); |
| 1911 | |
| 1912 | bool isVolatile = |
| 1913 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 1914 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); |
| 1915 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, |
| 1916 | isVolatile); |
| 1917 | return RValue::get(HasOverflow); |
| 1918 | } |
| 1919 | |
| 1920 | /// Determine if a binop is a checked mixed-sign multiply we can specialize. |
| 1921 | static bool isSpecialMixedSignMultiply(unsigned BuiltinID, |
| 1922 | WidthAndSignedness Op1Info, |
| 1923 | WidthAndSignedness Op2Info, |
| 1924 | WidthAndSignedness ResultInfo) { |
| 1925 | return BuiltinID == Builtin::BI__builtin_mul_overflow && |
| 1926 | std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width && |
| 1927 | Op1Info.Signed != Op2Info.Signed; |
| 1928 | } |
| 1929 | |
| 1930 | /// Emit a checked mixed-sign multiply. This is a cheaper specialization of |
| 1931 | /// the generic checked-binop irgen. |
| 1932 | static RValue |
| 1933 | EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1, |
| 1934 | WidthAndSignedness Op1Info, const clang::Expr *Op2, |
| 1935 | WidthAndSignedness Op2Info, |
| 1936 | const clang::Expr *ResultArg, QualType ResultQTy, |
| 1937 | WidthAndSignedness ResultInfo) { |
| 1938 | assert(isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info,((void)0) |
| 1939 | Op2Info, ResultInfo) &&((void)0) |
| 1940 | "Not a mixed-sign multipliction we can specialize")((void)0); |
| 1941 | |
| 1942 | // Emit the signed and unsigned operands. |
| 1943 | const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2; |
| 1944 | const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1; |
| 1945 | llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp); |
| 1946 | llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp); |
| 1947 | unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width; |
| 1948 | unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width; |
| 1949 | |
| 1950 | // One of the operands may be smaller than the other. If so, [s|z]ext it. |
| 1951 | if (SignedOpWidth < UnsignedOpWidth) |
| 1952 | Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext"); |
| 1953 | if (UnsignedOpWidth < SignedOpWidth) |
| 1954 | Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext"); |
| 1955 | |
| 1956 | llvm::Type *OpTy = Signed->getType(); |
| 1957 | llvm::Value *Zero = llvm::Constant::getNullValue(OpTy); |
| 1958 | Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg); |
| 1959 | llvm::Type *ResTy = ResultPtr.getElementType(); |
| 1960 | unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width); |
| 1961 | |
| 1962 | // Take the absolute value of the signed operand. |
| 1963 | llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero); |
| 1964 | llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed); |
| 1965 | llvm::Value *AbsSigned = |
| 1966 | CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed); |
| 1967 | |
| 1968 | // Perform a checked unsigned multiplication. |
| 1969 | llvm::Value *UnsignedOverflow; |
| 1970 | llvm::Value *UnsignedResult = |
| 1971 | EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned, |
| 1972 | Unsigned, UnsignedOverflow); |
| 1973 | |
| 1974 | llvm::Value *Overflow, *Result; |
| 1975 | if (ResultInfo.Signed) { |
| 1976 | // Signed overflow occurs if the result is greater than INT_MAX or lesser |
| 1977 | // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative). |
| 1978 | auto IntMax = |
| 1979 | llvm::APInt::getSignedMaxValue(ResultInfo.Width).zextOrSelf(OpWidth); |
| 1980 | llvm::Value *MaxResult = |
| 1981 | CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax), |
| 1982 | CGF.Builder.CreateZExt(IsNegative, OpTy)); |
| 1983 | llvm::Value *SignedOverflow = |
| 1984 | CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult); |
| 1985 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow); |
| 1986 | |
| 1987 | // Prepare the signed result (possibly by negating it). |
| 1988 | llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult); |
| 1989 | llvm::Value *SignedResult = |
| 1990 | CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult); |
| 1991 | Result = CGF.Builder.CreateTrunc(SignedResult, ResTy); |
| 1992 | } else { |
| 1993 | // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX. |
| 1994 | llvm::Value *Underflow = CGF.Builder.CreateAnd( |
| 1995 | IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult)); |
| 1996 | Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow); |
| 1997 | if (ResultInfo.Width < OpWidth) { |
| 1998 | auto IntMax = |
| 1999 | llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth); |
| 2000 | llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT( |
| 2001 | UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax)); |
| 2002 | Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow); |
| 2003 | } |
| 2004 | |
| 2005 | // Negate the product if it would be negative in infinite precision. |
| 2006 | Result = CGF.Builder.CreateSelect( |
| 2007 | IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult); |
| 2008 | |
| 2009 | Result = CGF.Builder.CreateTrunc(Result, ResTy); |
| 2010 | } |
| 2011 | assert(Overflow && Result && "Missing overflow or result")((void)0); |
| 2012 | |
| 2013 | bool isVolatile = |
| 2014 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 2015 | CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr, |
| 2016 | isVolatile); |
| 2017 | return RValue::get(Overflow); |
| 2018 | } |
| 2019 | |
| 2020 | static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType, |
| 2021 | Value *&RecordPtr, CharUnits Align, |
| 2022 | llvm::FunctionCallee Func, int Lvl) { |
| 2023 | ASTContext &Context = CGF.getContext(); |
| 2024 | RecordDecl *RD = RType->castAs<RecordType>()->getDecl()->getDefinition(); |
| 2025 | std::string Pad = std::string(Lvl * 4, ' '); |
| 2026 | |
| 2027 | Value *GString = |
| 2028 | CGF.Builder.CreateGlobalStringPtr(RType.getAsString() + " {\n"); |
| 2029 | Value *Res = CGF.Builder.CreateCall(Func, {GString}); |
| 2030 | |
| 2031 | static llvm::DenseMap<QualType, const char *> Types; |
| 2032 | if (Types.empty()) { |
| 2033 | Types[Context.CharTy] = "%c"; |
| 2034 | Types[Context.BoolTy] = "%d"; |
| 2035 | Types[Context.SignedCharTy] = "%hhd"; |
| 2036 | Types[Context.UnsignedCharTy] = "%hhu"; |
| 2037 | Types[Context.IntTy] = "%d"; |
| 2038 | Types[Context.UnsignedIntTy] = "%u"; |
| 2039 | Types[Context.LongTy] = "%ld"; |
| 2040 | Types[Context.UnsignedLongTy] = "%lu"; |
| 2041 | Types[Context.LongLongTy] = "%lld"; |
| 2042 | Types[Context.UnsignedLongLongTy] = "%llu"; |
| 2043 | Types[Context.ShortTy] = "%hd"; |
| 2044 | Types[Context.UnsignedShortTy] = "%hu"; |
| 2045 | Types[Context.VoidPtrTy] = "%p"; |
| 2046 | Types[Context.FloatTy] = "%f"; |
| 2047 | Types[Context.DoubleTy] = "%f"; |
| 2048 | Types[Context.LongDoubleTy] = "%Lf"; |
| 2049 | Types[Context.getPointerType(Context.CharTy)] = "%s"; |
| 2050 | Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s"; |
| 2051 | } |
| 2052 | |
| 2053 | for (const auto *FD : RD->fields()) { |
| 2054 | Value *FieldPtr = RecordPtr; |
| 2055 | if (RD->isUnion()) |
| 2056 | FieldPtr = CGF.Builder.CreatePointerCast( |
| 2057 | FieldPtr, CGF.ConvertType(Context.getPointerType(FD->getType()))); |
| 2058 | else |
| 2059 | FieldPtr = CGF.Builder.CreateStructGEP(CGF.ConvertType(RType), FieldPtr, |
| 2060 | FD->getFieldIndex()); |
| 2061 | |
| 2062 | GString = CGF.Builder.CreateGlobalStringPtr( |
| 2063 | llvm::Twine(Pad) |
| 2064 | .concat(FD->getType().getAsString()) |
| 2065 | .concat(llvm::Twine(' ')) |
| 2066 | .concat(FD->getNameAsString()) |
| 2067 | .concat(" : ") |
| 2068 | .str()); |
| 2069 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); |
| 2070 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
| 2071 | |
| 2072 | QualType CanonicalType = |
| 2073 | FD->getType().getUnqualifiedType().getCanonicalType(); |
| 2074 | |
| 2075 | // We check whether we are in a recursive type |
| 2076 | if (CanonicalType->isRecordType()) { |
| 2077 | TmpRes = dumpRecord(CGF, CanonicalType, FieldPtr, Align, Func, Lvl + 1); |
| 2078 | Res = CGF.Builder.CreateAdd(TmpRes, Res); |
| 2079 | continue; |
| 2080 | } |
| 2081 | |
| 2082 | // We try to determine the best format to print the current field |
| 2083 | llvm::Twine Format = Types.find(CanonicalType) == Types.end() |
| 2084 | ? Types[Context.VoidPtrTy] |
| 2085 | : Types[CanonicalType]; |
| 2086 | |
| 2087 | Address FieldAddress = Address(FieldPtr, Align); |
| 2088 | FieldPtr = CGF.Builder.CreateLoad(FieldAddress); |
| 2089 | |
| 2090 | // FIXME Need to handle bitfield here |
| 2091 | GString = CGF.Builder.CreateGlobalStringPtr( |
| 2092 | Format.concat(llvm::Twine('\n')).str()); |
| 2093 | TmpRes = CGF.Builder.CreateCall(Func, {GString, FieldPtr}); |
| 2094 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
| 2095 | } |
| 2096 | |
| 2097 | GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n"); |
| 2098 | Value *TmpRes = CGF.Builder.CreateCall(Func, {GString}); |
| 2099 | Res = CGF.Builder.CreateAdd(Res, TmpRes); |
| 2100 | return Res; |
| 2101 | } |
| 2102 | |
| 2103 | static bool |
| 2104 | TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty, |
| 2105 | llvm::SmallPtrSetImpl<const Decl *> &Seen) { |
| 2106 | if (const auto *Arr = Ctx.getAsArrayType(Ty)) |
| 2107 | Ty = Ctx.getBaseElementType(Arr); |
| 2108 | |
| 2109 | const auto *Record = Ty->getAsCXXRecordDecl(); |
| 2110 | if (!Record) |
| 2111 | return false; |
| 2112 | |
| 2113 | // We've already checked this type, or are in the process of checking it. |
| 2114 | if (!Seen.insert(Record).second) |
| 2115 | return false; |
| 2116 | |
| 2117 | assert(Record->hasDefinition() &&((void)0) |
| 2118 | "Incomplete types should already be diagnosed")((void)0); |
| 2119 | |
| 2120 | if (Record->isDynamicClass()) |
| 2121 | return true; |
| 2122 | |
| 2123 | for (FieldDecl *F : Record->fields()) { |
| 2124 | if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen)) |
| 2125 | return true; |
| 2126 | } |
| 2127 | return false; |
| 2128 | } |
| 2129 | |
| 2130 | /// Determine if the specified type requires laundering by checking if it is a |
| 2131 | /// dynamic class type or contains a subobject which is a dynamic class type. |
| 2132 | static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) { |
| 2133 | if (!CGM.getCodeGenOpts().StrictVTablePointers) |
| 2134 | return false; |
| 2135 | llvm::SmallPtrSet<const Decl *, 16> Seen; |
| 2136 | return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen); |
| 2137 | } |
| 2138 | |
| 2139 | RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) { |
| 2140 | llvm::Value *Src = EmitScalarExpr(E->getArg(0)); |
| 2141 | llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1)); |
| 2142 | |
| 2143 | // The builtin's shift arg may have a different type than the source arg and |
| 2144 | // result, but the LLVM intrinsic uses the same type for all values. |
| 2145 | llvm::Type *Ty = Src->getType(); |
| 2146 | ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false); |
| 2147 | |
| 2148 | // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same. |
| 2149 | unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl; |
| 2150 | Function *F = CGM.getIntrinsic(IID, Ty); |
| 2151 | return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt })); |
| 2152 | } |
| 2153 | |
| 2154 | // Map math builtins for long-double to f128 version. |
| 2155 | static unsigned mutateLongDoubleBuiltin(unsigned BuiltinID) { |
| 2156 | switch (BuiltinID) { |
| 2157 | #define MUTATE_LDBL(func) \ |
| 2158 | case Builtin::BI__builtin_##func##l: \ |
| 2159 | return Builtin::BI__builtin_##func##f128; |
| 2160 | MUTATE_LDBL(sqrt) |
| 2161 | MUTATE_LDBL(cbrt) |
| 2162 | MUTATE_LDBL(fabs) |
| 2163 | MUTATE_LDBL(log) |
| 2164 | MUTATE_LDBL(log2) |
| 2165 | MUTATE_LDBL(log10) |
| 2166 | MUTATE_LDBL(log1p) |
| 2167 | MUTATE_LDBL(logb) |
| 2168 | MUTATE_LDBL(exp) |
| 2169 | MUTATE_LDBL(exp2) |
| 2170 | MUTATE_LDBL(expm1) |
| 2171 | MUTATE_LDBL(fdim) |
| 2172 | MUTATE_LDBL(hypot) |
| 2173 | MUTATE_LDBL(ilogb) |
| 2174 | MUTATE_LDBL(pow) |
| 2175 | MUTATE_LDBL(fmin) |
| 2176 | MUTATE_LDBL(fmax) |
| 2177 | MUTATE_LDBL(ceil) |
| 2178 | MUTATE_LDBL(trunc) |
| 2179 | MUTATE_LDBL(rint) |
| 2180 | MUTATE_LDBL(nearbyint) |
| 2181 | MUTATE_LDBL(round) |
| 2182 | MUTATE_LDBL(floor) |
| 2183 | MUTATE_LDBL(lround) |
| 2184 | MUTATE_LDBL(llround) |
| 2185 | MUTATE_LDBL(lrint) |
| 2186 | MUTATE_LDBL(llrint) |
| 2187 | MUTATE_LDBL(fmod) |
| 2188 | MUTATE_LDBL(modf) |
| 2189 | MUTATE_LDBL(nan) |
| 2190 | MUTATE_LDBL(nans) |
| 2191 | MUTATE_LDBL(inf) |
| 2192 | MUTATE_LDBL(fma) |
| 2193 | MUTATE_LDBL(sin) |
| 2194 | MUTATE_LDBL(cos) |
| 2195 | MUTATE_LDBL(tan) |
| 2196 | MUTATE_LDBL(sinh) |
| 2197 | MUTATE_LDBL(cosh) |
| 2198 | MUTATE_LDBL(tanh) |
| 2199 | MUTATE_LDBL(asin) |
| 2200 | MUTATE_LDBL(acos) |
| 2201 | MUTATE_LDBL(atan) |
| 2202 | MUTATE_LDBL(asinh) |
| 2203 | MUTATE_LDBL(acosh) |
| 2204 | MUTATE_LDBL(atanh) |
| 2205 | MUTATE_LDBL(atan2) |
| 2206 | MUTATE_LDBL(erf) |
| 2207 | MUTATE_LDBL(erfc) |
| 2208 | MUTATE_LDBL(ldexp) |
| 2209 | MUTATE_LDBL(frexp) |
| 2210 | MUTATE_LDBL(huge_val) |
| 2211 | MUTATE_LDBL(copysign) |
| 2212 | MUTATE_LDBL(nextafter) |
| 2213 | MUTATE_LDBL(nexttoward) |
| 2214 | MUTATE_LDBL(remainder) |
| 2215 | MUTATE_LDBL(remquo) |
| 2216 | MUTATE_LDBL(scalbln) |
| 2217 | MUTATE_LDBL(scalbn) |
| 2218 | MUTATE_LDBL(tgamma) |
| 2219 | MUTATE_LDBL(lgamma) |
| 2220 | #undef MUTATE_LDBL |
| 2221 | default: |
| 2222 | return BuiltinID; |
| 2223 | } |
| 2224 | } |
| 2225 | |
| 2226 | RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, |
| 2227 | const CallExpr *E, |
| 2228 | ReturnValueSlot ReturnValue) { |
| 2229 | const FunctionDecl *FD = GD.getDecl()->getAsFunction(); |
| 2230 | // See if we can constant fold this builtin. If so, don't emit it at all. |
| 2231 | Expr::EvalResult Result; |
| 2232 | if (E->EvaluateAsRValue(Result, CGM.getContext()) && |
| 2233 | !Result.hasSideEffects()) { |
| 2234 | if (Result.Val.isInt()) |
| 2235 | return RValue::get(llvm::ConstantInt::get(getLLVMContext(), |
| 2236 | Result.Val.getInt())); |
| 2237 | if (Result.Val.isFloat()) |
| 2238 | return RValue::get(llvm::ConstantFP::get(getLLVMContext(), |
| 2239 | Result.Val.getFloat())); |
| 2240 | } |
| 2241 | |
| 2242 | // If current long-double semantics is IEEE 128-bit, replace math builtins |
| 2243 | // of long-double with f128 equivalent. |
| 2244 | // TODO: This mutation should also be applied to other targets other than PPC, |
| 2245 | // after backend supports IEEE 128-bit style libcalls. |
| 2246 | if (getTarget().getTriple().isPPC64() && |
| 2247 | &getTarget().getLongDoubleFormat() == &llvm::APFloat::IEEEquad()) |
| 2248 | BuiltinID = mutateLongDoubleBuiltin(BuiltinID); |
| 2249 | |
| 2250 | // If the builtin has been declared explicitly with an assembler label, |
| 2251 | // disable the specialized emitting below. Ideally we should communicate the |
| 2252 | // rename in IR, or at least avoid generating the intrinsic calls that are |
| 2253 | // likely to get lowered to the renamed library functions. |
| 2254 | const unsigned BuiltinIDIfNoAsmLabel = |
| 2255 | FD->hasAttr<AsmLabelAttr>() ? 0 : BuiltinID; |
| 2256 | |
| 2257 | // There are LLVM math intrinsics/instructions corresponding to math library |
| 2258 | // functions except the LLVM op will never set errno while the math library |
| 2259 | // might. Also, math builtins have the same semantics as their math library |
| 2260 | // twins. Thus, we can transform math library and builtin calls to their |
| 2261 | // LLVM counterparts if the call is marked 'const' (known to never set errno). |
| 2262 | if (FD->hasAttr<ConstAttr>()) { |
| 2263 | switch (BuiltinIDIfNoAsmLabel) { |
| 2264 | case Builtin::BIceil: |
| 2265 | case Builtin::BIceilf: |
| 2266 | case Builtin::BIceill: |
| 2267 | case Builtin::BI__builtin_ceil: |
| 2268 | case Builtin::BI__builtin_ceilf: |
| 2269 | case Builtin::BI__builtin_ceilf16: |
| 2270 | case Builtin::BI__builtin_ceill: |
| 2271 | case Builtin::BI__builtin_ceilf128: |
| 2272 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2273 | Intrinsic::ceil, |
| 2274 | Intrinsic::experimental_constrained_ceil)); |
| 2275 | |
| 2276 | case Builtin::BIcopysign: |
| 2277 | case Builtin::BIcopysignf: |
| 2278 | case Builtin::BIcopysignl: |
| 2279 | case Builtin::BI__builtin_copysign: |
| 2280 | case Builtin::BI__builtin_copysignf: |
| 2281 | case Builtin::BI__builtin_copysignf16: |
| 2282 | case Builtin::BI__builtin_copysignl: |
| 2283 | case Builtin::BI__builtin_copysignf128: |
| 2284 | return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign)); |
| 2285 | |
| 2286 | case Builtin::BIcos: |
| 2287 | case Builtin::BIcosf: |
| 2288 | case Builtin::BIcosl: |
| 2289 | case Builtin::BI__builtin_cos: |
| 2290 | case Builtin::BI__builtin_cosf: |
| 2291 | case Builtin::BI__builtin_cosf16: |
| 2292 | case Builtin::BI__builtin_cosl: |
| 2293 | case Builtin::BI__builtin_cosf128: |
| 2294 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2295 | Intrinsic::cos, |
| 2296 | Intrinsic::experimental_constrained_cos)); |
| 2297 | |
| 2298 | case Builtin::BIexp: |
| 2299 | case Builtin::BIexpf: |
| 2300 | case Builtin::BIexpl: |
| 2301 | case Builtin::BI__builtin_exp: |
| 2302 | case Builtin::BI__builtin_expf: |
| 2303 | case Builtin::BI__builtin_expf16: |
| 2304 | case Builtin::BI__builtin_expl: |
| 2305 | case Builtin::BI__builtin_expf128: |
| 2306 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2307 | Intrinsic::exp, |
| 2308 | Intrinsic::experimental_constrained_exp)); |
| 2309 | |
| 2310 | case Builtin::BIexp2: |
| 2311 | case Builtin::BIexp2f: |
| 2312 | case Builtin::BIexp2l: |
| 2313 | case Builtin::BI__builtin_exp2: |
| 2314 | case Builtin::BI__builtin_exp2f: |
| 2315 | case Builtin::BI__builtin_exp2f16: |
| 2316 | case Builtin::BI__builtin_exp2l: |
| 2317 | case Builtin::BI__builtin_exp2f128: |
| 2318 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2319 | Intrinsic::exp2, |
| 2320 | Intrinsic::experimental_constrained_exp2)); |
| 2321 | |
| 2322 | case Builtin::BIfabs: |
| 2323 | case Builtin::BIfabsf: |
| 2324 | case Builtin::BIfabsl: |
| 2325 | case Builtin::BI__builtin_fabs: |
| 2326 | case Builtin::BI__builtin_fabsf: |
| 2327 | case Builtin::BI__builtin_fabsf16: |
| 2328 | case Builtin::BI__builtin_fabsl: |
| 2329 | case Builtin::BI__builtin_fabsf128: |
| 2330 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs)); |
| 2331 | |
| 2332 | case Builtin::BIfloor: |
| 2333 | case Builtin::BIfloorf: |
| 2334 | case Builtin::BIfloorl: |
| 2335 | case Builtin::BI__builtin_floor: |
| 2336 | case Builtin::BI__builtin_floorf: |
| 2337 | case Builtin::BI__builtin_floorf16: |
| 2338 | case Builtin::BI__builtin_floorl: |
| 2339 | case Builtin::BI__builtin_floorf128: |
| 2340 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2341 | Intrinsic::floor, |
| 2342 | Intrinsic::experimental_constrained_floor)); |
| 2343 | |
| 2344 | case Builtin::BIfma: |
| 2345 | case Builtin::BIfmaf: |
| 2346 | case Builtin::BIfmal: |
| 2347 | case Builtin::BI__builtin_fma: |
| 2348 | case Builtin::BI__builtin_fmaf: |
| 2349 | case Builtin::BI__builtin_fmaf16: |
| 2350 | case Builtin::BI__builtin_fmal: |
| 2351 | case Builtin::BI__builtin_fmaf128: |
| 2352 | return RValue::get(emitTernaryMaybeConstrainedFPBuiltin(*this, E, |
| 2353 | Intrinsic::fma, |
| 2354 | Intrinsic::experimental_constrained_fma)); |
| 2355 | |
| 2356 | case Builtin::BIfmax: |
| 2357 | case Builtin::BIfmaxf: |
| 2358 | case Builtin::BIfmaxl: |
| 2359 | case Builtin::BI__builtin_fmax: |
| 2360 | case Builtin::BI__builtin_fmaxf: |
| 2361 | case Builtin::BI__builtin_fmaxf16: |
| 2362 | case Builtin::BI__builtin_fmaxl: |
| 2363 | case Builtin::BI__builtin_fmaxf128: |
| 2364 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
| 2365 | Intrinsic::maxnum, |
| 2366 | Intrinsic::experimental_constrained_maxnum)); |
| 2367 | |
| 2368 | case Builtin::BIfmin: |
| 2369 | case Builtin::BIfminf: |
| 2370 | case Builtin::BIfminl: |
| 2371 | case Builtin::BI__builtin_fmin: |
| 2372 | case Builtin::BI__builtin_fminf: |
| 2373 | case Builtin::BI__builtin_fminf16: |
| 2374 | case Builtin::BI__builtin_fminl: |
| 2375 | case Builtin::BI__builtin_fminf128: |
| 2376 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
| 2377 | Intrinsic::minnum, |
| 2378 | Intrinsic::experimental_constrained_minnum)); |
| 2379 | |
| 2380 | // fmod() is a special-case. It maps to the frem instruction rather than an |
| 2381 | // LLVM intrinsic. |
| 2382 | case Builtin::BIfmod: |
| 2383 | case Builtin::BIfmodf: |
| 2384 | case Builtin::BIfmodl: |
| 2385 | case Builtin::BI__builtin_fmod: |
| 2386 | case Builtin::BI__builtin_fmodf: |
| 2387 | case Builtin::BI__builtin_fmodf16: |
| 2388 | case Builtin::BI__builtin_fmodl: |
| 2389 | case Builtin::BI__builtin_fmodf128: { |
| 2390 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 2391 | Value *Arg1 = EmitScalarExpr(E->getArg(0)); |
| 2392 | Value *Arg2 = EmitScalarExpr(E->getArg(1)); |
| 2393 | return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod")); |
| 2394 | } |
| 2395 | |
| 2396 | case Builtin::BIlog: |
| 2397 | case Builtin::BIlogf: |
| 2398 | case Builtin::BIlogl: |
| 2399 | case Builtin::BI__builtin_log: |
| 2400 | case Builtin::BI__builtin_logf: |
| 2401 | case Builtin::BI__builtin_logf16: |
| 2402 | case Builtin::BI__builtin_logl: |
| 2403 | case Builtin::BI__builtin_logf128: |
| 2404 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2405 | Intrinsic::log, |
| 2406 | Intrinsic::experimental_constrained_log)); |
| 2407 | |
| 2408 | case Builtin::BIlog10: |
| 2409 | case Builtin::BIlog10f: |
| 2410 | case Builtin::BIlog10l: |
| 2411 | case Builtin::BI__builtin_log10: |
| 2412 | case Builtin::BI__builtin_log10f: |
| 2413 | case Builtin::BI__builtin_log10f16: |
| 2414 | case Builtin::BI__builtin_log10l: |
| 2415 | case Builtin::BI__builtin_log10f128: |
| 2416 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2417 | Intrinsic::log10, |
| 2418 | Intrinsic::experimental_constrained_log10)); |
| 2419 | |
| 2420 | case Builtin::BIlog2: |
| 2421 | case Builtin::BIlog2f: |
| 2422 | case Builtin::BIlog2l: |
| 2423 | case Builtin::BI__builtin_log2: |
| 2424 | case Builtin::BI__builtin_log2f: |
| 2425 | case Builtin::BI__builtin_log2f16: |
| 2426 | case Builtin::BI__builtin_log2l: |
| 2427 | case Builtin::BI__builtin_log2f128: |
| 2428 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2429 | Intrinsic::log2, |
| 2430 | Intrinsic::experimental_constrained_log2)); |
| 2431 | |
| 2432 | case Builtin::BInearbyint: |
| 2433 | case Builtin::BInearbyintf: |
| 2434 | case Builtin::BInearbyintl: |
| 2435 | case Builtin::BI__builtin_nearbyint: |
| 2436 | case Builtin::BI__builtin_nearbyintf: |
| 2437 | case Builtin::BI__builtin_nearbyintl: |
| 2438 | case Builtin::BI__builtin_nearbyintf128: |
| 2439 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2440 | Intrinsic::nearbyint, |
| 2441 | Intrinsic::experimental_constrained_nearbyint)); |
| 2442 | |
| 2443 | case Builtin::BIpow: |
| 2444 | case Builtin::BIpowf: |
| 2445 | case Builtin::BIpowl: |
| 2446 | case Builtin::BI__builtin_pow: |
| 2447 | case Builtin::BI__builtin_powf: |
| 2448 | case Builtin::BI__builtin_powf16: |
| 2449 | case Builtin::BI__builtin_powl: |
| 2450 | case Builtin::BI__builtin_powf128: |
| 2451 | return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, |
| 2452 | Intrinsic::pow, |
| 2453 | Intrinsic::experimental_constrained_pow)); |
| 2454 | |
| 2455 | case Builtin::BIrint: |
| 2456 | case Builtin::BIrintf: |
| 2457 | case Builtin::BIrintl: |
| 2458 | case Builtin::BI__builtin_rint: |
| 2459 | case Builtin::BI__builtin_rintf: |
| 2460 | case Builtin::BI__builtin_rintf16: |
| 2461 | case Builtin::BI__builtin_rintl: |
| 2462 | case Builtin::BI__builtin_rintf128: |
| 2463 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2464 | Intrinsic::rint, |
| 2465 | Intrinsic::experimental_constrained_rint)); |
| 2466 | |
| 2467 | case Builtin::BIround: |
| 2468 | case Builtin::BIroundf: |
| 2469 | case Builtin::BIroundl: |
| 2470 | case Builtin::BI__builtin_round: |
| 2471 | case Builtin::BI__builtin_roundf: |
| 2472 | case Builtin::BI__builtin_roundf16: |
| 2473 | case Builtin::BI__builtin_roundl: |
| 2474 | case Builtin::BI__builtin_roundf128: |
| 2475 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2476 | Intrinsic::round, |
| 2477 | Intrinsic::experimental_constrained_round)); |
| 2478 | |
| 2479 | case Builtin::BIsin: |
| 2480 | case Builtin::BIsinf: |
| 2481 | case Builtin::BIsinl: |
| 2482 | case Builtin::BI__builtin_sin: |
| 2483 | case Builtin::BI__builtin_sinf: |
| 2484 | case Builtin::BI__builtin_sinf16: |
| 2485 | case Builtin::BI__builtin_sinl: |
| 2486 | case Builtin::BI__builtin_sinf128: |
| 2487 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2488 | Intrinsic::sin, |
| 2489 | Intrinsic::experimental_constrained_sin)); |
| 2490 | |
| 2491 | case Builtin::BIsqrt: |
| 2492 | case Builtin::BIsqrtf: |
| 2493 | case Builtin::BIsqrtl: |
| 2494 | case Builtin::BI__builtin_sqrt: |
| 2495 | case Builtin::BI__builtin_sqrtf: |
| 2496 | case Builtin::BI__builtin_sqrtf16: |
| 2497 | case Builtin::BI__builtin_sqrtl: |
| 2498 | case Builtin::BI__builtin_sqrtf128: |
| 2499 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2500 | Intrinsic::sqrt, |
