| File: | src/gnu/usr.bin/clang/libLLVM/../../../llvm/llvm/include/llvm/IR/IRBuilder.h |
| Warning: | line 192, column 10 Called C++ object pointer is null |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===- OpenMPIRBuilder.cpp - Builder for LLVM-IR for OpenMP directives ----===// | ||||||
| 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 | /// \file | ||||||
| 9 | /// | ||||||
| 10 | /// This file implements the OpenMPIRBuilder class, which is used as a | ||||||
| 11 | /// convenient way to create LLVM instructions for OpenMP directives. | ||||||
| 12 | /// | ||||||
| 13 | //===----------------------------------------------------------------------===// | ||||||
| 14 | |||||||
| 15 | #include "llvm/Frontend/OpenMP/OMPIRBuilder.h" | ||||||
| 16 | |||||||
| 17 | #include "llvm/ADT/StringRef.h" | ||||||
| 18 | #include "llvm/ADT/Triple.h" | ||||||
| 19 | #include "llvm/IR/CFG.h" | ||||||
| 20 | #include "llvm/IR/DebugInfo.h" | ||||||
| 21 | #include "llvm/IR/IRBuilder.h" | ||||||
| 22 | #include "llvm/IR/MDBuilder.h" | ||||||
| 23 | #include "llvm/IR/Value.h" | ||||||
| 24 | #include "llvm/Support/CommandLine.h" | ||||||
| 25 | #include "llvm/Support/Error.h" | ||||||
| 26 | #include "llvm/Transforms/Utils/BasicBlockUtils.h" | ||||||
| 27 | #include "llvm/Transforms/Utils/CodeExtractor.h" | ||||||
| 28 | |||||||
| 29 | #include <sstream> | ||||||
| 30 | |||||||
| 31 | #define DEBUG_TYPE"openmp-ir-builder" "openmp-ir-builder" | ||||||
| 32 | |||||||
| 33 | using namespace llvm; | ||||||
| 34 | using namespace omp; | ||||||
| 35 | |||||||
| 36 | static cl::opt<bool> | ||||||
| 37 | OptimisticAttributes("openmp-ir-builder-optimistic-attributes", cl::Hidden, | ||||||
| 38 | cl::desc("Use optimistic attributes describing " | ||||||
| 39 | "'as-if' properties of runtime calls."), | ||||||
| 40 | cl::init(false)); | ||||||
| 41 | |||||||
| 42 | void OpenMPIRBuilder::addAttributes(omp::RuntimeFunction FnID, Function &Fn) { | ||||||
| 43 | LLVMContext &Ctx = Fn.getContext(); | ||||||
| 44 | |||||||
| 45 | // Get the function's current attributes. | ||||||
| 46 | auto Attrs = Fn.getAttributes(); | ||||||
| 47 | auto FnAttrs = Attrs.getFnAttributes(); | ||||||
| 48 | auto RetAttrs = Attrs.getRetAttributes(); | ||||||
| 49 | SmallVector<AttributeSet, 4> ArgAttrs; | ||||||
| 50 | for (size_t ArgNo = 0; ArgNo < Fn.arg_size(); ++ArgNo) | ||||||
| 51 | ArgAttrs.emplace_back(Attrs.getParamAttributes(ArgNo)); | ||||||
| 52 | |||||||
| 53 | #define OMP_ATTRS_SET(VarName, AttrSet) AttributeSet VarName = AttrSet; | ||||||
| 54 | #include "llvm/Frontend/OpenMP/OMPKinds.def" | ||||||
| 55 | |||||||
| 56 | // Add attributes to the function declaration. | ||||||
| 57 | switch (FnID) { | ||||||
| 58 | #define OMP_RTL_ATTRS(Enum, FnAttrSet, RetAttrSet, ArgAttrSets) \ | ||||||
| 59 | case Enum: \ | ||||||
| 60 | FnAttrs = FnAttrs.addAttributes(Ctx, FnAttrSet); \ | ||||||
| 61 | RetAttrs = RetAttrs.addAttributes(Ctx, RetAttrSet); \ | ||||||
| 62 | for (size_t ArgNo = 0; ArgNo < ArgAttrSets.size(); ++ArgNo) \ | ||||||
| 63 | ArgAttrs[ArgNo] = \ | ||||||
| 64 | ArgAttrs[ArgNo].addAttributes(Ctx, ArgAttrSets[ArgNo]); \ | ||||||
| 65 | Fn.setAttributes(AttributeList::get(Ctx, FnAttrs, RetAttrs, ArgAttrs)); \ | ||||||
| 66 | break; | ||||||
| 67 | #include "llvm/Frontend/OpenMP/OMPKinds.def" | ||||||
| 68 | default: | ||||||
| 69 | // Attributes are optional. | ||||||
| 70 | break; | ||||||
| 71 | } | ||||||
| 72 | } | ||||||
| 73 | |||||||
| 74 | FunctionCallee | ||||||
| 75 | OpenMPIRBuilder::getOrCreateRuntimeFunction(Module &M, RuntimeFunction FnID) { | ||||||
| 76 | FunctionType *FnTy = nullptr; | ||||||
| 77 | Function *Fn = nullptr; | ||||||
| 78 | |||||||
| 79 | // Try to find the declation in the module first. | ||||||
| 80 | switch (FnID) { | ||||||
| 81 | #define OMP_RTL(Enum, Str, IsVarArg, ReturnType, ...) \ | ||||||
| 82 | case Enum: \ | ||||||
| 83 | FnTy = FunctionType::get(ReturnType, ArrayRef<Type *>{__VA_ARGS__}, \ | ||||||
| 84 | IsVarArg); \ | ||||||
| 85 | Fn = M.getFunction(Str); \ | ||||||
| 86 | break; | ||||||
| 87 | #include "llvm/Frontend/OpenMP/OMPKinds.def" | ||||||
| 88 | } | ||||||
| 89 | |||||||
| 90 | if (!Fn) { | ||||||
| 91 | // Create a new declaration if we need one. | ||||||
| 92 | switch (FnID) { | ||||||
| 93 | #define OMP_RTL(Enum, Str, ...) \ | ||||||
| 94 | case Enum: \ | ||||||
| 95 | Fn = Function::Create(FnTy, GlobalValue::ExternalLinkage, Str, M); \ | ||||||
| 96 | break; | ||||||
| 97 | #include "llvm/Frontend/OpenMP/OMPKinds.def" | ||||||
| 98 | } | ||||||
| 99 | |||||||
| 100 | // Add information if the runtime function takes a callback function | ||||||
| 101 | if (FnID == OMPRTL___kmpc_fork_call || FnID == OMPRTL___kmpc_fork_teams) { | ||||||
| 102 | if (!Fn->hasMetadata(LLVMContext::MD_callback)) { | ||||||
| 103 | LLVMContext &Ctx = Fn->getContext(); | ||||||
| 104 | MDBuilder MDB(Ctx); | ||||||
| 105 | // Annotate the callback behavior of the runtime function: | ||||||
| 106 | // - The callback callee is argument number 2 (microtask). | ||||||
| 107 | // - The first two arguments of the callback callee are unknown (-1). | ||||||
| 108 | // - All variadic arguments to the runtime function are passed to the | ||||||
| 109 | // callback callee. | ||||||
| 110 | Fn->addMetadata( | ||||||
| 111 | LLVMContext::MD_callback, | ||||||
| 112 | *MDNode::get(Ctx, {MDB.createCallbackEncoding( | ||||||
| 113 | 2, {-1, -1}, /* VarArgsArePassed */ true)})); | ||||||
| 114 | } | ||||||
| 115 | } | ||||||
| 116 | |||||||
| 117 | LLVM_DEBUG(dbgs() << "Created OpenMP runtime function " << Fn->getName()do { } while (false) | ||||||
| 118 | << " with type " << *Fn->getFunctionType() << "\n")do { } while (false); | ||||||
| 119 | addAttributes(FnID, *Fn); | ||||||
| 120 | |||||||
| 121 | } else { | ||||||
| 122 | LLVM_DEBUG(dbgs() << "Found OpenMP runtime function " << Fn->getName()do { } while (false) | ||||||
| 123 | << " with type " << *Fn->getFunctionType() << "\n")do { } while (false); | ||||||
| 124 | } | ||||||
| 125 | |||||||
| 126 | assert(Fn && "Failed to create OpenMP runtime function")((void)0); | ||||||
| 127 | |||||||
| 128 | // Cast the function to the expected type if necessary | ||||||
| 129 | Constant *C = ConstantExpr::getBitCast(Fn, FnTy->getPointerTo()); | ||||||
| 130 | return {FnTy, C}; | ||||||
| 131 | } | ||||||
| 132 | |||||||
| 133 | Function *OpenMPIRBuilder::getOrCreateRuntimeFunctionPtr(RuntimeFunction FnID) { | ||||||
| 134 | FunctionCallee RTLFn = getOrCreateRuntimeFunction(M, FnID); | ||||||
| 135 | auto *Fn = dyn_cast<llvm::Function>(RTLFn.getCallee()); | ||||||
| 136 | assert(Fn && "Failed to create OpenMP runtime function pointer")((void)0); | ||||||
| 137 | return Fn; | ||||||
| 138 | } | ||||||
| 139 | |||||||
| 140 | void OpenMPIRBuilder::initialize() { initializeTypes(M); } | ||||||
| 141 | |||||||
| 142 | void OpenMPIRBuilder::finalize(Function *Fn, bool AllowExtractorSinking) { | ||||||
| 143 | SmallPtrSet<BasicBlock *, 32> ParallelRegionBlockSet; | ||||||
| 144 | SmallVector<BasicBlock *, 32> Blocks; | ||||||
| 145 | SmallVector<OutlineInfo, 16> DeferredOutlines; | ||||||
| 146 | for (OutlineInfo &OI : OutlineInfos) { | ||||||
| 147 | // Skip functions that have not finalized yet; may happen with nested | ||||||
| 148 | // function generation. | ||||||
| 149 | if (Fn && OI.getFunction() != Fn) { | ||||||
| 150 | DeferredOutlines.push_back(OI); | ||||||
| 151 | continue; | ||||||
| 152 | } | ||||||
| 153 | |||||||
| 154 | ParallelRegionBlockSet.clear(); | ||||||
| 155 | Blocks.clear(); | ||||||
| 156 | OI.collectBlocks(ParallelRegionBlockSet, Blocks); | ||||||
| 157 | |||||||
| 158 | Function *OuterFn = OI.getFunction(); | ||||||
| 159 | CodeExtractorAnalysisCache CEAC(*OuterFn); | ||||||
| 160 | CodeExtractor Extractor(Blocks, /* DominatorTree */ nullptr, | ||||||
| 161 | /* AggregateArgs */ false, | ||||||
| 162 | /* BlockFrequencyInfo */ nullptr, | ||||||
| 163 | /* BranchProbabilityInfo */ nullptr, | ||||||
| 164 | /* AssumptionCache */ nullptr, | ||||||
| 165 | /* AllowVarArgs */ true, | ||||||
| 166 | /* AllowAlloca */ true, | ||||||
| 167 | /* Suffix */ ".omp_par"); | ||||||
| 168 | |||||||
| 169 | LLVM_DEBUG(dbgs() << "Before outlining: " << *OuterFn << "\n")do { } while (false); | ||||||
| 170 | LLVM_DEBUG(dbgs() << "Entry " << OI.EntryBB->getName()do { } while (false) | ||||||
| 171 | << " Exit: " << OI.ExitBB->getName() << "\n")do { } while (false); | ||||||
| 172 | assert(Extractor.isEligible() &&((void)0) | ||||||
| 173 | "Expected OpenMP outlining to be possible!")((void)0); | ||||||
| 174 | |||||||
| 175 | Function *OutlinedFn = Extractor.extractCodeRegion(CEAC); | ||||||
| 176 | |||||||
| 177 | LLVM_DEBUG(dbgs() << "After outlining: " << *OuterFn << "\n")do { } while (false); | ||||||
| 178 | LLVM_DEBUG(dbgs() << " Outlined function: " << *OutlinedFn << "\n")do { } while (false); | ||||||
| 179 | assert(OutlinedFn->getReturnType()->isVoidTy() &&((void)0) | ||||||
| 180 | "OpenMP outlined functions should not return a value!")((void)0); | ||||||
| 181 | |||||||
| 182 | // For compability with the clang CG we move the outlined function after the | ||||||
| 183 | // one with the parallel region. | ||||||
| 184 | OutlinedFn->removeFromParent(); | ||||||
| 185 | M.getFunctionList().insertAfter(OuterFn->getIterator(), OutlinedFn); | ||||||
| 186 | |||||||
| 187 | // Remove the artificial entry introduced by the extractor right away, we | ||||||
| 188 | // made our own entry block after all. | ||||||
| 189 | { | ||||||
| 190 | BasicBlock &ArtificialEntry = OutlinedFn->getEntryBlock(); | ||||||
| 191 | assert(ArtificialEntry.getUniqueSuccessor() == OI.EntryBB)((void)0); | ||||||
| 192 | assert(OI.EntryBB->getUniquePredecessor() == &ArtificialEntry)((void)0); | ||||||
| 193 | if (AllowExtractorSinking) { | ||||||
| 194 | // Move instructions from the to-be-deleted ArtificialEntry to the entry | ||||||
| 195 | // basic block of the parallel region. CodeExtractor may have sunk | ||||||
| 196 | // allocas/bitcasts for values that are solely used in the outlined | ||||||
| 197 | // region and do not escape. | ||||||
| 198 | assert(!ArtificialEntry.empty() &&((void)0) | ||||||
| 199 | "Expected instructions to sink in the outlined region")((void)0); | ||||||
| 200 | for (BasicBlock::iterator It = ArtificialEntry.begin(), | ||||||
| 201 | End = ArtificialEntry.end(); | ||||||
| 202 | It != End;) { | ||||||
| 203 | Instruction &I = *It; | ||||||
| 204 | It++; | ||||||
| 205 | |||||||
| 206 | if (I.isTerminator()) | ||||||
| 207 | continue; | ||||||
| 208 | |||||||
| 209 | I.moveBefore(*OI.EntryBB, OI.EntryBB->getFirstInsertionPt()); | ||||||
| 210 | } | ||||||
| 211 | } | ||||||
| 212 | OI.EntryBB->moveBefore(&ArtificialEntry); | ||||||
| 213 | ArtificialEntry.eraseFromParent(); | ||||||
| 214 | } | ||||||
| 215 | assert(&OutlinedFn->getEntryBlock() == OI.EntryBB)((void)0); | ||||||
| 216 | assert(OutlinedFn && OutlinedFn->getNumUses() == 1)((void)0); | ||||||
| 217 | |||||||
| 218 | // Run a user callback, e.g. to add attributes. | ||||||
| 219 | if (OI.PostOutlineCB) | ||||||
| 220 | OI.PostOutlineCB(*OutlinedFn); | ||||||
| 221 | } | ||||||
| 222 | |||||||
| 223 | // Remove work items that have been completed. | ||||||
| 224 | OutlineInfos = std::move(DeferredOutlines); | ||||||
| 225 | } | ||||||
| 226 | |||||||
| 227 | OpenMPIRBuilder::~OpenMPIRBuilder() { | ||||||
| 228 | assert(OutlineInfos.empty() && "There must be no outstanding outlinings")((void)0); | ||||||
| 229 | } | ||||||
| 230 | |||||||
| 231 | Value *OpenMPIRBuilder::getOrCreateIdent(Constant *SrcLocStr, | ||||||
| 232 | IdentFlag LocFlags, | ||||||
| 233 | unsigned Reserve2Flags) { | ||||||
| 234 | // Enable "C-mode". | ||||||
| 235 | LocFlags |= OMP_IDENT_FLAG_KMPC; | ||||||
| 236 | |||||||
| 237 | Value *&Ident = | ||||||
| 238 | IdentMap[{SrcLocStr, uint64_t(LocFlags) << 31 | Reserve2Flags}]; | ||||||
| 239 | if (!Ident) { | ||||||
| 240 | Constant *I32Null = ConstantInt::getNullValue(Int32); | ||||||
| 241 | Constant *IdentData[] = { | ||||||
| 242 | I32Null, ConstantInt::get(Int32, uint32_t(LocFlags)), | ||||||
| 243 | ConstantInt::get(Int32, Reserve2Flags), I32Null, SrcLocStr}; | ||||||
| 244 | Constant *Initializer = ConstantStruct::get( | ||||||
| 245 | cast<StructType>(IdentPtr->getPointerElementType()), IdentData); | ||||||
| 246 | |||||||
| 247 | // Look for existing encoding of the location + flags, not needed but | ||||||
| 248 | // minimizes the difference to the existing solution while we transition. | ||||||
| 249 | for (GlobalVariable &GV : M.getGlobalList()) | ||||||
| 250 | if (GV.getType() == IdentPtr && GV.hasInitializer()) | ||||||
| 251 | if (GV.getInitializer() == Initializer) | ||||||
| 252 | return Ident = &GV; | ||||||
| 253 | |||||||
| 254 | auto *GV = new GlobalVariable(M, IdentPtr->getPointerElementType(), | ||||||
| 255 | /* isConstant = */ true, | ||||||
| 256 | GlobalValue::PrivateLinkage, Initializer); | ||||||
| 257 | GV->setUnnamedAddr(GlobalValue::UnnamedAddr::Global); | ||||||
| 258 | GV->setAlignment(Align(8)); | ||||||
| 259 | Ident = GV; | ||||||
| 260 | } | ||||||
| 261 | return Builder.CreatePointerCast(Ident, IdentPtr); | ||||||
| 262 | } | ||||||
| 263 | |||||||
| 264 | Type *OpenMPIRBuilder::getLanemaskType() { | ||||||
| 265 | LLVMContext &Ctx = M.getContext(); | ||||||
| 266 | Triple triple(M.getTargetTriple()); | ||||||
| 267 | |||||||
| 268 | // This test is adequate until deviceRTL has finer grained lane widths | ||||||
| 269 | return triple.isAMDGCN() ? Type::getInt64Ty(Ctx) : Type::getInt32Ty(Ctx); | ||||||
| 270 | } | ||||||
| 271 | |||||||
| 272 | Constant *OpenMPIRBuilder::getOrCreateSrcLocStr(StringRef LocStr) { | ||||||
| 273 | Constant *&SrcLocStr = SrcLocStrMap[LocStr]; | ||||||
| 274 | if (!SrcLocStr) { | ||||||
| 275 | Constant *Initializer = | ||||||
| 276 | ConstantDataArray::getString(M.getContext(), LocStr); | ||||||
| 277 | |||||||
| 278 | // Look for existing encoding of the location, not needed but minimizes the | ||||||
| 279 | // difference to the existing solution while we transition. | ||||||
| 280 | for (GlobalVariable &GV : M.getGlobalList()) | ||||||
| 281 | if (GV.isConstant() && GV.hasInitializer() && | ||||||
| 282 | GV.getInitializer() == Initializer) | ||||||
| 283 | return SrcLocStr = ConstantExpr::getPointerCast(&GV, Int8Ptr); | ||||||
| 284 | |||||||
| 285 | SrcLocStr = Builder.CreateGlobalStringPtr(LocStr, /* Name */ "", | ||||||
| 286 | /* AddressSpace */ 0, &M); | ||||||
| 287 | } | ||||||
| 288 | return SrcLocStr; | ||||||
| 289 | } | ||||||
| 290 | |||||||
| 291 | Constant *OpenMPIRBuilder::getOrCreateSrcLocStr(StringRef FunctionName, | ||||||
| 292 | StringRef FileName, | ||||||
| 293 | unsigned Line, | ||||||
| 294 | unsigned Column) { | ||||||
| 295 | SmallString<128> Buffer; | ||||||
| 296 | Buffer.push_back(';'); | ||||||
| 297 | Buffer.append(FileName); | ||||||
| 298 | Buffer.push_back(';'); | ||||||
| 299 | Buffer.append(FunctionName); | ||||||
| 300 | Buffer.push_back(';'); | ||||||
| 301 | Buffer.append(std::to_string(Line)); | ||||||
| 302 | Buffer.push_back(';'); | ||||||
| 303 | Buffer.append(std::to_string(Column)); | ||||||
| 304 | Buffer.push_back(';'); | ||||||
| 305 | Buffer.push_back(';'); | ||||||
| 306 | return getOrCreateSrcLocStr(Buffer.str()); | ||||||
| 307 | } | ||||||
| 308 | |||||||
| 309 | Constant *OpenMPIRBuilder::getOrCreateDefaultSrcLocStr() { | ||||||
| 310 | return getOrCreateSrcLocStr(";unknown;unknown;0;0;;"); | ||||||
| 311 | } | ||||||
| 312 | |||||||
| 313 | Constant * | ||||||
| 314 | OpenMPIRBuilder::getOrCreateSrcLocStr(const LocationDescription &Loc) { | ||||||
| 315 | DILocation *DIL = Loc.DL.get(); | ||||||
| 316 | if (!DIL) | ||||||
| 317 | return getOrCreateDefaultSrcLocStr(); | ||||||
| 318 | StringRef FileName = M.getName(); | ||||||
| 319 | if (DIFile *DIF = DIL->getFile()) | ||||||
| 320 | if (Optional<StringRef> Source = DIF->getSource()) | ||||||
| 321 | FileName = *Source; | ||||||
| 322 | StringRef Function = DIL->getScope()->getSubprogram()->getName(); | ||||||
| 323 | Function = | ||||||
| 324 | !Function.empty() ? Function : Loc.IP.getBlock()->getParent()->getName(); | ||||||
| 325 | return getOrCreateSrcLocStr(Function, FileName, DIL->getLine(), | ||||||
| 326 | DIL->getColumn()); | ||||||
| 327 | } | ||||||
| 328 | |||||||
| 329 | Value *OpenMPIRBuilder::getOrCreateThreadID(Value *Ident) { | ||||||
| 330 | return Builder.CreateCall( | ||||||
| 331 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_global_thread_num), Ident, | ||||||
| 332 | "omp_global_thread_num"); | ||||||
| 333 | } | ||||||
| 334 | |||||||
| 335 | OpenMPIRBuilder::InsertPointTy | ||||||
| 336 | OpenMPIRBuilder::createBarrier(const LocationDescription &Loc, Directive DK, | ||||||
| 337 | bool ForceSimpleCall, bool CheckCancelFlag) { | ||||||
| 338 | if (!updateToLocation(Loc)) | ||||||
| 339 | return Loc.IP; | ||||||
| 340 | return emitBarrierImpl(Loc, DK, ForceSimpleCall, CheckCancelFlag); | ||||||
| 341 | } | ||||||
| 342 | |||||||
| 343 | OpenMPIRBuilder::InsertPointTy | ||||||
| 344 | OpenMPIRBuilder::emitBarrierImpl(const LocationDescription &Loc, Directive Kind, | ||||||
| 345 | bool ForceSimpleCall, bool CheckCancelFlag) { | ||||||
| 346 | // Build call __kmpc_cancel_barrier(loc, thread_id) or | ||||||
| 347 | // __kmpc_barrier(loc, thread_id); | ||||||
| 348 | |||||||
| 349 | IdentFlag BarrierLocFlags; | ||||||
| 350 | switch (Kind) { | ||||||
| 351 | case OMPD_for: | ||||||
| 352 | BarrierLocFlags = OMP_IDENT_FLAG_BARRIER_IMPL_FOR; | ||||||
| 353 | break; | ||||||
| 354 | case OMPD_sections: | ||||||
| 355 | BarrierLocFlags = OMP_IDENT_FLAG_BARRIER_IMPL_SECTIONS; | ||||||
| 356 | break; | ||||||
| 357 | case OMPD_single: | ||||||
| 358 | BarrierLocFlags = OMP_IDENT_FLAG_BARRIER_IMPL_SINGLE; | ||||||
| 359 | break; | ||||||
| 360 | case OMPD_barrier: | ||||||
| 361 | BarrierLocFlags = OMP_IDENT_FLAG_BARRIER_EXPL; | ||||||
| 362 | break; | ||||||
| 363 | default: | ||||||
| 364 | BarrierLocFlags = OMP_IDENT_FLAG_BARRIER_IMPL; | ||||||
| 365 | break; | ||||||
| 366 | } | ||||||
| 367 | |||||||
| 368 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 369 | Value *Args[] = {getOrCreateIdent(SrcLocStr, BarrierLocFlags), | ||||||
| 370 | getOrCreateThreadID(getOrCreateIdent(SrcLocStr))}; | ||||||
| 371 | |||||||
| 372 | // If we are in a cancellable parallel region, barriers are cancellation | ||||||
| 373 | // points. | ||||||
| 374 | // TODO: Check why we would force simple calls or to ignore the cancel flag. | ||||||
| 375 | bool UseCancelBarrier = | ||||||
| 376 | !ForceSimpleCall && isLastFinalizationInfoCancellable(OMPD_parallel); | ||||||
| 377 | |||||||
| 378 | Value *Result = | ||||||
| 379 | Builder.CreateCall(getOrCreateRuntimeFunctionPtr( | ||||||
| 380 | UseCancelBarrier ? OMPRTL___kmpc_cancel_barrier | ||||||
| 381 | : OMPRTL___kmpc_barrier), | ||||||
| 382 | Args); | ||||||
| 383 | |||||||
| 384 | if (UseCancelBarrier && CheckCancelFlag) | ||||||
| 385 | emitCancelationCheckImpl(Result, OMPD_parallel); | ||||||
| 386 | |||||||
| 387 | return Builder.saveIP(); | ||||||
| 388 | } | ||||||
| 389 | |||||||
| 390 | OpenMPIRBuilder::InsertPointTy | ||||||
| 391 | OpenMPIRBuilder::createCancel(const LocationDescription &Loc, | ||||||
| 392 | Value *IfCondition, | ||||||
| 393 | omp::Directive CanceledDirective) { | ||||||
| 394 | if (!updateToLocation(Loc)) | ||||||
| 395 | return Loc.IP; | ||||||
| 396 | |||||||
| 397 | // LLVM utilities like blocks with terminators. | ||||||
| 398 | auto *UI = Builder.CreateUnreachable(); | ||||||
| 399 | |||||||
| 400 | Instruction *ThenTI = UI, *ElseTI = nullptr; | ||||||
| 401 | if (IfCondition) | ||||||
| 402 | SplitBlockAndInsertIfThenElse(IfCondition, UI, &ThenTI, &ElseTI); | ||||||
| 403 | Builder.SetInsertPoint(ThenTI); | ||||||
| 404 | |||||||
| 405 | Value *CancelKind = nullptr; | ||||||
| 406 | switch (CanceledDirective) { | ||||||
| 407 | #define OMP_CANCEL_KIND(Enum, Str, DirectiveEnum, Value) \ | ||||||
| 408 | case DirectiveEnum: \ | ||||||
| 409 | CancelKind = Builder.getInt32(Value); \ | ||||||
| 410 | break; | ||||||
| 411 | #include "llvm/Frontend/OpenMP/OMPKinds.def" | ||||||
| 412 | default: | ||||||
| 413 | llvm_unreachable("Unknown cancel kind!")__builtin_unreachable(); | ||||||
| 414 | } | ||||||
| 415 | |||||||
| 416 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 417 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 418 | Value *Args[] = {Ident, getOrCreateThreadID(Ident), CancelKind}; | ||||||
| 419 | Value *Result = Builder.CreateCall( | ||||||
| 420 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_cancel), Args); | ||||||
| 421 | auto ExitCB = [this, CanceledDirective, Loc](InsertPointTy IP) { | ||||||
| 422 | if (CanceledDirective == OMPD_parallel) { | ||||||
| 423 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 424 | Builder.restoreIP(IP); | ||||||
| 425 | createBarrier(LocationDescription(Builder.saveIP(), Loc.DL), | ||||||
| 426 | omp::Directive::OMPD_unknown, /* ForceSimpleCall */ false, | ||||||
| 427 | /* CheckCancelFlag */ false); | ||||||
| 428 | } | ||||||
| 429 | }; | ||||||
| 430 | |||||||
| 431 | // The actual cancel logic is shared with others, e.g., cancel_barriers. | ||||||
| 432 | emitCancelationCheckImpl(Result, CanceledDirective, ExitCB); | ||||||
| 433 | |||||||
| 434 | // Update the insertion point and remove the terminator we introduced. | ||||||
| 435 | Builder.SetInsertPoint(UI->getParent()); | ||||||
| 436 | UI->eraseFromParent(); | ||||||
| 437 | |||||||
| 438 | return Builder.saveIP(); | ||||||
| 439 | } | ||||||
| 440 | |||||||
| 441 | void OpenMPIRBuilder::emitCancelationCheckImpl(Value *CancelFlag, | ||||||
| 442 | omp::Directive CanceledDirective, | ||||||
| 443 | FinalizeCallbackTy ExitCB) { | ||||||
| 444 | assert(isLastFinalizationInfoCancellable(CanceledDirective) &&((void)0) | ||||||
| 445 | "Unexpected cancellation!")((void)0); | ||||||
| 446 | |||||||
| 447 | // For a cancel barrier we create two new blocks. | ||||||
| 448 | BasicBlock *BB = Builder.GetInsertBlock(); | ||||||
| 449 | BasicBlock *NonCancellationBlock; | ||||||
| 450 | if (Builder.GetInsertPoint() == BB->end()) { | ||||||
| 451 | // TODO: This branch will not be needed once we moved to the | ||||||
| 452 | // OpenMPIRBuilder codegen completely. | ||||||
| 453 | NonCancellationBlock = BasicBlock::Create( | ||||||
| 454 | BB->getContext(), BB->getName() + ".cont", BB->getParent()); | ||||||
| 455 | } else { | ||||||
| 456 | NonCancellationBlock = SplitBlock(BB, &*Builder.GetInsertPoint()); | ||||||
| 457 | BB->getTerminator()->eraseFromParent(); | ||||||
| 458 | Builder.SetInsertPoint(BB); | ||||||
| 459 | } | ||||||
| 460 | BasicBlock *CancellationBlock = BasicBlock::Create( | ||||||
| 461 | BB->getContext(), BB->getName() + ".cncl", BB->getParent()); | ||||||
| 462 | |||||||
| 463 | // Jump to them based on the return value. | ||||||
| 464 | Value *Cmp = Builder.CreateIsNull(CancelFlag); | ||||||
| 465 | Builder.CreateCondBr(Cmp, NonCancellationBlock, CancellationBlock, | ||||||
| 466 | /* TODO weight */ nullptr, nullptr); | ||||||
| 467 | |||||||
| 468 | // From the cancellation block we finalize all variables and go to the | ||||||
| 469 | // post finalization block that is known to the FiniCB callback. | ||||||
| 470 | Builder.SetInsertPoint(CancellationBlock); | ||||||
| 471 | if (ExitCB) | ||||||
| 472 | ExitCB(Builder.saveIP()); | ||||||
| 473 | auto &FI = FinalizationStack.back(); | ||||||
| 474 | FI.FiniCB(Builder.saveIP()); | ||||||
| 475 | |||||||
| 476 | // The continuation block is where code generation continues. | ||||||
| 477 | Builder.SetInsertPoint(NonCancellationBlock, NonCancellationBlock->begin()); | ||||||
| 478 | } | ||||||
| 479 | |||||||
| 480 | IRBuilder<>::InsertPoint OpenMPIRBuilder::createParallel( | ||||||
| 481 | const LocationDescription &Loc, InsertPointTy OuterAllocaIP, | ||||||
| 482 | BodyGenCallbackTy BodyGenCB, PrivatizeCallbackTy PrivCB, | ||||||
| 483 | FinalizeCallbackTy FiniCB, Value *IfCondition, Value *NumThreads, | ||||||
| 484 | omp::ProcBindKind ProcBind, bool IsCancellable) { | ||||||
| 485 | if (!updateToLocation(Loc)) | ||||||
| 486 | return Loc.IP; | ||||||
| 487 | |||||||
| 488 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 489 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 490 | Value *ThreadID = getOrCreateThreadID(Ident); | ||||||
| 491 | |||||||
| 492 | if (NumThreads) { | ||||||
| 493 | // Build call __kmpc_push_num_threads(&Ident, global_tid, num_threads) | ||||||
| 494 | Value *Args[] = { | ||||||
| 495 | Ident, ThreadID, | ||||||
| 496 | Builder.CreateIntCast(NumThreads, Int32, /*isSigned*/ false)}; | ||||||
| 497 | Builder.CreateCall( | ||||||
| 498 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_push_num_threads), Args); | ||||||
| 499 | } | ||||||
| 500 | |||||||
| 501 | if (ProcBind != OMP_PROC_BIND_default) { | ||||||
| 502 | // Build call __kmpc_push_proc_bind(&Ident, global_tid, proc_bind) | ||||||
| 503 | Value *Args[] = { | ||||||
| 504 | Ident, ThreadID, | ||||||
| 505 | ConstantInt::get(Int32, unsigned(ProcBind), /*isSigned=*/true)}; | ||||||
| 506 | Builder.CreateCall( | ||||||
| 507 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_push_proc_bind), Args); | ||||||
| 508 | } | ||||||
| 509 | |||||||
| 510 | BasicBlock *InsertBB = Builder.GetInsertBlock(); | ||||||
| 511 | Function *OuterFn = InsertBB->getParent(); | ||||||
| 512 | |||||||
| 513 | // Save the outer alloca block because the insertion iterator may get | ||||||
| 514 | // invalidated and we still need this later. | ||||||
| 515 | BasicBlock *OuterAllocaBlock = OuterAllocaIP.getBlock(); | ||||||
| 516 | |||||||
| 517 | // Vector to remember instructions we used only during the modeling but which | ||||||
| 518 | // we want to delete at the end. | ||||||
| 519 | SmallVector<Instruction *, 4> ToBeDeleted; | ||||||
| 520 | |||||||
| 521 | // Change the location to the outer alloca insertion point to create and | ||||||
| 522 | // initialize the allocas we pass into the parallel region. | ||||||
| 523 | Builder.restoreIP(OuterAllocaIP); | ||||||
| 524 | AllocaInst *TIDAddr = Builder.CreateAlloca(Int32, nullptr, "tid.addr"); | ||||||
| 525 | AllocaInst *ZeroAddr = Builder.CreateAlloca(Int32, nullptr, "zero.addr"); | ||||||
| 526 | |||||||
| 527 | // If there is an if condition we actually use the TIDAddr and ZeroAddr in the | ||||||
| 528 | // program, otherwise we only need them for modeling purposes to get the | ||||||
| 529 | // associated arguments in the outlined function. In the former case, | ||||||
| 530 | // initialize the allocas properly, in the latter case, delete them later. | ||||||
| 531 | if (IfCondition) { | ||||||
| 532 | Builder.CreateStore(Constant::getNullValue(Int32), TIDAddr); | ||||||
| 533 | Builder.CreateStore(Constant::getNullValue(Int32), ZeroAddr); | ||||||
| 534 | } else { | ||||||
| 535 | ToBeDeleted.push_back(TIDAddr); | ||||||
| 536 | ToBeDeleted.push_back(ZeroAddr); | ||||||
| 537 | } | ||||||
| 538 | |||||||
| 539 | // Create an artificial insertion point that will also ensure the blocks we | ||||||
| 540 | // are about to split are not degenerated. | ||||||
| 541 | auto *UI = new UnreachableInst(Builder.getContext(), InsertBB); | ||||||
| 542 | |||||||
| 543 | Instruction *ThenTI = UI, *ElseTI = nullptr; | ||||||
| 544 | if (IfCondition) | ||||||
| 545 | SplitBlockAndInsertIfThenElse(IfCondition, UI, &ThenTI, &ElseTI); | ||||||
| 546 | |||||||
| 547 | BasicBlock *ThenBB = ThenTI->getParent(); | ||||||
| 548 | BasicBlock *PRegEntryBB = ThenBB->splitBasicBlock(ThenTI, "omp.par.entry"); | ||||||
| 549 | BasicBlock *PRegBodyBB = | ||||||
| 550 | PRegEntryBB->splitBasicBlock(ThenTI, "omp.par.region"); | ||||||
| 551 | BasicBlock *PRegPreFiniBB = | ||||||
| 552 | PRegBodyBB->splitBasicBlock(ThenTI, "omp.par.pre_finalize"); | ||||||
| 553 | BasicBlock *PRegExitBB = | ||||||
| 554 | PRegPreFiniBB->splitBasicBlock(ThenTI, "omp.par.exit"); | ||||||
| 555 | |||||||
| 556 | auto FiniCBWrapper = [&](InsertPointTy IP) { | ||||||
| 557 | // Hide "open-ended" blocks from the given FiniCB by setting the right jump | ||||||
| 558 | // target to the region exit block. | ||||||
| 559 | if (IP.getBlock()->end() == IP.getPoint()) { | ||||||
| 560 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 561 | Builder.restoreIP(IP); | ||||||
| 562 | Instruction *I = Builder.CreateBr(PRegExitBB); | ||||||
| 563 | IP = InsertPointTy(I->getParent(), I->getIterator()); | ||||||
| 564 | } | ||||||
| 565 | assert(IP.getBlock()->getTerminator()->getNumSuccessors() == 1 &&((void)0) | ||||||
| 566 | IP.getBlock()->getTerminator()->getSuccessor(0) == PRegExitBB &&((void)0) | ||||||
| 567 | "Unexpected insertion point for finalization call!")((void)0); | ||||||
| 568 | return FiniCB(IP); | ||||||
| 569 | }; | ||||||
| 570 | |||||||
| 571 | FinalizationStack.push_back({FiniCBWrapper, OMPD_parallel, IsCancellable}); | ||||||
| 572 | |||||||
| 573 | // Generate the privatization allocas in the block that will become the entry | ||||||
| 574 | // of the outlined function. | ||||||
| 575 | Builder.SetInsertPoint(PRegEntryBB->getTerminator()); | ||||||
| 576 | InsertPointTy InnerAllocaIP = Builder.saveIP(); | ||||||
| 577 | |||||||
| 578 | AllocaInst *PrivTIDAddr = | ||||||
| 579 | Builder.CreateAlloca(Int32, nullptr, "tid.addr.local"); | ||||||
| 580 | Instruction *PrivTID = Builder.CreateLoad(Int32, PrivTIDAddr, "tid"); | ||||||
| 581 | |||||||
| 582 | // Add some fake uses for OpenMP provided arguments. | ||||||
| 583 | ToBeDeleted.push_back(Builder.CreateLoad(Int32, TIDAddr, "tid.addr.use")); | ||||||
| 584 | Instruction *ZeroAddrUse = Builder.CreateLoad(Int32, ZeroAddr, | ||||||
| 585 | "zero.addr.use"); | ||||||
| 586 | ToBeDeleted.push_back(ZeroAddrUse); | ||||||
| 587 | |||||||
| 588 | // ThenBB | ||||||
| 589 | // | | ||||||
| 590 | // V | ||||||
| 591 | // PRegionEntryBB <- Privatization allocas are placed here. | ||||||
| 592 | // | | ||||||
| 593 | // V | ||||||
| 594 | // PRegionBodyBB <- BodeGen is invoked here. | ||||||
| 595 | // | | ||||||
| 596 | // V | ||||||
| 597 | // PRegPreFiniBB <- The block we will start finalization from. | ||||||
| 598 | // | | ||||||
| 599 | // V | ||||||
| 600 | // PRegionExitBB <- A common exit to simplify block collection. | ||||||
| 601 | // | ||||||
| 602 | |||||||
| 603 | LLVM_DEBUG(dbgs() << "Before body codegen: " << *OuterFn << "\n")do { } while (false); | ||||||
| 604 | |||||||
| 605 | // Let the caller create the body. | ||||||
| 606 | assert(BodyGenCB && "Expected body generation callback!")((void)0); | ||||||
| 607 | InsertPointTy CodeGenIP(PRegBodyBB, PRegBodyBB->begin()); | ||||||
| 608 | BodyGenCB(InnerAllocaIP, CodeGenIP, *PRegPreFiniBB); | ||||||
| 609 | |||||||
| 610 | LLVM_DEBUG(dbgs() << "After body codegen: " << *OuterFn << "\n")do { } while (false); | ||||||
| 611 | |||||||
| 612 | FunctionCallee RTLFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_fork_call); | ||||||
| 613 | if (auto *F = dyn_cast<llvm::Function>(RTLFn.getCallee())) { | ||||||
| 614 | if (!F->hasMetadata(llvm::LLVMContext::MD_callback)) { | ||||||
| 615 | llvm::LLVMContext &Ctx = F->getContext(); | ||||||
| 616 | MDBuilder MDB(Ctx); | ||||||
| 617 | // Annotate the callback behavior of the __kmpc_fork_call: | ||||||
| 618 | // - The callback callee is argument number 2 (microtask). | ||||||
| 619 | // - The first two arguments of the callback callee are unknown (-1). | ||||||
| 620 | // - All variadic arguments to the __kmpc_fork_call are passed to the | ||||||
| 621 | // callback callee. | ||||||
| 622 | F->addMetadata( | ||||||
| 623 | llvm::LLVMContext::MD_callback, | ||||||
| 624 | *llvm::MDNode::get( | ||||||
| 625 | Ctx, {MDB.createCallbackEncoding(2, {-1, -1}, | ||||||
| 626 | /* VarArgsArePassed */ true)})); | ||||||
| 627 | } | ||||||
| 628 | } | ||||||
| 629 | |||||||
| 630 | OutlineInfo OI; | ||||||
| 631 | OI.PostOutlineCB = [=](Function &OutlinedFn) { | ||||||
| 632 | // Add some known attributes. | ||||||
| 633 | OutlinedFn.addParamAttr(0, Attribute::NoAlias); | ||||||
| 634 | OutlinedFn.addParamAttr(1, Attribute::NoAlias); | ||||||
| 635 | OutlinedFn.addFnAttr(Attribute::NoUnwind); | ||||||
| 636 | OutlinedFn.addFnAttr(Attribute::NoRecurse); | ||||||
| 637 | |||||||
| 638 | assert(OutlinedFn.arg_size() >= 2 &&((void)0) | ||||||
| 639 | "Expected at least tid and bounded tid as arguments")((void)0); | ||||||
| 640 | unsigned NumCapturedVars = | ||||||
| 641 | OutlinedFn.arg_size() - /* tid & bounded tid */ 2; | ||||||
| 642 | |||||||
| 643 | CallInst *CI = cast<CallInst>(OutlinedFn.user_back()); | ||||||
| 644 | CI->getParent()->setName("omp_parallel"); | ||||||
| 645 | Builder.SetInsertPoint(CI); | ||||||
| 646 | |||||||
| 647 | // Build call __kmpc_fork_call(Ident, n, microtask, var1, .., varn); | ||||||
| 648 | Value *ForkCallArgs[] = { | ||||||
| 649 | Ident, Builder.getInt32(NumCapturedVars), | ||||||
| 650 | Builder.CreateBitCast(&OutlinedFn, ParallelTaskPtr)}; | ||||||
| 651 | |||||||
| 652 | SmallVector<Value *, 16> RealArgs; | ||||||
| 653 | RealArgs.append(std::begin(ForkCallArgs), std::end(ForkCallArgs)); | ||||||
| 654 | RealArgs.append(CI->arg_begin() + /* tid & bound tid */ 2, CI->arg_end()); | ||||||
| 655 | |||||||
| 656 | Builder.CreateCall(RTLFn, RealArgs); | ||||||
| 657 | |||||||