| 2501 | Intrinsic::experimental_constrained_sqrt)); |
| 2502 | |
| 2503 | case Builtin::BItrunc: |
| 2504 | case Builtin::BItruncf: |
| 2505 | case Builtin::BItruncl: |
| 2506 | case Builtin::BI__builtin_trunc: |
| 2507 | case Builtin::BI__builtin_truncf: |
| 2508 | case Builtin::BI__builtin_truncf16: |
| 2509 | case Builtin::BI__builtin_truncl: |
| 2510 | case Builtin::BI__builtin_truncf128: |
| 2511 | return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(*this, E, |
| 2512 | Intrinsic::trunc, |
| 2513 | Intrinsic::experimental_constrained_trunc)); |
| 2514 | |
| 2515 | case Builtin::BIlround: |
| 2516 | case Builtin::BIlroundf: |
| 2517 | case Builtin::BIlroundl: |
| 2518 | case Builtin::BI__builtin_lround: |
| 2519 | case Builtin::BI__builtin_lroundf: |
| 2520 | case Builtin::BI__builtin_lroundl: |
| 2521 | case Builtin::BI__builtin_lroundf128: |
| 2522 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2523 | *this, E, Intrinsic::lround, |
| 2524 | Intrinsic::experimental_constrained_lround)); |
| 2525 | |
| 2526 | case Builtin::BIllround: |
| 2527 | case Builtin::BIllroundf: |
| 2528 | case Builtin::BIllroundl: |
| 2529 | case Builtin::BI__builtin_llround: |
| 2530 | case Builtin::BI__builtin_llroundf: |
| 2531 | case Builtin::BI__builtin_llroundl: |
| 2532 | case Builtin::BI__builtin_llroundf128: |
| 2533 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2534 | *this, E, Intrinsic::llround, |
| 2535 | Intrinsic::experimental_constrained_llround)); |
| 2536 | |
| 2537 | case Builtin::BIlrint: |
| 2538 | case Builtin::BIlrintf: |
| 2539 | case Builtin::BIlrintl: |
| 2540 | case Builtin::BI__builtin_lrint: |
| 2541 | case Builtin::BI__builtin_lrintf: |
| 2542 | case Builtin::BI__builtin_lrintl: |
| 2543 | case Builtin::BI__builtin_lrintf128: |
| 2544 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2545 | *this, E, Intrinsic::lrint, |
| 2546 | Intrinsic::experimental_constrained_lrint)); |
| 2547 | |
| 2548 | case Builtin::BIllrint: |
| 2549 | case Builtin::BIllrintf: |
| 2550 | case Builtin::BIllrintl: |
| 2551 | case Builtin::BI__builtin_llrint: |
| 2552 | case Builtin::BI__builtin_llrintf: |
| 2553 | case Builtin::BI__builtin_llrintl: |
| 2554 | case Builtin::BI__builtin_llrintf128: |
| 2555 | return RValue::get(emitMaybeConstrainedFPToIntRoundBuiltin( |
| 2556 | *this, E, Intrinsic::llrint, |
| 2557 | Intrinsic::experimental_constrained_llrint)); |
| 2558 | |
| 2559 | default: |
| 2560 | break; |
| 2561 | } |
| 2562 | } |
| 2563 | |
| 2564 | switch (BuiltinIDIfNoAsmLabel) { |
| 2565 | default: break; |
| 2566 | case Builtin::BI__builtin___CFStringMakeConstantString: |
| 2567 | case Builtin::BI__builtin___NSStringMakeConstantString: |
| 2568 | return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType())); |
| 2569 | case Builtin::BI__builtin_stdarg_start: |
| 2570 | case Builtin::BI__builtin_va_start: |
| 2571 | case Builtin::BI__va_start: |
| 2572 | case Builtin::BI__builtin_va_end: |
| 2573 | return RValue::get( |
| 2574 | EmitVAStartEnd(BuiltinID == Builtin::BI__va_start |
| 2575 | ? EmitScalarExpr(E->getArg(0)) |
| 2576 | : EmitVAListRef(E->getArg(0)).getPointer(), |
| 2577 | BuiltinID != Builtin::BI__builtin_va_end)); |
| 2578 | case Builtin::BI__builtin_va_copy: { |
| 2579 | Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer(); |
| 2580 | Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer(); |
| 2581 | |
| 2582 | llvm::Type *Type = Int8PtrTy; |
| 2583 | |
| 2584 | DstPtr = Builder.CreateBitCast(DstPtr, Type); |
| 2585 | SrcPtr = Builder.CreateBitCast(SrcPtr, Type); |
| 2586 | return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy), |
| 2587 | {DstPtr, SrcPtr})); |
| 2588 | } |
| 2589 | case Builtin::BI__builtin_abs: |
| 2590 | case Builtin::BI__builtin_labs: |
| 2591 | case Builtin::BI__builtin_llabs: { |
| 2592 | // X < 0 ? -X : X |
| 2593 | // The negation has 'nsw' because abs of INT_MIN is undefined. |
| 2594 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2595 | Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg"); |
| 2596 | Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType()); |
| 2597 | Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond"); |
| 2598 | Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs"); |
| 2599 | return RValue::get(Result); |
| 2600 | } |
| 2601 | case Builtin::BI__builtin_complex: { |
| 2602 | Value *Real = EmitScalarExpr(E->getArg(0)); |
| 2603 | Value *Imag = EmitScalarExpr(E->getArg(1)); |
| 2604 | return RValue::getComplex({Real, Imag}); |
| 2605 | } |
| 2606 | case Builtin::BI__builtin_conj: |
| 2607 | case Builtin::BI__builtin_conjf: |
| 2608 | case Builtin::BI__builtin_conjl: |
| 2609 | case Builtin::BIconj: |
| 2610 | case Builtin::BIconjf: |
| 2611 | case Builtin::BIconjl: { |
| 2612 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2613 | Value *Real = ComplexVal.first; |
| 2614 | Value *Imag = ComplexVal.second; |
| 2615 | Imag = Builder.CreateFNeg(Imag, "neg"); |
| 2616 | return RValue::getComplex(std::make_pair(Real, Imag)); |
| 2617 | } |
| 2618 | case Builtin::BI__builtin_creal: |
| 2619 | case Builtin::BI__builtin_crealf: |
| 2620 | case Builtin::BI__builtin_creall: |
| 2621 | case Builtin::BIcreal: |
| 2622 | case Builtin::BIcrealf: |
| 2623 | case Builtin::BIcreall: { |
| 2624 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2625 | return RValue::get(ComplexVal.first); |
| 2626 | } |
| 2627 | |
| 2628 | case Builtin::BI__builtin_dump_struct: { |
| 2629 | llvm::Type *LLVMIntTy = getTypes().ConvertType(getContext().IntTy); |
| 2630 | llvm::FunctionType *LLVMFuncType = llvm::FunctionType::get( |
| 2631 | LLVMIntTy, {llvm::Type::getInt8PtrTy(getLLVMContext())}, true); |
| 2632 | |
| 2633 | Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts()); |
| 2634 | CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment(); |
| 2635 | |
| 2636 | const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts(); |
| 2637 | QualType Arg0Type = Arg0->getType()->getPointeeType(); |
| 2638 | |
| 2639 | Value *RecordPtr = EmitScalarExpr(Arg0); |
| 2640 | Value *Res = dumpRecord(*this, Arg0Type, RecordPtr, Arg0Align, |
| 2641 | {LLVMFuncType, Func}, 0); |
| 2642 | return RValue::get(Res); |
| 2643 | } |
| 2644 | |
| 2645 | case Builtin::BI__builtin_preserve_access_index: { |
| 2646 | // Only enabled preserved access index region when debuginfo |
| 2647 | // is available as debuginfo is needed to preserve user-level |
| 2648 | // access pattern. |
| 2649 | if (!getDebugInfo()) { |
| 2650 | CGM.Error(E->getExprLoc(), "using builtin_preserve_access_index() without -g"); |
| 2651 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
| 2652 | } |
| 2653 | |
| 2654 | // Nested builtin_preserve_access_index() not supported |
| 2655 | if (IsInPreservedAIRegion) { |
| 2656 | CGM.Error(E->getExprLoc(), "nested builtin_preserve_access_index() not supported"); |
| 2657 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
| 2658 | } |
| 2659 | |
| 2660 | IsInPreservedAIRegion = true; |
| 2661 | Value *Res = EmitScalarExpr(E->getArg(0)); |
| 2662 | IsInPreservedAIRegion = false; |
| 2663 | return RValue::get(Res); |
| 2664 | } |
| 2665 | |
| 2666 | case Builtin::BI__builtin_cimag: |
| 2667 | case Builtin::BI__builtin_cimagf: |
| 2668 | case Builtin::BI__builtin_cimagl: |
| 2669 | case Builtin::BIcimag: |
| 2670 | case Builtin::BIcimagf: |
| 2671 | case Builtin::BIcimagl: { |
| 2672 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2673 | return RValue::get(ComplexVal.second); |
| 2674 | } |
| 2675 | |
| 2676 | case Builtin::BI__builtin_clrsb: |
| 2677 | case Builtin::BI__builtin_clrsbl: |
| 2678 | case Builtin::BI__builtin_clrsbll: { |
| 2679 | // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or |
| 2680 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2681 | |
| 2682 | llvm::Type *ArgType = ArgValue->getType(); |
| 2683 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 2684 | |
| 2685 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2686 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
| 2687 | Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg"); |
| 2688 | Value *Inverse = Builder.CreateNot(ArgValue, "not"); |
| 2689 | Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue); |
| 2690 | Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()}); |
| 2691 | Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1)); |
| 2692 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2693 | "cast"); |
| 2694 | return RValue::get(Result); |
| 2695 | } |
| 2696 | case Builtin::BI__builtin_ctzs: |
| 2697 | case Builtin::BI__builtin_ctz: |
| 2698 | case Builtin::BI__builtin_ctzl: |
| 2699 | case Builtin::BI__builtin_ctzll: { |
| 2700 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero); |
| 2701 | |
| 2702 | llvm::Type *ArgType = ArgValue->getType(); |
| 2703 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 2704 | |
| 2705 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2706 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
| 2707 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
| 2708 | if (Result->getType() != ResultType) |
| 2709 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2710 | "cast"); |
| 2711 | return RValue::get(Result); |
| 2712 | } |
| 2713 | case Builtin::BI__builtin_clzs: |
| 2714 | case Builtin::BI__builtin_clz: |
| 2715 | case Builtin::BI__builtin_clzl: |
| 2716 | case Builtin::BI__builtin_clzll: { |
| 2717 | Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero); |
| 2718 | |
| 2719 | llvm::Type *ArgType = ArgValue->getType(); |
| 2720 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 2721 | |
| 2722 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2723 | Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef()); |
| 2724 | Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef}); |
| 2725 | if (Result->getType() != ResultType) |
| 2726 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2727 | "cast"); |
| 2728 | return RValue::get(Result); |
| 2729 | } |
| 2730 | case Builtin::BI__builtin_ffs: |
| 2731 | case Builtin::BI__builtin_ffsl: |
| 2732 | case Builtin::BI__builtin_ffsll: { |
| 2733 | // ffs(x) -> x ? cttz(x) + 1 : 0 |
| 2734 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2735 | |
| 2736 | llvm::Type *ArgType = ArgValue->getType(); |
| 2737 | Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType); |
| 2738 | |
| 2739 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2740 | Value *Tmp = |
| 2741 | Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}), |
| 2742 | llvm::ConstantInt::get(ArgType, 1)); |
| 2743 | Value *Zero = llvm::Constant::getNullValue(ArgType); |
| 2744 | Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero"); |
| 2745 | Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs"); |
| 2746 | if (Result->getType() != ResultType) |
| 2747 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2748 | "cast"); |
| 2749 | return RValue::get(Result); |
| 2750 | } |
| 2751 | case Builtin::BI__builtin_parity: |
| 2752 | case Builtin::BI__builtin_parityl: |
| 2753 | case Builtin::BI__builtin_parityll: { |
| 2754 | // parity(x) -> ctpop(x) & 1 |
| 2755 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2756 | |
| 2757 | llvm::Type *ArgType = ArgValue->getType(); |
| 2758 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
| 2759 | |
| 2760 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2761 | Value *Tmp = Builder.CreateCall(F, ArgValue); |
| 2762 | Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1)); |
| 2763 | if (Result->getType() != ResultType) |
| 2764 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2765 | "cast"); |
| 2766 | return RValue::get(Result); |
| 2767 | } |
| 2768 | case Builtin::BI__lzcnt16: |
| 2769 | case Builtin::BI__lzcnt: |
| 2770 | case Builtin::BI__lzcnt64: { |
| 2771 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2772 | |
| 2773 | llvm::Type *ArgType = ArgValue->getType(); |
| 2774 | Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType); |
| 2775 | |
| 2776 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2777 | Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()}); |
| 2778 | if (Result->getType() != ResultType) |
| 2779 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2780 | "cast"); |
| 2781 | return RValue::get(Result); |
| 2782 | } |
| 2783 | case Builtin::BI__popcnt16: |
| 2784 | case Builtin::BI__popcnt: |
| 2785 | case Builtin::BI__popcnt64: |
| 2786 | case Builtin::BI__builtin_popcount: |
| 2787 | case Builtin::BI__builtin_popcountl: |
| 2788 | case Builtin::BI__builtin_popcountll: { |
| 2789 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2790 | |
| 2791 | llvm::Type *ArgType = ArgValue->getType(); |
| 2792 | Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType); |
| 2793 | |
| 2794 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2795 | Value *Result = Builder.CreateCall(F, ArgValue); |
| 2796 | if (Result->getType() != ResultType) |
| 2797 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 2798 | "cast"); |
| 2799 | return RValue::get(Result); |
| 2800 | } |
| 2801 | case Builtin::BI__builtin_unpredictable: { |
| 2802 | // Always return the argument of __builtin_unpredictable. LLVM does not |
| 2803 | // handle this builtin. Metadata for this builtin should be added directly |
| 2804 | // to instructions such as branches or switches that use it. |
| 2805 | return RValue::get(EmitScalarExpr(E->getArg(0))); |
| 2806 | } |
| 2807 | case Builtin::BI__builtin_expect: { |
| 2808 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2809 | llvm::Type *ArgType = ArgValue->getType(); |
| 2810 | |
| 2811 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); |
| 2812 | // Don't generate llvm.expect on -O0 as the backend won't use it for |
| 2813 | // anything. |
| 2814 | // Note, we still IRGen ExpectedValue because it could have side-effects. |
| 2815 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
| 2816 | return RValue::get(ArgValue); |
| 2817 | |
| 2818 | Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType); |
| 2819 | Value *Result = |
| 2820 | Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval"); |
| 2821 | return RValue::get(Result); |
| 2822 | } |
| 2823 | case Builtin::BI__builtin_expect_with_probability: { |
| 2824 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2825 | llvm::Type *ArgType = ArgValue->getType(); |
| 2826 | |
| 2827 | Value *ExpectedValue = EmitScalarExpr(E->getArg(1)); |
| 2828 | llvm::APFloat Probability(0.0); |
| 2829 | const Expr *ProbArg = E->getArg(2); |
| 2830 | bool EvalSucceed = ProbArg->EvaluateAsFloat(Probability, CGM.getContext()); |
| 2831 | assert(EvalSucceed && "probability should be able to evaluate as float")((void)0); |
| 2832 | (void)EvalSucceed; |
| 2833 | bool LoseInfo = false; |
| 2834 | Probability.convert(llvm::APFloat::IEEEdouble(), |
| 2835 | llvm::RoundingMode::Dynamic, &LoseInfo); |
| 2836 | llvm::Type *Ty = ConvertType(ProbArg->getType()); |
| 2837 | Constant *Confidence = ConstantFP::get(Ty, Probability); |
| 2838 | // Don't generate llvm.expect.with.probability on -O0 as the backend |
| 2839 | // won't use it for anything. |
| 2840 | // Note, we still IRGen ExpectedValue because it could have side-effects. |
| 2841 | if (CGM.getCodeGenOpts().OptimizationLevel == 0) |
| 2842 | return RValue::get(ArgValue); |
| 2843 | |
| 2844 | Function *FnExpect = |
| 2845 | CGM.getIntrinsic(Intrinsic::expect_with_probability, ArgType); |
| 2846 | Value *Result = Builder.CreateCall( |
| 2847 | FnExpect, {ArgValue, ExpectedValue, Confidence}, "expval"); |
| 2848 | return RValue::get(Result); |
| 2849 | } |
| 2850 | case Builtin::BI__builtin_assume_aligned: { |
| 2851 | const Expr *Ptr = E->getArg(0); |
| 2852 | Value *PtrValue = EmitScalarExpr(Ptr); |
| 2853 | Value *OffsetValue = |
| 2854 | (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr; |
| 2855 | |
| 2856 | Value *AlignmentValue = EmitScalarExpr(E->getArg(1)); |
| 2857 | ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue); |
| 2858 | if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment)) |
| 2859 | AlignmentCI = ConstantInt::get(AlignmentCI->getType(), |
| 2860 | llvm::Value::MaximumAlignment); |
| 2861 | |
| 2862 | emitAlignmentAssumption(PtrValue, Ptr, |
| 2863 | /*The expr loc is sufficient.*/ SourceLocation(), |
| 2864 | AlignmentCI, OffsetValue); |
| 2865 | return RValue::get(PtrValue); |
| 2866 | } |
| 2867 | case Builtin::BI__assume: |
| 2868 | case Builtin::BI__builtin_assume: { |
| 2869 | if (E->getArg(0)->HasSideEffects(getContext())) |
| 2870 | return RValue::get(nullptr); |
| 2871 | |
| 2872 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2873 | Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume); |
| 2874 | return RValue::get(Builder.CreateCall(FnAssume, ArgValue)); |
| 2875 | } |
| 2876 | case Builtin::BI__arithmetic_fence: { |
| 2877 | // Create the builtin call if FastMath is selected, and the target |
| 2878 | // supports the builtin, otherwise just return the argument. |
| 2879 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 2880 | llvm::FastMathFlags FMF = Builder.getFastMathFlags(); |
| 2881 | bool isArithmeticFenceEnabled = |
| 2882 | FMF.allowReassoc() && |
| 2883 | getContext().getTargetInfo().checkArithmeticFenceSupported(); |
| 2884 | QualType ArgType = E->getArg(0)->getType(); |
| 2885 | if (ArgType->isComplexType()) { |
| 2886 | if (isArithmeticFenceEnabled) { |
| 2887 | QualType ElementType = ArgType->castAs<ComplexType>()->getElementType(); |
| 2888 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2889 | Value *Real = Builder.CreateArithmeticFence(ComplexVal.first, |
| 2890 | ConvertType(ElementType)); |
| 2891 | Value *Imag = Builder.CreateArithmeticFence(ComplexVal.second, |
| 2892 | ConvertType(ElementType)); |
| 2893 | return RValue::getComplex(std::make_pair(Real, Imag)); |
| 2894 | } |
| 2895 | ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0)); |
| 2896 | Value *Real = ComplexVal.first; |
| 2897 | Value *Imag = ComplexVal.second; |
| 2898 | return RValue::getComplex(std::make_pair(Real, Imag)); |
| 2899 | } |
| 2900 | Value *ArgValue = EmitScalarExpr(E->getArg(0)); |
| 2901 | if (isArithmeticFenceEnabled) |
| 2902 | return RValue::get( |
| 2903 | Builder.CreateArithmeticFence(ArgValue, ConvertType(ArgType))); |
| 2904 | return RValue::get(ArgValue); |
| 2905 | } |
| 2906 | case Builtin::BI__builtin_bswap16: |
| 2907 | case Builtin::BI__builtin_bswap32: |
| 2908 | case Builtin::BI__builtin_bswap64: { |
| 2909 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap)); |
| 2910 | } |
| 2911 | case Builtin::BI__builtin_bitreverse8: |
| 2912 | case Builtin::BI__builtin_bitreverse16: |
| 2913 | case Builtin::BI__builtin_bitreverse32: |
| 2914 | case Builtin::BI__builtin_bitreverse64: { |
| 2915 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse)); |
| 2916 | } |
| 2917 | case Builtin::BI__builtin_rotateleft8: |
| 2918 | case Builtin::BI__builtin_rotateleft16: |
| 2919 | case Builtin::BI__builtin_rotateleft32: |
| 2920 | case Builtin::BI__builtin_rotateleft64: |
| 2921 | case Builtin::BI_rotl8: // Microsoft variants of rotate left |
| 2922 | case Builtin::BI_rotl16: |
| 2923 | case Builtin::BI_rotl: |
| 2924 | case Builtin::BI_lrotl: |
| 2925 | case Builtin::BI_rotl64: |
| 2926 | return emitRotate(E, false); |
| 2927 | |
| 2928 | case Builtin::BI__builtin_rotateright8: |
| 2929 | case Builtin::BI__builtin_rotateright16: |
| 2930 | case Builtin::BI__builtin_rotateright32: |
| 2931 | case Builtin::BI__builtin_rotateright64: |
| 2932 | case Builtin::BI_rotr8: // Microsoft variants of rotate right |
| 2933 | case Builtin::BI_rotr16: |
| 2934 | case Builtin::BI_rotr: |
| 2935 | case Builtin::BI_lrotr: |
| 2936 | case Builtin::BI_rotr64: |
| 2937 | return emitRotate(E, true); |
| 2938 | |
| 2939 | case Builtin::BI__builtin_constant_p: { |
| 2940 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 2941 | |
| 2942 | const Expr *Arg = E->getArg(0); |
| 2943 | QualType ArgType = Arg->getType(); |
| 2944 | // FIXME: The allowance for Obj-C pointers and block pointers is historical |
| 2945 | // and likely a mistake. |
| 2946 | if (!ArgType->isIntegralOrEnumerationType() && !ArgType->isFloatingType() && |
| 2947 | !ArgType->isObjCObjectPointerType() && !ArgType->isBlockPointerType()) |
| 2948 | // Per the GCC documentation, only numeric constants are recognized after |
| 2949 | // inlining. |
| 2950 | return RValue::get(ConstantInt::get(ResultType, 0)); |
| 2951 | |
| 2952 | if (Arg->HasSideEffects(getContext())) |
| 2953 | // The argument is unevaluated, so be conservative if it might have |
| 2954 | // side-effects. |
| 2955 | return RValue::get(ConstantInt::get(ResultType, 0)); |
| 2956 | |
| 2957 | Value *ArgValue = EmitScalarExpr(Arg); |
| 2958 | if (ArgType->isObjCObjectPointerType()) { |
| 2959 | // Convert Objective-C objects to id because we cannot distinguish between |
| 2960 | // LLVM types for Obj-C classes as they are opaque. |
| 2961 | ArgType = CGM.getContext().getObjCIdType(); |
| 2962 | ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType)); |
| 2963 | } |
| 2964 | Function *F = |
| 2965 | CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType)); |
| 2966 | Value *Result = Builder.CreateCall(F, ArgValue); |
| 2967 | if (Result->getType() != ResultType) |
| 2968 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false); |
| 2969 | return RValue::get(Result); |
| 2970 | } |
| 2971 | case Builtin::BI__builtin_dynamic_object_size: |
| 2972 | case Builtin::BI__builtin_object_size: { |
| 2973 | unsigned Type = |
| 2974 | E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
| 2975 | auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType())); |
| 2976 | |
| 2977 | // We pass this builtin onto the optimizer so that it can figure out the |
| 2978 | // object size in more complex cases. |
| 2979 | bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size; |
| 2980 | return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType, |
| 2981 | /*EmittedE=*/nullptr, IsDynamic)); |
| 2982 | } |
| 2983 | case Builtin::BI__builtin_prefetch: { |
| 2984 | Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0)); |
| 2985 | // FIXME: Technically these constants should of type 'int', yes? |
| 2986 | RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) : |
| 2987 | llvm::ConstantInt::get(Int32Ty, 0); |
| 2988 | Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : |
| 2989 | llvm::ConstantInt::get(Int32Ty, 3); |
| 2990 | Value *Data = llvm::ConstantInt::get(Int32Ty, 1); |
| 2991 | Function *F = CGM.getIntrinsic(Intrinsic::prefetch, Address->getType()); |
| 2992 | return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data})); |
| 2993 | } |
| 2994 | case Builtin::BI__builtin_readcyclecounter: { |
| 2995 | Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter); |
| 2996 | return RValue::get(Builder.CreateCall(F)); |
| 2997 | } |
| 2998 | case Builtin::BI__builtin___clear_cache: { |
| 2999 | Value *Begin = EmitScalarExpr(E->getArg(0)); |
| 3000 | Value *End = EmitScalarExpr(E->getArg(1)); |
| 3001 | Function *F = CGM.getIntrinsic(Intrinsic::clear_cache); |
| 3002 | return RValue::get(Builder.CreateCall(F, {Begin, End})); |
| 3003 | } |
| 3004 | case Builtin::BI__builtin_trap: |
| 3005 | return RValue::get(EmitTrapCall(Intrinsic::trap)); |
| 3006 | case Builtin::BI__debugbreak: |
| 3007 | return RValue::get(EmitTrapCall(Intrinsic::debugtrap)); |
| 3008 | case Builtin::BI__builtin_unreachable: { |
| 3009 | EmitUnreachable(E->getExprLoc()); |
| 3010 | |
| 3011 | // We do need to preserve an insertion point. |
| 3012 | EmitBlock(createBasicBlock("unreachable.cont")); |
| 3013 | |
| 3014 | return RValue::get(nullptr); |
| 3015 | } |
| 3016 | |
| 3017 | case Builtin::BI__builtin_powi: |
| 3018 | case Builtin::BI__builtin_powif: |
| 3019 | case Builtin::BI__builtin_powil: { |
| 3020 | llvm::Value *Src0 = EmitScalarExpr(E->getArg(0)); |
| 3021 | llvm::Value *Src1 = EmitScalarExpr(E->getArg(1)); |
| 3022 | |
| 3023 | if (Builder.getIsFPConstrained()) { |
| 3024 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3025 | Function *F = CGM.getIntrinsic(Intrinsic::experimental_constrained_powi, |
| 3026 | Src0->getType()); |
| 3027 | return RValue::get(Builder.CreateConstrainedFPCall(F, { Src0, Src1 })); |
| 3028 | } |
| 3029 | |
| 3030 | Function *F = CGM.getIntrinsic(Intrinsic::powi, |
| 3031 | { Src0->getType(), Src1->getType() }); |
| 3032 | return RValue::get(Builder.CreateCall(F, { Src0, Src1 })); |
| 3033 | } |
| 3034 | case Builtin::BI__builtin_isgreater: |
| 3035 | case Builtin::BI__builtin_isgreaterequal: |
| 3036 | case Builtin::BI__builtin_isless: |
| 3037 | case Builtin::BI__builtin_islessequal: |
| 3038 | case Builtin::BI__builtin_islessgreater: |
| 3039 | case Builtin::BI__builtin_isunordered: { |
| 3040 | // Ordered comparisons: we know the arguments to these are matching scalar |
| 3041 | // floating point values. |
| 3042 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3043 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
| 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")__builtin_unreachable(); |
| 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_matrix_transpose: { |
| 3105 | const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>(); |
| 3106 | Value *MatValue = EmitScalarExpr(E->getArg(0)); |
| 3107 | MatrixBuilder<CGBuilderTy> MB(Builder); |