| 658 | LLVM_DEBUG(dbgs() << "With fork_call placed: "do { } while (false) | ||||||
| 659 | << *Builder.GetInsertBlock()->getParent() << "\n")do { } while (false); | ||||||
| 660 | |||||||
| 661 | InsertPointTy ExitIP(PRegExitBB, PRegExitBB->end()); | ||||||
| 662 | |||||||
| 663 | // Initialize the local TID stack location with the argument value. | ||||||
| 664 | Builder.SetInsertPoint(PrivTID); | ||||||
| 665 | Function::arg_iterator OutlinedAI = OutlinedFn.arg_begin(); | ||||||
| 666 | Builder.CreateStore(Builder.CreateLoad(Int32, OutlinedAI), PrivTIDAddr); | ||||||
| 667 | |||||||
| 668 | // If no "if" clause was present we do not need the call created during | ||||||
| 669 | // outlining, otherwise we reuse it in the serialized parallel region. | ||||||
| 670 | if (!ElseTI) { | ||||||
| 671 | CI->eraseFromParent(); | ||||||
| 672 | } else { | ||||||
| 673 | |||||||
| 674 | // If an "if" clause was present we are now generating the serialized | ||||||
| 675 | // version into the "else" branch. | ||||||
| 676 | Builder.SetInsertPoint(ElseTI); | ||||||
| 677 | |||||||
| 678 | // Build calls __kmpc_serialized_parallel(&Ident, GTid); | ||||||
| 679 | Value *SerializedParallelCallArgs[] = {Ident, ThreadID}; | ||||||
| 680 | Builder.CreateCall( | ||||||
| 681 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_serialized_parallel), | ||||||
| 682 | SerializedParallelCallArgs); | ||||||
| 683 | |||||||
| 684 | // OutlinedFn(>id, &zero, CapturedStruct); | ||||||
| 685 | CI->removeFromParent(); | ||||||
| 686 | Builder.Insert(CI); | ||||||
| 687 | |||||||
| 688 | // __kmpc_end_serialized_parallel(&Ident, GTid); | ||||||
| 689 | Value *EndArgs[] = {Ident, ThreadID}; | ||||||
| 690 | Builder.CreateCall( | ||||||
| 691 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_end_serialized_parallel), | ||||||
| 692 | EndArgs); | ||||||
| 693 | |||||||
| 694 | LLVM_DEBUG(dbgs() << "With serialized parallel region: "do { } while (false) | ||||||
| 695 | << *Builder.GetInsertBlock()->getParent() << "\n")do { } while (false); | ||||||
| 696 | } | ||||||
| 697 | |||||||
| 698 | for (Instruction *I : ToBeDeleted) | ||||||
| 699 | I->eraseFromParent(); | ||||||
| 700 | }; | ||||||
| 701 | |||||||
| 702 | // Adjust the finalization stack, verify the adjustment, and call the | ||||||
| 703 | // finalize function a last time to finalize values between the pre-fini | ||||||
| 704 | // block and the exit block if we left the parallel "the normal way". | ||||||
| 705 | auto FiniInfo = FinalizationStack.pop_back_val(); | ||||||
| 706 | (void)FiniInfo; | ||||||
| 707 | assert(FiniInfo.DK == OMPD_parallel &&((void)0) | ||||||
| 708 | "Unexpected finalization stack state!")((void)0); | ||||||
| 709 | |||||||
| 710 | Instruction *PRegPreFiniTI = PRegPreFiniBB->getTerminator(); | ||||||
| 711 | |||||||
| 712 | InsertPointTy PreFiniIP(PRegPreFiniBB, PRegPreFiniTI->getIterator()); | ||||||
| 713 | FiniCB(PreFiniIP); | ||||||
| 714 | |||||||
| 715 | OI.EntryBB = PRegEntryBB; | ||||||
| 716 | OI.ExitBB = PRegExitBB; | ||||||
| 717 | |||||||
| 718 | SmallPtrSet<BasicBlock *, 32> ParallelRegionBlockSet; | ||||||
| 719 | SmallVector<BasicBlock *, 32> Blocks; | ||||||
| 720 | OI.collectBlocks(ParallelRegionBlockSet, Blocks); | ||||||
| 721 | |||||||
| 722 | // Ensure a single exit node for the outlined region by creating one. | ||||||
| 723 | // We might have multiple incoming edges to the exit now due to finalizations, | ||||||
| 724 | // e.g., cancel calls that cause the control flow to leave the region. | ||||||
| 725 | BasicBlock *PRegOutlinedExitBB = PRegExitBB; | ||||||
| 726 | PRegExitBB = SplitBlock(PRegExitBB, &*PRegExitBB->getFirstInsertionPt()); | ||||||
| 727 | PRegOutlinedExitBB->setName("omp.par.outlined.exit"); | ||||||
| 728 | Blocks.push_back(PRegOutlinedExitBB); | ||||||
| 729 | |||||||
| 730 | CodeExtractorAnalysisCache CEAC(*OuterFn); | ||||||
| 731 | CodeExtractor Extractor(Blocks, /* DominatorTree */ nullptr, | ||||||
| 732 | /* AggregateArgs */ false, | ||||||
| 733 | /* BlockFrequencyInfo */ nullptr, | ||||||
| 734 | /* BranchProbabilityInfo */ nullptr, | ||||||
| 735 | /* AssumptionCache */ nullptr, | ||||||
| 736 | /* AllowVarArgs */ true, | ||||||
| 737 | /* AllowAlloca */ true, | ||||||
| 738 | /* Suffix */ ".omp_par"); | ||||||
| 739 | |||||||
| 740 | // Find inputs to, outputs from the code region. | ||||||
| 741 | BasicBlock *CommonExit = nullptr; | ||||||
| 742 | SetVector<Value *> Inputs, Outputs, SinkingCands, HoistingCands; | ||||||
| 743 | Extractor.findAllocas(CEAC, SinkingCands, HoistingCands, CommonExit); | ||||||
| 744 | Extractor.findInputsOutputs(Inputs, Outputs, SinkingCands); | ||||||
| 745 | |||||||
| 746 | LLVM_DEBUG(dbgs() << "Before privatization: " << *OuterFn << "\n")do { } while (false); | ||||||
| 747 | |||||||
| 748 | FunctionCallee TIDRTLFn = | ||||||
| 749 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_global_thread_num); | ||||||
| 750 | |||||||
| 751 | auto PrivHelper = [&](Value &V) { | ||||||
| 752 | if (&V == TIDAddr || &V == ZeroAddr) | ||||||
| 753 | return; | ||||||
| 754 | |||||||
| 755 | SetVector<Use *> Uses; | ||||||
| 756 | for (Use &U : V.uses()) | ||||||
| 757 | if (auto *UserI = dyn_cast<Instruction>(U.getUser())) | ||||||
| 758 | if (ParallelRegionBlockSet.count(UserI->getParent())) | ||||||
| 759 | Uses.insert(&U); | ||||||
| 760 | |||||||
| 761 | // __kmpc_fork_call expects extra arguments as pointers. If the input | ||||||
| 762 | // already has a pointer type, everything is fine. Otherwise, store the | ||||||
| 763 | // value onto stack and load it back inside the to-be-outlined region. This | ||||||
| 764 | // will ensure only the pointer will be passed to the function. | ||||||
| 765 | // FIXME: if there are more than 15 trailing arguments, they must be | ||||||
| 766 | // additionally packed in a struct. | ||||||
| 767 | Value *Inner = &V; | ||||||
| 768 | if (!V.getType()->isPointerTy()) { | ||||||
| 769 | IRBuilder<>::InsertPointGuard Guard(Builder); | ||||||
| 770 | LLVM_DEBUG(llvm::dbgs() << "Forwarding input as pointer: " << V << "\n")do { } while (false); | ||||||
| 771 | |||||||
| 772 | Builder.restoreIP(OuterAllocaIP); | ||||||
| 773 | Value *Ptr = | ||||||
| 774 | Builder.CreateAlloca(V.getType(), nullptr, V.getName() + ".reloaded"); | ||||||
| 775 | |||||||
| 776 | // Store to stack at end of the block that currently branches to the entry | ||||||
| 777 | // block of the to-be-outlined region. | ||||||
| 778 | Builder.SetInsertPoint(InsertBB, | ||||||
| 779 | InsertBB->getTerminator()->getIterator()); | ||||||
| 780 | Builder.CreateStore(&V, Ptr); | ||||||
| 781 | |||||||
| 782 | // Load back next to allocations in the to-be-outlined region. | ||||||
| 783 | Builder.restoreIP(InnerAllocaIP); | ||||||
| 784 | Inner = Builder.CreateLoad(V.getType(), Ptr); | ||||||
| 785 | } | ||||||
| 786 | |||||||
| 787 | Value *ReplacementValue = nullptr; | ||||||
| 788 | CallInst *CI = dyn_cast<CallInst>(&V); | ||||||
| 789 | if (CI && CI->getCalledFunction() == TIDRTLFn.getCallee()) { | ||||||
| 790 | ReplacementValue = PrivTID; | ||||||
| 791 | } else { | ||||||
| 792 | Builder.restoreIP( | ||||||
| 793 | PrivCB(InnerAllocaIP, Builder.saveIP(), V, *Inner, ReplacementValue)); | ||||||
| 794 | assert(ReplacementValue &&((void)0) | ||||||
| 795 | "Expected copy/create callback to set replacement value!")((void)0); | ||||||
| 796 | if (ReplacementValue == &V) | ||||||
| 797 | return; | ||||||
| 798 | } | ||||||
| 799 | |||||||
| 800 | for (Use *UPtr : Uses) | ||||||
| 801 | UPtr->set(ReplacementValue); | ||||||
| 802 | }; | ||||||
| 803 | |||||||
| 804 | // Reset the inner alloca insertion as it will be used for loading the values | ||||||
| 805 | // wrapped into pointers before passing them into the to-be-outlined region. | ||||||
| 806 | // Configure it to insert immediately after the fake use of zero address so | ||||||
| 807 | // that they are available in the generated body and so that the | ||||||
| 808 | // OpenMP-related values (thread ID and zero address pointers) remain leading | ||||||
| 809 | // in the argument list. | ||||||
| 810 | InnerAllocaIP = IRBuilder<>::InsertPoint( | ||||||
| 811 | ZeroAddrUse->getParent(), ZeroAddrUse->getNextNode()->getIterator()); | ||||||
| 812 | |||||||
| 813 | // Reset the outer alloca insertion point to the entry of the relevant block | ||||||
| 814 | // in case it was invalidated. | ||||||
| 815 | OuterAllocaIP = IRBuilder<>::InsertPoint( | ||||||
| 816 | OuterAllocaBlock, OuterAllocaBlock->getFirstInsertionPt()); | ||||||
| 817 | |||||||
| 818 | for (Value *Input : Inputs) { | ||||||
| 819 | LLVM_DEBUG(dbgs() << "Captured input: " << *Input << "\n")do { } while (false); | ||||||
| 820 | PrivHelper(*Input); | ||||||
| 821 | } | ||||||
| 822 | LLVM_DEBUG({do { } while (false) | ||||||
| 823 | for (Value *Output : Outputs)do { } while (false) | ||||||
| 824 | LLVM_DEBUG(dbgs() << "Captured output: " << *Output << "\n");do { } while (false) | ||||||
| 825 | })do { } while (false); | ||||||
| 826 | assert(Outputs.empty() &&((void)0) | ||||||
| 827 | "OpenMP outlining should not produce live-out values!")((void)0); | ||||||
| 828 | |||||||
| 829 | LLVM_DEBUG(dbgs() << "After privatization: " << *OuterFn << "\n")do { } while (false); | ||||||
| 830 | LLVM_DEBUG({do { } while (false) | ||||||
| 831 | for (auto *BB : Blocks)do { } while (false) | ||||||
| 832 | dbgs() << " PBR: " << BB->getName() << "\n";do { } while (false) | ||||||
| 833 | })do { } while (false); | ||||||
| 834 | |||||||
| 835 | // Register the outlined info. | ||||||
| 836 | addOutlineInfo(std::move(OI)); | ||||||
| 837 | |||||||
| 838 | InsertPointTy AfterIP(UI->getParent(), UI->getParent()->end()); | ||||||
| 839 | UI->eraseFromParent(); | ||||||
| 840 | |||||||
| 841 | return AfterIP; | ||||||
| 842 | } | ||||||
| 843 | |||||||
| 844 | void OpenMPIRBuilder::emitFlush(const LocationDescription &Loc) { | ||||||
| 845 | // Build call void __kmpc_flush(ident_t *loc) | ||||||
| 846 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 847 | Value *Args[] = {getOrCreateIdent(SrcLocStr)}; | ||||||
| 848 | |||||||
| 849 | Builder.CreateCall(getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_flush), Args); | ||||||
| 850 | } | ||||||
| 851 | |||||||
| 852 | void OpenMPIRBuilder::createFlush(const LocationDescription &Loc) { | ||||||
| 853 | if (!updateToLocation(Loc)) | ||||||
| 854 | return; | ||||||
| 855 | emitFlush(Loc); | ||||||
| 856 | } | ||||||
| 857 | |||||||
| 858 | void OpenMPIRBuilder::emitTaskwaitImpl(const LocationDescription &Loc) { | ||||||
| 859 | // Build call kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32 | ||||||
| 860 | // global_tid); | ||||||
| 861 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 862 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 863 | Value *Args[] = {Ident, getOrCreateThreadID(Ident)}; | ||||||
| 864 | |||||||
| 865 | // Ignore return result until untied tasks are supported. | ||||||
| 866 | Builder.CreateCall(getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_omp_taskwait), | ||||||
| 867 | Args); | ||||||
| 868 | } | ||||||
| 869 | |||||||
| 870 | void OpenMPIRBuilder::createTaskwait(const LocationDescription &Loc) { | ||||||
| 871 | if (!updateToLocation(Loc)) | ||||||
| 872 | return; | ||||||
| 873 | emitTaskwaitImpl(Loc); | ||||||
| 874 | } | ||||||
| 875 | |||||||
| 876 | void OpenMPIRBuilder::emitTaskyieldImpl(const LocationDescription &Loc) { | ||||||
| 877 | // Build call __kmpc_omp_taskyield(loc, thread_id, 0); | ||||||
| 878 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 879 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 880 | Constant *I32Null = ConstantInt::getNullValue(Int32); | ||||||
| 881 | Value *Args[] = {Ident, getOrCreateThreadID(Ident), I32Null}; | ||||||
| 882 | |||||||
| 883 | Builder.CreateCall(getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_omp_taskyield), | ||||||
| 884 | Args); | ||||||
| 885 | } | ||||||
| 886 | |||||||
| 887 | void OpenMPIRBuilder::createTaskyield(const LocationDescription &Loc) { | ||||||
| 888 | if (!updateToLocation(Loc)) | ||||||
| 889 | return; | ||||||
| 890 | emitTaskyieldImpl(Loc); | ||||||
| 891 | } | ||||||
| 892 | |||||||
| 893 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createSections( | ||||||
| 894 | const LocationDescription &Loc, InsertPointTy AllocaIP, | ||||||
| 895 | ArrayRef<StorableBodyGenCallbackTy> SectionCBs, PrivatizeCallbackTy PrivCB, | ||||||
| 896 | FinalizeCallbackTy FiniCB, bool IsCancellable, bool IsNowait) { | ||||||
| 897 | if (!updateToLocation(Loc)) | ||||||
| 898 | return Loc.IP; | ||||||
| 899 | |||||||
| 900 | auto FiniCBWrapper = [&](InsertPointTy IP) { | ||||||
| 901 | if (IP.getBlock()->end() != IP.getPoint()) | ||||||
| 902 | return FiniCB(IP); | ||||||
| 903 | // This must be done otherwise any nested constructs using FinalizeOMPRegion | ||||||
| 904 | // will fail because that function requires the Finalization Basic Block to | ||||||
| 905 | // have a terminator, which is already removed by EmitOMPRegionBody. | ||||||
| 906 | // IP is currently at cancelation block. | ||||||
| 907 | // We need to backtrack to the condition block to fetch | ||||||
| 908 | // the exit block and create a branch from cancelation | ||||||
| 909 | // to exit block. | ||||||
| 910 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 911 | Builder.restoreIP(IP); | ||||||
| 912 | auto *CaseBB = IP.getBlock()->getSinglePredecessor(); | ||||||
| 913 | auto *CondBB = CaseBB->getSinglePredecessor()->getSinglePredecessor(); | ||||||
| 914 | auto *ExitBB = CondBB->getTerminator()->getSuccessor(1); | ||||||
| 915 | Instruction *I = Builder.CreateBr(ExitBB); | ||||||
| 916 | IP = InsertPointTy(I->getParent(), I->getIterator()); | ||||||
| 917 | return FiniCB(IP); | ||||||
| 918 | }; | ||||||
| 919 | |||||||
| 920 | FinalizationStack.push_back({FiniCBWrapper, OMPD_sections, IsCancellable}); | ||||||
| 921 | |||||||
| 922 | // Each section is emitted as a switch case | ||||||
| 923 | // Each finalization callback is handled from clang.EmitOMPSectionDirective() | ||||||
| 924 | // -> OMP.createSection() which generates the IR for each section | ||||||
| 925 | // Iterate through all sections and emit a switch construct: | ||||||
| 926 | // switch (IV) { | ||||||
| 927 | // case 0: | ||||||
| 928 | // <SectionStmt[0]>; | ||||||
| 929 | // break; | ||||||
| 930 | // ... | ||||||
| 931 | // case <NumSection> - 1: | ||||||
| 932 | // <SectionStmt[<NumSection> - 1]>; | ||||||
| 933 | // break; | ||||||
| 934 | // } | ||||||
| 935 | // ... | ||||||
| 936 | // section_loop.after: | ||||||
| 937 | // <FiniCB>; | ||||||
| 938 | auto LoopBodyGenCB = [&](InsertPointTy CodeGenIP, Value *IndVar) { | ||||||
| 939 | auto *CurFn = CodeGenIP.getBlock()->getParent(); | ||||||
| 940 | auto *ForIncBB = CodeGenIP.getBlock()->getSingleSuccessor(); | ||||||
| 941 | auto *ForExitBB = CodeGenIP.getBlock() | ||||||
| 942 | ->getSinglePredecessor() | ||||||
| 943 | ->getTerminator() | ||||||
| 944 | ->getSuccessor(1); | ||||||
| 945 | SwitchInst *SwitchStmt = Builder.CreateSwitch(IndVar, ForIncBB); | ||||||
| 946 | Builder.restoreIP(CodeGenIP); | ||||||
| 947 | unsigned CaseNumber = 0; | ||||||
| 948 | for (auto SectionCB : SectionCBs) { | ||||||
| 949 | auto *CaseBB = BasicBlock::Create(M.getContext(), | ||||||
| 950 | "omp_section_loop.body.case", CurFn); | ||||||
| 951 | SwitchStmt->addCase(Builder.getInt32(CaseNumber), CaseBB); | ||||||
| 952 | Builder.SetInsertPoint(CaseBB); | ||||||
| 953 | SectionCB(InsertPointTy(), Builder.saveIP(), *ForExitBB); | ||||||
| 954 | CaseNumber++; | ||||||
| 955 | } | ||||||
| 956 | // remove the existing terminator from body BB since there can be no | ||||||
| 957 | // terminators after switch/case | ||||||
| 958 | CodeGenIP.getBlock()->getTerminator()->eraseFromParent(); | ||||||
| 959 | }; | ||||||
| 960 | // Loop body ends here | ||||||
| 961 | // LowerBound, UpperBound, and STride for createCanonicalLoop | ||||||
| 962 | Type *I32Ty = Type::getInt32Ty(M.getContext()); | ||||||
| 963 | Value *LB = ConstantInt::get(I32Ty, 0); | ||||||
| 964 | Value *UB = ConstantInt::get(I32Ty, SectionCBs.size()); | ||||||
| 965 | Value *ST = ConstantInt::get(I32Ty, 1); | ||||||
| 966 | llvm::CanonicalLoopInfo *LoopInfo = createCanonicalLoop( | ||||||
| 967 | Loc, LoopBodyGenCB, LB, UB, ST, true, false, AllocaIP, "section_loop"); | ||||||
| 968 | LoopInfo = createStaticWorkshareLoop(Loc, LoopInfo, AllocaIP, true); | ||||||
| 969 | BasicBlock *LoopAfterBB = LoopInfo->getAfter(); | ||||||
| 970 | Instruction *SplitPos = LoopAfterBB->getTerminator(); | ||||||
| 971 | if (!isa_and_nonnull<BranchInst>(SplitPos)) | ||||||