| 3108 | Value *Result = MB.CreateMatrixTranspose(MatValue, MatrixTy->getNumRows(), |
| 3109 | MatrixTy->getNumColumns()); |
| 3110 | return RValue::get(Result); |
| 3111 | } |
| 3112 | |
| 3113 | case Builtin::BI__builtin_matrix_column_major_load: { |
| 3114 | MatrixBuilder<CGBuilderTy> MB(Builder); |
| 3115 | // Emit everything that isn't dependent on the first parameter type |
| 3116 | Value *Stride = EmitScalarExpr(E->getArg(3)); |
| 3117 | const auto *ResultTy = E->getType()->getAs<ConstantMatrixType>(); |
| 3118 | auto *PtrTy = E->getArg(0)->getType()->getAs<PointerType>(); |
| 3119 | assert(PtrTy && "arg0 must be of pointer type")((void)0); |
| 3120 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); |
| 3121 | |
| 3122 | Address Src = EmitPointerWithAlignment(E->getArg(0)); |
| 3123 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(0)->getType(), |
| 3124 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3125 | Value *Result = MB.CreateColumnMajorLoad( |
| 3126 | Src.getPointer(), Align(Src.getAlignment().getQuantity()), Stride, |
| 3127 | IsVolatile, ResultTy->getNumRows(), ResultTy->getNumColumns(), |
| 3128 | "matrix"); |
| 3129 | return RValue::get(Result); |
| 3130 | } |
| 3131 | |
| 3132 | case Builtin::BI__builtin_matrix_column_major_store: { |
| 3133 | MatrixBuilder<CGBuilderTy> MB(Builder); |
| 3134 | Value *Matrix = EmitScalarExpr(E->getArg(0)); |
| 3135 | Address Dst = EmitPointerWithAlignment(E->getArg(1)); |
| 3136 | Value *Stride = EmitScalarExpr(E->getArg(2)); |
| 3137 | |
| 3138 | const auto *MatrixTy = E->getArg(0)->getType()->getAs<ConstantMatrixType>(); |
| 3139 | auto *PtrTy = E->getArg(1)->getType()->getAs<PointerType>(); |
| 3140 | assert(PtrTy && "arg1 must be of pointer type")((void)0); |
| 3141 | bool IsVolatile = PtrTy->getPointeeType().isVolatileQualified(); |
| 3142 | |
| 3143 | EmitNonNullArgCheck(RValue::get(Dst.getPointer()), E->getArg(1)->getType(), |
| 3144 | E->getArg(1)->getExprLoc(), FD, 0); |
| 3145 | Value *Result = MB.CreateColumnMajorStore( |
| 3146 | Matrix, Dst.getPointer(), Align(Dst.getAlignment().getQuantity()), |
| 3147 | Stride, IsVolatile, MatrixTy->getNumRows(), MatrixTy->getNumColumns()); |
| 3148 | return RValue::get(Result); |
| 3149 | } |
| 3150 | |
| 3151 | case Builtin::BIfinite: |
| 3152 | case Builtin::BI__finite: |
| 3153 | case Builtin::BIfinitef: |
| 3154 | case Builtin::BI__finitef: |
| 3155 | case Builtin::BIfinitel: |
| 3156 | case Builtin::BI__finitel: |
| 3157 | case Builtin::BI__builtin_isinf: |
| 3158 | case Builtin::BI__builtin_isfinite: { |
| 3159 | // isinf(x) --> fabs(x) == infinity |
| 3160 | // isfinite(x) --> fabs(x) != infinity |
| 3161 | // x != NaN via the ordered compare in either case. |
| 3162 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3163 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 3164 | llvm::Type *Ty = V->getType(); |
| 3165 | if (!Builder.getIsFPConstrained() || |
| 3166 | Builder.getDefaultConstrainedExcept() == fp::ebIgnore || |
| 3167 | !Ty->isIEEE()) { |
| 3168 | Value *Fabs = EmitFAbs(*this, V); |
| 3169 | Constant *Infinity = ConstantFP::getInfinity(V->getType()); |
| 3170 | CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf) |
| 3171 | ? CmpInst::FCMP_OEQ |
| 3172 | : CmpInst::FCMP_ONE; |
| 3173 | Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf"); |
| 3174 | return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType()))); |
| 3175 | } |
| 3176 | |
| 3177 | if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM)) |
| 3178 | return RValue::get(Result); |
| 3179 | |
| 3180 | // Inf values have all exp bits set and a zero significand. Therefore: |
| 3181 | // isinf(V) == ((V << 1) == ((exp mask) << 1)) |
| 3182 | // isfinite(V) == ((V << 1) < ((exp mask) << 1)) using unsigned comparison |
| 3183 | unsigned bitsize = Ty->getScalarSizeInBits(); |
| 3184 | llvm::IntegerType *IntTy = Builder.getIntNTy(bitsize); |
| 3185 | Value *IntV = Builder.CreateBitCast(V, IntTy); |
| 3186 | Value *Shl1 = Builder.CreateShl(IntV, 1); |
| 3187 | const llvm::fltSemantics &Semantics = Ty->getFltSemantics(); |
| 3188 | APInt ExpMask = APFloat::getInf(Semantics).bitcastToAPInt(); |
| 3189 | Value *ExpMaskShl1 = llvm::ConstantInt::get(IntTy, ExpMask.shl(1)); |
| 3190 | if (BuiltinID == Builtin::BI__builtin_isinf) |
| 3191 | V = Builder.CreateICmpEQ(Shl1, ExpMaskShl1); |
| 3192 | else |
| 3193 | V = Builder.CreateICmpULT(Shl1, ExpMaskShl1); |
| 3194 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
| 3195 | } |
| 3196 | |
| 3197 | case Builtin::BI__builtin_isinf_sign: { |
| 3198 | // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0 |
| 3199 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3200 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
| 3201 | Value *Arg = EmitScalarExpr(E->getArg(0)); |
| 3202 | Value *AbsArg = EmitFAbs(*this, Arg); |
| 3203 | Value *IsInf = Builder.CreateFCmpOEQ( |
| 3204 | AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf"); |
| 3205 | Value *IsNeg = EmitSignBit(*this, Arg); |
| 3206 | |
| 3207 | llvm::Type *IntTy = ConvertType(E->getType()); |
| 3208 | Value *Zero = Constant::getNullValue(IntTy); |
| 3209 | Value *One = ConstantInt::get(IntTy, 1); |
| 3210 | Value *NegativeOne = ConstantInt::get(IntTy, -1); |
| 3211 | Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One); |
| 3212 | Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero); |
| 3213 | return RValue::get(Result); |
| 3214 | } |
| 3215 | |
| 3216 | case Builtin::BI__builtin_isnormal: { |
| 3217 | // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min |
| 3218 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3219 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
| 3220 | Value *V = EmitScalarExpr(E->getArg(0)); |
| 3221 | Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq"); |
| 3222 | |
| 3223 | Value *Abs = EmitFAbs(*this, V); |
| 3224 | Value *IsLessThanInf = |
| 3225 | Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf"); |
| 3226 | APFloat Smallest = APFloat::getSmallestNormalized( |
| 3227 | getContext().getFloatTypeSemantics(E->getArg(0)->getType())); |
| 3228 | Value *IsNormal = |
| 3229 | Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest), |
| 3230 | "isnormal"); |
| 3231 | V = Builder.CreateAnd(Eq, IsLessThanInf, "and"); |
| 3232 | V = Builder.CreateAnd(V, IsNormal, "and"); |
| 3233 | return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType()))); |
| 3234 | } |
| 3235 | |
| 3236 | case Builtin::BI__builtin_flt_rounds: { |
| 3237 | Function *F = CGM.getIntrinsic(Intrinsic::flt_rounds); |
| 3238 | |
| 3239 | llvm::Type *ResultType = ConvertType(E->getType()); |
| 3240 | Value *Result = Builder.CreateCall(F); |
| 3241 | if (Result->getType() != ResultType) |
| 3242 | Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true, |
| 3243 | "cast"); |
| 3244 | return RValue::get(Result); |
| 3245 | } |
| 3246 | |
| 3247 | case Builtin::BI__builtin_fpclassify: { |
| 3248 | CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E); |
| 3249 | // FIXME: for strictfp/IEEE-754 we need to not trap on SNaN here. |
| 3250 | Value *V = EmitScalarExpr(E->getArg(5)); |
| 3251 | llvm::Type *Ty = ConvertType(E->getArg(5)->getType()); |
| 3252 | |
| 3253 | // Create Result |
| 3254 | BasicBlock *Begin = Builder.GetInsertBlock(); |
| 3255 | BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn); |
| 3256 | Builder.SetInsertPoint(End); |
| 3257 | PHINode *Result = |
| 3258 | Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4, |
| 3259 | "fpclassify_result"); |
| 3260 | |
| 3261 | // if (V==0) return FP_ZERO |
| 3262 | Builder.SetInsertPoint(Begin); |
| 3263 | Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty), |
| 3264 | "iszero"); |
| 3265 | Value *ZeroLiteral = EmitScalarExpr(E->getArg(4)); |
| 3266 | BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn); |
| 3267 | Builder.CreateCondBr(IsZero, End, NotZero); |
| 3268 | Result->addIncoming(ZeroLiteral, Begin); |
| 3269 | |
| 3270 | // if (V != V) return FP_NAN |
| 3271 | Builder.SetInsertPoint(NotZero); |
| 3272 | Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp"); |
| 3273 | Value *NanLiteral = EmitScalarExpr(E->getArg(0)); |
| 3274 | BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn); |
| 3275 | Builder.CreateCondBr(IsNan, End, NotNan); |
| 3276 | Result->addIncoming(NanLiteral, NotZero); |
| 3277 | |
| 3278 | // if (fabs(V) == infinity) return FP_INFINITY |
| 3279 | Builder.SetInsertPoint(NotNan); |
| 3280 | Value *VAbs = EmitFAbs(*this, V); |
| 3281 | Value *IsInf = |
| 3282 | Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()), |
| 3283 | "isinf"); |
| 3284 | Value *InfLiteral = EmitScalarExpr(E->getArg(1)); |
| 3285 | BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn); |
| 3286 | Builder.CreateCondBr(IsInf, End, NotInf); |
| 3287 | Result->addIncoming(InfLiteral, NotNan); |
| 3288 | |
| 3289 | // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL |
| 3290 | Builder.SetInsertPoint(NotInf); |
| 3291 | APFloat Smallest = APFloat::getSmallestNormalized( |
| 3292 | getContext().getFloatTypeSemantics(E->getArg(5)->getType())); |
| 3293 | Value *IsNormal = |
| 3294 | Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest), |
| 3295 | "isnormal"); |
| 3296 | Value *NormalResult = |
| 3297 | Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)), |
| 3298 | EmitScalarExpr(E->getArg(3))); |
| 3299 | Builder.CreateBr(End); |
| 3300 | Result->addIncoming(NormalResult, NotInf); |
| 3301 | |
| 3302 | // return Result |
| 3303 | Builder.SetInsertPoint(End); |
| 3304 | return RValue::get(Result); |
| 3305 | } |
| 3306 | |
| 3307 | case Builtin::BIalloca: |
| 3308 | case Builtin::BI_alloca: |
| 3309 | case Builtin::BI__builtin_alloca: { |
| 3310 | Value *Size = EmitScalarExpr(E->getArg(0)); |
| 3311 | const TargetInfo &TI = getContext().getTargetInfo(); |
| 3312 | // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__. |
| 3313 | const Align SuitableAlignmentInBytes = |
| 3314 | CGM.getContext() |
| 3315 | .toCharUnitsFromBits(TI.getSuitableAlign()) |
| 3316 | .getAsAlign(); |
| 3317 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
| 3318 | AI->setAlignment(SuitableAlignmentInBytes); |
| 3319 | initializeAlloca(*this, AI, Size, SuitableAlignmentInBytes); |
| 3320 | return RValue::get(AI); |
| 3321 | } |
| 3322 | |
| 3323 | case Builtin::BI__builtin_alloca_with_align: { |
| 3324 | Value *Size = EmitScalarExpr(E->getArg(0)); |
| 3325 | Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1)); |
| 3326 | auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue); |
| 3327 | unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue(); |
| 3328 | const Align AlignmentInBytes = |
| 3329 | CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getAsAlign(); |
| 3330 | AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size); |
| 3331 | AI->setAlignment(AlignmentInBytes); |
| 3332 | initializeAlloca(*this, AI, Size, AlignmentInBytes); |
| 3333 | return RValue::get(AI); |
| 3334 | } |
| 3335 | |
| 3336 | case Builtin::BIbzero: |
| 3337 | case Builtin::BI__builtin_bzero: { |
| 3338 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3339 | Value *SizeVal = EmitScalarExpr(E->getArg(1)); |
| 3340 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3341 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3342 | Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false); |
| 3343 | return RValue::get(nullptr); |
| 3344 | } |
| 3345 | case Builtin::BImemcpy: |
| 3346 | case Builtin::BI__builtin_memcpy: |
| 3347 | case Builtin::BImempcpy: |
| 3348 | case Builtin::BI__builtin_mempcpy: { |
| 3349 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3350 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3351 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3352 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3353 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3354 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 3355 | E->getArg(1)->getExprLoc(), FD, 1); |
| 3356 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
| 3357 | if (BuiltinID == Builtin::BImempcpy || |
| 3358 | BuiltinID == Builtin::BI__builtin_mempcpy) |
| 3359 | return RValue::get(Builder.CreateInBoundsGEP(Dest.getElementType(), |
| 3360 | Dest.getPointer(), SizeVal)); |
| 3361 | else |
| 3362 | return RValue::get(Dest.getPointer()); |
| 3363 | } |
| 3364 | |
| 3365 | case Builtin::BI__builtin_memcpy_inline: { |
| 3366 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3367 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3368 | uint64_t Size = |
| 3369 | E->getArg(2)->EvaluateKnownConstInt(getContext()).getZExtValue(); |
| 3370 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3371 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3372 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 3373 | E->getArg(1)->getExprLoc(), FD, 1); |
| 3374 | Builder.CreateMemCpyInline(Dest, Src, Size); |
| 3375 | return RValue::get(nullptr); |
| 3376 | } |
| 3377 | |
| 3378 | case Builtin::BI__builtin_char_memchr: |
| 3379 | BuiltinID = Builtin::BI__builtin_memchr; |
| 3380 | break; |
| 3381 | |
| 3382 | case Builtin::BI__builtin___memcpy_chk: { |
| 3383 | // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2. |
| 3384 | Expr::EvalResult SizeResult, DstSizeResult; |
| 3385 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 3386 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 3387 | break; |
| 3388 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 3389 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 3390 | if (Size.ugt(DstSize)) |
| 3391 | break; |
| 3392 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3393 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3394 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 3395 | Builder.CreateMemCpy(Dest, Src, SizeVal, false); |
| 3396 | return RValue::get(Dest.getPointer()); |
| 3397 | } |
| 3398 | |
| 3399 | case Builtin::BI__builtin_objc_memmove_collectable: { |
| 3400 | Address DestAddr = EmitPointerWithAlignment(E->getArg(0)); |
| 3401 | Address SrcAddr = EmitPointerWithAlignment(E->getArg(1)); |
| 3402 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3403 | CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, |
| 3404 | DestAddr, SrcAddr, SizeVal); |
| 3405 | return RValue::get(DestAddr.getPointer()); |
| 3406 | } |
| 3407 | |
| 3408 | case Builtin::BI__builtin___memmove_chk: { |
| 3409 | // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2. |
| 3410 | Expr::EvalResult SizeResult, DstSizeResult; |
| 3411 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 3412 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 3413 | break; |
| 3414 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 3415 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 3416 | if (Size.ugt(DstSize)) |
| 3417 | break; |
| 3418 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3419 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3420 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 3421 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
| 3422 | return RValue::get(Dest.getPointer()); |
| 3423 | } |
| 3424 | |
| 3425 | case Builtin::BImemmove: |
| 3426 | case Builtin::BI__builtin_memmove: { |
| 3427 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3428 | Address Src = EmitPointerWithAlignment(E->getArg(1)); |
| 3429 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3430 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3431 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3432 | EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(), |
| 3433 | E->getArg(1)->getExprLoc(), FD, 1); |
| 3434 | Builder.CreateMemMove(Dest, Src, SizeVal, false); |
| 3435 | return RValue::get(Dest.getPointer()); |
| 3436 | } |
| 3437 | case Builtin::BImemset: |
| 3438 | case Builtin::BI__builtin_memset: { |
| 3439 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3440 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
| 3441 | Builder.getInt8Ty()); |
| 3442 | Value *SizeVal = EmitScalarExpr(E->getArg(2)); |
| 3443 | EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(), |
| 3444 | E->getArg(0)->getExprLoc(), FD, 0); |
| 3445 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
| 3446 | return RValue::get(Dest.getPointer()); |
| 3447 | } |
| 3448 | case Builtin::BI__builtin___memset_chk: { |
| 3449 | // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2. |
| 3450 | Expr::EvalResult SizeResult, DstSizeResult; |
| 3451 | if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) || |
| 3452 | !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext())) |
| 3453 | break; |
| 3454 | llvm::APSInt Size = SizeResult.Val.getInt(); |
| 3455 | llvm::APSInt DstSize = DstSizeResult.Val.getInt(); |
| 3456 | if (Size.ugt(DstSize)) |
| 3457 | break; |
| 3458 | Address Dest = EmitPointerWithAlignment(E->getArg(0)); |
| 3459 | Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)), |
| 3460 | Builder.getInt8Ty()); |
| 3461 | Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size); |
| 3462 | Builder.CreateMemSet(Dest, ByteVal, SizeVal, false); |
| 3463 | return RValue::get(Dest.getPointer()); |
| 3464 | } |
| 3465 | case Builtin::BI__builtin_wmemchr: { |
| 3466 | // The MSVC runtime library does not provide a definition of wmemchr, so we |
| 3467 | // need an inline implementation. |
| 3468 | if (!getTarget().getTriple().isOSMSVCRT()) |
| 3469 | break; |
| 3470 | |
| 3471 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); |
| 3472 | Value *Str = EmitScalarExpr(E->getArg(0)); |
| 3473 | Value *Chr = EmitScalarExpr(E->getArg(1)); |
| 3474 | Value *Size = EmitScalarExpr(E->getArg(2)); |
| 3475 | |
| 3476 | BasicBlock *Entry = Builder.GetInsertBlock(); |
| 3477 | BasicBlock *CmpEq = createBasicBlock("wmemchr.eq"); |
| 3478 | BasicBlock *Next = createBasicBlock("wmemchr.next"); |
| 3479 | BasicBlock *Exit = createBasicBlock("wmemchr.exit"); |
| 3480 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); |
| 3481 | Builder.CreateCondBr(SizeEq0, Exit, CmpEq); |
| 3482 | |
| 3483 | EmitBlock(CmpEq); |
| 3484 | PHINode *StrPhi = Builder.CreatePHI(Str->getType(), 2); |
| 3485 | StrPhi->addIncoming(Str, Entry); |
| 3486 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); |
| 3487 | SizePhi->addIncoming(Size, Entry); |
| 3488 | CharUnits WCharAlign = |
| 3489 | getContext().getTypeAlignInChars(getContext().WCharTy); |
| 3490 | Value *StrCh = Builder.CreateAlignedLoad(WCharTy, StrPhi, WCharAlign); |
| 3491 | Value *FoundChr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 0); |
| 3492 | Value *StrEqChr = Builder.CreateICmpEQ(StrCh, Chr); |
| 3493 | Builder.CreateCondBr(StrEqChr, Exit, Next); |
| 3494 | |
| 3495 | EmitBlock(Next); |
| 3496 | Value *NextStr = Builder.CreateConstInBoundsGEP1_32(WCharTy, StrPhi, 1); |
| 3497 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); |
| 3498 | Value *NextSizeEq0 = |
| 3499 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); |
| 3500 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpEq); |
| 3501 | StrPhi->addIncoming(NextStr, Next); |
| 3502 | SizePhi->addIncoming(NextSize, Next); |
| 3503 | |
| 3504 | EmitBlock(Exit); |
| 3505 | PHINode *Ret = Builder.CreatePHI(Str->getType(), 3); |
| 3506 | Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Entry); |
| 3507 | Ret->addIncoming(llvm::Constant::getNullValue(Str->getType()), Next); |
| 3508 | Ret->addIncoming(FoundChr, CmpEq); |
| 3509 | return RValue::get(Ret); |
| 3510 | } |
| 3511 | case Builtin::BI__builtin_wmemcmp: { |
| 3512 | // The MSVC runtime library does not provide a definition of wmemcmp, so we |
| 3513 | // need an inline implementation. |
| 3514 | if (!getTarget().getTriple().isOSMSVCRT()) |
| 3515 | break; |
| 3516 | |
| 3517 | llvm::Type *WCharTy = ConvertType(getContext().WCharTy); |
| 3518 | |
| 3519 | Value *Dst = EmitScalarExpr(E->getArg(0)); |
| 3520 | Value *Src = EmitScalarExpr(E->getArg(1)); |
| 3521 | Value *Size = EmitScalarExpr(E->getArg(2)); |
| 3522 | |
| 3523 | BasicBlock *Entry = Builder.GetInsertBlock(); |
| 3524 | BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt"); |
| 3525 | BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt"); |
| 3526 | BasicBlock *Next = createBasicBlock("wmemcmp.next"); |
| 3527 | BasicBlock *Exit = createBasicBlock("wmemcmp.exit"); |
| 3528 | Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0)); |
| 3529 | Builder.CreateCondBr(SizeEq0, Exit, CmpGT); |
| 3530 | |
| 3531 | EmitBlock(CmpGT); |
| 3532 | PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2); |
| 3533 | DstPhi->addIncoming(Dst, Entry); |
| 3534 | PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2); |
| 3535 | SrcPhi->addIncoming(Src, Entry); |
| 3536 | PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2); |
| 3537 | SizePhi->addIncoming(Size, Entry); |
| 3538 | CharUnits WCharAlign = |
| 3539 | getContext().getTypeAlignInChars(getContext().WCharTy); |
| 3540 | Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign); |
| 3541 | Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign); |
| 3542 | Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh); |
| 3543 | Builder.CreateCondBr(DstGtSrc, Exit, CmpLT); |
| 3544 | |
| 3545 | EmitBlock(CmpLT); |
| 3546 | Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh); |
| 3547 | Builder.CreateCondBr(DstLtSrc, Exit, Next); |
| 3548 | |
| 3549 | EmitBlock(Next); |
| 3550 | Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1); |
| 3551 | Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1); |
| 3552 | Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1)); |
| 3553 | Value *NextSizeEq0 = |
| 3554 | Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0)); |
| 3555 | Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT); |
| 3556 | DstPhi->addIncoming(NextDst, Next); |
| 3557 | SrcPhi->addIncoming(NextSrc, Next); |
| 3558 | SizePhi->addIncoming(NextSize, Next); |
| 3559 | |
| 3560 | EmitBlock(Exit); |
| 3561 | PHINode *Ret = Builder.CreatePHI(IntTy, 4); |
| 3562 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry); |
| 3563 | Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT); |
| 3564 | Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT); |
| 3565 | Ret->addIncoming(ConstantInt::get(IntTy, 0), Next); |
| 3566 | return RValue::get(Ret); |
| 3567 | } |
| 3568 | case Builtin::BI__builtin_dwarf_cfa: { |
| 3569 | // The offset in bytes from the first argument to the CFA. |
| 3570 | // |
| 3571 | // Why on earth is this in the frontend? Is there any reason at |
| 3572 | // all that the backend can't reasonably determine this while |
| 3573 | // lowering llvm.eh.dwarf.cfa()? |
| 3574 | // |
| 3575 | // TODO: If there's a satisfactory reason, add a target hook for |
| 3576 | // this instead of hard-coding 0, which is correct for most targets. |
| 3577 | int32_t Offset = 0; |
| 3578 | |
| 3579 | Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa); |
| 3580 | return RValue::get(Builder.CreateCall(F, |
| 3581 | llvm::ConstantInt::get(Int32Ty, Offset))); |
| 3582 | } |
| 3583 | case Builtin::BI__builtin_return_address: { |
| 3584 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
| 3585 | getContext().UnsignedIntTy); |
| 3586 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
| 3587 | return RValue::get(Builder.CreateCall(F, Depth)); |
| 3588 | } |
| 3589 | case Builtin::BI_ReturnAddress: { |
| 3590 | Function *F = CGM.getIntrinsic(Intrinsic::returnaddress); |
| 3591 | return RValue::get(Builder.CreateCall(F, Builder.getInt32(0))); |
| 3592 | } |
| 3593 | case Builtin::BI__builtin_frame_address: { |
| 3594 | Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0), |
| 3595 | getContext().UnsignedIntTy); |
| 3596 | Function *F = CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy); |
| 3597 | return RValue::get(Builder.CreateCall(F, Depth)); |
| 3598 | } |
| 3599 | case Builtin::BI__builtin_extract_return_addr: { |
| 3600 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 3601 | Value *Result = getTargetHooks().decodeReturnAddress(*this, Address); |
| 3602 | return RValue::get(Result); |
| 3603 | } |
| 3604 | case Builtin::BI__builtin_frob_return_addr: { |
| 3605 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 3606 | Value *Result = getTargetHooks().encodeReturnAddress(*this, Address); |
| 3607 | return RValue::get(Result); |
| 3608 | } |
| 3609 | case Builtin::BI__builtin_dwarf_sp_column: { |
| 3610 | llvm::IntegerType *Ty |
| 3611 | = cast<llvm::IntegerType>(ConvertType(E->getType())); |
| 3612 | int Column = getTargetHooks().getDwarfEHStackPointer(CGM); |
| 3613 | if (Column == -1) { |
| 3614 | CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column"); |
| 3615 | return RValue::get(llvm::UndefValue::get(Ty)); |
| 3616 | } |
| 3617 | return RValue::get(llvm::ConstantInt::get(Ty, Column, true)); |
| 3618 | } |
| 3619 | case Builtin::BI__builtin_init_dwarf_reg_size_table: { |
| 3620 | Value *Address = EmitScalarExpr(E->getArg(0)); |
| 3621 | if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address)) |
| 3622 | CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table"); |
| 3623 | return RValue::get(llvm::UndefValue::get(ConvertType(E->getType()))); |
| 3624 | } |
| 3625 | case Builtin::BI__builtin_eh_return: { |
| 3626 | Value *Int = EmitScalarExpr(E->getArg(0)); |
| 3627 | Value *Ptr = EmitScalarExpr(E->getArg(1)); |
| 3628 | |
| 3629 | llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType()); |