| 972 | SplitPos = new UnreachableInst(Builder.getContext(), LoopAfterBB); | ||||||
| 973 | // ExitBB after LoopAfterBB because LoopAfterBB is used for FinalizationCB, | ||||||
| 974 | // which requires a BB with branch | ||||||
| 975 | BasicBlock *ExitBB = | ||||||
| 976 | LoopAfterBB->splitBasicBlock(SplitPos, "omp_sections.end"); | ||||||
| 977 | SplitPos->eraseFromParent(); | ||||||
| 978 | |||||||
| 979 | // Apply the finalization callback in LoopAfterBB | ||||||
| 980 | auto FiniInfo = FinalizationStack.pop_back_val(); | ||||||
| 981 | assert(FiniInfo.DK == OMPD_sections &&((void)0) | ||||||
| 982 | "Unexpected finalization stack state!")((void)0); | ||||||
| 983 | Builder.SetInsertPoint(LoopAfterBB->getTerminator()); | ||||||
| 984 | FiniInfo.FiniCB(Builder.saveIP()); | ||||||
| 985 | Builder.SetInsertPoint(ExitBB); | ||||||
| 986 | |||||||
| 987 | return Builder.saveIP(); | ||||||
| 988 | } | ||||||
| 989 | |||||||
| 990 | OpenMPIRBuilder::InsertPointTy | ||||||
| 991 | OpenMPIRBuilder::createSection(const LocationDescription &Loc, | ||||||
| 992 | BodyGenCallbackTy BodyGenCB, | ||||||
| 993 | FinalizeCallbackTy FiniCB) { | ||||||
| 994 | if (!updateToLocation(Loc)) | ||||||
| 995 | return Loc.IP; | ||||||
| 996 | |||||||
| 997 | auto FiniCBWrapper = [&](InsertPointTy IP) { | ||||||
| 998 | if (IP.getBlock()->end() != IP.getPoint()) | ||||||
| 999 | return FiniCB(IP); | ||||||
| 1000 | // This must be done otherwise any nested constructs using FinalizeOMPRegion | ||||||
| 1001 | // will fail because that function requires the Finalization Basic Block to | ||||||
| 1002 | // have a terminator, which is already removed by EmitOMPRegionBody. | ||||||
| 1003 | // IP is currently at cancelation block. | ||||||
| 1004 | // We need to backtrack to the condition block to fetch | ||||||
| 1005 | // the exit block and create a branch from cancelation | ||||||
| 1006 | // to exit block. | ||||||
| 1007 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 1008 | Builder.restoreIP(IP); | ||||||
| 1009 | auto *CaseBB = Loc.IP.getBlock(); | ||||||
| 1010 | auto *CondBB = CaseBB->getSinglePredecessor()->getSinglePredecessor(); | ||||||
| 1011 | auto *ExitBB = CondBB->getTerminator()->getSuccessor(1); | ||||||
| 1012 | Instruction *I = Builder.CreateBr(ExitBB); | ||||||
| 1013 | IP = InsertPointTy(I->getParent(), I->getIterator()); | ||||||
| 1014 | return FiniCB(IP); | ||||||
| 1015 | }; | ||||||
| 1016 | |||||||
| 1017 | Directive OMPD = Directive::OMPD_sections; | ||||||
| 1018 | // Since we are using Finalization Callback here, HasFinalize | ||||||
| 1019 | // and IsCancellable have to be true | ||||||
| 1020 | return EmitOMPInlinedRegion(OMPD, nullptr, nullptr, BodyGenCB, FiniCBWrapper, | ||||||
| 1021 | /*Conditional*/ false, /*hasFinalize*/ true, | ||||||
| 1022 | /*IsCancellable*/ true); | ||||||
| 1023 | } | ||||||
| 1024 | |||||||
| 1025 | OpenMPIRBuilder::InsertPointTy | ||||||
| 1026 | OpenMPIRBuilder::createMaster(const LocationDescription &Loc, | ||||||
| 1027 | BodyGenCallbackTy BodyGenCB, | ||||||
| 1028 | FinalizeCallbackTy FiniCB) { | ||||||
| 1029 | |||||||
| 1030 | if (!updateToLocation(Loc)) | ||||||
| 1031 | return Loc.IP; | ||||||
| 1032 | |||||||
| 1033 | Directive OMPD = Directive::OMPD_master; | ||||||
| 1034 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 1035 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 1036 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 1037 | Value *Args[] = {Ident, ThreadId}; | ||||||
| 1038 | |||||||
| 1039 | Function *EntryRTLFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_master); | ||||||
| 1040 | Instruction *EntryCall = Builder.CreateCall(EntryRTLFn, Args); | ||||||
| 1041 | |||||||
| 1042 | Function *ExitRTLFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_end_master); | ||||||
| 1043 | Instruction *ExitCall = Builder.CreateCall(ExitRTLFn, Args); | ||||||
| 1044 | |||||||
| 1045 | return EmitOMPInlinedRegion(OMPD, EntryCall, ExitCall, BodyGenCB, FiniCB, | ||||||
| 1046 | /*Conditional*/ true, /*hasFinalize*/ true); | ||||||
| 1047 | } | ||||||
| 1048 | |||||||
| 1049 | OpenMPIRBuilder::InsertPointTy | ||||||
| 1050 | OpenMPIRBuilder::createMasked(const LocationDescription &Loc, | ||||||
| 1051 | BodyGenCallbackTy BodyGenCB, | ||||||
| 1052 | FinalizeCallbackTy FiniCB, Value *Filter) { | ||||||
| 1053 | if (!updateToLocation(Loc)) | ||||||
| 1054 | return Loc.IP; | ||||||
| 1055 | |||||||
| 1056 | Directive OMPD = Directive::OMPD_masked; | ||||||
| 1057 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 1058 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 1059 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 1060 | Value *Args[] = {Ident, ThreadId, Filter}; | ||||||
| 1061 | Value *ArgsEnd[] = {Ident, ThreadId}; | ||||||
| 1062 | |||||||
| 1063 | Function *EntryRTLFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_masked); | ||||||
| 1064 | Instruction *EntryCall = Builder.CreateCall(EntryRTLFn, Args); | ||||||
| 1065 | |||||||
| 1066 | Function *ExitRTLFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_end_masked); | ||||||
| 1067 | Instruction *ExitCall = Builder.CreateCall(ExitRTLFn, ArgsEnd); | ||||||
| 1068 | |||||||
| 1069 | return EmitOMPInlinedRegion(OMPD, EntryCall, ExitCall, BodyGenCB, FiniCB, | ||||||
| 1070 | /*Conditional*/ true, /*hasFinalize*/ true); | ||||||
| 1071 | } | ||||||
| 1072 | |||||||
| 1073 | CanonicalLoopInfo *OpenMPIRBuilder::createLoopSkeleton( | ||||||
| 1074 | DebugLoc DL, Value *TripCount, Function *F, BasicBlock *PreInsertBefore, | ||||||
| 1075 | BasicBlock *PostInsertBefore, const Twine &Name) { | ||||||
| 1076 | Module *M = F->getParent(); | ||||||
| 1077 | LLVMContext &Ctx = M->getContext(); | ||||||
| 1078 | Type *IndVarTy = TripCount->getType(); | ||||||
| 1079 | |||||||
| 1080 | // Create the basic block structure. | ||||||
| 1081 | BasicBlock *Preheader = | ||||||
| 1082 | BasicBlock::Create(Ctx, "omp_" + Name + ".preheader", F, PreInsertBefore); | ||||||
| 1083 | BasicBlock *Header = | ||||||
| 1084 | BasicBlock::Create(Ctx, "omp_" + Name + ".header", F, PreInsertBefore); | ||||||
| 1085 | BasicBlock *Cond = | ||||||
| 1086 | BasicBlock::Create(Ctx, "omp_" + Name + ".cond", F, PreInsertBefore); | ||||||
| 1087 | BasicBlock *Body = | ||||||
| 1088 | BasicBlock::Create(Ctx, "omp_" + Name + ".body", F, PreInsertBefore); | ||||||
| 1089 | BasicBlock *Latch = | ||||||
| 1090 | BasicBlock::Create(Ctx, "omp_" + Name + ".inc", F, PostInsertBefore); | ||||||
| 1091 | BasicBlock *Exit = | ||||||
| 1092 | BasicBlock::Create(Ctx, "omp_" + Name + ".exit", F, PostInsertBefore); | ||||||
| 1093 | BasicBlock *After = | ||||||
| 1094 | BasicBlock::Create(Ctx, "omp_" + Name + ".after", F, PostInsertBefore); | ||||||
| 1095 | |||||||
| 1096 | // Use specified DebugLoc for new instructions. | ||||||
| 1097 | Builder.SetCurrentDebugLocation(DL); | ||||||
| 1098 | |||||||
| 1099 | Builder.SetInsertPoint(Preheader); | ||||||
| 1100 | Builder.CreateBr(Header); | ||||||
| 1101 | |||||||
| 1102 | Builder.SetInsertPoint(Header); | ||||||
| 1103 | PHINode *IndVarPHI = Builder.CreatePHI(IndVarTy, 2, "omp_" + Name + ".iv"); | ||||||
| 1104 | IndVarPHI->addIncoming(ConstantInt::get(IndVarTy, 0), Preheader); | ||||||
| 1105 | Builder.CreateBr(Cond); | ||||||
| 1106 | |||||||
| 1107 | Builder.SetInsertPoint(Cond); | ||||||
| 1108 | Value *Cmp = | ||||||
| 1109 | Builder.CreateICmpULT(IndVarPHI, TripCount, "omp_" + Name + ".cmp"); | ||||||
| 1110 | Builder.CreateCondBr(Cmp, Body, Exit); | ||||||
| 1111 | |||||||
| 1112 | Builder.SetInsertPoint(Body); | ||||||
| 1113 | Builder.CreateBr(Latch); | ||||||
| 1114 | |||||||
| 1115 | Builder.SetInsertPoint(Latch); | ||||||
| 1116 | Value *Next = Builder.CreateAdd(IndVarPHI, ConstantInt::get(IndVarTy, 1), | ||||||
| 1117 | "omp_" + Name + ".next", /*HasNUW=*/true); | ||||||
| 1118 | Builder.CreateBr(Header); | ||||||
| 1119 | IndVarPHI->addIncoming(Next, Latch); | ||||||
| 1120 | |||||||
| 1121 | Builder.SetInsertPoint(Exit); | ||||||
| 1122 | Builder.CreateBr(After); | ||||||
| 1123 | |||||||
| 1124 | // Remember and return the canonical control flow. | ||||||
| 1125 | LoopInfos.emplace_front(); | ||||||
| 1126 | CanonicalLoopInfo *CL = &LoopInfos.front(); | ||||||
| 1127 | |||||||
| 1128 | CL->Preheader = Preheader; | ||||||
| 1129 | CL->Header = Header; | ||||||
| 1130 | CL->Cond = Cond; | ||||||
| 1131 | CL->Body = Body; | ||||||
| 1132 | CL->Latch = Latch; | ||||||
| 1133 | CL->Exit = Exit; | ||||||
| 1134 | CL->After = After; | ||||||
| 1135 | |||||||
| 1136 | CL->IsValid = true; | ||||||
| 1137 | |||||||
| 1138 | #ifndef NDEBUG1 | ||||||
| 1139 | CL->assertOK(); | ||||||
| 1140 | #endif | ||||||
| 1141 | return CL; | ||||||
| 1142 | } | ||||||
| 1143 | |||||||
| 1144 | CanonicalLoopInfo * | ||||||
| 1145 | OpenMPIRBuilder::createCanonicalLoop(const LocationDescription &Loc, | ||||||
| 1146 | LoopBodyGenCallbackTy BodyGenCB, | ||||||
| 1147 | Value *TripCount, const Twine &Name) { | ||||||
| 1148 | BasicBlock *BB = Loc.IP.getBlock(); | ||||||
| 1149 | BasicBlock *NextBB = BB->getNextNode(); | ||||||
| 1150 | |||||||
| 1151 | CanonicalLoopInfo *CL = createLoopSkeleton(Loc.DL, TripCount, BB->getParent(), | ||||||
| 1152 | NextBB, NextBB, Name); | ||||||
| 1153 | BasicBlock *After = CL->getAfter(); | ||||||
| 1154 | |||||||
| 1155 | // If location is not set, don't connect the loop. | ||||||
| 1156 | if (updateToLocation(Loc)) { | ||||||
| 1157 | // Split the loop at the insertion point: Branch to the preheader and move | ||||||
| 1158 | // every following instruction to after the loop (the After BB). Also, the | ||||||
| 1159 | // new successor is the loop's after block. | ||||||
| 1160 | Builder.CreateBr(CL->Preheader); | ||||||
| 1161 | After->getInstList().splice(After->begin(), BB->getInstList(), | ||||||
| 1162 | Builder.GetInsertPoint(), BB->end()); | ||||||
| 1163 | After->replaceSuccessorsPhiUsesWith(BB, After); | ||||||
| 1164 | } | ||||||
| 1165 | |||||||
| 1166 | // Emit the body content. We do it after connecting the loop to the CFG to | ||||||
| 1167 | // avoid that the callback encounters degenerate BBs. | ||||||
| 1168 | BodyGenCB(CL->getBodyIP(), CL->getIndVar()); | ||||||
| 1169 | |||||||
| 1170 | #ifndef NDEBUG1 | ||||||
| 1171 | CL->assertOK(); | ||||||
| 1172 | #endif | ||||||
| 1173 | return CL; | ||||||
| 1174 | } | ||||||
| 1175 | |||||||
| 1176 | CanonicalLoopInfo *OpenMPIRBuilder::createCanonicalLoop( | ||||||
| 1177 | const LocationDescription &Loc, LoopBodyGenCallbackTy BodyGenCB, | ||||||
| 1178 | Value *Start, Value *Stop, Value *Step, bool IsSigned, bool InclusiveStop, | ||||||
| 1179 | InsertPointTy ComputeIP, const Twine &Name) { | ||||||
| 1180 | |||||||
| 1181 | // Consider the following difficulties (assuming 8-bit signed integers): | ||||||
| 1182 | // * Adding \p Step to the loop counter which passes \p Stop may overflow: | ||||||
| 1183 | // DO I = 1, 100, 50 | ||||||
| 1184 | /// * A \p Step of INT_MIN cannot not be normalized to a positive direction: | ||||||
| 1185 | // DO I = 100, 0, -128 | ||||||
| 1186 | |||||||
| 1187 | // Start, Stop and Step must be of the same integer type. | ||||||
| 1188 | auto *IndVarTy = cast<IntegerType>(Start->getType()); | ||||||
| 1189 | assert(IndVarTy == Stop->getType() && "Stop type mismatch")((void)0); | ||||||
| 1190 | assert(IndVarTy == Step->getType() && "Step type mismatch")((void)0); | ||||||
| 1191 | |||||||
| 1192 | LocationDescription ComputeLoc = | ||||||
| 1193 | ComputeIP.isSet() ? LocationDescription(ComputeIP, Loc.DL) : Loc; | ||||||
| 1194 | updateToLocation(ComputeLoc); | ||||||
| 1195 | |||||||
| 1196 | ConstantInt *Zero = ConstantInt::get(IndVarTy, 0); | ||||||
| 1197 | ConstantInt *One = ConstantInt::get(IndVarTy, 1); | ||||||
| 1198 | |||||||
| 1199 | // Like Step, but always positive. | ||||||
| 1200 | Value *Incr = Step; | ||||||
| 1201 | |||||||
| 1202 | // Distance between Start and Stop; always positive. | ||||||
| 1203 | Value *Span; | ||||||
| 1204 | |||||||
| 1205 | // Condition whether there are no iterations are executed at all, e.g. because | ||||||
| 1206 | // UB < LB. | ||||||
| 1207 | Value *ZeroCmp; | ||||||
| 1208 | |||||||
| 1209 | if (IsSigned) { | ||||||
| 1210 | // Ensure that increment is positive. If not, negate and invert LB and UB. | ||||||
| 1211 | Value *IsNeg = Builder.CreateICmpSLT(Step, Zero); | ||||||
| 1212 | Incr = Builder.CreateSelect(IsNeg, Builder.CreateNeg(Step), Step); | ||||||
| 1213 | Value *LB = Builder.CreateSelect(IsNeg, Stop, Start); | ||||||
| 1214 | Value *UB = Builder.CreateSelect(IsNeg, Start, Stop); | ||||||
| 1215 | Span = Builder.CreateSub(UB, LB, "", false, true); | ||||||
| 1216 | ZeroCmp = Builder.CreateICmp( | ||||||
| 1217 | InclusiveStop ? CmpInst::ICMP_SLT : CmpInst::ICMP_SLE, UB, LB); | ||||||
| 1218 | } else { | ||||||
| 1219 | Span = Builder.CreateSub(Stop, Start, "", true); | ||||||
| 1220 | ZeroCmp = Builder.CreateICmp( | ||||||
| 1221 | InclusiveStop ? CmpInst::ICMP_ULT : CmpInst::ICMP_ULE, Stop, Start); | ||||||
| 1222 | } | ||||||
| 1223 | |||||||
| 1224 | Value *CountIfLooping; | ||||||
| 1225 | if (InclusiveStop) { | ||||||
| 1226 | CountIfLooping = Builder.CreateAdd(Builder.CreateUDiv(Span, Incr), One); | ||||||
| 1227 | } else { | ||||||
| 1228 | // Avoid incrementing past stop since it could overflow. | ||||||
| 1229 | Value *CountIfTwo = Builder.CreateAdd( | ||||||
| 1230 | Builder.CreateUDiv(Builder.CreateSub(Span, One), Incr), One); | ||||||
| 1231 | Value *OneCmp = Builder.CreateICmp( | ||||||
| 1232 | InclusiveStop ? CmpInst::ICMP_ULT : CmpInst::ICMP_ULE, Span, Incr); | ||||||
| 1233 | CountIfLooping = Builder.CreateSelect(OneCmp, One, CountIfTwo); | ||||||
| 1234 | } | ||||||
| 1235 | Value *TripCount = Builder.CreateSelect(ZeroCmp, Zero, CountIfLooping, | ||||||
| 1236 | "omp_" + Name + ".tripcount"); | ||||||
| 1237 | |||||||
| 1238 | auto BodyGen = [=](InsertPointTy CodeGenIP, Value *IV) { | ||||||
| 1239 | Builder.restoreIP(CodeGenIP); | ||||||
| 1240 | Value *Span = Builder.CreateMul(IV, Step); | ||||||
| 1241 | Value *IndVar = Builder.CreateAdd(Span, Start); | ||||||
| 1242 | BodyGenCB(Builder.saveIP(), IndVar); | ||||||
| 1243 | }; | ||||||
| 1244 | LocationDescription LoopLoc = ComputeIP.isSet() ? Loc.IP : Builder.saveIP(); | ||||||
| 1245 | return createCanonicalLoop(LoopLoc, BodyGen, TripCount, Name); | ||||||
| 1246 | } | ||||||
| 1247 | |||||||
| 1248 | // Returns an LLVM function to call for initializing loop bounds using OpenMP | ||||||
| 1249 | // static scheduling depending on `type`. Only i32 and i64 are supported by the | ||||||
| 1250 | // runtime. Always interpret integers as unsigned similarly to | ||||||
| 1251 | // CanonicalLoopInfo. | ||||||
| 1252 | static FunctionCallee getKmpcForStaticInitForType(Type *Ty, Module &M, | ||||||
| 1253 | OpenMPIRBuilder &OMPBuilder) { | ||||||
| 1254 | unsigned Bitwidth = Ty->getIntegerBitWidth(); | ||||||
| 1255 | if (Bitwidth == 32) | ||||||
| 1256 | return OMPBuilder.getOrCreateRuntimeFunction( | ||||||
| 1257 | M, omp::RuntimeFunction::OMPRTL___kmpc_for_static_init_4u); | ||||||
| 1258 | if (Bitwidth == 64) | ||||||
| 1259 | return OMPBuilder.getOrCreateRuntimeFunction( | ||||||
| 1260 | M, omp::RuntimeFunction::OMPRTL___kmpc_for_static_init_8u); | ||||||
| 1261 | llvm_unreachable("unknown OpenMP loop iterator bitwidth")__builtin_unreachable(); | ||||||
| 1262 | } | ||||||
| 1263 | |||||||
| 1264 | // Sets the number of loop iterations to the given value. This value must be | ||||||
| 1265 | // valid in the condition block (i.e., defined in the preheader) and is | ||||||
| 1266 | // interpreted as an unsigned integer. | ||||||
| 1267 | void setCanonicalLoopTripCount(CanonicalLoopInfo *CLI, Value *TripCount) { | ||||||
| 1268 | Instruction *CmpI = &CLI->getCond()->front(); | ||||||
| 1269 | assert(isa<CmpInst>(CmpI) && "First inst must compare IV with TripCount")((void)0); | ||||||
| 1270 | CmpI->setOperand(1, TripCount); | ||||||
| 1271 | CLI->assertOK(); | ||||||
| 1272 | } | ||||||
| 1273 | |||||||
| 1274 | CanonicalLoopInfo *OpenMPIRBuilder::createStaticWorkshareLoop( | ||||||
| 1275 | const LocationDescription &Loc, CanonicalLoopInfo *CLI, | ||||||
| 1276 | InsertPointTy AllocaIP, bool NeedsBarrier, Value *Chunk) { | ||||||
| 1277 | // Set up the source location value for OpenMP runtime. | ||||||
| 1278 | if (!updateToLocation(Loc)) | ||||||
| 1279 | return nullptr; | ||||||
| 1280 | |||||||
| 1281 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 1282 | Value *SrcLoc = getOrCreateIdent(SrcLocStr); | ||||||
| 1283 | |||||||
| 1284 | // Declare useful OpenMP runtime functions. | ||||||
| 1285 | Value *IV = CLI->getIndVar(); | ||||||