| 3630 | assert((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) &&((void)0) |
| 3631 | "LLVM's __builtin_eh_return only supports 32- and 64-bit variants")((void)0); |
| 3632 | Function *F = |
| 3633 | CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32 |
| 3634 | : Intrinsic::eh_return_i64); |
| 3635 | Builder.CreateCall(F, {Int, Ptr}); |
| 3636 | Builder.CreateUnreachable(); |
| 3637 | |
| 3638 | // We do need to preserve an insertion point. |
| 3639 | EmitBlock(createBasicBlock("builtin_eh_return.cont")); |
| 3640 | |
| 3641 | return RValue::get(nullptr); |
| 3642 | } |
| 3643 | case Builtin::BI__builtin_unwind_init: { |
| 3644 | Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init); |
| 3645 | return RValue::get(Builder.CreateCall(F)); |
| 3646 | } |
| 3647 | case Builtin::BI__builtin_extend_pointer: { |
| 3648 | // Extends a pointer to the size of an _Unwind_Word, which is |
| 3649 | // uint64_t on all platforms. Generally this gets poked into a |
| 3650 | // register and eventually used as an address, so if the |
| 3651 | // addressing registers are wider than pointers and the platform |
| 3652 | // doesn't implicitly ignore high-order bits when doing |
| 3653 | // addressing, we need to make sure we zext / sext based on |
| 3654 | // the platform's expectations. |
| 3655 | // |
| 3656 | // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html |
| 3657 | |
| 3658 | // Cast the pointer to intptr_t. |
| 3659 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 3660 | Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast"); |
| 3661 | |
| 3662 | // If that's 64 bits, we're done. |
| 3663 | if (IntPtrTy->getBitWidth() == 64) |
| 3664 | return RValue::get(Result); |
| 3665 | |
| 3666 | // Otherwise, ask the codegen data what to do. |
| 3667 | if (getTargetHooks().extendPointerWithSExt()) |
| 3668 | return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext")); |
| 3669 | else |
| 3670 | return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext")); |
| 3671 | } |
| 3672 | case Builtin::BI__builtin_setjmp: { |
| 3673 | // Buffer is a void**. |
| 3674 | Address Buf = EmitPointerWithAlignment(E->getArg(0)); |
| 3675 | |
| 3676 | // Store the frame pointer to the setjmp buffer. |
| 3677 | Value *FrameAddr = Builder.CreateCall( |
| 3678 | CGM.getIntrinsic(Intrinsic::frameaddress, AllocaInt8PtrTy), |
| 3679 | ConstantInt::get(Int32Ty, 0)); |
| 3680 | Builder.CreateStore(FrameAddr, Buf); |
| 3681 | |
| 3682 | // Store the stack pointer to the setjmp buffer. |
| 3683 | Value *StackAddr = |
| 3684 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave)); |
| 3685 | Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2); |
| 3686 | Builder.CreateStore(StackAddr, StackSaveSlot); |
| 3687 | |
| 3688 | // Call LLVM's EH setjmp, which is lightweight. |
| 3689 | Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp); |
| 3690 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); |
| 3691 | return RValue::get(Builder.CreateCall(F, Buf.getPointer())); |
| 3692 | } |
| 3693 | case Builtin::BI__builtin_longjmp: { |
| 3694 | Value *Buf = EmitScalarExpr(E->getArg(0)); |
| 3695 | Buf = Builder.CreateBitCast(Buf, Int8PtrTy); |
| 3696 | |
| 3697 | // Call LLVM's EH longjmp, which is lightweight. |
| 3698 | Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf); |
| 3699 | |
| 3700 | // longjmp doesn't return; mark this as unreachable. |
| 3701 | Builder.CreateUnreachable(); |
| 3702 | |
| 3703 | // We do need to preserve an insertion point. |
| 3704 | EmitBlock(createBasicBlock("longjmp.cont")); |
| 3705 | |
| 3706 | return RValue::get(nullptr); |
| 3707 | } |
| 3708 | case Builtin::BI__builtin_launder: { |
| 3709 | const Expr *Arg = E->getArg(0); |
| 3710 | QualType ArgTy = Arg->getType()->getPointeeType(); |
| 3711 | Value *Ptr = EmitScalarExpr(Arg); |
| 3712 | if (TypeRequiresBuiltinLaunder(CGM, ArgTy)) |
| 3713 | Ptr = Builder.CreateLaunderInvariantGroup(Ptr); |
| 3714 | |
| 3715 | return RValue::get(Ptr); |
| 3716 | } |
| 3717 | case Builtin::BI__sync_fetch_and_add: |
| 3718 | case Builtin::BI__sync_fetch_and_sub: |
| 3719 | case Builtin::BI__sync_fetch_and_or: |
| 3720 | case Builtin::BI__sync_fetch_and_and: |
| 3721 | case Builtin::BI__sync_fetch_and_xor: |
| 3722 | case Builtin::BI__sync_fetch_and_nand: |
| 3723 | case Builtin::BI__sync_add_and_fetch: |
| 3724 | case Builtin::BI__sync_sub_and_fetch: |
| 3725 | case Builtin::BI__sync_and_and_fetch: |
| 3726 | case Builtin::BI__sync_or_and_fetch: |
| 3727 | case Builtin::BI__sync_xor_and_fetch: |
| 3728 | case Builtin::BI__sync_nand_and_fetch: |
| 3729 | case Builtin::BI__sync_val_compare_and_swap: |
| 3730 | case Builtin::BI__sync_bool_compare_and_swap: |
| 3731 | case Builtin::BI__sync_lock_test_and_set: |
| 3732 | case Builtin::BI__sync_lock_release: |
| 3733 | case Builtin::BI__sync_swap: |
| 3734 | llvm_unreachable("Shouldn't make it through sema")__builtin_unreachable(); |
| 3735 | case Builtin::BI__sync_fetch_and_add_1: |
| 3736 | case Builtin::BI__sync_fetch_and_add_2: |
| 3737 | case Builtin::BI__sync_fetch_and_add_4: |
| 3738 | case Builtin::BI__sync_fetch_and_add_8: |
| 3739 | case Builtin::BI__sync_fetch_and_add_16: |
| 3740 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E); |
| 3741 | case Builtin::BI__sync_fetch_and_sub_1: |
| 3742 | case Builtin::BI__sync_fetch_and_sub_2: |
| 3743 | case Builtin::BI__sync_fetch_and_sub_4: |
| 3744 | case Builtin::BI__sync_fetch_and_sub_8: |
| 3745 | case Builtin::BI__sync_fetch_and_sub_16: |
| 3746 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E); |
| 3747 | case Builtin::BI__sync_fetch_and_or_1: |
| 3748 | case Builtin::BI__sync_fetch_and_or_2: |
| 3749 | case Builtin::BI__sync_fetch_and_or_4: |
| 3750 | case Builtin::BI__sync_fetch_and_or_8: |
| 3751 | case Builtin::BI__sync_fetch_and_or_16: |
| 3752 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E); |
| 3753 | case Builtin::BI__sync_fetch_and_and_1: |
| 3754 | case Builtin::BI__sync_fetch_and_and_2: |
| 3755 | case Builtin::BI__sync_fetch_and_and_4: |
| 3756 | case Builtin::BI__sync_fetch_and_and_8: |
| 3757 | case Builtin::BI__sync_fetch_and_and_16: |
| 3758 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E); |
| 3759 | case Builtin::BI__sync_fetch_and_xor_1: |
| 3760 | case Builtin::BI__sync_fetch_and_xor_2: |
| 3761 | case Builtin::BI__sync_fetch_and_xor_4: |
| 3762 | case Builtin::BI__sync_fetch_and_xor_8: |
| 3763 | case Builtin::BI__sync_fetch_and_xor_16: |
| 3764 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E); |
| 3765 | case Builtin::BI__sync_fetch_and_nand_1: |
| 3766 | case Builtin::BI__sync_fetch_and_nand_2: |
| 3767 | case Builtin::BI__sync_fetch_and_nand_4: |
| 3768 | case Builtin::BI__sync_fetch_and_nand_8: |
| 3769 | case Builtin::BI__sync_fetch_and_nand_16: |
| 3770 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E); |
| 3771 | |
| 3772 | // Clang extensions: not overloaded yet. |
| 3773 | case Builtin::BI__sync_fetch_and_min: |
| 3774 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E); |
| 3775 | case Builtin::BI__sync_fetch_and_max: |
| 3776 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E); |
| 3777 | case Builtin::BI__sync_fetch_and_umin: |
| 3778 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E); |
| 3779 | case Builtin::BI__sync_fetch_and_umax: |
| 3780 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E); |
| 3781 | |
| 3782 | case Builtin::BI__sync_add_and_fetch_1: |
| 3783 | case Builtin::BI__sync_add_and_fetch_2: |
| 3784 | case Builtin::BI__sync_add_and_fetch_4: |
| 3785 | case Builtin::BI__sync_add_and_fetch_8: |
| 3786 | case Builtin::BI__sync_add_and_fetch_16: |
| 3787 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E, |
| 3788 | llvm::Instruction::Add); |
| 3789 | case Builtin::BI__sync_sub_and_fetch_1: |
| 3790 | case Builtin::BI__sync_sub_and_fetch_2: |
| 3791 | case Builtin::BI__sync_sub_and_fetch_4: |
| 3792 | case Builtin::BI__sync_sub_and_fetch_8: |
| 3793 | case Builtin::BI__sync_sub_and_fetch_16: |
| 3794 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E, |
| 3795 | llvm::Instruction::Sub); |
| 3796 | case Builtin::BI__sync_and_and_fetch_1: |
| 3797 | case Builtin::BI__sync_and_and_fetch_2: |
| 3798 | case Builtin::BI__sync_and_and_fetch_4: |
| 3799 | case Builtin::BI__sync_and_and_fetch_8: |
| 3800 | case Builtin::BI__sync_and_and_fetch_16: |
| 3801 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E, |
| 3802 | llvm::Instruction::And); |
| 3803 | case Builtin::BI__sync_or_and_fetch_1: |
| 3804 | case Builtin::BI__sync_or_and_fetch_2: |
| 3805 | case Builtin::BI__sync_or_and_fetch_4: |
| 3806 | case Builtin::BI__sync_or_and_fetch_8: |
| 3807 | case Builtin::BI__sync_or_and_fetch_16: |
| 3808 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E, |
| 3809 | llvm::Instruction::Or); |
| 3810 | case Builtin::BI__sync_xor_and_fetch_1: |
| 3811 | case Builtin::BI__sync_xor_and_fetch_2: |
| 3812 | case Builtin::BI__sync_xor_and_fetch_4: |
| 3813 | case Builtin::BI__sync_xor_and_fetch_8: |
| 3814 | case Builtin::BI__sync_xor_and_fetch_16: |
| 3815 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E, |
| 3816 | llvm::Instruction::Xor); |
| 3817 | case Builtin::BI__sync_nand_and_fetch_1: |
| 3818 | case Builtin::BI__sync_nand_and_fetch_2: |
| 3819 | case Builtin::BI__sync_nand_and_fetch_4: |
| 3820 | case Builtin::BI__sync_nand_and_fetch_8: |
| 3821 | case Builtin::BI__sync_nand_and_fetch_16: |
| 3822 | return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E, |
| 3823 | llvm::Instruction::And, true); |
| 3824 | |
| 3825 | case Builtin::BI__sync_val_compare_and_swap_1: |
| 3826 | case Builtin::BI__sync_val_compare_and_swap_2: |
| 3827 | case Builtin::BI__sync_val_compare_and_swap_4: |
| 3828 | case Builtin::BI__sync_val_compare_and_swap_8: |
| 3829 | case Builtin::BI__sync_val_compare_and_swap_16: |
| 3830 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, false)); |
| 3831 | |
| 3832 | case Builtin::BI__sync_bool_compare_and_swap_1: |
| 3833 | case Builtin::BI__sync_bool_compare_and_swap_2: |
| 3834 | case Builtin::BI__sync_bool_compare_and_swap_4: |
| 3835 | case Builtin::BI__sync_bool_compare_and_swap_8: |
| 3836 | case Builtin::BI__sync_bool_compare_and_swap_16: |
| 3837 | return RValue::get(MakeAtomicCmpXchgValue(*this, E, true)); |
| 3838 | |
| 3839 | case Builtin::BI__sync_swap_1: |
| 3840 | case Builtin::BI__sync_swap_2: |
| 3841 | case Builtin::BI__sync_swap_4: |
| 3842 | case Builtin::BI__sync_swap_8: |
| 3843 | case Builtin::BI__sync_swap_16: |
| 3844 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
| 3845 | |
| 3846 | case Builtin::BI__sync_lock_test_and_set_1: |
| 3847 | case Builtin::BI__sync_lock_test_and_set_2: |
| 3848 | case Builtin::BI__sync_lock_test_and_set_4: |
| 3849 | case Builtin::BI__sync_lock_test_and_set_8: |
| 3850 | case Builtin::BI__sync_lock_test_and_set_16: |
| 3851 | return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E); |
| 3852 | |
| 3853 | case Builtin::BI__sync_lock_release_1: |
| 3854 | case Builtin::BI__sync_lock_release_2: |
| 3855 | case Builtin::BI__sync_lock_release_4: |
| 3856 | case Builtin::BI__sync_lock_release_8: |
| 3857 | case Builtin::BI__sync_lock_release_16: { |
| 3858 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 3859 | QualType ElTy = E->getArg(0)->getType()->getPointeeType(); |
| 3860 | CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy); |
| 3861 | llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(), |
| 3862 | StoreSize.getQuantity() * 8); |
| 3863 | Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo()); |
| 3864 | llvm::StoreInst *Store = |
| 3865 | Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr, |
| 3866 | StoreSize); |
| 3867 | Store->setAtomic(llvm::AtomicOrdering::Release); |
| 3868 | return RValue::get(nullptr); |
| 3869 | } |
| 3870 | |
| 3871 | case Builtin::BI__sync_synchronize: { |
| 3872 | // We assume this is supposed to correspond to a C++0x-style |
| 3873 | // sequentially-consistent fence (i.e. this is only usable for |
| 3874 | // synchronization, not device I/O or anything like that). This intrinsic |
| 3875 | // is really badly designed in the sense that in theory, there isn't |
| 3876 | // any way to safely use it... but in practice, it mostly works |
| 3877 | // to use it with non-atomic loads and stores to get acquire/release |
| 3878 | // semantics. |
| 3879 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent); |
| 3880 | return RValue::get(nullptr); |
| 3881 | } |
| 3882 | |
| 3883 | case Builtin::BI__builtin_nontemporal_load: |
| 3884 | return RValue::get(EmitNontemporalLoad(*this, E)); |
| 3885 | case Builtin::BI__builtin_nontemporal_store: |
| 3886 | return RValue::get(EmitNontemporalStore(*this, E)); |
| 3887 | case Builtin::BI__c11_atomic_is_lock_free: |
| 3888 | case Builtin::BI__atomic_is_lock_free: { |
| 3889 | // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the |
| 3890 | // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since |
| 3891 | // _Atomic(T) is always properly-aligned. |
| 3892 | const char *LibCallName = "__atomic_is_lock_free"; |
| 3893 | CallArgList Args; |
| 3894 | Args.add(RValue::get(EmitScalarExpr(E->getArg(0))), |
| 3895 | getContext().getSizeType()); |
| 3896 | if (BuiltinID == Builtin::BI__atomic_is_lock_free) |
| 3897 | Args.add(RValue::get(EmitScalarExpr(E->getArg(1))), |
| 3898 | getContext().VoidPtrTy); |
| 3899 | else |
| 3900 | Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)), |
| 3901 | getContext().VoidPtrTy); |
| 3902 | const CGFunctionInfo &FuncInfo = |
| 3903 | CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args); |
| 3904 | llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo); |
| 3905 | llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName); |
| 3906 | return EmitCall(FuncInfo, CGCallee::forDirect(Func), |
| 3907 | ReturnValueSlot(), Args); |
| 3908 | } |
| 3909 | |
| 3910 | case Builtin::BI__atomic_test_and_set: { |
| 3911 | // Look at the argument type to determine whether this is a volatile |
| 3912 | // operation. The parameter type is always volatile. |
| 3913 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
| 3914 | bool Volatile = |
| 3915 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
| 3916 | |
| 3917 | Value *Ptr = EmitScalarExpr(E->getArg(0)); |
| 3918 | unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace(); |
| 3919 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); |
| 3920 | Value *NewVal = Builder.getInt8(1); |
| 3921 | Value *Order = EmitScalarExpr(E->getArg(1)); |
| 3922 | if (isa<llvm::ConstantInt>(Order)) { |
| 3923 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 3924 | AtomicRMWInst *Result = nullptr; |
| 3925 | switch (ord) { |
| 3926 | case 0: // memory_order_relaxed |
| 3927 | default: // invalid order |
| 3928 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 3929 | llvm::AtomicOrdering::Monotonic); |
| 3930 | break; |
| 3931 | case 1: // memory_order_consume |
| 3932 | case 2: // memory_order_acquire |
| 3933 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 3934 | llvm::AtomicOrdering::Acquire); |
| 3935 | break; |
| 3936 | case 3: // memory_order_release |
| 3937 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 3938 | llvm::AtomicOrdering::Release); |
| 3939 | break; |
| 3940 | case 4: // memory_order_acq_rel |
| 3941 | |
| 3942 | Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 3943 | llvm::AtomicOrdering::AcquireRelease); |
| 3944 | break; |
| 3945 | case 5: // memory_order_seq_cst |
| 3946 | Result = Builder.CreateAtomicRMW( |
| 3947 | llvm::AtomicRMWInst::Xchg, Ptr, NewVal, |
| 3948 | llvm::AtomicOrdering::SequentiallyConsistent); |
| 3949 | break; |
| 3950 | } |
| 3951 | Result->setVolatile(Volatile); |
| 3952 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
| 3953 | } |
| 3954 | |
| 3955 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 3956 | |
| 3957 | llvm::BasicBlock *BBs[5] = { |
| 3958 | createBasicBlock("monotonic", CurFn), |
| 3959 | createBasicBlock("acquire", CurFn), |
| 3960 | createBasicBlock("release", CurFn), |
| 3961 | createBasicBlock("acqrel", CurFn), |
| 3962 | createBasicBlock("seqcst", CurFn) |
| 3963 | }; |
| 3964 | llvm::AtomicOrdering Orders[5] = { |
| 3965 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire, |
| 3966 | llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease, |
| 3967 | llvm::AtomicOrdering::SequentiallyConsistent}; |
| 3968 | |
| 3969 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 3970 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
| 3971 | |
| 3972 | Builder.SetInsertPoint(ContBB); |
| 3973 | PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set"); |
| 3974 | |
| 3975 | for (unsigned i = 0; i < 5; ++i) { |
| 3976 | Builder.SetInsertPoint(BBs[i]); |
| 3977 | AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, |
| 3978 | Ptr, NewVal, Orders[i]); |
| 3979 | RMW->setVolatile(Volatile); |
| 3980 | Result->addIncoming(RMW, BBs[i]); |
| 3981 | Builder.CreateBr(ContBB); |
| 3982 | } |
| 3983 | |
| 3984 | SI->addCase(Builder.getInt32(0), BBs[0]); |
| 3985 | SI->addCase(Builder.getInt32(1), BBs[1]); |
| 3986 | SI->addCase(Builder.getInt32(2), BBs[1]); |
| 3987 | SI->addCase(Builder.getInt32(3), BBs[2]); |
| 3988 | SI->addCase(Builder.getInt32(4), BBs[3]); |
| 3989 | SI->addCase(Builder.getInt32(5), BBs[4]); |
| 3990 | |
| 3991 | Builder.SetInsertPoint(ContBB); |
| 3992 | return RValue::get(Builder.CreateIsNotNull(Result, "tobool")); |
| 3993 | } |
| 3994 | |
| 3995 | case Builtin::BI__atomic_clear: { |
| 3996 | QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType(); |
| 3997 | bool Volatile = |
| 3998 | PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified(); |
| 3999 | |
| 4000 | Address Ptr = EmitPointerWithAlignment(E->getArg(0)); |
| 4001 | unsigned AddrSpace = Ptr.getPointer()->getType()->getPointerAddressSpace(); |
| 4002 | Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace)); |
| 4003 | Value *NewVal = Builder.getInt8(0); |
| 4004 | Value *Order = EmitScalarExpr(E->getArg(1)); |
| 4005 | if (isa<llvm::ConstantInt>(Order)) { |
| 4006 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 4007 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
| 4008 | switch (ord) { |
| 4009 | case 0: // memory_order_relaxed |
| 4010 | default: // invalid order |
| 4011 | Store->setOrdering(llvm::AtomicOrdering::Monotonic); |
| 4012 | break; |
| 4013 | case 3: // memory_order_release |
| 4014 | Store->setOrdering(llvm::AtomicOrdering::Release); |
| 4015 | break; |
| 4016 | case 5: // memory_order_seq_cst |
| 4017 | Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent); |
| 4018 | break; |
| 4019 | } |
| 4020 | return RValue::get(nullptr); |
| 4021 | } |
| 4022 | |
| 4023 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 4024 | |
| 4025 | llvm::BasicBlock *BBs[3] = { |
| 4026 | createBasicBlock("monotonic", CurFn), |
| 4027 | createBasicBlock("release", CurFn), |
| 4028 | createBasicBlock("seqcst", CurFn) |
| 4029 | }; |
| 4030 | llvm::AtomicOrdering Orders[3] = { |
| 4031 | llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release, |
| 4032 | llvm::AtomicOrdering::SequentiallyConsistent}; |
| 4033 | |
| 4034 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 4035 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]); |
| 4036 | |
| 4037 | for (unsigned i = 0; i < 3; ++i) { |
| 4038 | Builder.SetInsertPoint(BBs[i]); |
| 4039 | StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile); |
| 4040 | Store->setOrdering(Orders[i]); |
| 4041 | Builder.CreateBr(ContBB); |
| 4042 | } |
| 4043 | |
| 4044 | SI->addCase(Builder.getInt32(0), BBs[0]); |
| 4045 | SI->addCase(Builder.getInt32(3), BBs[1]); |
| 4046 | SI->addCase(Builder.getInt32(5), BBs[2]); |
| 4047 | |
| 4048 | Builder.SetInsertPoint(ContBB); |
| 4049 | return RValue::get(nullptr); |
| 4050 | } |
| 4051 | |
| 4052 | case Builtin::BI__atomic_thread_fence: |
| 4053 | case Builtin::BI__atomic_signal_fence: |
| 4054 | case Builtin::BI__c11_atomic_thread_fence: |
| 4055 | case Builtin::BI__c11_atomic_signal_fence: { |
| 4056 | llvm::SyncScope::ID SSID; |
| 4057 | if (BuiltinID == Builtin::BI__atomic_signal_fence || |
| 4058 | BuiltinID == Builtin::BI__c11_atomic_signal_fence) |
| 4059 | SSID = llvm::SyncScope::SingleThread; |
| 4060 | else |
| 4061 | SSID = llvm::SyncScope::System; |
| 4062 | Value *Order = EmitScalarExpr(E->getArg(0)); |
| 4063 | if (isa<llvm::ConstantInt>(Order)) { |
| 4064 | int ord = cast<llvm::ConstantInt>(Order)->getZExtValue(); |
| 4065 | switch (ord) { |
| 4066 | case 0: // memory_order_relaxed |
| 4067 | default: // invalid order |
| 4068 | break; |
| 4069 | case 1: // memory_order_consume |
| 4070 | case 2: // memory_order_acquire |
| 4071 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
| 4072 | break; |
| 4073 | case 3: // memory_order_release |
| 4074 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
| 4075 | break; |
| 4076 | case 4: // memory_order_acq_rel |
| 4077 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
| 4078 | break; |
| 4079 | case 5: // memory_order_seq_cst |
| 4080 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
| 4081 | break; |
| 4082 | } |
| 4083 | return RValue::get(nullptr); |
| 4084 | } |
| 4085 | |
| 4086 | llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB; |
| 4087 | AcquireBB = createBasicBlock("acquire", CurFn); |
| 4088 | ReleaseBB = createBasicBlock("release", CurFn); |
| 4089 | AcqRelBB = createBasicBlock("acqrel", CurFn); |
| 4090 | SeqCstBB = createBasicBlock("seqcst", CurFn); |
| 4091 | llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn); |
| 4092 | |
| 4093 | Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false); |
| 4094 | llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB); |
| 4095 | |
| 4096 | Builder.SetInsertPoint(AcquireBB); |
| 4097 | Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID); |
| 4098 | Builder.CreateBr(ContBB); |
| 4099 | SI->addCase(Builder.getInt32(1), AcquireBB); |
| 4100 | SI->addCase(Builder.getInt32(2), AcquireBB); |
| 4101 | |
| 4102 | Builder.SetInsertPoint(ReleaseBB); |
| 4103 | Builder.CreateFence(llvm::AtomicOrdering::Release, SSID); |
| 4104 | Builder.CreateBr(ContBB); |
| 4105 | SI->addCase(Builder.getInt32(3), ReleaseBB); |
| 4106 | |
| 4107 | Builder.SetInsertPoint(AcqRelBB); |
| 4108 | Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID); |
| 4109 | Builder.CreateBr(ContBB); |
| 4110 | SI->addCase(Builder.getInt32(4), AcqRelBB); |
| 4111 | |
| 4112 | Builder.SetInsertPoint(SeqCstBB); |
| 4113 | Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID); |
| 4114 | Builder.CreateBr(ContBB); |
| 4115 | SI->addCase(Builder.getInt32(5), SeqCstBB); |
| 4116 | |
| 4117 | Builder.SetInsertPoint(ContBB); |
| 4118 | return RValue::get(nullptr); |
| 4119 | } |
| 4120 | |
| 4121 | case Builtin::BI__builtin_signbit: |
| 4122 | case Builtin::BI__builtin_signbitf: |
| 4123 | case Builtin::BI__builtin_signbitl: { |
| 4124 | return RValue::get( |
| 4125 | Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))), |
| 4126 | ConvertType(E->getType()))); |
| 4127 | } |
| 4128 | case Builtin::BI__warn_memset_zero_len: |
| 4129 | return RValue::getIgnored(); |
| 4130 | case Builtin::BI__annotation: { |
| 4131 | // Re-encode each wide string to UTF8 and make an MDString. |
| 4132 | SmallVector<Metadata *, 1> Strings; |
| 4133 | for (const Expr *Arg : E->arguments()) { |
| 4134 | const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts()); |
| 4135 | assert(Str->getCharByteWidth() == 2)((void)0); |
| 4136 | StringRef WideBytes = Str->getBytes(); |
| 4137 | std::string StrUtf8; |
| 4138 | if (!convertUTF16ToUTF8String( |
| 4139 | makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) { |
| 4140 | CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument"); |
| 4141 | continue; |
| 4142 | } |
| 4143 | Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8)); |
| 4144 | } |
| 4145 | |
| 4146 | // Build and MDTuple of MDStrings and emit the intrinsic call. |
| 4147 | llvm::Function *F = |
| 4148 | CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {}); |
| 4149 | MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings); |
| 4150 | Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple)); |
| 4151 | return RValue::getIgnored(); |
| 4152 | } |
| 4153 | case Builtin::BI__builtin_annotation: { |
| 4154 | llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0)); |
| 4155 | llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::annotation, |
| 4156 | AnnVal->getType()); |
| 4157 | |
| 4158 | // Get the annotation string, go through casts. Sema requires this to be a |
| 4159 | // non-wide string literal, potentially casted, so the cast<> is safe. |
| 4160 | const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts(); |
| 4161 | StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString(); |
| 4162 | return RValue::get( |
| 4163 | EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc(), nullptr)); |
| 4164 | } |
| 4165 | case Builtin::BI__builtin_addcb: |
| 4166 | case Builtin::BI__builtin_addcs: |
| 4167 | case Builtin::BI__builtin_addc: |
| 4168 | case Builtin::BI__builtin_addcl: |
| 4169 | case Builtin::BI__builtin_addcll: |
| 4170 | case Builtin::BI__builtin_subcb: |
| 4171 | case Builtin::BI__builtin_subcs: |
| 4172 | case Builtin::BI__builtin_subc: |
| 4173 | case Builtin::BI__builtin_subcl: |
| 4174 | case Builtin::BI__builtin_subcll: { |
| 4175 | |
| 4176 | // We translate all of these builtins from expressions of the form: |
| 4177 | // int x = ..., y = ..., carryin = ..., carryout, result; |
| 4178 | // result = __builtin_addc(x, y, carryin, &carryout); |
| 4179 | // |
| 4180 | // to LLVM IR of the form: |
| 4181 | // |
| 4182 | // %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y) |
| 4183 | // %tmpsum1 = extractvalue {i32, i1} %tmp1, 0 |
| 4184 | // %carry1 = extractvalue {i32, i1} %tmp1, 1 |
| 4185 | // %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1, |
| 4186 | // i32 %carryin) |
| 4187 | // %result = extractvalue {i32, i1} %tmp2, 0 |
| 4188 | // %carry2 = extractvalue {i32, i1} %tmp2, 1 |
| 4189 | // %tmp3 = or i1 %carry1, %carry2 |
| 4190 | // %tmp4 = zext i1 %tmp3 to i32 |
| 4191 | // store i32 %tmp4, i32* %carryout |
| 4192 | |
| 4193 | // Scalarize our inputs. |
| 4194 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
| 4195 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
| 4196 | llvm::Value *Carryin = EmitScalarExpr(E->getArg(2)); |
| 4197 | Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3)); |
| 4198 | |
| 4199 | // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow. |
| 4200 | llvm::Intrinsic::ID IntrinsicId; |
| 4201 | switch (BuiltinID) { |
| 4202 | default: llvm_unreachable("Unknown multiprecision builtin id.")__builtin_unreachable(); |
| 4203 | case Builtin::BI__builtin_addcb: |
| 4204 | case Builtin::BI__builtin_addcs: |
| 4205 | case Builtin::BI__builtin_addc: |
| 4206 | case Builtin::BI__builtin_addcl: |
| 4207 | case Builtin::BI__builtin_addcll: |
| 4208 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
| 4209 | break; |
| 4210 | case Builtin::BI__builtin_subcb: |
| 4211 | case Builtin::BI__builtin_subcs: |
| 4212 | case Builtin::BI__builtin_subc: |
| 4213 | case Builtin::BI__builtin_subcl: |
| 4214 | case Builtin::BI__builtin_subcll: |
| 4215 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
| 4216 | break; |
| 4217 | } |
| 4218 | |
| 4219 | // Construct our resulting LLVM IR expression. |
| 4220 | llvm::Value *Carry1; |
| 4221 | llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId, |
| 4222 | X, Y, Carry1); |
| 4223 | llvm::Value *Carry2; |
| 4224 | llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId, |
| 4225 | Sum1, Carryin, Carry2); |
| 4226 | llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2), |
| 4227 | X->getType()); |
| 4228 | Builder.CreateStore(CarryOut, CarryOutPtr); |
| 4229 | return RValue::get(Sum2); |
| 4230 | } |
| 4231 | |
| 4232 | case Builtin::BI__builtin_add_overflow: |
| 4233 | case Builtin::BI__builtin_sub_overflow: |
| 4234 | case Builtin::BI__builtin_mul_overflow: { |
| 4235 | const clang::Expr *LeftArg = E->getArg(0); |
| 4236 | const clang::Expr *RightArg = E->getArg(1); |
| 4237 | const clang::Expr *ResultArg = E->getArg(2); |
| 4238 | |
| 4239 | clang::QualType ResultQTy = |
| 4240 | ResultArg->getType()->castAs<PointerType>()->getPointeeType(); |
| 4241 | |
| 4242 | WidthAndSignedness LeftInfo = |
| 4243 | getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType()); |
| 4244 | WidthAndSignedness RightInfo = |
| 4245 | getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType()); |
| 4246 | WidthAndSignedness ResultInfo = |
| 4247 | getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy); |
| 4248 | |
| 4249 | // Handle mixed-sign multiplication as a special case, because adding |
| 4250 | // runtime or backend support for our generic irgen would be too expensive. |
| 4251 | if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo)) |
| 4252 | return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg, |
| 4253 | RightInfo, ResultArg, ResultQTy, |
| 4254 | ResultInfo); |
| 4255 | |
| 4256 | if (isSpecialUnsignedMultiplySignedResult(BuiltinID, LeftInfo, RightInfo, |
| 4257 | ResultInfo)) |
| 4258 | return EmitCheckedUnsignedMultiplySignedResult( |
| 4259 | *this, LeftArg, LeftInfo, RightArg, RightInfo, ResultArg, ResultQTy, |
| 4260 | ResultInfo); |
| 4261 | |
| 4262 | WidthAndSignedness EncompassingInfo = |
| 4263 | EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo}); |
| 4264 | |
| 4265 | llvm::Type *EncompassingLLVMTy = |
| 4266 | llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width); |
| 4267 | |
| 4268 | llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy); |
| 4269 | |
| 4270 | llvm::Intrinsic::ID IntrinsicId; |
| 4271 | switch (BuiltinID) { |
| 4272 | default: |
| 4273 | llvm_unreachable("Unknown overflow builtin id.")__builtin_unreachable(); |
| 4274 | case Builtin::BI__builtin_add_overflow: |
| 4275 | IntrinsicId = EncompassingInfo.Signed |
| 4276 | ? llvm::Intrinsic::sadd_with_overflow |
| 4277 | : llvm::Intrinsic::uadd_with_overflow; |
| 4278 | break; |
| 4279 | case Builtin::BI__builtin_sub_overflow: |
| 4280 | IntrinsicId = EncompassingInfo.Signed |
| 4281 | ? llvm::Intrinsic::ssub_with_overflow |
| 4282 | : llvm::Intrinsic::usub_with_overflow; |
| 4283 | break; |
| 4284 | case Builtin::BI__builtin_mul_overflow: |
| 4285 | IntrinsicId = EncompassingInfo.Signed |
| 4286 | ? llvm::Intrinsic::smul_with_overflow |
| 4287 | : llvm::Intrinsic::umul_with_overflow; |
| 4288 | break; |
| 4289 | } |
| 4290 | |
| 4291 | llvm::Value *Left = EmitScalarExpr(LeftArg); |
| 4292 | llvm::Value *Right = EmitScalarExpr(RightArg); |
| 4293 | Address ResultPtr = EmitPointerWithAlignment(ResultArg); |
| 4294 | |
| 4295 | // Extend each operand to the encompassing type. |
| 4296 | Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed); |
| 4297 | Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed); |
| 4298 | |
| 4299 | // Perform the operation on the extended values. |
| 4300 | llvm::Value *Overflow, *Result; |
| 4301 | Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow); |
| 4302 | |
| 4303 | if (EncompassingInfo.Width > ResultInfo.Width) { |
| 4304 | // The encompassing type is wider than the result type, so we need to |
| 4305 | // truncate it. |
| 4306 | llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy); |
| 4307 | |
| 4308 | // To see if the truncation caused an overflow, we will extend |
| 4309 | // the result and then compare it to the original result. |
| 4310 | llvm::Value *ResultTruncExt = Builder.CreateIntCast( |
| 4311 | ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed); |
| 4312 | llvm::Value *TruncationOverflow = |
| 4313 | Builder.CreateICmpNE(Result, ResultTruncExt); |
| 4314 | |
| 4315 | Overflow = Builder.CreateOr(Overflow, TruncationOverflow); |
| 4316 | Result = ResultTrunc; |
| 4317 | } |
| 4318 | |
| 4319 | // Finally, store the result using the pointer. |
| 4320 | bool isVolatile = |
| 4321 | ResultArg->getType()->getPointeeType().isVolatileQualified(); |
| 4322 | Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile); |
| 4323 | |
| 4324 | return RValue::get(Overflow); |
| 4325 | } |
| 4326 | |
| 4327 | case Builtin::BI__builtin_uadd_overflow: |
| 4328 | case Builtin::BI__builtin_uaddl_overflow: |
| 4329 | case Builtin::BI__builtin_uaddll_overflow: |
| 4330 | case Builtin::BI__builtin_usub_overflow: |
| 4331 | case Builtin::BI__builtin_usubl_overflow: |
| 4332 | case Builtin::BI__builtin_usubll_overflow: |
| 4333 | case Builtin::BI__builtin_umul_overflow: |
| 4334 | case Builtin::BI__builtin_umull_overflow: |
| 4335 | case Builtin::BI__builtin_umulll_overflow: |
| 4336 | case Builtin::BI__builtin_sadd_overflow: |
| 4337 | case Builtin::BI__builtin_saddl_overflow: |
| 4338 | case Builtin::BI__builtin_saddll_overflow: |
| 4339 | case Builtin::BI__builtin_ssub_overflow: |
| 4340 | case Builtin::BI__builtin_ssubl_overflow: |
| 4341 | case Builtin::BI__builtin_ssubll_overflow: |
| 4342 | case Builtin::BI__builtin_smul_overflow: |
| 4343 | case Builtin::BI__builtin_smull_overflow: |
| 4344 | case Builtin::BI__builtin_smulll_overflow: { |
| 4345 | |
| 4346 | // We translate all of these builtins directly to the relevant llvm IR node. |
| 4347 | |
| 4348 | // Scalarize our inputs. |
| 4349 | llvm::Value *X = EmitScalarExpr(E->getArg(0)); |
| 4350 | llvm::Value *Y = EmitScalarExpr(E->getArg(1)); |
| 4351 | Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2)); |
| 4352 | |
| 4353 | // Decide which of the overflow intrinsics we are lowering to: |
| 4354 | llvm::Intrinsic::ID IntrinsicId; |
| 4355 | switch (BuiltinID) { |
| 4356 | default: llvm_unreachable("Unknown overflow builtin id.")__builtin_unreachable(); |
| 4357 | case Builtin::BI__builtin_uadd_overflow: |
| 4358 | case Builtin::BI__builtin_uaddl_overflow: |
| 4359 | case Builtin::BI__builtin_uaddll_overflow: |
| 4360 | IntrinsicId = llvm::Intrinsic::uadd_with_overflow; |
| 4361 | break; |
| 4362 | case Builtin::BI__builtin_usub_overflow: |
| 4363 | case Builtin::BI__builtin_usubl_overflow: |
| 4364 | case Builtin::BI__builtin_usubll_overflow: |
| 4365 | IntrinsicId = llvm::Intrinsic::usub_with_overflow; |
| 4366 | break; |
| 4367 | case Builtin::BI__builtin_umul_overflow: |
| 4368 | case Builtin::BI__builtin_umull_overflow: |
| 4369 | case Builtin::BI__builtin_umulll_overflow: |
| 4370 | IntrinsicId = llvm::Intrinsic::umul_with_overflow; |
| 4371 | break; |
| 4372 | case Builtin::BI__builtin_sadd_overflow: |
| 4373 | case Builtin::BI__builtin_saddl_overflow: |
| 4374 | case Builtin::BI__builtin_saddll_overflow: |
| 4375 | IntrinsicId = llvm::Intrinsic::sadd_with_overflow; |
| 4376 | break; |
| 4377 | case Builtin::BI__builtin_ssub_overflow: |
| 4378 | case Builtin::BI__builtin_ssubl_overflow: |
| 4379 | case Builtin::BI__builtin_ssubll_overflow: |
| 4380 | IntrinsicId = llvm::Intrinsic::ssub_with_overflow; |
| 4381 | break; |
| 4382 | case Builtin::BI__builtin_smul_overflow: |
| 4383 | case Builtin::BI__builtin_smull_overflow: |
| 4384 | case Builtin::BI__builtin_smulll_overflow: |
| 4385 | IntrinsicId = llvm::Intrinsic::smul_with_overflow; |
| 4386 | break; |
| 4387 | } |
| 4388 | |
| 4389 | |
| 4390 | llvm::Value *Carry; |
| 4391 | llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry); |
| 4392 | Builder.CreateStore(Sum, SumOutPtr); |
| 4393 | |
| 4394 | return RValue::get(Carry); |
| 4395 | } |
| 4396 | case Builtin::BI__builtin_addressof: |
| 4397 | return RValue::get(EmitLValue(E->getArg(0)).getPointer(*this)); |
| 4398 | case Builtin::BI__builtin_operator_new: |
| 4399 | return EmitBuiltinNewDeleteCall( |
| 4400 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false); |
| 4401 | case Builtin::BI__builtin_operator_delete: |
| 4402 | return EmitBuiltinNewDeleteCall( |
| 4403 | E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true); |
| 4404 | |
| 4405 | case Builtin::BI__builtin_is_aligned: |
| 4406 | return EmitBuiltinIsAligned(E); |
| 4407 | case Builtin::BI__builtin_align_up: |
| 4408 | return EmitBuiltinAlignTo(E, true); |
| 4409 | case Builtin::BI__builtin_align_down: |
| 4410 | return EmitBuiltinAlignTo(E, false); |
| 4411 | |
| 4412 | case Builtin::BI__noop: |
| 4413 | // __noop always evaluates to an integer literal zero. |
| 4414 | return RValue::get(ConstantInt::get(IntTy, 0)); |
| 4415 | case Builtin::BI__builtin_call_with_static_chain: { |
| 4416 | const CallExpr *Call = cast<CallExpr>(E->getArg(0)); |
| 4417 | const Expr *Chain = E->getArg(1); |
| 4418 | return EmitCall(Call->getCallee()->getType(), |
| 4419 | EmitCallee(Call->getCallee()), Call, ReturnValue, |
| 4420 | EmitScalarExpr(Chain)); |
| 4421 | } |
| 4422 | case Builtin::BI_InterlockedExchange8: |
| 4423 | case Builtin::BI_InterlockedExchange16: |
| 4424 | case Builtin::BI_InterlockedExchange: |
| 4425 | case Builtin::BI_InterlockedExchangePointer: |
| 4426 | return RValue::get( |
| 4427 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E)); |
| 4428 | case Builtin::BI_InterlockedCompareExchangePointer: |
| 4429 | case Builtin::BI_InterlockedCompareExchangePointer_nf: { |
| 4430 | llvm::Type *RTy; |
| 4431 | llvm::IntegerType *IntType = |
| 4432 | IntegerType::get(getLLVMContext(), |
| 4433 | getContext().getTypeSize(E->getType())); |
| 4434 | llvm::Type *IntPtrType = IntType->getPointerTo(); |
| 4435 | |
| 4436 | llvm::Value *Destination = |
| 4437 | Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType); |
| 4438 | |
| 4439 | llvm::Value *Exchange = EmitScalarExpr(E->getArg(1)); |
| 4440 | RTy = Exchange->getType(); |
| 4441 | Exchange = Builder.CreatePtrToInt(Exchange, IntType); |
| 4442 | |
| 4443 | llvm::Value *Comparand = |
| 4444 | Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType); |
| 4445 | |
| 4446 | auto Ordering = |
| 4447 | BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ? |
| 4448 | AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent; |
| 4449 | |
| 4450 | auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange, |
| 4451 | Ordering, Ordering); |
| 4452 | Result->setVolatile(true); |
| 4453 | |
| 4454 | return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result, |
| 4455 | 0), |
| 4456 | RTy)); |
| 4457 | } |
| 4458 | case Builtin::BI_InterlockedCompareExchange8: |
| 4459 | case Builtin::BI_InterlockedCompareExchange16: |
| 4460 | case Builtin::BI_InterlockedCompareExchange: |
| 4461 | case Builtin::BI_InterlockedCompareExchange64: |
| 4462 | return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E)); |
| 4463 | case Builtin::BI_InterlockedIncrement16: |
| 4464 | case Builtin::BI_InterlockedIncrement: |
| 4465 | return RValue::get( |
| 4466 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E)); |
| 4467 | case Builtin::BI_InterlockedDecrement16: |
| 4468 | case Builtin::BI_InterlockedDecrement: |
| 4469 | return RValue::get( |
| 4470 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E)); |
| 4471 | case Builtin::BI_InterlockedAnd8: |
| 4472 | case Builtin::BI_InterlockedAnd16: |
| 4473 | case Builtin::BI_InterlockedAnd: |
| 4474 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E)); |
| 4475 | case Builtin::BI_InterlockedExchangeAdd8: |
| 4476 | case Builtin::BI_InterlockedExchangeAdd16: |
| 4477 | case Builtin::BI_InterlockedExchangeAdd: |
| 4478 | return RValue::get( |
| 4479 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E)); |
| 4480 | case Builtin::BI_InterlockedExchangeSub8: |
| 4481 | case Builtin::BI_InterlockedExchangeSub16: |
| 4482 | case Builtin::BI_InterlockedExchangeSub: |
| 4483 | return RValue::get( |
| 4484 | EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E)); |
| 4485 | case Builtin::BI_InterlockedOr8: |
| 4486 | case Builtin::BI_InterlockedOr16: |
| 4487 | case Builtin::BI_InterlockedOr: |
| 4488 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E)); |
| 4489 | case Builtin::BI_InterlockedXor8: |
| 4490 | case Builtin::BI_InterlockedXor16: |
| 4491 | case Builtin::BI_InterlockedXor: |
| 4492 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E)); |
| 4493 | |
| 4494 | case Builtin::BI_bittest64: |
| 4495 | case Builtin::BI_bittest: |
| 4496 | case Builtin::BI_bittestandcomplement64: |
| 4497 | case Builtin::BI_bittestandcomplement: |
| 4498 | case Builtin::BI_bittestandreset64: |
| 4499 | case Builtin::BI_bittestandreset: |
| 4500 | case Builtin::BI_bittestandset64: |
| 4501 | case Builtin::BI_bittestandset: |
| 4502 | case Builtin::BI_interlockedbittestandreset: |
| 4503 | case Builtin::BI_interlockedbittestandreset64: |
| 4504 | case Builtin::BI_interlockedbittestandset64: |
| 4505 | case Builtin::BI_interlockedbittestandset: |
| 4506 | case Builtin::BI_interlockedbittestandset_acq: |
| 4507 | case Builtin::BI_interlockedbittestandset_rel: |
| 4508 | case Builtin::BI_interlockedbittestandset_nf: |
| 4509 | case Builtin::BI_interlockedbittestandreset_acq: |
| 4510 | case Builtin::BI_interlockedbittestandreset_rel: |
| 4511 | case Builtin::BI_interlockedbittestandreset_nf: |
| 4512 | return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E)); |
| 4513 | |
| 4514 | // These builtins exist to emit regular volatile loads and stores not |
| 4515 | // affected by the -fms-volatile setting. |
| 4516 | case Builtin::BI__iso_volatile_load8: |
| 4517 | case Builtin::BI__iso_volatile_load16: |
| 4518 | case Builtin::BI__iso_volatile_load32: |
| 4519 | case Builtin::BI__iso_volatile_load64: |
| 4520 | return RValue::get(EmitISOVolatileLoad(*this, E)); |
| 4521 | case Builtin::BI__iso_volatile_store8: |
| 4522 | case Builtin::BI__iso_volatile_store16: |
| 4523 | case Builtin::BI__iso_volatile_store32: |
| 4524 | case Builtin::BI__iso_volatile_store64: |
| 4525 | return RValue::get(EmitISOVolatileStore(*this, E)); |
| 4526 | |
| 4527 | case Builtin::BI__exception_code: |
| 4528 | case Builtin::BI_exception_code: |
| 4529 | return RValue::get(EmitSEHExceptionCode()); |
| 4530 | case Builtin::BI__exception_info: |
| 4531 | case Builtin::BI_exception_info: |
| 4532 | return RValue::get(EmitSEHExceptionInfo()); |
| 4533 | case Builtin::BI__abnormal_termination: |
| 4534 | case Builtin::BI_abnormal_termination: |
| 4535 | return RValue::get(EmitSEHAbnormalTermination()); |
| 4536 | case Builtin::BI_setjmpex: |
| 4537 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && |
| 4538 | E->getArg(0)->getType()->isPointerType()) |
| 4539 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
| 4540 | break; |
| 4541 | case Builtin::BI_setjmp: |
| 4542 | if (getTarget().getTriple().isOSMSVCRT() && E->getNumArgs() == 1 && |
| 4543 | E->getArg(0)->getType()->isPointerType()) { |
| 4544 | if (getTarget().getTriple().getArch() == llvm::Triple::x86) |
| 4545 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E); |
| 4546 | else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64) |
| 4547 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E); |
| 4548 | return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E); |
| 4549 | } |
| 4550 | break; |
| 4551 | |
| 4552 | case Builtin::BI__GetExceptionInfo: { |
| 4553 | if (llvm::GlobalVariable *GV = |
| 4554 | CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType())) |
| 4555 | return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy)); |
| 4556 | break; |
| 4557 | } |
| 4558 | |
| 4559 | case Builtin::BI__fastfail: |
| 4560 | return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E)); |
| 4561 | |
| 4562 | case Builtin::BI__builtin_coro_size: { |
| 4563 | auto & Context = getContext(); |
| 4564 | auto SizeTy = Context.getSizeType(); |
| 4565 | auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy)); |
| 4566 | Function *F = CGM.getIntrinsic(Intrinsic::coro_size, T); |
| 4567 | return RValue::get(Builder.CreateCall(F)); |
| 4568 | } |
| 4569 | |
| 4570 | case Builtin::BI__builtin_coro_id: |
| 4571 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_id); |
| 4572 | case Builtin::BI__builtin_coro_promise: |
| 4573 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise); |
| 4574 | case Builtin::BI__builtin_coro_resume: |
| 4575 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume); |
| 4576 | case Builtin::BI__builtin_coro_frame: |
| 4577 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame); |
| 4578 | case Builtin::BI__builtin_coro_noop: |
| 4579 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop); |
| 4580 | case Builtin::BI__builtin_coro_free: |
| 4581 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_free); |
| 4582 | case Builtin::BI__builtin_coro_destroy: |
| 4583 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy); |
| 4584 | case Builtin::BI__builtin_coro_done: |
| 4585 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_done); |
| 4586 | case Builtin::BI__builtin_coro_alloc: |
| 4587 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc); |
| 4588 | case Builtin::BI__builtin_coro_begin: |
| 4589 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin); |
| 4590 | case Builtin::BI__builtin_coro_end: |
| 4591 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_end); |
| 4592 | case Builtin::BI__builtin_coro_suspend: |
| 4593 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend); |
| 4594 | case Builtin::BI__builtin_coro_param: |
| 4595 | return EmitCoroutineIntrinsic(E, Intrinsic::coro_param); |
| 4596 | |
| 4597 | // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions |
| 4598 | case Builtin::BIread_pipe: |
| 4599 | case Builtin::BIwrite_pipe: { |
| 4600 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 4601 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 4602 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4603 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4604 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4605 | |
| 4606 | // Type of the generic packet parameter. |
| 4607 | unsigned GenericAS = |
| 4608 | getContext().getTargetAddressSpace(LangAS::opencl_generic); |
| 4609 | llvm::Type *I8PTy = llvm::PointerType::get( |
| 4610 | llvm::Type::getInt8Ty(getLLVMContext()), GenericAS); |
| 4611 | |
| 4612 | // Testing which overloaded version we should generate the call for. |
| 4613 | if (2U == E->getNumArgs()) { |
| 4614 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2" |
| 4615 | : "__write_pipe_2"; |
| 4616 | // Creating a generic function type to be able to call with any builtin or |
| 4617 | // user defined type. |
| 4618 | llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty}; |
| 4619 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4620 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4621 | Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy); |
| 4622 | return RValue::get( |
| 4623 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4624 | {Arg0, BCast, PacketSize, PacketAlign})); |
| 4625 | } else { |
| 4626 | assert(4 == E->getNumArgs() &&((void)0) |
| 4627 | "Illegal number of parameters to pipe function")((void)0); |
| 4628 | const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4" |
| 4629 | : "__write_pipe_4"; |
| 4630 | |
| 4631 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy, |
| 4632 | Int32Ty, Int32Ty}; |
| 4633 | Value *Arg2 = EmitScalarExpr(E->getArg(2)), |
| 4634 | *Arg3 = EmitScalarExpr(E->getArg(3)); |
| 4635 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4636 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4637 | Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy); |
| 4638 | // We know the third argument is an integer type, but we may need to cast |
| 4639 | // it to i32. |
| 4640 | if (Arg2->getType() != Int32Ty) |
| 4641 | Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty); |
| 4642 | return RValue::get( |
| 4643 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4644 | {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign})); |
| 4645 | } |
| 4646 | } |
| 4647 | // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write |
| 4648 | // functions |
| 4649 | case Builtin::BIreserve_read_pipe: |
| 4650 | case Builtin::BIreserve_write_pipe: |
| 4651 | case Builtin::BIwork_group_reserve_read_pipe: |
| 4652 | case Builtin::BIwork_group_reserve_write_pipe: |
| 4653 | case Builtin::BIsub_group_reserve_read_pipe: |
| 4654 | case Builtin::BIsub_group_reserve_write_pipe: { |
| 4655 | // Composing the mangled name for the function. |
| 4656 | const char *Name; |
| 4657 | if (BuiltinID == Builtin::BIreserve_read_pipe) |
| 4658 | Name = "__reserve_read_pipe"; |
| 4659 | else if (BuiltinID == Builtin::BIreserve_write_pipe) |
| 4660 | Name = "__reserve_write_pipe"; |
| 4661 | else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe) |
| 4662 | Name = "__work_group_reserve_read_pipe"; |
| 4663 | else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe) |
| 4664 | Name = "__work_group_reserve_write_pipe"; |
| 4665 | else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe) |
| 4666 | Name = "__sub_group_reserve_read_pipe"; |
| 4667 | else |
| 4668 | Name = "__sub_group_reserve_write_pipe"; |
| 4669 | |
| 4670 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 4671 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 4672 | llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy); |
| 4673 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4674 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4675 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4676 | |
| 4677 | // Building the generic function prototype. |
| 4678 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty}; |
| 4679 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4680 | ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4681 | // We know the second argument is an integer type, but we may need to cast |
| 4682 | // it to i32. |
| 4683 | if (Arg1->getType() != Int32Ty) |
| 4684 | Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty); |
| 4685 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4686 | {Arg0, Arg1, PacketSize, PacketAlign})); |
| 4687 | } |
| 4688 | // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write |
| 4689 | // functions |
| 4690 | case Builtin::BIcommit_read_pipe: |
| 4691 | case Builtin::BIcommit_write_pipe: |
| 4692 | case Builtin::BIwork_group_commit_read_pipe: |
| 4693 | case Builtin::BIwork_group_commit_write_pipe: |
| 4694 | case Builtin::BIsub_group_commit_read_pipe: |
| 4695 | case Builtin::BIsub_group_commit_write_pipe: { |
| 4696 | const char *Name; |
| 4697 | if (BuiltinID == Builtin::BIcommit_read_pipe) |
| 4698 | Name = "__commit_read_pipe"; |
| 4699 | else if (BuiltinID == Builtin::BIcommit_write_pipe) |
| 4700 | Name = "__commit_write_pipe"; |
| 4701 | else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe) |
| 4702 | Name = "__work_group_commit_read_pipe"; |
| 4703 | else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe) |