| 1286 | Type *IVTy = IV->getType(); | ||||||
| 1287 | FunctionCallee StaticInit = getKmpcForStaticInitForType(IVTy, M, *this); | ||||||
| 1288 | FunctionCallee StaticFini = | ||||||
| 1289 | getOrCreateRuntimeFunction(M, omp::OMPRTL___kmpc_for_static_fini); | ||||||
| 1290 | |||||||
| 1291 | // Allocate space for computed loop bounds as expected by the "init" function. | ||||||
| 1292 | Builder.restoreIP(AllocaIP); | ||||||
| 1293 | Type *I32Type = Type::getInt32Ty(M.getContext()); | ||||||
| 1294 | Value *PLastIter = Builder.CreateAlloca(I32Type, nullptr, "p.lastiter"); | ||||||
| 1295 | Value *PLowerBound = Builder.CreateAlloca(IVTy, nullptr, "p.lowerbound"); | ||||||
| 1296 | Value *PUpperBound = Builder.CreateAlloca(IVTy, nullptr, "p.upperbound"); | ||||||
| 1297 | Value *PStride = Builder.CreateAlloca(IVTy, nullptr, "p.stride"); | ||||||
| 1298 | |||||||
| 1299 | // At the end of the preheader, prepare for calling the "init" function by | ||||||
| 1300 | // storing the current loop bounds into the allocated space. A canonical loop | ||||||
| 1301 | // always iterates from 0 to trip-count with step 1. Note that "init" expects | ||||||
| 1302 | // and produces an inclusive upper bound. | ||||||
| 1303 | Builder.SetInsertPoint(CLI->getPreheader()->getTerminator()); | ||||||
| 1304 | Constant *Zero = ConstantInt::get(IVTy, 0); | ||||||
| 1305 | Constant *One = ConstantInt::get(IVTy, 1); | ||||||
| 1306 | Builder.CreateStore(Zero, PLowerBound); | ||||||
| 1307 | Value *UpperBound = Builder.CreateSub(CLI->getTripCount(), One); | ||||||
| 1308 | Builder.CreateStore(UpperBound, PUpperBound); | ||||||
| 1309 | Builder.CreateStore(One, PStride); | ||||||
| 1310 | |||||||
| 1311 | if (!Chunk) | ||||||
| 1312 | Chunk = One; | ||||||
| 1313 | |||||||
| 1314 | Value *ThreadNum = getOrCreateThreadID(SrcLoc); | ||||||
| 1315 | |||||||
| 1316 | Constant *SchedulingType = | ||||||
| 1317 | ConstantInt::get(I32Type, static_cast<int>(OMPScheduleType::Static)); | ||||||
| 1318 | |||||||
| 1319 | // Call the "init" function and update the trip count of the loop with the | ||||||
| 1320 | // value it produced. | ||||||
| 1321 | Builder.CreateCall(StaticInit, | ||||||
| 1322 | {SrcLoc, ThreadNum, SchedulingType, PLastIter, PLowerBound, | ||||||
| 1323 | PUpperBound, PStride, One, Chunk}); | ||||||
| 1324 | Value *LowerBound = Builder.CreateLoad(IVTy, PLowerBound); | ||||||
| 1325 | Value *InclusiveUpperBound = Builder.CreateLoad(IVTy, PUpperBound); | ||||||
| 1326 | Value *TripCountMinusOne = Builder.CreateSub(InclusiveUpperBound, LowerBound); | ||||||
| 1327 | Value *TripCount = Builder.CreateAdd(TripCountMinusOne, One); | ||||||
| 1328 | setCanonicalLoopTripCount(CLI, TripCount); | ||||||
| 1329 | |||||||
| 1330 | // Update all uses of the induction variable except the one in the condition | ||||||
| 1331 | // block that compares it with the actual upper bound, and the increment in | ||||||
| 1332 | // the latch block. | ||||||
| 1333 | // TODO: this can eventually move to CanonicalLoopInfo or to a new | ||||||
| 1334 | // CanonicalLoopInfoUpdater interface. | ||||||
| 1335 | Builder.SetInsertPoint(CLI->getBody(), CLI->getBody()->getFirstInsertionPt()); | ||||||
| 1336 | Value *UpdatedIV = Builder.CreateAdd(IV, LowerBound); | ||||||
| 1337 | IV->replaceUsesWithIf(UpdatedIV, [&](Use &U) { | ||||||
| 1338 | auto *Instr = dyn_cast<Instruction>(U.getUser()); | ||||||
| 1339 | return !Instr || | ||||||
| 1340 | (Instr->getParent() != CLI->getCond() && | ||||||
| 1341 | Instr->getParent() != CLI->getLatch() && Instr != UpdatedIV); | ||||||
| 1342 | }); | ||||||
| 1343 | |||||||
| 1344 | // In the "exit" block, call the "fini" function. | ||||||
| 1345 | Builder.SetInsertPoint(CLI->getExit(), | ||||||
| 1346 | CLI->getExit()->getTerminator()->getIterator()); | ||||||
| 1347 | Builder.CreateCall(StaticFini, {SrcLoc, ThreadNum}); | ||||||
| 1348 | |||||||
| 1349 | // Add the barrier if requested. | ||||||
| 1350 | if (NeedsBarrier) | ||||||
| 1351 | createBarrier(LocationDescription(Builder.saveIP(), Loc.DL), | ||||||
| 1352 | omp::Directive::OMPD_for, /* ForceSimpleCall */ false, | ||||||
| 1353 | /* CheckCancelFlag */ false); | ||||||
| 1354 | |||||||
| 1355 | CLI->assertOK(); | ||||||
| 1356 | return CLI; | ||||||
| 1357 | } | ||||||
| 1358 | |||||||
| 1359 | CanonicalLoopInfo *OpenMPIRBuilder::createWorkshareLoop( | ||||||
| 1360 | const LocationDescription &Loc, CanonicalLoopInfo *CLI, | ||||||
| 1361 | InsertPointTy AllocaIP, bool NeedsBarrier) { | ||||||
| 1362 | // Currently only supports static schedules. | ||||||
| 1363 | return createStaticWorkshareLoop(Loc, CLI, AllocaIP, NeedsBarrier); | ||||||
| 1364 | } | ||||||
| 1365 | |||||||
| 1366 | /// Returns an LLVM function to call for initializing loop bounds using OpenMP | ||||||
| 1367 | /// dynamic scheduling depending on `type`. Only i32 and i64 are supported by | ||||||
| 1368 | /// the runtime. Always interpret integers as unsigned similarly to | ||||||
| 1369 | /// CanonicalLoopInfo. | ||||||
| 1370 | static FunctionCallee | ||||||
| 1371 | getKmpcForDynamicInitForType(Type *Ty, Module &M, OpenMPIRBuilder &OMPBuilder) { | ||||||
| 1372 | unsigned Bitwidth = Ty->getIntegerBitWidth(); | ||||||
| 1373 | if (Bitwidth == 32) | ||||||
| 1374 | return OMPBuilder.getOrCreateRuntimeFunction( | ||||||
| 1375 | M, omp::RuntimeFunction::OMPRTL___kmpc_dispatch_init_4u); | ||||||
| 1376 | if (Bitwidth == 64) | ||||||
| 1377 | return OMPBuilder.getOrCreateRuntimeFunction( | ||||||
| 1378 | M, omp::RuntimeFunction::OMPRTL___kmpc_dispatch_init_8u); | ||||||
| 1379 | llvm_unreachable("unknown OpenMP loop iterator bitwidth")__builtin_unreachable(); | ||||||
| 1380 | } | ||||||
| 1381 | |||||||
| 1382 | /// Returns an LLVM function to call for updating the next loop using OpenMP | ||||||
| 1383 | /// dynamic scheduling depending on `type`. Only i32 and i64 are supported by | ||||||
| 1384 | /// the runtime. Always interpret integers as unsigned similarly to | ||||||
| 1385 | /// CanonicalLoopInfo. | ||||||
| 1386 | static FunctionCallee | ||||||
| 1387 | getKmpcForDynamicNextForType(Type *Ty, Module &M, OpenMPIRBuilder &OMPBuilder) { | ||||||
| 1388 | unsigned Bitwidth = Ty->getIntegerBitWidth(); | ||||||
| 1389 | if (Bitwidth == 32) | ||||||
| 1390 | return OMPBuilder.getOrCreateRuntimeFunction( | ||||||
| 1391 | M, omp::RuntimeFunction::OMPRTL___kmpc_dispatch_next_4u); | ||||||
| 1392 | if (Bitwidth == 64) | ||||||
| 1393 | return OMPBuilder.getOrCreateRuntimeFunction( | ||||||
| 1394 | M, omp::RuntimeFunction::OMPRTL___kmpc_dispatch_next_8u); | ||||||
| 1395 | llvm_unreachable("unknown OpenMP loop iterator bitwidth")__builtin_unreachable(); | ||||||
| 1396 | } | ||||||
| 1397 | |||||||
| 1398 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createDynamicWorkshareLoop( | ||||||
| 1399 | const LocationDescription &Loc, CanonicalLoopInfo *CLI, | ||||||
| 1400 | InsertPointTy AllocaIP, OMPScheduleType SchedType, bool NeedsBarrier, | ||||||
| 1401 | Value *Chunk) { | ||||||
| 1402 | // Set up the source location value for OpenMP runtime. | ||||||
| 1403 | Builder.SetCurrentDebugLocation(Loc.DL); | ||||||
| 1404 | |||||||
| 1405 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 1406 | Value *SrcLoc = getOrCreateIdent(SrcLocStr); | ||||||
| 1407 | |||||||
| 1408 | // Declare useful OpenMP runtime functions. | ||||||
| 1409 | Value *IV = CLI->getIndVar(); | ||||||
| 1410 | Type *IVTy = IV->getType(); | ||||||
| 1411 | FunctionCallee DynamicInit = getKmpcForDynamicInitForType(IVTy, M, *this); | ||||||
| 1412 | FunctionCallee DynamicNext = getKmpcForDynamicNextForType(IVTy, M, *this); | ||||||
| 1413 | |||||||
| 1414 | // Allocate space for computed loop bounds as expected by the "init" function. | ||||||
| 1415 | Builder.restoreIP(AllocaIP); | ||||||
| 1416 | Type *I32Type = Type::getInt32Ty(M.getContext()); | ||||||
| 1417 | Value *PLastIter = Builder.CreateAlloca(I32Type, nullptr, "p.lastiter"); | ||||||
| 1418 | Value *PLowerBound = Builder.CreateAlloca(IVTy, nullptr, "p.lowerbound"); | ||||||
| 1419 | Value *PUpperBound = Builder.CreateAlloca(IVTy, nullptr, "p.upperbound"); | ||||||
| 1420 | Value *PStride = Builder.CreateAlloca(IVTy, nullptr, "p.stride"); | ||||||
| 1421 | |||||||
| 1422 | // At the end of the preheader, prepare for calling the "init" function by | ||||||
| 1423 | // storing the current loop bounds into the allocated space. A canonical loop | ||||||
| 1424 | // always iterates from 0 to trip-count with step 1. Note that "init" expects | ||||||
| 1425 | // and produces an inclusive upper bound. | ||||||
| 1426 | BasicBlock *PreHeader = CLI->getPreheader(); | ||||||
| 1427 | Builder.SetInsertPoint(PreHeader->getTerminator()); | ||||||
| 1428 | Constant *One = ConstantInt::get(IVTy, 1); | ||||||
| 1429 | Builder.CreateStore(One, PLowerBound); | ||||||
| 1430 | Value *UpperBound = CLI->getTripCount(); | ||||||
| 1431 | Builder.CreateStore(UpperBound, PUpperBound); | ||||||
| 1432 | Builder.CreateStore(One, PStride); | ||||||
| 1433 | |||||||
| 1434 | BasicBlock *Header = CLI->getHeader(); | ||||||
| 1435 | BasicBlock *Exit = CLI->getExit(); | ||||||
| 1436 | BasicBlock *Cond = CLI->getCond(); | ||||||
| 1437 | InsertPointTy AfterIP = CLI->getAfterIP(); | ||||||
| 1438 | |||||||
| 1439 | // The CLI will be "broken" in the code below, as the loop is no longer | ||||||
| 1440 | // a valid canonical loop. | ||||||
| 1441 | |||||||
| 1442 | if (!Chunk) | ||||||
| 1443 | Chunk = One; | ||||||
| 1444 | |||||||
| 1445 | Value *ThreadNum = getOrCreateThreadID(SrcLoc); | ||||||
| 1446 | |||||||
| 1447 | Constant *SchedulingType = | ||||||
| 1448 | ConstantInt::get(I32Type, static_cast<int>(SchedType)); | ||||||
| 1449 | |||||||
| 1450 | // Call the "init" function. | ||||||
| 1451 | Builder.CreateCall(DynamicInit, | ||||||
| 1452 | {SrcLoc, ThreadNum, SchedulingType, /* LowerBound */ One, | ||||||
| 1453 | UpperBound, /* step */ One, Chunk}); | ||||||
| 1454 | |||||||
| 1455 | // An outer loop around the existing one. | ||||||
| 1456 | BasicBlock *OuterCond = BasicBlock::Create( | ||||||
| 1457 | PreHeader->getContext(), Twine(PreHeader->getName()) + ".outer.cond", | ||||||
| 1458 | PreHeader->getParent()); | ||||||
| 1459 | // This needs to be 32-bit always, so can't use the IVTy Zero above. | ||||||
| 1460 | Builder.SetInsertPoint(OuterCond, OuterCond->getFirstInsertionPt()); | ||||||
| 1461 | Value *Res = | ||||||
| 1462 | Builder.CreateCall(DynamicNext, {SrcLoc, ThreadNum, PLastIter, | ||||||
| 1463 | PLowerBound, PUpperBound, PStride}); | ||||||
| 1464 | Constant *Zero32 = ConstantInt::get(I32Type, 0); | ||||||
| 1465 | Value *MoreWork = Builder.CreateCmp(CmpInst::ICMP_NE, Res, Zero32); | ||||||
| 1466 | Value *LowerBound = | ||||||
| 1467 | Builder.CreateSub(Builder.CreateLoad(IVTy, PLowerBound), One, "lb"); | ||||||
| 1468 | Builder.CreateCondBr(MoreWork, Header, Exit); | ||||||
| 1469 | |||||||
| 1470 | // Change PHI-node in loop header to use outer cond rather than preheader, | ||||||
| 1471 | // and set IV to the LowerBound. | ||||||
| 1472 | Instruction *Phi = &Header->front(); | ||||||
| 1473 | auto *PI = cast<PHINode>(Phi); | ||||||
| 1474 | PI->setIncomingBlock(0, OuterCond); | ||||||
| 1475 | PI->setIncomingValue(0, LowerBound); | ||||||
| 1476 | |||||||
| 1477 | // Then set the pre-header to jump to the OuterCond | ||||||
| 1478 | Instruction *Term = PreHeader->getTerminator(); | ||||||
| 1479 | auto *Br = cast<BranchInst>(Term); | ||||||
| 1480 | Br->setSuccessor(0, OuterCond); | ||||||
| 1481 | |||||||
| 1482 | // Modify the inner condition: | ||||||
| 1483 | // * Use the UpperBound returned from the DynamicNext call. | ||||||
| 1484 | // * jump to the loop outer loop when done with one of the inner loops. | ||||||
| 1485 | Builder.SetInsertPoint(Cond, Cond->getFirstInsertionPt()); | ||||||
| 1486 | UpperBound = Builder.CreateLoad(IVTy, PUpperBound, "ub"); | ||||||
| 1487 | Instruction *Comp = &*Builder.GetInsertPoint(); | ||||||
| 1488 | auto *CI = cast<CmpInst>(Comp); | ||||||
| 1489 | CI->setOperand(1, UpperBound); | ||||||
| 1490 | // Redirect the inner exit to branch to outer condition. | ||||||
| 1491 | Instruction *Branch = &Cond->back(); | ||||||
| 1492 | auto *BI = cast<BranchInst>(Branch); | ||||||
| 1493 | assert(BI->getSuccessor(1) == Exit)((void)0); | ||||||
| 1494 | BI->setSuccessor(1, OuterCond); | ||||||
| 1495 | |||||||
| 1496 | // Add the barrier if requested. | ||||||
| 1497 | if (NeedsBarrier) { | ||||||
| 1498 | Builder.SetInsertPoint(&Exit->back()); | ||||||
| 1499 | createBarrier(LocationDescription(Builder.saveIP(), Loc.DL), | ||||||
| 1500 | omp::Directive::OMPD_for, /* ForceSimpleCall */ false, | ||||||
| 1501 | /* CheckCancelFlag */ false); | ||||||
| 1502 | } | ||||||
| 1503 | |||||||
| 1504 | return AfterIP; | ||||||
| 1505 | } | ||||||
| 1506 | |||||||
| 1507 | /// Make \p Source branch to \p Target. | ||||||
| 1508 | /// | ||||||
| 1509 | /// Handles two situations: | ||||||
| 1510 | /// * \p Source already has an unconditional branch. | ||||||
| 1511 | /// * \p Source is a degenerate block (no terminator because the BB is | ||||||
| 1512 | /// the current head of the IR construction). | ||||||
| 1513 | static void redirectTo(BasicBlock *Source, BasicBlock *Target, DebugLoc DL) { | ||||||
| 1514 | if (Instruction *Term = Source->getTerminator()) { | ||||||
| 1515 | auto *Br = cast<BranchInst>(Term); | ||||||
| 1516 | assert(!Br->isConditional() &&((void)0) | ||||||
| 1517 | "BB's terminator must be an unconditional branch (or degenerate)")((void)0); | ||||||
| 1518 | BasicBlock *Succ = Br->getSuccessor(0); | ||||||
| 1519 | Succ->removePredecessor(Source, /*KeepOneInputPHIs=*/true); | ||||||
| 1520 | Br->setSuccessor(0, Target); | ||||||
| 1521 | return; | ||||||
| 1522 | } | ||||||
| 1523 | |||||||
| 1524 | auto *NewBr = BranchInst::Create(Target, Source); | ||||||
| 1525 | NewBr->setDebugLoc(DL); | ||||||
| 1526 | } | ||||||
| 1527 | |||||||
| 1528 | /// Redirect all edges that branch to \p OldTarget to \p NewTarget. That is, | ||||||
| 1529 | /// after this \p OldTarget will be orphaned. | ||||||
| 1530 | static void redirectAllPredecessorsTo(BasicBlock *OldTarget, | ||||||
| 1531 | BasicBlock *NewTarget, DebugLoc DL) { | ||||||
| 1532 | for (BasicBlock *Pred : make_early_inc_range(predecessors(OldTarget))) | ||||||
| 1533 | redirectTo(Pred, NewTarget, DL); | ||||||
| 1534 | } | ||||||
| 1535 | |||||||
| 1536 | /// Determine which blocks in \p BBs are reachable from outside and remove the | ||||||
| 1537 | /// ones that are not reachable from the function. | ||||||
| 1538 | static void removeUnusedBlocksFromParent(ArrayRef<BasicBlock *> BBs) { | ||||||
| 1539 | SmallPtrSet<BasicBlock *, 6> BBsToErase{BBs.begin(), BBs.end()}; | ||||||
| 1540 | auto HasRemainingUses = [&BBsToErase](BasicBlock *BB) { | ||||||
| 1541 | for (Use &U : BB->uses()) { | ||||||
| 1542 | auto *UseInst = dyn_cast<Instruction>(U.getUser()); | ||||||
| 1543 | if (!UseInst) | ||||||
| 1544 | continue; | ||||||
| 1545 | if (BBsToErase.count(UseInst->getParent())) | ||||||
| 1546 | continue; | ||||||
| 1547 | return true; | ||||||
| 1548 | } | ||||||
| 1549 | return false; | ||||||
| 1550 | }; | ||||||
| 1551 | |||||||
| 1552 | while (true) { | ||||||
| 1553 | bool Changed = false; | ||||||
| 1554 | for (BasicBlock *BB : make_early_inc_range(BBsToErase)) { | ||||||
| 1555 | if (HasRemainingUses(BB)) { | ||||||
| 1556 | BBsToErase.erase(BB); | ||||||
| 1557 | Changed = true; | ||||||
| 1558 | } | ||||||
| 1559 | } | ||||||
| 1560 | if (!Changed) | ||||||
| 1561 | break; | ||||||
| 1562 | } | ||||||
| 1563 | |||||||
| 1564 | SmallVector<BasicBlock *, 7> BBVec(BBsToErase.begin(), BBsToErase.end()); | ||||||
| 1565 | DeleteDeadBlocks(BBVec); | ||||||
| 1566 | } | ||||||
| 1567 | |||||||
| 1568 | CanonicalLoopInfo * | ||||||
| 1569 | OpenMPIRBuilder::collapseLoops(DebugLoc DL, ArrayRef<CanonicalLoopInfo *> Loops, | ||||||
| 1570 | InsertPointTy ComputeIP) { | ||||||
| 1571 | assert(Loops.size() >= 1 && "At least one loop required")((void)0); | ||||||
| 1572 | size_t NumLoops = Loops.size(); | ||||||
| 1573 | |||||||
| 1574 | // Nothing to do if there is already just one loop. | ||||||
| 1575 | if (NumLoops == 1) | ||||||
| 1576 | return Loops.front(); | ||||||
| 1577 | |||||||
| 1578 | CanonicalLoopInfo *Outermost = Loops.front(); | ||||||
| 1579 | CanonicalLoopInfo *Innermost = Loops.back(); | ||||||
| 1580 | BasicBlock *OrigPreheader = Outermost->getPreheader(); | ||||||
| 1581 | BasicBlock *OrigAfter = Outermost->getAfter(); | ||||||
| 1582 | Function *F = OrigPreheader->getParent(); | ||||||
| 1583 | |||||||
| 1584 | // Setup the IRBuilder for inserting the trip count computation. | ||||||