| 4704 | Name = "__work_group_commit_write_pipe"; |
| 4705 | else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe) |
| 4706 | Name = "__sub_group_commit_read_pipe"; |
| 4707 | else |
| 4708 | Name = "__sub_group_commit_write_pipe"; |
| 4709 | |
| 4710 | Value *Arg0 = EmitScalarExpr(E->getArg(0)), |
| 4711 | *Arg1 = EmitScalarExpr(E->getArg(1)); |
| 4712 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4713 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4714 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4715 | |
| 4716 | // Building the generic function prototype. |
| 4717 | llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty}; |
| 4718 | llvm::FunctionType *FTy = |
| 4719 | llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()), |
| 4720 | llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4721 | |
| 4722 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4723 | {Arg0, Arg1, PacketSize, PacketAlign})); |
| 4724 | } |
| 4725 | // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions |
| 4726 | case Builtin::BIget_pipe_num_packets: |
| 4727 | case Builtin::BIget_pipe_max_packets: { |
| 4728 | const char *BaseName; |
| 4729 | const auto *PipeTy = E->getArg(0)->getType()->castAs<PipeType>(); |
| 4730 | if (BuiltinID == Builtin::BIget_pipe_num_packets) |
| 4731 | BaseName = "__get_pipe_num_packets"; |
| 4732 | else |
| 4733 | BaseName = "__get_pipe_max_packets"; |
| 4734 | std::string Name = std::string(BaseName) + |
| 4735 | std::string(PipeTy->isReadOnly() ? "_ro" : "_wo"); |
| 4736 | |
| 4737 | // Building the generic function prototype. |
| 4738 | Value *Arg0 = EmitScalarExpr(E->getArg(0)); |
| 4739 | CGOpenCLRuntime OpenCLRT(CGM); |
| 4740 | Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0)); |
| 4741 | Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0)); |
| 4742 | llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty}; |
| 4743 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4744 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4745 | |
| 4746 | return RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4747 | {Arg0, PacketSize, PacketAlign})); |
| 4748 | } |
| 4749 | |
| 4750 | // OpenCL v2.0 s6.13.9 - Address space qualifier functions. |
| 4751 | case Builtin::BIto_global: |
| 4752 | case Builtin::BIto_local: |
| 4753 | case Builtin::BIto_private: { |
| 4754 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
| 4755 | auto NewArgT = llvm::PointerType::get(Int8Ty, |
| 4756 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4757 | auto NewRetT = llvm::PointerType::get(Int8Ty, |
| 4758 | CGM.getContext().getTargetAddressSpace( |
| 4759 | E->getType()->getPointeeType().getAddressSpace())); |
| 4760 | auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false); |
| 4761 | llvm::Value *NewArg; |
| 4762 | if (Arg0->getType()->getPointerAddressSpace() != |
| 4763 | NewArgT->getPointerAddressSpace()) |
| 4764 | NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT); |
| 4765 | else |
| 4766 | NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT); |
| 4767 | auto NewName = std::string("__") + E->getDirectCallee()->getName().str(); |
| 4768 | auto NewCall = |
| 4769 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg}); |
| 4770 | return RValue::get(Builder.CreateBitOrPointerCast(NewCall, |
| 4771 | ConvertType(E->getType()))); |
| 4772 | } |
| 4773 | |
| 4774 | // OpenCL v2.0, s6.13.17 - Enqueue kernel function. |
| 4775 | // It contains four different overload formats specified in Table 6.13.17.1. |
| 4776 | case Builtin::BIenqueue_kernel: { |
| 4777 | StringRef Name; // Generated function call name |
| 4778 | unsigned NumArgs = E->getNumArgs(); |
| 4779 | |
| 4780 | llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy); |
| 4781 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 4782 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4783 | |
| 4784 | llvm::Value *Queue = EmitScalarExpr(E->getArg(0)); |
| 4785 | llvm::Value *Flags = EmitScalarExpr(E->getArg(1)); |
| 4786 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(2)); |
| 4787 | llvm::Value *Range = NDRangeL.getAddress(*this).getPointer(); |
| 4788 | llvm::Type *RangeTy = NDRangeL.getAddress(*this).getType(); |
| 4789 | |
| 4790 | if (NumArgs == 4) { |
| 4791 | // The most basic form of the call with parameters: |
| 4792 | // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void) |
| 4793 | Name = "__enqueue_kernel_basic"; |
| 4794 | llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy, |
| 4795 | GenericVoidPtrTy}; |
| 4796 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4797 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4798 | |
| 4799 | auto Info = |
| 4800 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
| 4801 | llvm::Value *Kernel = |
| 4802 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 4803 | llvm::Value *Block = |
| 4804 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 4805 | |
| 4806 | AttrBuilder B; |
| 4807 | B.addByValAttr(NDRangeL.getAddress(*this).getElementType()); |
| 4808 | llvm::AttributeList ByValAttrSet = |
| 4809 | llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B); |
| 4810 | |
| 4811 | auto RTCall = |
| 4812 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet), |
| 4813 | {Queue, Flags, Range, Kernel, Block}); |
| 4814 | RTCall->setAttributes(ByValAttrSet); |
| 4815 | return RValue::get(RTCall); |
| 4816 | } |
| 4817 | assert(NumArgs >= 5 && "Invalid enqueue_kernel signature")((void)0); |
| 4818 | |
| 4819 | // Create a temporary array to hold the sizes of local pointer arguments |
| 4820 | // for the block. \p First is the position of the first size argument. |
| 4821 | auto CreateArrayForSizeVar = [=](unsigned First) |
| 4822 | -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> { |
| 4823 | llvm::APInt ArraySize(32, NumArgs - First); |
| 4824 | QualType SizeArrayTy = getContext().getConstantArrayType( |
| 4825 | getContext().getSizeType(), ArraySize, nullptr, ArrayType::Normal, |
| 4826 | /*IndexTypeQuals=*/0); |
| 4827 | auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes"); |
| 4828 | llvm::Value *TmpPtr = Tmp.getPointer(); |
| 4829 | llvm::Value *TmpSize = EmitLifetimeStart( |
| 4830 | CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr); |
| 4831 | llvm::Value *ElemPtr; |
| 4832 | // Each of the following arguments specifies the size of the corresponding |
| 4833 | // argument passed to the enqueued block. |
| 4834 | auto *Zero = llvm::ConstantInt::get(IntTy, 0); |
| 4835 | for (unsigned I = First; I < NumArgs; ++I) { |
| 4836 | auto *Index = llvm::ConstantInt::get(IntTy, I - First); |
| 4837 | auto *GEP = Builder.CreateGEP(Tmp.getElementType(), TmpPtr, |
| 4838 | {Zero, Index}); |
| 4839 | if (I == First) |
| 4840 | ElemPtr = GEP; |
| 4841 | auto *V = |
| 4842 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy); |
| 4843 | Builder.CreateAlignedStore( |
| 4844 | V, GEP, CGM.getDataLayout().getPrefTypeAlign(SizeTy)); |
| 4845 | } |
| 4846 | return std::tie(ElemPtr, TmpSize, TmpPtr); |
| 4847 | }; |
| 4848 | |
| 4849 | // Could have events and/or varargs. |
| 4850 | if (E->getArg(3)->getType()->isBlockPointerType()) { |
| 4851 | // No events passed, but has variadic arguments. |
| 4852 | Name = "__enqueue_kernel_varargs"; |
| 4853 | auto Info = |
| 4854 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3)); |
| 4855 | llvm::Value *Kernel = |
| 4856 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 4857 | auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 4858 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
| 4859 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4); |
| 4860 | |
| 4861 | // Create a vector of the arguments, as well as a constant value to |
| 4862 | // express to the runtime the number of variadic arguments. |
| 4863 | llvm::Value *const Args[] = {Queue, Flags, |
| 4864 | Range, Kernel, |
| 4865 | Block, ConstantInt::get(IntTy, NumArgs - 4), |
| 4866 | ElemPtr}; |
| 4867 | llvm::Type *const ArgTys[] = { |
| 4868 | QueueTy, IntTy, RangeTy, GenericVoidPtrTy, |
| 4869 | GenericVoidPtrTy, IntTy, ElemPtr->getType()}; |
| 4870 | |
| 4871 | llvm::FunctionType *FTy = llvm::FunctionType::get(Int32Ty, ArgTys, false); |
| 4872 | auto Call = RValue::get( |
| 4873 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Args)); |
| 4874 | if (TmpSize) |
| 4875 | EmitLifetimeEnd(TmpSize, TmpPtr); |
| 4876 | return Call; |
| 4877 | } |
| 4878 | // Any calls now have event arguments passed. |
| 4879 | if (NumArgs >= 7) { |
| 4880 | llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy); |
| 4881 | llvm::PointerType *EventPtrTy = EventTy->getPointerTo( |
| 4882 | CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4883 | |
| 4884 | llvm::Value *NumEvents = |
| 4885 | Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty); |
| 4886 | |
| 4887 | // Since SemaOpenCLBuiltinEnqueueKernel allows fifth and sixth arguments |
| 4888 | // to be a null pointer constant (including `0` literal), we can take it |
| 4889 | // into account and emit null pointer directly. |
| 4890 | llvm::Value *EventWaitList = nullptr; |
| 4891 | if (E->getArg(4)->isNullPointerConstant( |
| 4892 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
| 4893 | EventWaitList = llvm::ConstantPointerNull::get(EventPtrTy); |
| 4894 | } else { |
| 4895 | EventWaitList = E->getArg(4)->getType()->isArrayType() |
| 4896 | ? EmitArrayToPointerDecay(E->getArg(4)).getPointer() |
| 4897 | : EmitScalarExpr(E->getArg(4)); |
| 4898 | // Convert to generic address space. |
| 4899 | EventWaitList = Builder.CreatePointerCast(EventWaitList, EventPtrTy); |
| 4900 | } |
| 4901 | llvm::Value *EventRet = nullptr; |
| 4902 | if (E->getArg(5)->isNullPointerConstant( |
| 4903 | getContext(), Expr::NPC_ValueDependentIsNotNull)) { |
| 4904 | EventRet = llvm::ConstantPointerNull::get(EventPtrTy); |
| 4905 | } else { |
| 4906 | EventRet = |
| 4907 | Builder.CreatePointerCast(EmitScalarExpr(E->getArg(5)), EventPtrTy); |
| 4908 | } |
| 4909 | |
| 4910 | auto Info = |
| 4911 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6)); |
| 4912 | llvm::Value *Kernel = |
| 4913 | Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 4914 | llvm::Value *Block = |
| 4915 | Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 4916 | |
| 4917 | std::vector<llvm::Type *> ArgTys = { |
| 4918 | QueueTy, Int32Ty, RangeTy, Int32Ty, |
| 4919 | EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy}; |
| 4920 | |
| 4921 | std::vector<llvm::Value *> Args = {Queue, Flags, Range, |
| 4922 | NumEvents, EventWaitList, EventRet, |
| 4923 | Kernel, Block}; |
| 4924 | |
| 4925 | if (NumArgs == 7) { |
| 4926 | // Has events but no variadics. |
| 4927 | Name = "__enqueue_kernel_basic_events"; |
| 4928 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4929 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4930 | return RValue::get( |
| 4931 | EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4932 | llvm::ArrayRef<llvm::Value *>(Args))); |
| 4933 | } |
| 4934 | // Has event info and variadics |
| 4935 | // Pass the number of variadics to the runtime function too. |
| 4936 | Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7)); |
| 4937 | ArgTys.push_back(Int32Ty); |
| 4938 | Name = "__enqueue_kernel_events_varargs"; |
| 4939 | |
| 4940 | llvm::Value *ElemPtr, *TmpSize, *TmpPtr; |
| 4941 | std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7); |
| 4942 | Args.push_back(ElemPtr); |
| 4943 | ArgTys.push_back(ElemPtr->getType()); |
| 4944 | |
| 4945 | llvm::FunctionType *FTy = llvm::FunctionType::get( |
| 4946 | Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false); |
| 4947 | auto Call = |
| 4948 | RValue::get(EmitRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), |
| 4949 | llvm::ArrayRef<llvm::Value *>(Args))); |
| 4950 | if (TmpSize) |
| 4951 | EmitLifetimeEnd(TmpSize, TmpPtr); |
| 4952 | return Call; |
| 4953 | } |
| 4954 | LLVM_FALLTHROUGH[[gnu::fallthrough]]; |
| 4955 | } |
| 4956 | // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block |
| 4957 | // parameter. |
| 4958 | case Builtin::BIget_kernel_work_group_size: { |
| 4959 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 4960 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4961 | auto Info = |
| 4962 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
| 4963 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 4964 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 4965 | return RValue::get(EmitRuntimeCall( |
| 4966 | CGM.CreateRuntimeFunction( |
| 4967 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
| 4968 | false), |
| 4969 | "__get_kernel_work_group_size_impl"), |
| 4970 | {Kernel, Arg})); |
| 4971 | } |
| 4972 | case Builtin::BIget_kernel_preferred_work_group_size_multiple: { |
| 4973 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 4974 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4975 | auto Info = |
| 4976 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0)); |
| 4977 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 4978 | Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 4979 | return RValue::get(EmitRuntimeCall( |
| 4980 | CGM.CreateRuntimeFunction( |
| 4981 | llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy}, |
| 4982 | false), |
| 4983 | "__get_kernel_preferred_work_group_size_multiple_impl"), |
| 4984 | {Kernel, Arg})); |
| 4985 | } |
| 4986 | case Builtin::BIget_kernel_max_sub_group_size_for_ndrange: |
| 4987 | case Builtin::BIget_kernel_sub_group_count_for_ndrange: { |
| 4988 | llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy( |
| 4989 | getContext().getTargetAddressSpace(LangAS::opencl_generic)); |
| 4990 | LValue NDRangeL = EmitAggExprToLValue(E->getArg(0)); |
| 4991 | llvm::Value *NDRange = NDRangeL.getAddress(*this).getPointer(); |
| 4992 | auto Info = |
| 4993 | CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1)); |
| 4994 | Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy); |
| 4995 | Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy); |
| 4996 | const char *Name = |
| 4997 | BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange |
| 4998 | ? "__get_kernel_max_sub_group_size_for_ndrange_impl" |
| 4999 | : "__get_kernel_sub_group_count_for_ndrange_impl"; |
| 5000 | return RValue::get(EmitRuntimeCall( |
| 5001 | CGM.CreateRuntimeFunction( |
| 5002 | llvm::FunctionType::get( |
| 5003 | IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy}, |
| 5004 | false), |
| 5005 | Name), |
| 5006 | {NDRange, Kernel, Block})); |
| 5007 | } |
| 5008 | |
| 5009 | case Builtin::BI__builtin_store_half: |
| 5010 | case Builtin::BI__builtin_store_halff: { |
| 5011 | Value *Val = EmitScalarExpr(E->getArg(0)); |
| 5012 | Address Address = EmitPointerWithAlignment(E->getArg(1)); |
| 5013 | Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy()); |
| 5014 | return RValue::get(Builder.CreateStore(HalfVal, Address)); |
| 5015 | } |
| 5016 | case Builtin::BI__builtin_load_half: { |
| 5017 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
| 5018 | Value *HalfVal = Builder.CreateLoad(Address); |
| 5019 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy())); |
| 5020 | } |
| 5021 | case Builtin::BI__builtin_load_halff: { |
| 5022 | Address Address = EmitPointerWithAlignment(E->getArg(0)); |
| 5023 | Value *HalfVal = Builder.CreateLoad(Address); |
| 5024 | return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy())); |
| 5025 | } |
| 5026 | case Builtin::BIprintf: |
| 5027 | if (getTarget().getTriple().isNVPTX()) |
| 5028 | return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue); |
| 5029 | if (getTarget().getTriple().getArch() == Triple::amdgcn && |
| 5030 | getLangOpts().HIP) |
| 5031 | return EmitAMDGPUDevicePrintfCallExpr(E, ReturnValue); |
| 5032 | break; |
| 5033 | case Builtin::BI__builtin_canonicalize: |
| 5034 | case Builtin::BI__builtin_canonicalizef: |
| 5035 | case Builtin::BI__builtin_canonicalizef16: |
| 5036 | case Builtin::BI__builtin_canonicalizel: |
| 5037 | return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize)); |
| 5038 | |
| 5039 | case Builtin::BI__builtin_thread_pointer: { |
| 5040 | if (!getContext().getTargetInfo().isTLSSupported()) |
| 5041 | CGM.ErrorUnsupported(E, "__builtin_thread_pointer"); |
| 5042 | // Fall through - it's already mapped to the intrinsic by GCCBuiltin. |
| 5043 | break; |
| 5044 | } |
| 5045 | case Builtin::BI__builtin_os_log_format: |
| 5046 | return emitBuiltinOSLogFormat(*E); |
| 5047 | |
| 5048 | case Builtin::BI__xray_customevent: { |
| 5049 | if (!ShouldXRayInstrumentFunction()) |
| 5050 | return RValue::getIgnored(); |
| 5051 | |
| 5052 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
| 5053 | XRayInstrKind::Custom)) |
| 5054 | return RValue::getIgnored(); |
| 5055 | |
| 5056 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
| 5057 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents()) |
| 5058 | return RValue::getIgnored(); |
| 5059 | |
| 5060 | Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent); |
| 5061 | auto FTy = F->getFunctionType(); |
| 5062 | auto Arg0 = E->getArg(0); |
| 5063 | auto Arg0Val = EmitScalarExpr(Arg0); |
| 5064 | auto Arg0Ty = Arg0->getType(); |
| 5065 | auto PTy0 = FTy->getParamType(0); |
| 5066 | if (PTy0 != Arg0Val->getType()) { |
| 5067 | if (Arg0Ty->isArrayType()) |
| 5068 | Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer(); |
| 5069 | else |
| 5070 | Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0); |
| 5071 | } |
| 5072 | auto Arg1 = EmitScalarExpr(E->getArg(1)); |
| 5073 | auto PTy1 = FTy->getParamType(1); |
| 5074 | if (PTy1 != Arg1->getType()) |
| 5075 | Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1); |
| 5076 | return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1})); |
| 5077 | } |
| 5078 | |
| 5079 | case Builtin::BI__xray_typedevent: { |
| 5080 | // TODO: There should be a way to always emit events even if the current |
| 5081 | // function is not instrumented. Losing events in a stream can cripple |
| 5082 | // a trace. |
| 5083 | if (!ShouldXRayInstrumentFunction()) |
| 5084 | return RValue::getIgnored(); |
| 5085 | |
| 5086 | if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has( |
| 5087 | XRayInstrKind::Typed)) |
| 5088 | return RValue::getIgnored(); |
| 5089 | |
| 5090 | if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>()) |
| 5091 | if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents()) |
| 5092 | return RValue::getIgnored(); |
| 5093 | |
| 5094 | Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent); |
| 5095 | auto FTy = F->getFunctionType(); |
| 5096 | auto Arg0 = EmitScalarExpr(E->getArg(0)); |
| 5097 | auto PTy0 = FTy->getParamType(0); |
| 5098 | if (PTy0 != Arg0->getType()) |
| 5099 | Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0); |
| 5100 | auto Arg1 = E->getArg(1); |
| 5101 | auto Arg1Val = EmitScalarExpr(Arg1); |
| 5102 | auto Arg1Ty = Arg1->getType(); |
| 5103 | auto PTy1 = FTy->getParamType(1); |
| 5104 | if (PTy1 != Arg1Val->getType()) { |
| 5105 | if (Arg1Ty->isArrayType()) |
| 5106 | Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer(); |
| 5107 | else |
| 5108 | Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1); |
| 5109 | } |
| 5110 | auto Arg2 = EmitScalarExpr(E->getArg(2)); |
| 5111 | auto PTy2 = FTy->getParamType(2); |
| 5112 | if (PTy2 != Arg2->getType()) |
| 5113 | Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2); |
| 5114 | return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2})); |
| 5115 | } |
| 5116 | |
| 5117 | case Builtin::BI__builtin_ms_va_start: |
| 5118 | case Builtin::BI__builtin_ms_va_end: |
| 5119 | return RValue::get( |
| 5120 | EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(), |
| 5121 | BuiltinID == Builtin::BI__builtin_ms_va_start)); |
| 5122 | |
| 5123 | case Builtin::BI__builtin_ms_va_copy: { |
| 5124 | // Lower this manually. We can't reliably determine whether or not any |
| 5125 | // given va_copy() is for a Win64 va_list from the calling convention |
| 5126 | // alone, because it's legal to do this from a System V ABI function. |
| 5127 | // With opaque pointer types, we won't have enough information in LLVM |
| 5128 | // IR to determine this from the argument types, either. Best to do it |
| 5129 | // now, while we have enough information. |
| 5130 | Address DestAddr = EmitMSVAListRef(E->getArg(0)); |
| 5131 | Address SrcAddr = EmitMSVAListRef(E->getArg(1)); |
| 5132 | |
| 5133 | llvm::Type *BPP = Int8PtrPtrTy; |
| 5134 | |
| 5135 | DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"), |
| 5136 | DestAddr.getAlignment()); |
| 5137 | SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"), |
| 5138 | SrcAddr.getAlignment()); |
| 5139 | |
| 5140 | Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val"); |
| 5141 | return RValue::get(Builder.CreateStore(ArgPtr, DestAddr)); |
| 5142 | } |
| 5143 | |
| 5144 | case Builtin::BI__builtin_get_device_side_mangled_name: { |
| 5145 | auto Name = CGM.getCUDARuntime().getDeviceSideName( |
| 5146 | cast<DeclRefExpr>(E->getArg(0)->IgnoreImpCasts())->getDecl()); |
| 5147 | auto Str = CGM.GetAddrOfConstantCString(Name, ""); |
| 5148 | llvm::Constant *Zeros[] = {llvm::ConstantInt::get(SizeTy, 0), |
| 5149 | llvm::ConstantInt::get(SizeTy, 0)}; |
| 5150 | auto *Ptr = llvm::ConstantExpr::getGetElementPtr(Str.getElementType(), |
| 5151 | Str.getPointer(), Zeros); |
| 5152 | return RValue::get(Ptr); |
| 5153 | } |
| 5154 | } |
| 5155 | |
| 5156 | // If this is an alias for a lib function (e.g. __builtin_sin), emit |
| 5157 | // the call using the normal call path, but using the unmangled |
| 5158 | // version of the function name. |
| 5159 | if (getContext().BuiltinInfo.isLibFunction(BuiltinID)) |
| 5160 | return emitLibraryCall(*this, FD, E, |
| 5161 | CGM.getBuiltinLibFunction(FD, BuiltinID)); |
| 5162 | |
| 5163 | // If this is a predefined lib function (e.g. malloc), emit the call |
| 5164 | // using exactly the normal call path. |
| 5165 | if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID)) |
| 5166 | return emitLibraryCall(*this, FD, E, |
| 5167 | cast<llvm::Constant>(EmitScalarExpr(E->getCallee()))); |
| 5168 | |
| 5169 | // Check that a call to a target specific builtin has the correct target |
| 5170 | // features. |
| 5171 | // This is down here to avoid non-target specific builtins, however, if |
| 5172 | // generic builtins start to require generic target features then we |
| 5173 | // can move this up to the beginning of the function. |
| 5174 | checkTargetFeatures(E, FD); |
| 5175 | |
| 5176 | if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID)) |
| 5177 | LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth); |
| 5178 | |
| 5179 | // See if we have a target specific intrinsic. |
| 5180 | const char *Name = getContext().BuiltinInfo.getName(BuiltinID); |
| 5181 | Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic; |
| 5182 | StringRef Prefix = |
| 5183 | llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch()); |
| 5184 | if (!Prefix.empty()) { |
| 5185 | IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name); |
| 5186 | // NOTE we don't need to perform a compatibility flag check here since the |
| 5187 | // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the |
| 5188 | // MS builtins via ALL_MS_LANGUAGES and are filtered earlier. |
| 5189 | if (IntrinsicID == Intrinsic::not_intrinsic) |
| 5190 | IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name); |
| 5191 | } |
| 5192 | |
| 5193 | if (IntrinsicID != Intrinsic::not_intrinsic) { |
| 5194 | SmallVector<Value*, 16> Args; |
| 5195 | |
| 5196 | // Find out if any arguments are required to be integer constant |
| 5197 | // expressions. |
| 5198 | unsigned ICEArguments = 0; |
| 5199 | ASTContext::GetBuiltinTypeError Error; |
| 5200 | getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); |
| 5201 | assert(Error == ASTContext::GE_None && "Should not codegen an error")((void)0); |
| 5202 | |
| 5203 | Function *F = CGM.getIntrinsic(IntrinsicID); |
| 5204 | llvm::FunctionType *FTy = F->getFunctionType(); |
| 5205 | |
| 5206 | for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) { |
| 5207 | Value *ArgValue; |
| 5208 | // If this is a normal argument, just emit it as a scalar. |
| 5209 | if ((ICEArguments & (1 << i)) == 0) { |
| 5210 | ArgValue = EmitScalarExpr(E->getArg(i)); |
| 5211 | } else { |
| 5212 | // If this is required to be a constant, constant fold it so that we |
| 5213 | // know that the generated intrinsic gets a ConstantInt. |
| 5214 | ArgValue = llvm::ConstantInt::get( |
| 5215 | getLLVMContext(), |
| 5216 | *E->getArg(i)->getIntegerConstantExpr(getContext())); |
| 5217 | } |
| 5218 | |
| 5219 | // If the intrinsic arg type is different from the builtin arg type |
| 5220 | // we need to do a bit cast. |
| 5221 | llvm::Type *PTy = FTy->getParamType(i); |
| 5222 | if (PTy != ArgValue->getType()) { |
| 5223 | // XXX - vector of pointers? |
| 5224 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) { |
| 5225 | if (PtrTy->getAddressSpace() != |
| 5226 | ArgValue->getType()->getPointerAddressSpace()) { |
| 5227 | ArgValue = Builder.CreateAddrSpaceCast( |
| 5228 | ArgValue, |
| 5229 | ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace())); |
| 5230 | } |
| 5231 | } |
| 5232 | |
| 5233 | assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&((void)0) |