| 1585 | Builder.SetCurrentDebugLocation(DL); | ||||||
| 1586 | if (ComputeIP.isSet()) | ||||||
| 1587 | Builder.restoreIP(ComputeIP); | ||||||
| 1588 | else | ||||||
| 1589 | Builder.restoreIP(Outermost->getPreheaderIP()); | ||||||
| 1590 | |||||||
| 1591 | // Derive the collapsed' loop trip count. | ||||||
| 1592 | // TODO: Find common/largest indvar type. | ||||||
| 1593 | Value *CollapsedTripCount = nullptr; | ||||||
| 1594 | for (CanonicalLoopInfo *L : Loops) { | ||||||
| 1595 | Value *OrigTripCount = L->getTripCount(); | ||||||
| 1596 | if (!CollapsedTripCount) { | ||||||
| 1597 | CollapsedTripCount = OrigTripCount; | ||||||
| 1598 | continue; | ||||||
| 1599 | } | ||||||
| 1600 | |||||||
| 1601 | // TODO: Enable UndefinedSanitizer to diagnose an overflow here. | ||||||
| 1602 | CollapsedTripCount = Builder.CreateMul(CollapsedTripCount, OrigTripCount, | ||||||
| 1603 | {}, /*HasNUW=*/true); | ||||||
| 1604 | } | ||||||
| 1605 | |||||||
| 1606 | // Create the collapsed loop control flow. | ||||||
| 1607 | CanonicalLoopInfo *Result = | ||||||
| 1608 | createLoopSkeleton(DL, CollapsedTripCount, F, | ||||||
| 1609 | OrigPreheader->getNextNode(), OrigAfter, "collapsed"); | ||||||
| 1610 | |||||||
| 1611 | // Build the collapsed loop body code. | ||||||
| 1612 | // Start with deriving the input loop induction variables from the collapsed | ||||||
| 1613 | // one, using a divmod scheme. To preserve the original loops' order, the | ||||||
| 1614 | // innermost loop use the least significant bits. | ||||||
| 1615 | Builder.restoreIP(Result->getBodyIP()); | ||||||
| 1616 | |||||||
| 1617 | Value *Leftover = Result->getIndVar(); | ||||||
| 1618 | SmallVector<Value *> NewIndVars; | ||||||
| 1619 | NewIndVars.set_size(NumLoops); | ||||||
| 1620 | for (int i = NumLoops - 1; i >= 1; --i) { | ||||||
| 1621 | Value *OrigTripCount = Loops[i]->getTripCount(); | ||||||
| 1622 | |||||||
| 1623 | Value *NewIndVar = Builder.CreateURem(Leftover, OrigTripCount); | ||||||
| 1624 | NewIndVars[i] = NewIndVar; | ||||||
| 1625 | |||||||
| 1626 | Leftover = Builder.CreateUDiv(Leftover, OrigTripCount); | ||||||
| 1627 | } | ||||||
| 1628 | // Outermost loop gets all the remaining bits. | ||||||
| 1629 | NewIndVars[0] = Leftover; | ||||||
| 1630 | |||||||
| 1631 | // Construct the loop body control flow. | ||||||
| 1632 | // We progressively construct the branch structure following in direction of | ||||||
| 1633 | // the control flow, from the leading in-between code, the loop nest body, the | ||||||
| 1634 | // trailing in-between code, and rejoining the collapsed loop's latch. | ||||||
| 1635 | // ContinueBlock and ContinuePred keep track of the source(s) of next edge. If | ||||||
| 1636 | // the ContinueBlock is set, continue with that block. If ContinuePred, use | ||||||
| 1637 | // its predecessors as sources. | ||||||
| 1638 | BasicBlock *ContinueBlock = Result->getBody(); | ||||||
| 1639 | BasicBlock *ContinuePred = nullptr; | ||||||
| 1640 | auto ContinueWith = [&ContinueBlock, &ContinuePred, DL](BasicBlock *Dest, | ||||||
| 1641 | BasicBlock *NextSrc) { | ||||||
| 1642 | if (ContinueBlock) | ||||||
| 1643 | redirectTo(ContinueBlock, Dest, DL); | ||||||
| 1644 | else | ||||||
| 1645 | redirectAllPredecessorsTo(ContinuePred, Dest, DL); | ||||||
| 1646 | |||||||
| 1647 | ContinueBlock = nullptr; | ||||||
| 1648 | ContinuePred = NextSrc; | ||||||
| 1649 | }; | ||||||
| 1650 | |||||||
| 1651 | // The code before the nested loop of each level. | ||||||
| 1652 | // Because we are sinking it into the nest, it will be executed more often | ||||||
| 1653 | // that the original loop. More sophisticated schemes could keep track of what | ||||||
| 1654 | // the in-between code is and instantiate it only once per thread. | ||||||
| 1655 | for (size_t i = 0; i < NumLoops - 1; ++i) | ||||||
| 1656 | ContinueWith(Loops[i]->getBody(), Loops[i + 1]->getHeader()); | ||||||
| 1657 | |||||||
| 1658 | // Connect the loop nest body. | ||||||
| 1659 | ContinueWith(Innermost->getBody(), Innermost->getLatch()); | ||||||
| 1660 | |||||||
| 1661 | // The code after the nested loop at each level. | ||||||
| 1662 | for (size_t i = NumLoops - 1; i > 0; --i) | ||||||
| 1663 | ContinueWith(Loops[i]->getAfter(), Loops[i - 1]->getLatch()); | ||||||
| 1664 | |||||||
| 1665 | // Connect the finished loop to the collapsed loop latch. | ||||||
| 1666 | ContinueWith(Result->getLatch(), nullptr); | ||||||
| 1667 | |||||||
| 1668 | // Replace the input loops with the new collapsed loop. | ||||||
| 1669 | redirectTo(Outermost->getPreheader(), Result->getPreheader(), DL); | ||||||
| 1670 | redirectTo(Result->getAfter(), Outermost->getAfter(), DL); | ||||||
| 1671 | |||||||
| 1672 | // Replace the input loop indvars with the derived ones. | ||||||
| 1673 | for (size_t i = 0; i < NumLoops; ++i) | ||||||
| 1674 | Loops[i]->getIndVar()->replaceAllUsesWith(NewIndVars[i]); | ||||||
| 1675 | |||||||
| 1676 | // Remove unused parts of the input loops. | ||||||
| 1677 | SmallVector<BasicBlock *, 12> OldControlBBs; | ||||||
| 1678 | OldControlBBs.reserve(6 * Loops.size()); | ||||||
| 1679 | for (CanonicalLoopInfo *Loop : Loops) | ||||||
| 1680 | Loop->collectControlBlocks(OldControlBBs); | ||||||
| 1681 | removeUnusedBlocksFromParent(OldControlBBs); | ||||||
| 1682 | |||||||
| 1683 | #ifndef NDEBUG1 | ||||||
| 1684 | Result->assertOK(); | ||||||
| 1685 | #endif | ||||||
| 1686 | return Result; | ||||||
| 1687 | } | ||||||
| 1688 | |||||||
| 1689 | std::vector<CanonicalLoopInfo *> | ||||||
| 1690 | OpenMPIRBuilder::tileLoops(DebugLoc DL, ArrayRef<CanonicalLoopInfo *> Loops, | ||||||
| 1691 | ArrayRef<Value *> TileSizes) { | ||||||
| 1692 | assert(TileSizes.size() == Loops.size() &&((void)0) | ||||||
| 1693 | "Must pass as many tile sizes as there are loops")((void)0); | ||||||
| 1694 | int NumLoops = Loops.size(); | ||||||
| 1695 | assert(NumLoops >= 1 && "At least one loop to tile required")((void)0); | ||||||
| 1696 | |||||||
| 1697 | CanonicalLoopInfo *OutermostLoop = Loops.front(); | ||||||
| 1698 | CanonicalLoopInfo *InnermostLoop = Loops.back(); | ||||||
| 1699 | Function *F = OutermostLoop->getBody()->getParent(); | ||||||
| 1700 | BasicBlock *InnerEnter = InnermostLoop->getBody(); | ||||||
| 1701 | BasicBlock *InnerLatch = InnermostLoop->getLatch(); | ||||||
| 1702 | |||||||
| 1703 | // Collect original trip counts and induction variable to be accessible by | ||||||
| 1704 | // index. Also, the structure of the original loops is not preserved during | ||||||
| 1705 | // the construction of the tiled loops, so do it before we scavenge the BBs of | ||||||
| 1706 | // any original CanonicalLoopInfo. | ||||||
| 1707 | SmallVector<Value *, 4> OrigTripCounts, OrigIndVars; | ||||||
| 1708 | for (CanonicalLoopInfo *L : Loops) { | ||||||
| 1709 | OrigTripCounts.push_back(L->getTripCount()); | ||||||
| 1710 | OrigIndVars.push_back(L->getIndVar()); | ||||||
| 1711 | } | ||||||
| 1712 | |||||||
| 1713 | // Collect the code between loop headers. These may contain SSA definitions | ||||||
| 1714 | // that are used in the loop nest body. To be usable with in the innermost | ||||||
| 1715 | // body, these BasicBlocks will be sunk into the loop nest body. That is, | ||||||
| 1716 | // these instructions may be executed more often than before the tiling. | ||||||
| 1717 | // TODO: It would be sufficient to only sink them into body of the | ||||||
| 1718 | // corresponding tile loop. | ||||||
| 1719 | SmallVector<std::pair<BasicBlock *, BasicBlock *>, 4> InbetweenCode; | ||||||
| 1720 | for (int i = 0; i < NumLoops - 1; ++i) { | ||||||
| 1721 | CanonicalLoopInfo *Surrounding = Loops[i]; | ||||||
| 1722 | CanonicalLoopInfo *Nested = Loops[i + 1]; | ||||||
| 1723 | |||||||
| 1724 | BasicBlock *EnterBB = Surrounding->getBody(); | ||||||
| 1725 | BasicBlock *ExitBB = Nested->getHeader(); | ||||||
| 1726 | InbetweenCode.emplace_back(EnterBB, ExitBB); | ||||||
| 1727 | } | ||||||
| 1728 | |||||||
| 1729 | // Compute the trip counts of the floor loops. | ||||||
| 1730 | Builder.SetCurrentDebugLocation(DL); | ||||||
| 1731 | Builder.restoreIP(OutermostLoop->getPreheaderIP()); | ||||||
| 1732 | SmallVector<Value *, 4> FloorCount, FloorRems; | ||||||
| 1733 | for (int i = 0; i < NumLoops; ++i) { | ||||||
| 1734 | Value *TileSize = TileSizes[i]; | ||||||
| 1735 | Value *OrigTripCount = OrigTripCounts[i]; | ||||||
| 1736 | Type *IVType = OrigTripCount->getType(); | ||||||
| 1737 | |||||||
| 1738 | Value *FloorTripCount = Builder.CreateUDiv(OrigTripCount, TileSize); | ||||||
| 1739 | Value *FloorTripRem = Builder.CreateURem(OrigTripCount, TileSize); | ||||||
| 1740 | |||||||
| 1741 | // 0 if tripcount divides the tilesize, 1 otherwise. | ||||||
| 1742 | // 1 means we need an additional iteration for a partial tile. | ||||||
| 1743 | // | ||||||
| 1744 | // Unfortunately we cannot just use the roundup-formula | ||||||
| 1745 | // (tripcount + tilesize - 1)/tilesize | ||||||
| 1746 | // because the summation might overflow. We do not want introduce undefined | ||||||
| 1747 | // behavior when the untiled loop nest did not. | ||||||
| 1748 | Value *FloorTripOverflow = | ||||||
| 1749 | Builder.CreateICmpNE(FloorTripRem, ConstantInt::get(IVType, 0)); | ||||||
| 1750 | |||||||
| 1751 | FloorTripOverflow = Builder.CreateZExt(FloorTripOverflow, IVType); | ||||||
| 1752 | FloorTripCount = | ||||||
| 1753 | Builder.CreateAdd(FloorTripCount, FloorTripOverflow, | ||||||
| 1754 | "omp_floor" + Twine(i) + ".tripcount", true); | ||||||
| 1755 | |||||||
| 1756 | // Remember some values for later use. | ||||||
| 1757 | FloorCount.push_back(FloorTripCount); | ||||||
| 1758 | FloorRems.push_back(FloorTripRem); | ||||||
| 1759 | } | ||||||
| 1760 | |||||||
| 1761 | // Generate the new loop nest, from the outermost to the innermost. | ||||||
| 1762 | std::vector<CanonicalLoopInfo *> Result; | ||||||
| 1763 | Result.reserve(NumLoops * 2); | ||||||
| 1764 | |||||||
| 1765 | // The basic block of the surrounding loop that enters the nest generated | ||||||
| 1766 | // loop. | ||||||
| 1767 | BasicBlock *Enter = OutermostLoop->getPreheader(); | ||||||
| 1768 | |||||||
| 1769 | // The basic block of the surrounding loop where the inner code should | ||||||
| 1770 | // continue. | ||||||
| 1771 | BasicBlock *Continue = OutermostLoop->getAfter(); | ||||||
| 1772 | |||||||
| 1773 | // Where the next loop basic block should be inserted. | ||||||
| 1774 | BasicBlock *OutroInsertBefore = InnermostLoop->getExit(); | ||||||
| 1775 | |||||||
| 1776 | auto EmbeddNewLoop = | ||||||
| 1777 | [this, DL, F, InnerEnter, &Enter, &Continue, &OutroInsertBefore]( | ||||||
| 1778 | Value *TripCount, const Twine &Name) -> CanonicalLoopInfo * { | ||||||
| 1779 | CanonicalLoopInfo *EmbeddedLoop = createLoopSkeleton( | ||||||
| 1780 | DL, TripCount, F, InnerEnter, OutroInsertBefore, Name); | ||||||
| 1781 | redirectTo(Enter, EmbeddedLoop->getPreheader(), DL); | ||||||
| 1782 | redirectTo(EmbeddedLoop->getAfter(), Continue, DL); | ||||||
| 1783 | |||||||
| 1784 | // Setup the position where the next embedded loop connects to this loop. | ||||||
| 1785 | Enter = EmbeddedLoop->getBody(); | ||||||
| 1786 | Continue = EmbeddedLoop->getLatch(); | ||||||
| 1787 | OutroInsertBefore = EmbeddedLoop->getLatch(); | ||||||
| 1788 | return EmbeddedLoop; | ||||||
| 1789 | }; | ||||||
| 1790 | |||||||
| 1791 | auto EmbeddNewLoops = [&Result, &EmbeddNewLoop](ArrayRef<Value *> TripCounts, | ||||||
| 1792 | const Twine &NameBase) { | ||||||
| 1793 | for (auto P : enumerate(TripCounts)) { | ||||||
| 1794 | CanonicalLoopInfo *EmbeddedLoop = | ||||||
| 1795 | EmbeddNewLoop(P.value(), NameBase + Twine(P.index())); | ||||||
| 1796 | Result.push_back(EmbeddedLoop); | ||||||
| 1797 | } | ||||||
| 1798 | }; | ||||||
| 1799 | |||||||
| 1800 | EmbeddNewLoops(FloorCount, "floor"); | ||||||
| 1801 | |||||||
| 1802 | // Within the innermost floor loop, emit the code that computes the tile | ||||||
| 1803 | // sizes. | ||||||
| 1804 | Builder.SetInsertPoint(Enter->getTerminator()); | ||||||
| 1805 | SmallVector<Value *, 4> TileCounts; | ||||||
| 1806 | for (int i = 0; i < NumLoops; ++i) { | ||||||
| 1807 | CanonicalLoopInfo *FloorLoop = Result[i]; | ||||||
| 1808 | Value *TileSize = TileSizes[i]; | ||||||
| 1809 | |||||||
| 1810 | Value *FloorIsEpilogue = | ||||||
| 1811 | Builder.CreateICmpEQ(FloorLoop->getIndVar(), FloorCount[i]); | ||||||
| 1812 | Value *TileTripCount = | ||||||
| 1813 | Builder.CreateSelect(FloorIsEpilogue, FloorRems[i], TileSize); | ||||||
| 1814 | |||||||
| 1815 | TileCounts.push_back(TileTripCount); | ||||||
| 1816 | } | ||||||
| 1817 | |||||||
| 1818 | // Create the tile loops. | ||||||
| 1819 | EmbeddNewLoops(TileCounts, "tile"); | ||||||
| 1820 | |||||||
| 1821 | // Insert the inbetween code into the body. | ||||||
| 1822 | BasicBlock *BodyEnter = Enter; | ||||||
| 1823 | BasicBlock *BodyEntered = nullptr; | ||||||
| 1824 | for (std::pair<BasicBlock *, BasicBlock *> P : InbetweenCode) { | ||||||
| 1825 | BasicBlock *EnterBB = P.first; | ||||||
| 1826 | BasicBlock *ExitBB = P.second; | ||||||
| 1827 | |||||||
| 1828 | if (BodyEnter) | ||||||
| 1829 | redirectTo(BodyEnter, EnterBB, DL); | ||||||
| 1830 | else | ||||||
| 1831 | redirectAllPredecessorsTo(BodyEntered, EnterBB, DL); | ||||||
| 1832 | |||||||
| 1833 | BodyEnter = nullptr; | ||||||
| 1834 | BodyEntered = ExitBB; | ||||||
| 1835 | } | ||||||
| 1836 | |||||||
| 1837 | // Append the original loop nest body into the generated loop nest body. | ||||||
| 1838 | if (BodyEnter) | ||||||
| 1839 | redirectTo(BodyEnter, InnerEnter, DL); | ||||||
| 1840 | else | ||||||
| 1841 | redirectAllPredecessorsTo(BodyEntered, InnerEnter, DL); | ||||||
| 1842 | redirectAllPredecessorsTo(InnerLatch, Continue, DL); | ||||||
| 1843 | |||||||
| 1844 | // Replace the original induction variable with an induction variable computed | ||||||
| 1845 | // from the tile and floor induction variables. | ||||||
| 1846 | Builder.restoreIP(Result.back()->getBodyIP()); | ||||||
| 1847 | for (int i = 0; i < NumLoops; ++i) { | ||||||
| 1848 | CanonicalLoopInfo *FloorLoop = Result[i]; | ||||||
| 1849 | CanonicalLoopInfo *TileLoop = Result[NumLoops + i]; | ||||||
| 1850 | Value *OrigIndVar = OrigIndVars[i]; | ||||||
| 1851 | Value *Size = TileSizes[i]; | ||||||
| 1852 | |||||||
| 1853 | Value *Scale = | ||||||
| 1854 | Builder.CreateMul(Size, FloorLoop->getIndVar(), {}, /*HasNUW=*/true); | ||||||
| 1855 | Value *Shift = | ||||||
| 1856 | Builder.CreateAdd(Scale, TileLoop->getIndVar(), {}, /*HasNUW=*/true); | ||||||
| 1857 | OrigIndVar->replaceAllUsesWith(Shift); | ||||||
| 1858 | } | ||||||
| 1859 | |||||||
| 1860 | // Remove unused parts of the original loops. | ||||||
| 1861 | SmallVector<BasicBlock *, 12> OldControlBBs; | ||||||
| 1862 | OldControlBBs.reserve(6 * Loops.size()); | ||||||
| 1863 | for (CanonicalLoopInfo *Loop : Loops) | ||||||
| 1864 | Loop->collectControlBlocks(OldControlBBs); | ||||||
| 1865 | removeUnusedBlocksFromParent(OldControlBBs); | ||||||
| 1866 | |||||||
| 1867 | #ifndef NDEBUG1 | ||||||
| 1868 | for (CanonicalLoopInfo *GenL : Result) | ||||||
| 1869 | GenL->assertOK(); | ||||||
| 1870 | #endif | ||||||
| 1871 | return Result; | ||||||
| 1872 | } | ||||||
| 1873 | |||||||
| 1874 | OpenMPIRBuilder::InsertPointTy | ||||||
| 1875 | OpenMPIRBuilder::createCopyPrivate(const LocationDescription &Loc, | ||||||
| 1876 | llvm::Value *BufSize, llvm::Value *CpyBuf, | ||||||
| 1877 | llvm::Value *CpyFn, llvm::Value *DidIt) { | ||||||
| 1878 | if (!updateToLocation(Loc)) | ||||||
| 1879 | return Loc.IP; | ||||||
| 1880 | |||||||
| 1881 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 1882 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 1883 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 1884 | |||||||
| 1885 | llvm::Value *DidItLD = Builder.CreateLoad(Builder.getInt32Ty(), DidIt); | ||||||
| 1886 | |||||||
| 1887 | Value *Args[] = {Ident, ThreadId, BufSize, CpyBuf, CpyFn, DidItLD}; | ||||||
| 1888 | |||||||
| 1889 | Function *Fn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_copyprivate); | ||||||
| 1890 | Builder.CreateCall(Fn, Args); | ||||||
| 1891 | |||||||
| 1892 | return Builder.saveIP(); | ||||||
| 1893 | } | ||||||
| 1894 | |||||||
| 1895 | OpenMPIRBuilder::InsertPointTy | ||||||
| 1896 | OpenMPIRBuilder::createSingle(const LocationDescription &Loc, | ||||||
| 1897 | BodyGenCallbackTy BodyGenCB, | ||||||
| 1898 | FinalizeCallbackTy FiniCB, llvm::Value *DidIt) { | ||||||
| 1899 | |||||||
| 1900 | if (!updateToLocation(Loc)) | ||||||
| 1901 | return Loc.IP; | ||||||
| 1902 | |||||||
| 1903 | // If needed (i.e. not null), initialize `DidIt` with 0 | ||||||
| 1904 | if (DidIt) { | ||||||
| 1905 | Builder.CreateStore(Builder.getInt32(0), DidIt); | ||||||
| 1906 | } | ||||||
| 1907 | |||||||
| 1908 | Directive OMPD = Directive::OMPD_single; | ||||||
| 1909 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 1910 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 1911 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 1912 | Value *Args[] = {Ident, ThreadId}; | ||||||
| 1913 | |||||||
| 1914 | Function *EntryRTLFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_single); | ||||||
| 1915 | Instruction *EntryCall = Builder.CreateCall(EntryRTLFn, Args); | ||||||
| 1916 | |||||||
| 1917 | Function *ExitRTLFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_end_single); | ||||||
| 1918 | Instruction *ExitCall = Builder.CreateCall(ExitRTLFn, Args); | ||||||
| 1919 | |||||||
| 1920 | // generates the following: | ||||||
| 1921 | // if (__kmpc_single()) { | ||||||
| 1922 | // .... single region ... | ||||||
| 1923 | // __kmpc_end_single | ||||||
| 1924 | // } | ||||||
| 1925 | |||||||
| 1926 | return EmitOMPInlinedRegion(OMPD, EntryCall, ExitCall, BodyGenCB, FiniCB, | ||||||
| 1927 | /*Conditional*/ true, /*hasFinalize*/ true); | ||||||
| 1928 | } | ||||||
| 1929 | |||||||
| 1930 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createCritical( | ||||||
| 1931 | const LocationDescription &Loc, BodyGenCallbackTy BodyGenCB, | ||||||
| 1932 | FinalizeCallbackTy FiniCB, StringRef CriticalName, Value *HintInst) { | ||||||
| 1933 | |||||||
| 1934 | if (!updateToLocation(Loc)) | ||||||
| 1935 | return Loc.IP; | ||||||
| 1936 | |||||||
| 1937 | Directive OMPD = Directive::OMPD_critical; | ||||||
| 1938 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 1939 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 1940 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 1941 | Value *LockVar = getOMPCriticalRegionLock(CriticalName); | ||||||
| 1942 | Value *Args[] = {Ident, ThreadId, LockVar}; | ||||||
| 1943 | |||||||
| 1944 | SmallVector<llvm::Value *, 4> EnterArgs(std::begin(Args), std::end(Args)); | ||||||
| 1945 | Function *RTFn = nullptr; | ||||||
| 1946 | if (HintInst) { | ||||||
| 1947 | // Add Hint to entry Args and create call | ||||||
| 1948 | EnterArgs.push_back(HintInst); | ||||||
| 1949 | RTFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_critical_with_hint); | ||||||
| 1950 | } else { | ||||||
| 1951 | RTFn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_critical); | ||||||
| 1952 | } | ||||||
| 1953 | Instruction *EntryCall = Builder.CreateCall(RTFn, EnterArgs); | ||||||
| 1954 | |||||||
| 1955 | Function *ExitRTLFn = | ||||||
| 1956 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_end_critical); | ||||||
| 1957 | Instruction *ExitCall = Builder.CreateCall(ExitRTLFn, Args); | ||||||
| 1958 | |||||||
| 1959 | return EmitOMPInlinedRegion(OMPD, EntryCall, ExitCall, BodyGenCB, FiniCB, | ||||||
| 1960 | /*Conditional*/ false, /*hasFinalize*/ true); | ||||||
| 1961 | } | ||||||
| 1962 | |||||||
| 1963 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::EmitOMPInlinedRegion( | ||||||
| 1964 | Directive OMPD, Instruction *EntryCall, Instruction *ExitCall, | ||||||
| 1965 | BodyGenCallbackTy BodyGenCB, FinalizeCallbackTy FiniCB, bool Conditional, | ||||||
| 1966 | bool HasFinalize, bool IsCancellable) { | ||||||
| 1967 | |||||||
| 1968 | if (HasFinalize) | ||||||
| 1969 | FinalizationStack.push_back({FiniCB, OMPD, IsCancellable}); | ||||||
| 1970 | |||||||
| 1971 | // Create inlined region's entry and body blocks, in preparation | ||||||
| 1972 | // for conditional creation | ||||||
| 1973 | BasicBlock *EntryBB = Builder.GetInsertBlock(); | ||||||
| 1974 | Instruction *SplitPos = EntryBB->getTerminator(); | ||||||
| 1975 | if (!isa_and_nonnull<BranchInst>(SplitPos)) | ||||||
| 1976 | SplitPos = new UnreachableInst(Builder.getContext(), EntryBB); | ||||||
| 1977 | BasicBlock *ExitBB = EntryBB->splitBasicBlock(SplitPos, "omp_region.end"); | ||||||
| 1978 | BasicBlock *FiniBB = | ||||||
| 1979 | EntryBB->splitBasicBlock(EntryBB->getTerminator(), "omp_region.finalize"); | ||||||
| 1980 | |||||||
| 1981 | Builder.SetInsertPoint(EntryBB->getTerminator()); | ||||||
| 1982 | emitCommonDirectiveEntry(OMPD, EntryCall, ExitBB, Conditional); | ||||||
| 1983 | |||||||
| 1984 | // generate body | ||||||
| 1985 | BodyGenCB(/* AllocaIP */ InsertPointTy(), | ||||||
| 1986 | /* CodeGenIP */ Builder.saveIP(), *FiniBB); | ||||||
| 1987 | |||||||
| 1988 | // If we didn't emit a branch to FiniBB during body generation, it means | ||||||
| 1989 | // FiniBB is unreachable (e.g. while(1);). stop generating all the | ||||||
| 1990 | // unreachable blocks, and remove anything we are not going to use. | ||||||
| 1991 | auto SkipEmittingRegion = FiniBB->hasNPredecessors(0); | ||||||
| 1992 | if (SkipEmittingRegion) { | ||||||
| 1993 | FiniBB->eraseFromParent(); | ||||||
| 1994 | ExitCall->eraseFromParent(); | ||||||
| 1995 | // Discard finalization if we have it. | ||||||
| 1996 | if (HasFinalize) { | ||||||
| 1997 | assert(!FinalizationStack.empty() &&((void)0) | ||||||
| 1998 | "Unexpected finalization stack state!")((void)0); | ||||||
| 1999 | FinalizationStack.pop_back(); | ||||||
| 2000 | } | ||||||
| 2001 | } else { | ||||||
| 2002 | // emit exit call and do any needed finalization. | ||||||
| 2003 | auto FinIP = InsertPointTy(FiniBB, FiniBB->getFirstInsertionPt()); | ||||||
| 2004 | assert(FiniBB->getTerminator()->getNumSuccessors() == 1 &&((void)0) | ||||||
| 2005 | FiniBB->getTerminator()->getSuccessor(0) == ExitBB &&((void)0) | ||||||
| 2006 | "Unexpected control flow graph state!!")((void)0); | ||||||
| 2007 | emitCommonDirectiveExit(OMPD, FinIP, ExitCall, HasFinalize); | ||||||
| 2008 | assert(FiniBB->getUniquePredecessor()->getUniqueSuccessor() == FiniBB &&((void)0) | ||||||
| 2009 | "Unexpected Control Flow State!")((void)0); | ||||||
| 2010 | MergeBlockIntoPredecessor(FiniBB); | ||||||
| 2011 | } | ||||||
| 2012 | |||||||
| 2013 | // If we are skipping the region of a non conditional, remove the exit | ||||||
| 2014 | // block, and clear the builder's insertion point. | ||||||
| 2015 | assert(SplitPos->getParent() == ExitBB &&((void)0) | ||||||
| 2016 | "Unexpected Insertion point location!")((void)0); | ||||||
| 2017 | if (!Conditional && SkipEmittingRegion) { | ||||||
| 2018 | ExitBB->eraseFromParent(); | ||||||
| 2019 | Builder.ClearInsertionPoint(); | ||||||
| 2020 | } else { | ||||||
| 2021 | auto merged = MergeBlockIntoPredecessor(ExitBB); | ||||||
| 2022 | BasicBlock *ExitPredBB = SplitPos->getParent(); | ||||||
| 2023 | auto InsertBB = merged ? ExitPredBB : ExitBB; | ||||||
| 2024 | if (!isa_and_nonnull<BranchInst>(SplitPos)) | ||||||
| 2025 | SplitPos->eraseFromParent(); | ||||||
| 2026 | Builder.SetInsertPoint(InsertBB); | ||||||
| 2027 | } | ||||||
| 2028 | |||||||
| 2029 | return Builder.saveIP(); | ||||||
| 2030 | } | ||||||
| 2031 | |||||||
| 2032 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::emitCommonDirectiveEntry( | ||||||
| 2033 | Directive OMPD, Value *EntryCall, BasicBlock *ExitBB, bool Conditional) { | ||||||
| 2034 | // if nothing to do, Return current insertion point. | ||||||
| 2035 | if (!Conditional || !EntryCall) | ||||||
| 2036 | return Builder.saveIP(); | ||||||
| 2037 | |||||||
| 2038 | BasicBlock *EntryBB = Builder.GetInsertBlock(); | ||||||
| 2039 | Value *CallBool = Builder.CreateIsNotNull(EntryCall); | ||||||
| 2040 | auto *ThenBB = BasicBlock::Create(M.getContext(), "omp_region.body"); | ||||||
| 2041 | auto *UI = new UnreachableInst(Builder.getContext(), ThenBB); | ||||||
| 2042 | |||||||
| 2043 | // Emit thenBB and set the Builder's insertion point there for | ||||||
| 2044 | // body generation next. Place the block after the current block. | ||||||
| 2045 | Function *CurFn = EntryBB->getParent(); | ||||||
| 2046 | CurFn->getBasicBlockList().insertAfter(EntryBB->getIterator(), ThenBB); | ||||||
| 2047 | |||||||
| 2048 | // Move Entry branch to end of ThenBB, and replace with conditional | ||||||
| 2049 | // branch (If-stmt) | ||||||
| 2050 | Instruction *EntryBBTI = EntryBB->getTerminator(); | ||||||
| 2051 | Builder.CreateCondBr(CallBool, ThenBB, ExitBB); | ||||||
| 2052 | EntryBBTI->removeFromParent(); | ||||||
| 2053 | Builder.SetInsertPoint(UI); | ||||||
| 2054 | Builder.Insert(EntryBBTI); | ||||||
| 2055 | UI->eraseFromParent(); | ||||||
| 2056 | Builder.SetInsertPoint(ThenBB->getTerminator()); | ||||||
| 2057 | |||||||
| 2058 | // return an insertion point to ExitBB. | ||||||
| 2059 | return IRBuilder<>::InsertPoint(ExitBB, ExitBB->getFirstInsertionPt()); | ||||||
| 2060 | } | ||||||
| 2061 | |||||||
| 2062 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::emitCommonDirectiveExit( | ||||||
| 2063 | omp::Directive OMPD, InsertPointTy FinIP, Instruction *ExitCall, | ||||||
| 2064 | bool HasFinalize) { | ||||||
| 2065 | |||||||
| 2066 | Builder.restoreIP(FinIP); | ||||||
| 2067 | |||||||
| 2068 | // If there is finalization to do, emit it before the exit call | ||||||
| 2069 | if (HasFinalize) { | ||||||
| 2070 | assert(!FinalizationStack.empty() &&((void)0) | ||||||
| 2071 | "Unexpected finalization stack state!")((void)0); | ||||||
| 2072 | |||||||
| 2073 | FinalizationInfo Fi = FinalizationStack.pop_back_val(); | ||||||
| 2074 | assert(Fi.DK == OMPD && "Unexpected Directive for Finalization call!")((void)0); | ||||||
| 2075 | |||||||
| 2076 | Fi.FiniCB(FinIP); | ||||||
| 2077 | |||||||
| 2078 | BasicBlock *FiniBB = FinIP.getBlock(); | ||||||
| 2079 | Instruction *FiniBBTI = FiniBB->getTerminator(); | ||||||
| 2080 | |||||||
| 2081 | // set Builder IP for call creation | ||||||
| 2082 | Builder.SetInsertPoint(FiniBBTI); | ||||||
| 2083 | } | ||||||
| 2084 | |||||||
| 2085 | if (!ExitCall) | ||||||
| 2086 | return Builder.saveIP(); | ||||||
| 2087 | |||||||
| 2088 | // place the Exitcall as last instruction before Finalization block terminator | ||||||
| 2089 | ExitCall->removeFromParent(); | ||||||
| 2090 | Builder.Insert(ExitCall); | ||||||
| 2091 | |||||||
| 2092 | return IRBuilder<>::InsertPoint(ExitCall->getParent(), | ||||||
| 2093 | ExitCall->getIterator()); | ||||||
| 2094 | } | ||||||
| 2095 | |||||||
| 2096 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createCopyinClauseBlocks( | ||||||
| 2097 | InsertPointTy IP, Value *MasterAddr, Value *PrivateAddr, | ||||||
| 2098 | llvm::IntegerType *IntPtrTy, bool BranchtoEnd) { | ||||||
| 2099 | if (!IP.isSet()) | ||||||
| 2100 | return IP; | ||||||
| 2101 | |||||||
| 2102 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 2103 | |||||||
| 2104 | // creates the following CFG structure | ||||||
| 2105 | // OMP_Entry : (MasterAddr != PrivateAddr)? | ||||||
| 2106 | // F T | ||||||
| 2107 | // | \ | ||||||
| 2108 | // | copin.not.master | ||||||
| 2109 | // | / | ||||||
| 2110 | // v / | ||||||
| 2111 | // copyin.not.master.end | ||||||
| 2112 | // | | ||||||
| 2113 | // v | ||||||
| 2114 | // OMP.Entry.Next | ||||||
| 2115 | |||||||
| 2116 | BasicBlock *OMP_Entry = IP.getBlock(); | ||||||
| 2117 | Function *CurFn = OMP_Entry->getParent(); | ||||||
| 2118 | BasicBlock *CopyBegin = | ||||||
| 2119 | BasicBlock::Create(M.getContext(), "copyin.not.master", CurFn); | ||||||
| 2120 | BasicBlock *CopyEnd = nullptr; | ||||||
| 2121 | |||||||
| 2122 | // If entry block is terminated, split to preserve the branch to following | ||||||
| 2123 | // basic block (i.e. OMP.Entry.Next), otherwise, leave everything as is. | ||||||
| 2124 | if (isa_and_nonnull<BranchInst>(OMP_Entry->getTerminator())) { | ||||||
| 2125 | CopyEnd = OMP_Entry->splitBasicBlock(OMP_Entry->getTerminator(), | ||||||
| 2126 | "copyin.not.master.end"); | ||||||
| 2127 | OMP_Entry->getTerminator()->eraseFromParent(); | ||||||
| 2128 | } else { | ||||||
| 2129 | CopyEnd = | ||||||
| 2130 | BasicBlock::Create(M.getContext(), "copyin.not.master.end", CurFn); | ||||||
| 2131 | } | ||||||
| 2132 | |||||||
| 2133 | Builder.SetInsertPoint(OMP_Entry); | ||||||
| 2134 | Value *MasterPtr = Builder.CreatePtrToInt(MasterAddr, IntPtrTy); | ||||||
| 2135 | Value *PrivatePtr = Builder.CreatePtrToInt(PrivateAddr, IntPtrTy); | ||||||
| 2136 | Value *cmp = Builder.CreateICmpNE(MasterPtr, PrivatePtr); | ||||||
| 2137 | Builder.CreateCondBr(cmp, CopyBegin, CopyEnd); | ||||||
| 2138 | |||||||
| 2139 | Builder.SetInsertPoint(CopyBegin); | ||||||
| 2140 | if (BranchtoEnd) | ||||||
| 2141 | Builder.SetInsertPoint(Builder.CreateBr(CopyEnd)); | ||||||
| 2142 | |||||||
| 2143 | return Builder.saveIP(); | ||||||
| 2144 | } | ||||||
| 2145 | |||||||
| 2146 | CallInst *OpenMPIRBuilder::createOMPAlloc(const LocationDescription &Loc, | ||||||
| 2147 | Value *Size, Value *Allocator, | ||||||
| 2148 | std::string Name) { | ||||||
| 2149 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 2150 | Builder.restoreIP(Loc.IP); | ||||||
| 2151 | |||||||
| 2152 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 2153 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 2154 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 2155 | Value *Args[] = {ThreadId, Size, Allocator}; | ||||||
| 2156 | |||||||
| 2157 | Function *Fn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_alloc); | ||||||
| 2158 | |||||||
| 2159 | return Builder.CreateCall(Fn, Args, Name); | ||||||
| 2160 | } | ||||||
| 2161 | |||||||
| 2162 | CallInst *OpenMPIRBuilder::createOMPFree(const LocationDescription &Loc, | ||||||
| 2163 | Value *Addr, Value *Allocator, | ||||||
| 2164 | std::string Name) { | ||||||
| 2165 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 2166 | Builder.restoreIP(Loc.IP); | ||||||
| 2167 | |||||||
| 2168 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 2169 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 2170 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 2171 | Value *Args[] = {ThreadId, Addr, Allocator}; | ||||||
| 2172 | Function *Fn = getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_free); | ||||||
| 2173 | return Builder.CreateCall(Fn, Args, Name); | ||||||
| 2174 | } | ||||||
| 2175 | |||||||
| 2176 | CallInst *OpenMPIRBuilder::createCachedThreadPrivate( | ||||||
| 2177 | const LocationDescription &Loc, llvm::Value *Pointer, | ||||||
| 2178 | llvm::ConstantInt *Size, const llvm::Twine &Name) { | ||||||
| 2179 | IRBuilder<>::InsertPointGuard IPG(Builder); | ||||||
| 2180 | Builder.restoreIP(Loc.IP); | ||||||
| 2181 | |||||||
| 2182 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 2183 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 2184 | Value *ThreadId = getOrCreateThreadID(Ident); | ||||||
| 2185 | Constant *ThreadPrivateCache = | ||||||
| 2186 | getOrCreateOMPInternalVariable(Int8PtrPtr, Name); | ||||||
| 2187 | llvm::Value *Args[] = {Ident, ThreadId, Pointer, Size, ThreadPrivateCache}; | ||||||
| 2188 | |||||||
| 2189 | Function *Fn = | ||||||
| 2190 | getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_threadprivate_cached); | ||||||
| 2191 | |||||||
| 2192 | return Builder.CreateCall(Fn, Args); | ||||||
| 2193 | } | ||||||
| 2194 | |||||||
| 2195 | OpenMPIRBuilder::InsertPointTy | ||||||
| 2196 | OpenMPIRBuilder::createTargetInit(const LocationDescription &Loc, bool IsSPMD, bool RequiresFullRuntime) { | ||||||
| 2197 | if (!updateToLocation(Loc)) | ||||||
| 2198 | return Loc.IP; | ||||||
| 2199 | |||||||
| 2200 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 2201 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 2202 | ConstantInt *IsSPMDVal = ConstantInt::getBool(Int32->getContext(), IsSPMD); | ||||||
| 2203 | ConstantInt *UseGenericStateMachine = | ||||||
| 2204 | ConstantInt::getBool(Int32->getContext(), !IsSPMD); | ||||||
| 2205 | ConstantInt *RequiresFullRuntimeVal = ConstantInt::getBool(Int32->getContext(), RequiresFullRuntime); | ||||||
| 2206 | |||||||
| 2207 | Function *Fn = getOrCreateRuntimeFunctionPtr( | ||||||
| 2208 | omp::RuntimeFunction::OMPRTL___kmpc_target_init); | ||||||
| 2209 | |||||||
| 2210 | CallInst *ThreadKind = | ||||||
| 2211 | Builder.CreateCall(Fn, {Ident, IsSPMDVal, UseGenericStateMachine, RequiresFullRuntimeVal}); | ||||||
| 2212 | |||||||
| 2213 | Value *ExecUserCode = Builder.CreateICmpEQ( | ||||||
| 2214 | ThreadKind, ConstantInt::get(ThreadKind->getType(), -1), "exec_user_code"); | ||||||
| 2215 | |||||||
| 2216 | // ThreadKind = __kmpc_target_init(...) | ||||||
| 2217 | // if (ThreadKind == -1) | ||||||
| 2218 | // user_code | ||||||
| 2219 | // else | ||||||
| 2220 | // return; | ||||||
| 2221 | |||||||
| 2222 | auto *UI = Builder.CreateUnreachable(); | ||||||
| 2223 | BasicBlock *CheckBB = UI->getParent(); | ||||||
| 2224 | BasicBlock *UserCodeEntryBB = CheckBB->splitBasicBlock(UI, "user_code.entry"); | ||||||
| 2225 | |||||||
| 2226 | BasicBlock *WorkerExitBB = BasicBlock::Create( | ||||||