| 5234 | "Must be able to losslessly bit cast to param")((void)0); |
| 5235 | ArgValue = Builder.CreateBitCast(ArgValue, PTy); |
| 5236 | } |
| 5237 | |
| 5238 | Args.push_back(ArgValue); |
| 5239 | } |
| 5240 | |
| 5241 | Value *V = Builder.CreateCall(F, Args); |
| 5242 | QualType BuiltinRetType = E->getType(); |
| 5243 | |
| 5244 | llvm::Type *RetTy = VoidTy; |
| 5245 | if (!BuiltinRetType->isVoidType()) |
| 5246 | RetTy = ConvertType(BuiltinRetType); |
| 5247 | |
| 5248 | if (RetTy != V->getType()) { |
| 5249 | // XXX - vector of pointers? |
| 5250 | if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) { |
| 5251 | if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) { |
| 5252 | V = Builder.CreateAddrSpaceCast( |
| 5253 | V, V->getType()->getPointerTo(PtrTy->getAddressSpace())); |
| 5254 | } |
| 5255 | } |
| 5256 | |
| 5257 | assert(V->getType()->canLosslesslyBitCastTo(RetTy) &&((void)0) |
| 5258 | "Must be able to losslessly bit cast result type")((void)0); |
| 5259 | V = Builder.CreateBitCast(V, RetTy); |
| 5260 | } |
| 5261 | |
| 5262 | return RValue::get(V); |
| 5263 | } |
| 5264 | |
| 5265 | // Some target-specific builtins can have aggregate return values, e.g. |
| 5266 | // __builtin_arm_mve_vld2q_u32. So if the result is an aggregate, force |
| 5267 | // ReturnValue to be non-null, so that the target-specific emission code can |
| 5268 | // always just emit into it. |
| 5269 | TypeEvaluationKind EvalKind = getEvaluationKind(E->getType()); |
| 5270 | if (EvalKind == TEK_Aggregate && ReturnValue.isNull()) { |
| 5271 | Address DestPtr = CreateMemTemp(E->getType(), "agg.tmp"); |
| 5272 | ReturnValue = ReturnValueSlot(DestPtr, false); |
| 5273 | } |
| 5274 | |
| 5275 | // Now see if we can emit a target-specific builtin. |
| 5276 | if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E, ReturnValue)) { |
| 5277 | switch (EvalKind) { |
| 5278 | case TEK_Scalar: |
| 5279 | return RValue::get(V); |
| 5280 | case TEK_Aggregate: |
| 5281 | return RValue::getAggregate(ReturnValue.getValue(), |
| 5282 | ReturnValue.isVolatile()); |
| 5283 | case TEK_Complex: |
| 5284 | llvm_unreachable("No current target builtin returns complex")__builtin_unreachable(); |
| 5285 | } |
| 5286 | llvm_unreachable("Bad evaluation kind in EmitBuiltinExpr")__builtin_unreachable(); |
| 5287 | } |
| 5288 | |
| 5289 | ErrorUnsupported(E, "builtin function"); |
| 5290 | |
| 5291 | // Unknown builtin, for now just dump it out and return undef. |
| 5292 | return GetUndefRValue(E->getType()); |
| 5293 | } |
| 5294 | |
| 5295 | static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF, |
| 5296 | unsigned BuiltinID, const CallExpr *E, |
| 5297 | ReturnValueSlot ReturnValue, |
| 5298 | llvm::Triple::ArchType Arch) { |
| 5299 | switch (Arch) { |
| 5300 | case llvm::Triple::arm: |
| 5301 | case llvm::Triple::armeb: |
| 5302 | case llvm::Triple::thumb: |
| 5303 | case llvm::Triple::thumbeb: |
| 5304 | return CGF->EmitARMBuiltinExpr(BuiltinID, E, ReturnValue, Arch); |
| 5305 | case llvm::Triple::aarch64: |
| 5306 | case llvm::Triple::aarch64_32: |
| 5307 | case llvm::Triple::aarch64_be: |
| 5308 | return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch); |
| 5309 | case llvm::Triple::bpfeb: |
| 5310 | case llvm::Triple::bpfel: |
| 5311 | return CGF->EmitBPFBuiltinExpr(BuiltinID, E); |
| 5312 | case llvm::Triple::x86: |
| 5313 | case llvm::Triple::x86_64: |
| 5314 | return CGF->EmitX86BuiltinExpr(BuiltinID, E); |
| 5315 | case llvm::Triple::ppc: |
| 5316 | case llvm::Triple::ppcle: |
| 5317 | case llvm::Triple::ppc64: |
| 5318 | case llvm::Triple::ppc64le: |
| 5319 | return CGF->EmitPPCBuiltinExpr(BuiltinID, E); |
| 5320 | case llvm::Triple::r600: |
| 5321 | case llvm::Triple::amdgcn: |
| 5322 | return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E); |
| 5323 | case llvm::Triple::systemz: |
| 5324 | return CGF->EmitSystemZBuiltinExpr(BuiltinID, E); |
| 5325 | case llvm::Triple::nvptx: |
| 5326 | case llvm::Triple::nvptx64: |
| 5327 | return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E); |
| 5328 | case llvm::Triple::wasm32: |
| 5329 | case llvm::Triple::wasm64: |
| 5330 | return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E); |
| 5331 | case llvm::Triple::hexagon: |
| 5332 | return CGF->EmitHexagonBuiltinExpr(BuiltinID, E); |
| 5333 | case llvm::Triple::riscv32: |
| 5334 | case llvm::Triple::riscv64: |
| 5335 | return CGF->EmitRISCVBuiltinExpr(BuiltinID, E, ReturnValue); |
| 5336 | default: |
| 5337 | return nullptr; |
| 5338 | } |
| 5339 | } |
| 5340 | |
| 5341 | Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID, |
| 5342 | const CallExpr *E, |
| 5343 | ReturnValueSlot ReturnValue) { |
| 5344 | if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) { |
| 5345 | assert(getContext().getAuxTargetInfo() && "Missing aux target info")((void)0); |
| 5346 | return EmitTargetArchBuiltinExpr( |
| 5347 | this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E, |
| 5348 | ReturnValue, getContext().getAuxTargetInfo()->getTriple().getArch()); |
| 5349 | } |
| 5350 | |
| 5351 | return EmitTargetArchBuiltinExpr(this, BuiltinID, E, ReturnValue, |
| 5352 | getTarget().getTriple().getArch()); |
| 5353 | } |
| 5354 | |
| 5355 | static llvm::FixedVectorType *GetNeonType(CodeGenFunction *CGF, |
| 5356 | NeonTypeFlags TypeFlags, |
| 5357 | bool HasLegalHalfType = true, |
| 5358 | bool V1Ty = false, |
| 5359 | bool AllowBFloatArgsAndRet = true) { |
| 5360 | int IsQuad = TypeFlags.isQuad(); |
| 5361 | switch (TypeFlags.getEltType()) { |
| 5362 | case NeonTypeFlags::Int8: |
| 5363 | case NeonTypeFlags::Poly8: |
| 5364 | return llvm::FixedVectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad)); |
| 5365 | case NeonTypeFlags::Int16: |
| 5366 | case NeonTypeFlags::Poly16: |
| 5367 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 5368 | case NeonTypeFlags::BFloat16: |
| 5369 | if (AllowBFloatArgsAndRet) |
| 5370 | return llvm::FixedVectorType::get(CGF->BFloatTy, V1Ty ? 1 : (4 << IsQuad)); |
| 5371 | else |
| 5372 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 5373 | case NeonTypeFlags::Float16: |
| 5374 | if (HasLegalHalfType) |
| 5375 | return llvm::FixedVectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad)); |
| 5376 | else |
| 5377 | return llvm::FixedVectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad)); |
| 5378 | case NeonTypeFlags::Int32: |
| 5379 | return llvm::FixedVectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad)); |
| 5380 | case NeonTypeFlags::Int64: |
| 5381 | case NeonTypeFlags::Poly64: |
| 5382 | return llvm::FixedVectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad)); |
| 5383 | case NeonTypeFlags::Poly128: |
| 5384 | // FIXME: i128 and f128 doesn't get fully support in Clang and llvm. |
| 5385 | // There is a lot of i128 and f128 API missing. |
| 5386 | // so we use v16i8 to represent poly128 and get pattern matched. |
| 5387 | return llvm::FixedVectorType::get(CGF->Int8Ty, 16); |
| 5388 | case NeonTypeFlags::Float32: |
| 5389 | return llvm::FixedVectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad)); |
| 5390 | case NeonTypeFlags::Float64: |
| 5391 | return llvm::FixedVectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad)); |
| 5392 | } |
| 5393 | llvm_unreachable("Unknown vector element type!")__builtin_unreachable(); |
| 5394 | } |
| 5395 | |
| 5396 | static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF, |
| 5397 | NeonTypeFlags IntTypeFlags) { |
| 5398 | int IsQuad = IntTypeFlags.isQuad(); |
| 5399 | switch (IntTypeFlags.getEltType()) { |
| 5400 | case NeonTypeFlags::Int16: |
| 5401 | return llvm::FixedVectorType::get(CGF->HalfTy, (4 << IsQuad)); |
| 5402 | case NeonTypeFlags::Int32: |
| 5403 | return llvm::FixedVectorType::get(CGF->FloatTy, (2 << IsQuad)); |
| 5404 | case NeonTypeFlags::Int64: |
| 5405 | return llvm::FixedVectorType::get(CGF->DoubleTy, (1 << IsQuad)); |
| 5406 | default: |
| 5407 | llvm_unreachable("Type can't be converted to floating-point!")__builtin_unreachable(); |
| 5408 | } |
| 5409 | } |
| 5410 | |
| 5411 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C, |
| 5412 | const ElementCount &Count) { |
| 5413 | Value *SV = llvm::ConstantVector::getSplat(Count, C); |
| 5414 | return Builder.CreateShuffleVector(V, V, SV, "lane"); |
| 5415 | } |
| 5416 | |
| 5417 | Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) { |
| 5418 | ElementCount EC = cast<llvm::VectorType>(V->getType())->getElementCount(); |
| 5419 | return EmitNeonSplat(V, C, EC); |
| 5420 | } |
| 5421 | |
| 5422 | Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops, |
| 5423 | const char *name, |
| 5424 | unsigned shift, bool rightshift) { |
| 5425 | unsigned j = 0; |
| 5426 | for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end(); |
| 5427 | ai != ae; ++ai, ++j) { |
| 5428 | if (F->isConstrainedFPIntrinsic()) |
| 5429 | if (ai->getType()->isMetadataTy()) |
| 5430 | continue; |
| 5431 | if (shift > 0 && shift == j) |
| 5432 | Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift); |
| 5433 | else |
| 5434 | Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name); |
| 5435 | } |
| 5436 | |
| 5437 | if (F->isConstrainedFPIntrinsic()) |
| 5438 | return Builder.CreateConstrainedFPCall(F, Ops, name); |
| 5439 | else |
| 5440 | return Builder.CreateCall(F, Ops, name); |
| 5441 | } |
| 5442 | |
| 5443 | Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty, |
| 5444 | bool neg) { |
| 5445 | int SV = cast<ConstantInt>(V)->getSExtValue(); |
| 5446 | return ConstantInt::get(Ty, neg ? -SV : SV); |
| 5447 | } |
| 5448 | |
| 5449 | // Right-shift a vector by a constant. |
| 5450 | Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift, |
| 5451 | llvm::Type *Ty, bool usgn, |
| 5452 | const char *name) { |
| 5453 | llvm::VectorType *VTy = cast<llvm::VectorType>(Ty); |
| 5454 | |
| 5455 | int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue(); |
| 5456 | int EltSize = VTy->getScalarSizeInBits(); |
| 5457 | |
| 5458 | Vec = Builder.CreateBitCast(Vec, Ty); |
| 5459 | |
| 5460 | // lshr/ashr are undefined when the shift amount is equal to the vector |
| 5461 | // element size. |
| 5462 | if (ShiftAmt == EltSize) { |
| 5463 | if (usgn) { |
| 5464 | // Right-shifting an unsigned value by its size yields 0. |
| 5465 | return llvm::ConstantAggregateZero::get(VTy); |
| 5466 | } else { |
| 5467 | // Right-shifting a signed value by its size is equivalent |
| 5468 | // to a shift of size-1. |
| 5469 | --ShiftAmt; |
| 5470 | Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt); |
| 5471 | } |
| 5472 | } |
| 5473 | |
| 5474 | Shift = EmitNeonShiftVector(Shift, Ty, false); |
| 5475 | if (usgn) |
| 5476 | return Builder.CreateLShr(Vec, Shift, name); |
| 5477 | else |
| 5478 | return Builder.CreateAShr(Vec, Shift, name); |
| 5479 | } |
| 5480 | |
| 5481 | enum { |
| 5482 | AddRetType = (1 << 0), |
| 5483 | Add1ArgType = (1 << 1), |
| 5484 | Add2ArgTypes = (1 << 2), |
| 5485 | |
| 5486 | VectorizeRetType = (1 << 3), |
| 5487 | VectorizeArgTypes = (1 << 4), |
| 5488 | |
| 5489 | InventFloatType = (1 << 5), |
| 5490 | UnsignedAlts = (1 << 6), |
| 5491 | |
| 5492 | Use64BitVectors = (1 << 7), |
| 5493 | Use128BitVectors = (1 << 8), |
| 5494 | |
| 5495 | Vectorize1ArgType = Add1ArgType | VectorizeArgTypes, |
| 5496 | VectorRet = AddRetType | VectorizeRetType, |
| 5497 | VectorRetGetArgs01 = |
| 5498 | AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes, |
| 5499 | FpCmpzModifiers = |
| 5500 | AddRetType | VectorizeRetType | Add1ArgType | InventFloatType |
| 5501 | }; |
| 5502 | |
| 5503 | namespace { |
| 5504 | struct ARMVectorIntrinsicInfo { |
| 5505 | const char *NameHint; |
| 5506 | unsigned BuiltinID; |
| 5507 | unsigned LLVMIntrinsic; |
| 5508 | unsigned AltLLVMIntrinsic; |
| 5509 | uint64_t TypeModifier; |
| 5510 | |
| 5511 | bool operator<(unsigned RHSBuiltinID) const { |
| 5512 | return BuiltinID < RHSBuiltinID; |
| 5513 | } |
| 5514 | bool operator<(const ARMVectorIntrinsicInfo &TE) const { |
| 5515 | return BuiltinID < TE.BuiltinID; |
| 5516 | } |
| 5517 | }; |
| 5518 | } // end anonymous namespace |
| 5519 | |
| 5520 | #define NEONMAP0(NameBase) \ |
| 5521 | { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 } |
| 5522 | |
| 5523 | #define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \ |
| 5524 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
| 5525 | Intrinsic::LLVMIntrinsic, 0, TypeModifier } |
| 5526 | |
| 5527 | #define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \ |
| 5528 | { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \ |
| 5529 | Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \ |
| 5530 | TypeModifier } |
| 5531 | |
| 5532 | static const ARMVectorIntrinsicInfo ARMSIMDIntrinsicMap [] = { |
| 5533 | NEONMAP1(__a32_vcvt_bf16_v, arm_neon_vcvtfp2bf, 0), |
| 5534 | NEONMAP0(splat_lane_v), |
| 5535 | NEONMAP0(splat_laneq_v), |
| 5536 | NEONMAP0(splatq_lane_v), |
| 5537 | NEONMAP0(splatq_laneq_v), |
| 5538 | NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
| 5539 | NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts), |
| 5540 | NEONMAP1(vabs_v, arm_neon_vabs, 0), |
| 5541 | NEONMAP1(vabsq_v, arm_neon_vabs, 0), |
| 5542 | NEONMAP0(vadd_v), |
| 5543 | NEONMAP0(vaddhn_v), |
| 5544 | NEONMAP0(vaddq_v), |
| 5545 | NEONMAP1(vaesdq_v, arm_neon_aesd, 0), |
| 5546 | NEONMAP1(vaeseq_v, arm_neon_aese, 0), |
| 5547 | NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0), |
| 5548 | NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0), |
| 5549 | NEONMAP1(vbfdot_v, arm_neon_bfdot, 0), |
| 5550 | NEONMAP1(vbfdotq_v, arm_neon_bfdot, 0), |
| 5551 | NEONMAP1(vbfmlalbq_v, arm_neon_bfmlalb, 0), |
| 5552 | NEONMAP1(vbfmlaltq_v, arm_neon_bfmlalt, 0), |
| 5553 | NEONMAP1(vbfmmlaq_v, arm_neon_bfmmla, 0), |
| 5554 | NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType), |
| 5555 | NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType), |
| 5556 | NEONMAP1(vcadd_rot270_v, arm_neon_vcadd_rot270, Add1ArgType), |
| 5557 | NEONMAP1(vcadd_rot90_v, arm_neon_vcadd_rot90, Add1ArgType), |
| 5558 | NEONMAP1(vcaddq_rot270_v, arm_neon_vcadd_rot270, Add1ArgType), |
| 5559 | NEONMAP1(vcaddq_rot90_v, arm_neon_vcadd_rot90, Add1ArgType), |
| 5560 | NEONMAP1(vcage_v, arm_neon_vacge, 0), |
| 5561 | NEONMAP1(vcageq_v, arm_neon_vacge, 0), |
| 5562 | NEONMAP1(vcagt_v, arm_neon_vacgt, 0), |
| 5563 | NEONMAP1(vcagtq_v, arm_neon_vacgt, 0), |
| 5564 | NEONMAP1(vcale_v, arm_neon_vacge, 0), |
| 5565 | NEONMAP1(vcaleq_v, arm_neon_vacge, 0), |
| 5566 | NEONMAP1(vcalt_v, arm_neon_vacgt, 0), |
| 5567 | NEONMAP1(vcaltq_v, arm_neon_vacgt, 0), |
| 5568 | NEONMAP0(vceqz_v), |
| 5569 | NEONMAP0(vceqzq_v), |
| 5570 | NEONMAP0(vcgez_v), |
| 5571 | NEONMAP0(vcgezq_v), |
| 5572 | NEONMAP0(vcgtz_v), |
| 5573 | NEONMAP0(vcgtzq_v), |
| 5574 | NEONMAP0(vclez_v), |
| 5575 | NEONMAP0(vclezq_v), |
| 5576 | NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType), |
| 5577 | NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType), |
| 5578 | NEONMAP0(vcltz_v), |
| 5579 | NEONMAP0(vcltzq_v), |
| 5580 | NEONMAP1(vclz_v, ctlz, Add1ArgType), |
| 5581 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), |
| 5582 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), |
| 5583 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), |
| 5584 | NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0), |
| 5585 | NEONMAP0(vcvt_f16_v), |
| 5586 | NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0), |
| 5587 | NEONMAP0(vcvt_f32_v), |
| 5588 | NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 5589 | NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 5590 | NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0), |
| 5591 | NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
| 5592 | NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
| 5593 | NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0), |
| 5594 | NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
| 5595 | NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
| 5596 | NEONMAP0(vcvt_s16_v), |
| 5597 | NEONMAP0(vcvt_s32_v), |
| 5598 | NEONMAP0(vcvt_s64_v), |
| 5599 | NEONMAP0(vcvt_u16_v), |
| 5600 | NEONMAP0(vcvt_u32_v), |
| 5601 | NEONMAP0(vcvt_u64_v), |
| 5602 | NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0), |
| 5603 | NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0), |
| 5604 | NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0), |
| 5605 | NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0), |
| 5606 | NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0), |
| 5607 | NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0), |
| 5608 | NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0), |
| 5609 | NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0), |
| 5610 | NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0), |
| 5611 | NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0), |
| 5612 | NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0), |
| 5613 | NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0), |
| 5614 | NEONMAP1(vcvth_bf16_f32, arm_neon_vcvtbfp2bf, 0), |
| 5615 | NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0), |
| 5616 | NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0), |
| 5617 | NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0), |
| 5618 | NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0), |
| 5619 | NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0), |
| 5620 | NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0), |
| 5621 | NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0), |
| 5622 | NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0), |
| 5623 | NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0), |
| 5624 | NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0), |
| 5625 | NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0), |
| 5626 | NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0), |
| 5627 | NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0), |
| 5628 | NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0), |
| 5629 | NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0), |
| 5630 | NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0), |
| 5631 | NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0), |
| 5632 | NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0), |
| 5633 | NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0), |
| 5634 | NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0), |
| 5635 | NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0), |
| 5636 | NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0), |
| 5637 | NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0), |
| 5638 | NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0), |
| 5639 | NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0), |
| 5640 | NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0), |
| 5641 | NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0), |
| 5642 | NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0), |
| 5643 | NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0), |
| 5644 | NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0), |
| 5645 | NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0), |
| 5646 | NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0), |
| 5647 | NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0), |
| 5648 | NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0), |
| 5649 | NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0), |
| 5650 | NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0), |
| 5651 | NEONMAP0(vcvtq_f16_v), |
| 5652 | NEONMAP0(vcvtq_f32_v), |
| 5653 | NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 5654 | NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0), |
| 5655 | NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0), |
| 5656 | NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0), |
| 5657 | NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0), |
| 5658 | NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0), |
| 5659 | NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0), |
| 5660 | NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0), |
| 5661 | NEONMAP0(vcvtq_s16_v), |
| 5662 | NEONMAP0(vcvtq_s32_v), |
| 5663 | NEONMAP0(vcvtq_s64_v), |
| 5664 | NEONMAP0(vcvtq_u16_v), |
| 5665 | NEONMAP0(vcvtq_u32_v), |
| 5666 | NEONMAP0(vcvtq_u64_v), |
| 5667 | NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0), |
| 5668 | NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0), |
| 5669 | NEONMAP0(vext_v), |
| 5670 | NEONMAP0(vextq_v), |
| 5671 | NEONMAP0(vfma_v), |
| 5672 | NEONMAP0(vfmaq_v), |
| 5673 | NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), |
| 5674 | NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts), |
| 5675 | NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), |
| 5676 | NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts), |
| 5677 | NEONMAP0(vld1_dup_v), |
| 5678 | NEONMAP1(vld1_v, arm_neon_vld1, 0), |
| 5679 | NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0), |
| 5680 | NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0), |
| 5681 | NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0), |
| 5682 | NEONMAP0(vld1q_dup_v), |
| 5683 | NEONMAP1(vld1q_v, arm_neon_vld1, 0), |
| 5684 | NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0), |
| 5685 | NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0), |
| 5686 | NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0), |
| 5687 | NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0), |
| 5688 | NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0), |
| 5689 | NEONMAP1(vld2_v, arm_neon_vld2, 0), |
| 5690 | NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0), |
| 5691 | NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0), |
| 5692 | NEONMAP1(vld2q_v, arm_neon_vld2, 0), |
| 5693 | NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0), |
| 5694 | NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0), |
| 5695 | NEONMAP1(vld3_v, arm_neon_vld3, 0), |
| 5696 | NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0), |
| 5697 | NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0), |
| 5698 | NEONMAP1(vld3q_v, arm_neon_vld3, 0), |
| 5699 | NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0), |
| 5700 | NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0), |
| 5701 | NEONMAP1(vld4_v, arm_neon_vld4, 0), |
| 5702 | NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0), |
| 5703 | NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0), |
| 5704 | NEONMAP1(vld4q_v, arm_neon_vld4, 0), |
| 5705 | NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), |
| 5706 | NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType), |
| 5707 | NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType), |
| 5708 | NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts), |
| 5709 | NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), |
| 5710 | NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType), |
| 5711 | NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType), |
| 5712 | NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts), |
| 5713 | NEONMAP2(vmmlaq_v, arm_neon_ummla, arm_neon_smmla, 0), |
| 5714 | NEONMAP0(vmovl_v), |
| 5715 | NEONMAP0(vmovn_v), |
| 5716 | NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType), |
| 5717 | NEONMAP0(vmull_v), |
| 5718 | NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType), |
| 5719 | NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), |
| 5720 | NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts), |
| 5721 | NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType), |
| 5722 | NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), |
| 5723 | NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts), |
| 5724 | NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType), |
| 5725 | NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts), |
| 5726 | NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts), |
| 5727 | NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType), |
| 5728 | NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType), |
| 5729 | NEONMAP2(vqadd_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts), |
| 5730 | NEONMAP2(vqaddq_v, uadd_sat, sadd_sat, Add1ArgType | UnsignedAlts), |
| 5731 | NEONMAP2(vqdmlal_v, arm_neon_vqdmull, sadd_sat, 0), |
| 5732 | NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, ssub_sat, 0), |
| 5733 | NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType), |
| 5734 | NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType), |
| 5735 | NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType), |
| 5736 | NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts), |
| 5737 | NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType), |
| 5738 | NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType), |
| 5739 | NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType), |
| 5740 | NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType), |
| 5741 | NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType), |
| 5742 | NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), |
| 5743 | NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts), |
| 5744 | NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), |
| 5745 | NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), |
| 5746 | NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts), |
| 5747 | NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts), |
| 5748 | NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0), |
| 5749 | NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0), |
| 5750 | NEONMAP2(vqsub_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts), |
| 5751 | NEONMAP2(vqsubq_v, usub_sat, ssub_sat, Add1ArgType | UnsignedAlts), |
| 5752 | NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType), |
| 5753 | NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), |
| 5754 | NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0), |
| 5755 | NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType), |
| 5756 | NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType), |
| 5757 | NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), |
| 5758 | NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts), |
| 5759 | NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType), |
| 5760 | NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType), |
| 5761 | NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType), |
| 5762 | NEONMAP0(vrndi_v), |
| 5763 | NEONMAP0(vrndiq_v), |
| 5764 | NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType), |
| 5765 | NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType), |
| 5766 | NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType), |
| 5767 | NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType), |
| 5768 | NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType), |
| 5769 | NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType), |
| 5770 | NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType), |
| 5771 | NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType), |
| 5772 | NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType), |
| 5773 | NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), |
| 5774 | NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts), |
| 5775 | NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), |
| 5776 | NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts), |
| 5777 | NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), |
| 5778 | NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0), |
| 5779 | NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType), |
| 5780 | NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType), |
| 5781 | NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType), |
| 5782 | NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0), |
| 5783 | NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0), |
| 5784 | NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0), |
| 5785 | NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0), |
| 5786 | NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0), |
| 5787 | NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0), |
| 5788 | NEONMAP0(vshl_n_v), |
| 5789 | NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), |
| 5790 | NEONMAP0(vshll_n_v), |
| 5791 | NEONMAP0(vshlq_n_v), |
| 5792 | NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts), |
| 5793 | NEONMAP0(vshr_n_v), |
| 5794 | NEONMAP0(vshrn_n_v), |
| 5795 | NEONMAP0(vshrq_n_v), |
| 5796 | NEONMAP1(vst1_v, arm_neon_vst1, 0), |
| 5797 | NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0), |
| 5798 | NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0), |
| 5799 | NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0), |
| 5800 | NEONMAP1(vst1q_v, arm_neon_vst1, 0), |
| 5801 | NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0), |
| 5802 | NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0), |
| 5803 | NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0), |
| 5804 | NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0), |
| 5805 | NEONMAP1(vst2_v, arm_neon_vst2, 0), |
| 5806 | NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0), |
| 5807 | NEONMAP1(vst2q_v, arm_neon_vst2, 0), |
| 5808 | NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0), |
| 5809 | NEONMAP1(vst3_v, arm_neon_vst3, 0), |
| 5810 | NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0), |
| 5811 | NEONMAP1(vst3q_v, arm_neon_vst3, 0), |
| 5812 | NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0), |
| 5813 | NEONMAP1(vst4_v, arm_neon_vst4, 0), |
| 5814 | NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0), |
| 5815 | NEONMAP1(vst4q_v, arm_neon_vst4, 0), |
| 5816 | NEONMAP0(vsubhn_v), |
| 5817 | NEONMAP0(vtrn_v), |
| 5818 | NEONMAP0(vtrnq_v), |
| 5819 | NEONMAP0(vtst_v), |
| 5820 | NEONMAP0(vtstq_v), |
| 5821 | NEONMAP1(vusdot_v, arm_neon_usdot, 0), |
| 5822 | NEONMAP1(vusdotq_v, arm_neon_usdot, 0), |
| 5823 | NEONMAP1(vusmmlaq_v, arm_neon_usmmla, 0), |
| 5824 | NEONMAP0(vuzp_v), |
| 5825 | NEONMAP0(vuzpq_v), |
| 5826 | NEONMAP0(vzip_v), |
| 5827 | NEONMAP0(vzipq_v) |
| 5828 | }; |
| 5829 | |
| 5830 | static const ARMVectorIntrinsicInfo AArch64SIMDIntrinsicMap[] = { |
| 5831 | NEONMAP1(__a64_vcvtq_low_bf16_v, aarch64_neon_bfcvtn, 0), |
| 5832 | NEONMAP0(splat_lane_v), |
| 5833 | NEONMAP0(splat_laneq_v), |
| 5834 | NEONMAP0(splatq_lane_v), |
| 5835 | NEONMAP0(splatq_laneq_v), |
| 5836 | NEONMAP1(vabs_v, aarch64_neon_abs, 0), |
| 5837 | NEONMAP1(vabsq_v, aarch64_neon_abs, 0), |
| 5838 | NEONMAP0(vadd_v), |
| 5839 | NEONMAP0(vaddhn_v), |
| 5840 | NEONMAP0(vaddq_p128), |
| 5841 | NEONMAP0(vaddq_v), |
| 5842 | NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0), |
| 5843 | NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0), |
| 5844 | NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0), |
| 5845 | NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0), |
| 5846 | NEONMAP2(vbcaxq_v, aarch64_crypto_bcaxu, aarch64_crypto_bcaxs, Add1ArgType | UnsignedAlts), |
| 5847 | NEONMAP1(vbfdot_v, aarch64_neon_bfdot, 0), |
| 5848 | NEONMAP1(vbfdotq_v, aarch64_neon_bfdot, 0), |
| 5849 | NEONMAP1(vbfmlalbq_v, aarch64_neon_bfmlalb, 0), |
| 5850 | NEONMAP1(vbfmlaltq_v, aarch64_neon_bfmlalt, 0), |
| 5851 | NEONMAP1(vbfmmlaq_v, aarch64_neon_bfmmla, 0), |
| 5852 | NEONMAP1(vcadd_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType), |
| 5853 | NEONMAP1(vcadd_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType), |
| 5854 | NEONMAP1(vcaddq_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType), |
| 5855 | NEONMAP1(vcaddq_rot90_v, aarch64_neon_vcadd_rot90, Add1ArgType), |
| 5856 | NEONMAP1(vcage_v, aarch64_neon_facge, 0), |
| 5857 | NEONMAP1(vcageq_v, aarch64_neon_facge, 0), |
| 5858 | NEONMAP1(vcagt_v, aarch64_neon_facgt, 0), |
| 5859 | NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0), |
| 5860 | NEONMAP1(vcale_v, aarch64_neon_facge, 0), |
| 5861 | NEONMAP1(vcaleq_v, aarch64_neon_facge, 0), |
| 5862 | NEONMAP1(vcalt_v, aarch64_neon_facgt, 0), |
| 5863 | NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0), |
| 5864 | NEONMAP0(vceqz_v), |
| 5865 | NEONMAP0(vceqzq_v), |
| 5866 | NEONMAP0(vcgez_v), |
| 5867 | NEONMAP0(vcgezq_v), |
| 5868 | NEONMAP0(vcgtz_v), |
| 5869 | NEONMAP0(vcgtzq_v), |
| 5870 | NEONMAP0(vclez_v), |
| 5871 | NEONMAP0(vclezq_v), |
| 5872 | NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType), |
| 5873 | NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType), |
| 5874 | NEONMAP0(vcltz_v), |
| 5875 | NEONMAP0(vcltzq_v), |
| 5876 | NEONMAP1(vclz_v, ctlz, Add1ArgType), |
| 5877 | NEONMAP1(vclzq_v, ctlz, Add1ArgType), |
| 5878 | NEONMAP1(vcmla_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType), |
| 5879 | NEONMAP1(vcmla_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType), |
| 5880 | NEONMAP1(vcmla_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType), |
| 5881 | NEONMAP1(vcmla_v, aarch64_neon_vcmla_rot0, Add1ArgType), |
| 5882 | NEONMAP1(vcmlaq_rot180_v, aarch64_neon_vcmla_rot180, Add1ArgType), |
| 5883 | NEONMAP1(vcmlaq_rot270_v, aarch64_neon_vcmla_rot270, Add1ArgType), |
| 5884 | NEONMAP1(vcmlaq_rot90_v, aarch64_neon_vcmla_rot90, Add1ArgType), |
| 5885 | NEONMAP1(vcmlaq_v, aarch64_neon_vcmla_rot0, Add1ArgType), |
| 5886 | NEONMAP1(vcnt_v, ctpop, Add1ArgType), |
| 5887 | NEONMAP1(vcntq_v, ctpop, Add1ArgType), |
| 5888 | NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0), |
| 5889 | NEONMAP0(vcvt_f16_v), |
| 5890 | NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0), |
| 5891 | NEONMAP0(vcvt_f32_v), |
| 5892 | NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 5893 | NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 5894 | NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 5895 | NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), |
| 5896 | NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), |
| 5897 | NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), |
| 5898 | NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), |
| 5899 | NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), |
| 5900 | NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), |
| 5901 | NEONMAP0(vcvtq_f16_v), |
| 5902 | NEONMAP0(vcvtq_f32_v), |
| 5903 | NEONMAP1(vcvtq_high_bf16_v, aarch64_neon_bfcvtn2, 0), |
| 5904 | NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 5905 | NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 5906 | NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0), |
| 5907 | NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0), |
| 5908 | NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0), |
| 5909 | NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0), |
| 5910 | NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0), |
| 5911 | NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0), |
| 5912 | NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0), |
| 5913 | NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType), |
| 5914 | NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0), |
| 5915 | NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0), |
| 5916 | NEONMAP2(veor3q_v, aarch64_crypto_eor3u, aarch64_crypto_eor3s, Add1ArgType | UnsignedAlts), |
| 5917 | NEONMAP0(vext_v), |
| 5918 | NEONMAP0(vextq_v), |
| 5919 | NEONMAP0(vfma_v), |
| 5920 | NEONMAP0(vfmaq_v), |
| 5921 | NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0), |
| 5922 | NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0), |
| 5923 | NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0), |
| 5924 | NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0), |
| 5925 | NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0), |
| 5926 | NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0), |
| 5927 | NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0), |
| 5928 | NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0), |
| 5929 | NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), |
| 5930 | NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts), |
| 5931 | NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), |
| 5932 | NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts), |
| 5933 | NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0), |
| 5934 | NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0), |
| 5935 | NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0), |
| 5936 | NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0), |
| 5937 | NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0), |
| 5938 | NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0), |
| 5939 | NEONMAP2(vmmlaq_v, aarch64_neon_ummla, aarch64_neon_smmla, 0), |
| 5940 | NEONMAP0(vmovl_v), |
| 5941 | NEONMAP0(vmovn_v), |
| 5942 | NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType), |
| 5943 | NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType), |
| 5944 | NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType), |
| 5945 | NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), |
| 5946 | NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts), |
| 5947 | NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType), |
| 5948 | NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType), |
| 5949 | NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType), |
| 5950 | NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), |
| 5951 | NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts), |
| 5952 | NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0), |
| 5953 | NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0), |
| 5954 | NEONMAP1(vqdmulh_lane_v, aarch64_neon_sqdmulh_lane, 0), |
| 5955 | NEONMAP1(vqdmulh_laneq_v, aarch64_neon_sqdmulh_laneq, 0), |
| 5956 | NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType), |
| 5957 | NEONMAP1(vqdmulhq_lane_v, aarch64_neon_sqdmulh_lane, 0), |
| 5958 | NEONMAP1(vqdmulhq_laneq_v, aarch64_neon_sqdmulh_laneq, 0), |
| 5959 | NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType), |
| 5960 | NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType), |
| 5961 | NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts), |
| 5962 | NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType), |
| 5963 | NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType), |
| 5964 | NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType), |
| 5965 | NEONMAP1(vqrdmulh_lane_v, aarch64_neon_sqrdmulh_lane, 0), |
| 5966 | NEONMAP1(vqrdmulh_laneq_v, aarch64_neon_sqrdmulh_laneq, 0), |
| 5967 | NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType), |
| 5968 | NEONMAP1(vqrdmulhq_lane_v, aarch64_neon_sqrdmulh_lane, 0), |
| 5969 | NEONMAP1(vqrdmulhq_laneq_v, aarch64_neon_sqrdmulh_laneq, 0), |
| 5970 | NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType), |
| 5971 | NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), |
| 5972 | NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts), |
| 5973 | NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts), |
| 5974 | NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), |
| 5975 | NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts), |
| 5976 | NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts), |
| 5977 | NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0), |
| 5978 | NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0), |
| 5979 | NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), |
| 5980 | NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts), |
| 5981 | NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType), |
| 5982 | NEONMAP1(vrax1q_v, aarch64_crypto_rax1, 0), |
| 5983 | NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), |
| 5984 | NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0), |
| 5985 | NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType), |
| 5986 | NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType), |
| 5987 | NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), |
| 5988 | NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts), |
| 5989 | NEONMAP1(vrnd32x_v, aarch64_neon_frint32x, Add1ArgType), |
| 5990 | NEONMAP1(vrnd32xq_v, aarch64_neon_frint32x, Add1ArgType), |
| 5991 | NEONMAP1(vrnd32z_v, aarch64_neon_frint32z, Add1ArgType), |
| 5992 | NEONMAP1(vrnd32zq_v, aarch64_neon_frint32z, Add1ArgType), |
| 5993 | NEONMAP1(vrnd64x_v, aarch64_neon_frint64x, Add1ArgType), |
| 5994 | NEONMAP1(vrnd64xq_v, aarch64_neon_frint64x, Add1ArgType), |
| 5995 | NEONMAP1(vrnd64z_v, aarch64_neon_frint64z, Add1ArgType), |
| 5996 | NEONMAP1(vrnd64zq_v, aarch64_neon_frint64z, Add1ArgType), |
| 5997 | NEONMAP0(vrndi_v), |
| 5998 | NEONMAP0(vrndiq_v), |
| 5999 | NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), |
| 6000 | NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts), |
| 6001 | NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), |
| 6002 | NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts), |
| 6003 | NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), |
| 6004 | NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0), |
| 6005 | NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType), |
| 6006 | NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType), |
| 6007 | NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType), |
| 6008 | NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0), |
| 6009 | NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0), |
| 6010 | NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0), |
| 6011 | NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0), |
| 6012 | NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0), |
| 6013 | NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0), |
| 6014 | NEONMAP1(vsha512h2q_v, aarch64_crypto_sha512h2, 0), |
| 6015 | NEONMAP1(vsha512hq_v, aarch64_crypto_sha512h, 0), |
| 6016 | NEONMAP1(vsha512su0q_v, aarch64_crypto_sha512su0, 0), |
| 6017 | NEONMAP1(vsha512su1q_v, aarch64_crypto_sha512su1, 0), |
| 6018 | NEONMAP0(vshl_n_v), |
| 6019 | NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), |
| 6020 | NEONMAP0(vshll_n_v), |
| 6021 | NEONMAP0(vshlq_n_v), |
| 6022 | NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts), |
| 6023 | NEONMAP0(vshr_n_v), |
| 6024 | NEONMAP0(vshrn_n_v), |
| 6025 | NEONMAP0(vshrq_n_v), |
| 6026 | NEONMAP1(vsm3partw1q_v, aarch64_crypto_sm3partw1, 0), |
| 6027 | NEONMAP1(vsm3partw2q_v, aarch64_crypto_sm3partw2, 0), |
| 6028 | NEONMAP1(vsm3ss1q_v, aarch64_crypto_sm3ss1, 0), |
| 6029 | NEONMAP1(vsm3tt1aq_v, aarch64_crypto_sm3tt1a, 0), |
| 6030 | NEONMAP1(vsm3tt1bq_v, aarch64_crypto_sm3tt1b, 0), |
| 6031 | NEONMAP1(vsm3tt2aq_v, aarch64_crypto_sm3tt2a, 0), |
| 6032 | NEONMAP1(vsm3tt2bq_v, aarch64_crypto_sm3tt2b, 0), |
| 6033 | NEONMAP1(vsm4ekeyq_v, aarch64_crypto_sm4ekey, 0), |
| 6034 | NEONMAP1(vsm4eq_v, aarch64_crypto_sm4e, 0), |
| 6035 | NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0), |
| 6036 | NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0), |
| 6037 | NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0), |
| 6038 | NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0), |
| 6039 | NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0), |
| 6040 | NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0), |
| 6041 | NEONMAP0(vsubhn_v), |
| 6042 | NEONMAP0(vtst_v), |
| 6043 | NEONMAP0(vtstq_v), |
| 6044 | NEONMAP1(vusdot_v, aarch64_neon_usdot, 0), |
| 6045 | NEONMAP1(vusdotq_v, aarch64_neon_usdot, 0), |
| 6046 | NEONMAP1(vusmmlaq_v, aarch64_neon_usmmla, 0), |
| 6047 | NEONMAP1(vxarq_v, aarch64_crypto_xar, 0), |
| 6048 | }; |
| 6049 | |
| 6050 | static const ARMVectorIntrinsicInfo AArch64SISDIntrinsicMap[] = { |
| 6051 | NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType), |
| 6052 | NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType), |
| 6053 | NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType), |
| 6054 | NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), |
| 6055 | NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), |
| 6056 | NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType), |
| 6057 | NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType), |
| 6058 | NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), |
| 6059 | NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), |
| 6060 | NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 6061 | NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType), |
| 6062 | NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType), |
| 6063 | NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType), |
| 6064 | NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType), |
| 6065 | NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 6066 | NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 6067 | NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 6068 | NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 6069 | NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 6070 | NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 6071 | NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 6072 | NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType), |
| 6073 | NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 6074 | NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType), |
| 6075 | NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
| 6076 | NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
| 6077 | NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType), |
| 6078 | NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType), |
| 6079 | NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
| 6080 | NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
| 6081 | NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
| 6082 | NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
| 6083 | NEONMAP1(vcvtd_s64_f64, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), |
| 6084 | NEONMAP1(vcvtd_u64_f64, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), |
| 6085 | NEONMAP1(vcvth_bf16_f32, aarch64_neon_bfcvt, 0), |
| 6086 | NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
| 6087 | NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
| 6088 | NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType), |
| 6089 | NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType), |
| 6090 | NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
| 6091 | NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
| 6092 | NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType), |
| 6093 | NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType), |
| 6094 | NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
| 6095 | NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
| 6096 | NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType), |
| 6097 | NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType), |
| 6098 | NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType), |
| 6099 | NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType), |
| 6100 | NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType), |
| 6101 | NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType), |
| 6102 | NEONMAP1(vcvts_s32_f32, aarch64_neon_fcvtzs, AddRetType | Add1ArgType), |
| 6103 | NEONMAP1(vcvts_u32_f32, aarch64_neon_fcvtzu, AddRetType | Add1ArgType), |
| 6104 | NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0), |
| 6105 | NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 6106 | NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 6107 | NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 6108 | NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 6109 | NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), |
| 6110 | NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), |
| 6111 | NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 6112 | NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 6113 | NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType), |
| 6114 | NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType), |
| 6115 | NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 6116 | NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 6117 | NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 6118 | NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 6119 | NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), |
| 6120 | NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), |
| 6121 | NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 6122 | NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 6123 | NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType), |
| 6124 | NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType), |
| 6125 | NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0), |
| 6126 | NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType), |
| 6127 | NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType), |
| 6128 | NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 6129 | NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType), |
| 6130 | NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 6131 | NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType), |
| 6132 | NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 6133 | NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType), |
| 6134 | NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 6135 | NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType), |
| 6136 | NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 6137 | NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType), |
| 6138 | NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), |
| 6139 | NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType), |
| 6140 | NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors), |
| 6141 | NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType), |
| 6142 | NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), |
| 6143 | NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), |
| 6144 | NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType), |
| 6145 | NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType), |
| 6146 | NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors), |
| 6147 | NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors), |
| 6148 | NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType), |
| 6149 | NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType), |
| 6150 | NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors), |
| 6151 | NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType), |
| 6152 | NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors), |
| 6153 | NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0), |
| 6154 | NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType), |