| 2227 | CheckBB->getContext(), "worker.exit", CheckBB->getParent()); | ||||||
| 2228 | Builder.SetInsertPoint(WorkerExitBB); | ||||||
| 2229 | Builder.CreateRetVoid(); | ||||||
| 2230 | |||||||
| 2231 | auto *CheckBBTI = CheckBB->getTerminator(); | ||||||
| 2232 | Builder.SetInsertPoint(CheckBBTI); | ||||||
| 2233 | Builder.CreateCondBr(ExecUserCode, UI->getParent(), WorkerExitBB); | ||||||
| 2234 | |||||||
| 2235 | CheckBBTI->eraseFromParent(); | ||||||
| 2236 | UI->eraseFromParent(); | ||||||
| 2237 | |||||||
| 2238 | // Continue in the "user_code" block, see diagram above and in | ||||||
| 2239 | // openmp/libomptarget/deviceRTLs/common/include/target.h . | ||||||
| 2240 | return InsertPointTy(UserCodeEntryBB, UserCodeEntryBB->getFirstInsertionPt()); | ||||||
| 2241 | } | ||||||
| 2242 | |||||||
| 2243 | void OpenMPIRBuilder::createTargetDeinit(const LocationDescription &Loc, | ||||||
| 2244 | bool IsSPMD, bool RequiresFullRuntime) { | ||||||
| 2245 | if (!updateToLocation(Loc)) | ||||||
| 2246 | return; | ||||||
| 2247 | |||||||
| 2248 | Constant *SrcLocStr = getOrCreateSrcLocStr(Loc); | ||||||
| 2249 | Value *Ident = getOrCreateIdent(SrcLocStr); | ||||||
| 2250 | ConstantInt *IsSPMDVal = ConstantInt::getBool(Int32->getContext(), IsSPMD); | ||||||
| 2251 | ConstantInt *RequiresFullRuntimeVal = ConstantInt::getBool(Int32->getContext(), RequiresFullRuntime); | ||||||
| 2252 | |||||||
| 2253 | Function *Fn = getOrCreateRuntimeFunctionPtr( | ||||||
| 2254 | omp::RuntimeFunction::OMPRTL___kmpc_target_deinit); | ||||||
| 2255 | |||||||
| 2256 | Builder.CreateCall(Fn, {Ident, IsSPMDVal, RequiresFullRuntimeVal}); | ||||||
| 2257 | } | ||||||
| 2258 | |||||||
| 2259 | std::string OpenMPIRBuilder::getNameWithSeparators(ArrayRef<StringRef> Parts, | ||||||
| 2260 | StringRef FirstSeparator, | ||||||
| 2261 | StringRef Separator) { | ||||||
| 2262 | SmallString<128> Buffer; | ||||||
| 2263 | llvm::raw_svector_ostream OS(Buffer); | ||||||
| 2264 | StringRef Sep = FirstSeparator; | ||||||
| 2265 | for (StringRef Part : Parts) { | ||||||
| 2266 | OS << Sep << Part; | ||||||
| 2267 | Sep = Separator; | ||||||
| 2268 | } | ||||||
| 2269 | return OS.str().str(); | ||||||
| 2270 | } | ||||||
| 2271 | |||||||
| 2272 | Constant *OpenMPIRBuilder::getOrCreateOMPInternalVariable( | ||||||
| 2273 | llvm::Type *Ty, const llvm::Twine &Name, unsigned AddressSpace) { | ||||||
| 2274 | // TODO: Replace the twine arg with stringref to get rid of the conversion | ||||||
| 2275 | // logic. However This is taken from current implementation in clang as is. | ||||||
| 2276 | // Since this method is used in many places exclusively for OMP internal use | ||||||
| 2277 | // we will keep it as is for temporarily until we move all users to the | ||||||
| 2278 | // builder and then, if possible, fix it everywhere in one go. | ||||||
| 2279 | SmallString<256> Buffer; | ||||||
| 2280 | llvm::raw_svector_ostream Out(Buffer); | ||||||
| 2281 | Out << Name; | ||||||
| 2282 | StringRef RuntimeName = Out.str(); | ||||||
| 2283 | auto &Elem = *InternalVars.try_emplace(RuntimeName, nullptr).first; | ||||||
| 2284 | if (Elem.second) { | ||||||
| 2285 | assert(Elem.second->getType()->getPointerElementType() == Ty &&((void)0) | ||||||
| 2286 | "OMP internal variable has different type than requested")((void)0); | ||||||
| 2287 | } else { | ||||||
| 2288 | // TODO: investigate the appropriate linkage type used for the global | ||||||
| 2289 | // variable for possibly changing that to internal or private, or maybe | ||||||
| 2290 | // create different versions of the function for different OMP internal | ||||||
| 2291 | // variables. | ||||||
| 2292 | Elem.second = new llvm::GlobalVariable( | ||||||
| 2293 | M, Ty, /*IsConstant*/ false, llvm::GlobalValue::CommonLinkage, | ||||||
| 2294 | llvm::Constant::getNullValue(Ty), Elem.first(), | ||||||
| 2295 | /*InsertBefore=*/nullptr, llvm::GlobalValue::NotThreadLocal, | ||||||
| 2296 | AddressSpace); | ||||||
| 2297 | } | ||||||
| 2298 | |||||||
| 2299 | return Elem.second; | ||||||
| 2300 | } | ||||||
| 2301 | |||||||
| 2302 | Value *OpenMPIRBuilder::getOMPCriticalRegionLock(StringRef CriticalName) { | ||||||
| 2303 | std::string Prefix = Twine("gomp_critical_user_", CriticalName).str(); | ||||||
| 2304 | std::string Name = getNameWithSeparators({Prefix, "var"}, ".", "."); | ||||||
| 2305 | return getOrCreateOMPInternalVariable(KmpCriticalNameTy, Name); | ||||||
| 2306 | } | ||||||
| 2307 | |||||||
| 2308 | GlobalVariable * | ||||||
| 2309 | OpenMPIRBuilder::createOffloadMaptypes(SmallVectorImpl<uint64_t> &Mappings, | ||||||
| 2310 | std::string VarName) { | ||||||
| 2311 | llvm::Constant *MaptypesArrayInit = | ||||||
| 2312 | llvm::ConstantDataArray::get(M.getContext(), Mappings); | ||||||
| 2313 | auto *MaptypesArrayGlobal = new llvm::GlobalVariable( | ||||||
| 2314 | M, MaptypesArrayInit->getType(), | ||||||
| 2315 | /*isConstant=*/true, llvm::GlobalValue::PrivateLinkage, MaptypesArrayInit, | ||||||
| 2316 | VarName); | ||||||
| 2317 | MaptypesArrayGlobal->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global); | ||||||
| 2318 | return MaptypesArrayGlobal; | ||||||
| 2319 | } | ||||||
| 2320 | |||||||
| 2321 | void OpenMPIRBuilder::createMapperAllocas(const LocationDescription &Loc, | ||||||
| 2322 | InsertPointTy AllocaIP, | ||||||
| 2323 | unsigned NumOperands, | ||||||
| 2324 | struct MapperAllocas &MapperAllocas) { | ||||||
| 2325 | if (!updateToLocation(Loc)) | ||||||
| 2326 | return; | ||||||
| 2327 | |||||||
| 2328 | auto *ArrI8PtrTy = ArrayType::get(Int8Ptr, NumOperands); | ||||||
| 2329 | auto *ArrI64Ty = ArrayType::get(Int64, NumOperands); | ||||||
| 2330 | Builder.restoreIP(AllocaIP); | ||||||
| 2331 | AllocaInst *ArgsBase = Builder.CreateAlloca(ArrI8PtrTy); | ||||||
| 2332 | AllocaInst *Args = Builder.CreateAlloca(ArrI8PtrTy); | ||||||
| 2333 | AllocaInst *ArgSizes = Builder.CreateAlloca(ArrI64Ty); | ||||||
| 2334 | Builder.restoreIP(Loc.IP); | ||||||
| 2335 | MapperAllocas.ArgsBase = ArgsBase; | ||||||
| 2336 | MapperAllocas.Args = Args; | ||||||
| 2337 | MapperAllocas.ArgSizes = ArgSizes; | ||||||
| 2338 | } | ||||||
| 2339 | |||||||
| 2340 | void OpenMPIRBuilder::emitMapperCall(const LocationDescription &Loc, | ||||||
| 2341 | Function *MapperFunc, Value *SrcLocInfo, | ||||||
| 2342 | Value *MaptypesArg, Value *MapnamesArg, | ||||||
| 2343 | struct MapperAllocas &MapperAllocas, | ||||||
| 2344 | int64_t DeviceID, unsigned NumOperands) { | ||||||
| 2345 | if (!updateToLocation(Loc)) | ||||||
| 2346 | return; | ||||||
| 2347 | |||||||
| 2348 | auto *ArrI8PtrTy = ArrayType::get(Int8Ptr, NumOperands); | ||||||
| 2349 | auto *ArrI64Ty = ArrayType::get(Int64, NumOperands); | ||||||
| 2350 | Value *ArgsBaseGEP = | ||||||
| 2351 | Builder.CreateInBoundsGEP(ArrI8PtrTy, MapperAllocas.ArgsBase, | ||||||
| 2352 | {Builder.getInt32(0), Builder.getInt32(0)}); | ||||||
| 2353 | Value *ArgsGEP = | ||||||
| 2354 | Builder.CreateInBoundsGEP(ArrI8PtrTy, MapperAllocas.Args, | ||||||
| 2355 | {Builder.getInt32(0), Builder.getInt32(0)}); | ||||||
| 2356 | Value *ArgSizesGEP = | ||||||
| 2357 | Builder.CreateInBoundsGEP(ArrI64Ty, MapperAllocas.ArgSizes, | ||||||
| 2358 | {Builder.getInt32(0), Builder.getInt32(0)}); | ||||||
| 2359 | Value *NullPtr = Constant::getNullValue(Int8Ptr->getPointerTo()); | ||||||
| 2360 | Builder.CreateCall(MapperFunc, | ||||||
| 2361 | {SrcLocInfo, Builder.getInt64(DeviceID), | ||||||
| 2362 | Builder.getInt32(NumOperands), ArgsBaseGEP, ArgsGEP, | ||||||
| 2363 | ArgSizesGEP, MaptypesArg, MapnamesArg, NullPtr}); | ||||||
| 2364 | } | ||||||
| 2365 | |||||||
| 2366 | bool OpenMPIRBuilder::checkAndEmitFlushAfterAtomic( | ||||||
| 2367 | const LocationDescription &Loc, llvm::AtomicOrdering AO, AtomicKind AK) { | ||||||
| 2368 | assert(!(AO == AtomicOrdering::NotAtomic ||((void)0) | ||||||
| 2369 | AO == llvm::AtomicOrdering::Unordered) &&((void)0) | ||||||
| 2370 | "Unexpected Atomic Ordering.")((void)0); | ||||||
| 2371 | |||||||
| 2372 | bool Flush = false; | ||||||
| 2373 | llvm::AtomicOrdering FlushAO = AtomicOrdering::Monotonic; | ||||||
| 2374 | |||||||
| 2375 | switch (AK) { | ||||||
| 2376 | case Read: | ||||||
| 2377 | if (AO == AtomicOrdering::Acquire || AO == AtomicOrdering::AcquireRelease || | ||||||
| 2378 | AO == AtomicOrdering::SequentiallyConsistent) { | ||||||
| 2379 | FlushAO = AtomicOrdering::Acquire; | ||||||
| 2380 | Flush = true; | ||||||
| 2381 | } | ||||||
| 2382 | break; | ||||||
| 2383 | case Write: | ||||||
| 2384 | case Update: | ||||||
| 2385 | if (AO == AtomicOrdering::Release || AO == AtomicOrdering::AcquireRelease || | ||||||
| 2386 | AO == AtomicOrdering::SequentiallyConsistent) { | ||||||
| 2387 | FlushAO = AtomicOrdering::Release; | ||||||
| 2388 | Flush = true; | ||||||
| 2389 | } | ||||||
| 2390 | break; | ||||||
| 2391 | case Capture: | ||||||
| 2392 | switch (AO) { | ||||||
| 2393 | case AtomicOrdering::Acquire: | ||||||
| 2394 | FlushAO = AtomicOrdering::Acquire; | ||||||
| 2395 | Flush = true; | ||||||
| 2396 | break; | ||||||
| 2397 | case AtomicOrdering::Release: | ||||||
| 2398 | FlushAO = AtomicOrdering::Release; | ||||||
| 2399 | Flush = true; | ||||||
| 2400 | break; | ||||||
| 2401 | case AtomicOrdering::AcquireRelease: | ||||||
| 2402 | case AtomicOrdering::SequentiallyConsistent: | ||||||
| 2403 | FlushAO = AtomicOrdering::AcquireRelease; | ||||||
| 2404 | Flush = true; | ||||||
| 2405 | break; | ||||||
| 2406 | default: | ||||||
| 2407 | // do nothing - leave silently. | ||||||
| 2408 | break; | ||||||
| 2409 | } | ||||||
| 2410 | } | ||||||
| 2411 | |||||||
| 2412 | if (Flush) { | ||||||
| 2413 | // Currently Flush RT call still doesn't take memory_ordering, so for when | ||||||
| 2414 | // that happens, this tries to do the resolution of which atomic ordering | ||||||
| 2415 | // to use with but issue the flush call | ||||||
| 2416 | // TODO: pass `FlushAO` after memory ordering support is added | ||||||
| 2417 | (void)FlushAO; | ||||||
| 2418 | emitFlush(Loc); | ||||||
| 2419 | } | ||||||
| 2420 | |||||||
| 2421 | // for AO == AtomicOrdering::Monotonic and all other case combinations | ||||||
| 2422 | // do nothing | ||||||
| 2423 | return Flush; | ||||||
| 2424 | } | ||||||
| 2425 | |||||||
| 2426 | OpenMPIRBuilder::InsertPointTy | ||||||
| 2427 | OpenMPIRBuilder::createAtomicRead(const LocationDescription &Loc, | ||||||
| 2428 | AtomicOpValue &X, AtomicOpValue &V, | ||||||
| 2429 | AtomicOrdering AO) { | ||||||
| 2430 | if (!updateToLocation(Loc)) | ||||||
| 2431 | return Loc.IP; | ||||||
| 2432 | |||||||
| 2433 | Type *XTy = X.Var->getType(); | ||||||
| 2434 | assert(XTy->isPointerTy() && "OMP Atomic expects a pointer to target memory")((void)0); | ||||||
| 2435 | Type *XElemTy = XTy->getPointerElementType(); | ||||||
| 2436 | assert((XElemTy->isFloatingPointTy() || XElemTy->isIntegerTy() ||((void)0) | ||||||
| 2437 | XElemTy->isPointerTy()) &&((void)0) | ||||||
| 2438 | "OMP atomic read expected a scalar type")((void)0); | ||||||
| 2439 | |||||||
| 2440 | Value *XRead = nullptr; | ||||||
| 2441 | |||||||
| 2442 | if (XElemTy->isIntegerTy()) { | ||||||
| 2443 | LoadInst *XLD = | ||||||
| 2444 | Builder.CreateLoad(XElemTy, X.Var, X.IsVolatile, "omp.atomic.read"); | ||||||
| 2445 | XLD->setAtomic(AO); | ||||||
| 2446 | XRead = cast<Value>(XLD); | ||||||
| 2447 | } else { | ||||||
| 2448 | // We need to bitcast and perform atomic op as integer | ||||||
| 2449 | unsigned Addrspace = cast<PointerType>(XTy)->getAddressSpace(); | ||||||
| 2450 | IntegerType *IntCastTy = | ||||||
| 2451 | IntegerType::get(M.getContext(), XElemTy->getScalarSizeInBits()); | ||||||
| 2452 | Value *XBCast = Builder.CreateBitCast( | ||||||
| 2453 | X.Var, IntCastTy->getPointerTo(Addrspace), "atomic.src.int.cast"); | ||||||
| 2454 | LoadInst *XLoad = | ||||||
| 2455 | Builder.CreateLoad(IntCastTy, XBCast, X.IsVolatile, "omp.atomic.load"); | ||||||
| 2456 | XLoad->setAtomic(AO); | ||||||
| 2457 | if (XElemTy->isFloatingPointTy()) { | ||||||
| 2458 | XRead = Builder.CreateBitCast(XLoad, XElemTy, "atomic.flt.cast"); | ||||||
| 2459 | } else { | ||||||
| 2460 | XRead = Builder.CreateIntToPtr(XLoad, XElemTy, "atomic.ptr.cast"); | ||||||
| 2461 | } | ||||||
| 2462 | } | ||||||
| 2463 | checkAndEmitFlushAfterAtomic(Loc, AO, AtomicKind::Read); | ||||||
| 2464 | Builder.CreateStore(XRead, V.Var, V.IsVolatile); | ||||||
| 2465 | return Builder.saveIP(); | ||||||
| 2466 | } | ||||||
| 2467 | |||||||
| 2468 | OpenMPIRBuilder::InsertPointTy | ||||||
| 2469 | OpenMPIRBuilder::createAtomicWrite(const LocationDescription &Loc, | ||||||
| 2470 | AtomicOpValue &X, Value *Expr, | ||||||
| 2471 | AtomicOrdering AO) { | ||||||
| 2472 | if (!updateToLocation(Loc)) | ||||||
| 2473 | return Loc.IP; | ||||||
| 2474 | |||||||
| 2475 | Type *XTy = X.Var->getType(); | ||||||
| 2476 | assert(XTy->isPointerTy() && "OMP Atomic expects a pointer to target memory")((void)0); | ||||||
| 2477 | Type *XElemTy = XTy->getPointerElementType(); | ||||||
| 2478 | assert((XElemTy->isFloatingPointTy() || XElemTy->isIntegerTy() ||((void)0) | ||||||
| 2479 | XElemTy->isPointerTy()) &&((void)0) | ||||||
| 2480 | "OMP atomic write expected a scalar type")((void)0); | ||||||
| 2481 | |||||||
| 2482 | if (XElemTy->isIntegerTy()) { | ||||||
| 2483 | StoreInst *XSt = Builder.CreateStore(Expr, X.Var, X.IsVolatile); | ||||||
| 2484 | XSt->setAtomic(AO); | ||||||
| 2485 | } else { | ||||||
| 2486 | // We need to bitcast and perform atomic op as integers | ||||||
| 2487 | unsigned Addrspace = cast<PointerType>(XTy)->getAddressSpace(); | ||||||
| 2488 | IntegerType *IntCastTy = | ||||||
| 2489 | IntegerType::get(M.getContext(), XElemTy->getScalarSizeInBits()); | ||||||
| 2490 | Value *XBCast = Builder.CreateBitCast( | ||||||
| 2491 | X.Var, IntCastTy->getPointerTo(Addrspace), "atomic.dst.int.cast"); | ||||||
| 2492 | Value *ExprCast = | ||||||
| 2493 | Builder.CreateBitCast(Expr, IntCastTy, "atomic.src.int.cast"); | ||||||
| 2494 | StoreInst *XSt = Builder.CreateStore(ExprCast, XBCast, X.IsVolatile); | ||||||
| 2495 | XSt->setAtomic(AO); | ||||||
| 2496 | } | ||||||
| 2497 | |||||||
| 2498 | checkAndEmitFlushAfterAtomic(Loc, AO, AtomicKind::Write); | ||||||
| 2499 | return Builder.saveIP(); | ||||||
| 2500 | } | ||||||
| 2501 | |||||||
| 2502 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createAtomicUpdate( | ||||||
| 2503 | const LocationDescription &Loc, Instruction *AllocIP, AtomicOpValue &X, | ||||||
| 2504 | Value *Expr, AtomicOrdering AO, AtomicRMWInst::BinOp RMWOp, | ||||||
| 2505 | AtomicUpdateCallbackTy &UpdateOp, bool IsXLHSInRHSPart) { | ||||||
| 2506 | if (!updateToLocation(Loc)) | ||||||
| |||||||
| 2507 | return Loc.IP; | ||||||
| 2508 | |||||||
| 2509 | LLVM_DEBUG({do { } while (false) | ||||||
| 2510 | Type *XTy = X.Var->getType();do { } while (false) | ||||||
| 2511 | assert(XTy->isPointerTy() &&do { } while (false) | ||||||
| 2512 | "OMP Atomic expects a pointer to target memory");do { } while (false) | ||||||
| 2513 | Type *XElemTy = XTy->getPointerElementType();do { } while (false) | ||||||
| 2514 | assert((XElemTy->isFloatingPointTy() || XElemTy->isIntegerTy() ||do { } while (false) | ||||||
| 2515 | XElemTy->isPointerTy()) &&do { } while (false) | ||||||
| 2516 | "OMP atomic update expected a scalar type");do { } while (false) | ||||||
| 2517 | assert((RMWOp != AtomicRMWInst::Max) && (RMWOp != AtomicRMWInst::Min) &&do { } while (false) | ||||||
| 2518 | (RMWOp != AtomicRMWInst::UMax) && (RMWOp != AtomicRMWInst::UMin) &&do { } while (false) | ||||||
| 2519 | "OpenMP atomic does not support LT or GT operations");do { } while (false) | ||||||
| 2520 | })do { } while (false); | ||||||
| 2521 | |||||||
| 2522 | emitAtomicUpdate(AllocIP, X.Var, Expr, AO, RMWOp, UpdateOp, X.IsVolatile, | ||||||
| 2523 | IsXLHSInRHSPart); | ||||||
| 2524 | checkAndEmitFlushAfterAtomic(Loc, AO, AtomicKind::Update); | ||||||
| 2525 | return Builder.saveIP(); | ||||||
| 2526 | } | ||||||
| 2527 | |||||||
| 2528 | Value *OpenMPIRBuilder::emitRMWOpAsInstruction(Value *Src1, Value *Src2, | ||||||
| 2529 | AtomicRMWInst::BinOp RMWOp) { | ||||||
| 2530 | switch (RMWOp) { | ||||||
| 2531 | case AtomicRMWInst::Add: | ||||||
| 2532 | return Builder.CreateAdd(Src1, Src2); | ||||||
| 2533 | case AtomicRMWInst::Sub: | ||||||
| 2534 | return Builder.CreateSub(Src1, Src2); | ||||||
| 2535 | case AtomicRMWInst::And: | ||||||
| 2536 | return Builder.CreateAnd(Src1, Src2); | ||||||
| 2537 | case AtomicRMWInst::Nand: | ||||||
| 2538 | return Builder.CreateNeg(Builder.CreateAnd(Src1, Src2)); | ||||||
| 2539 | case AtomicRMWInst::Or: | ||||||
| 2540 | return Builder.CreateOr(Src1, Src2); | ||||||
| 2541 | case AtomicRMWInst::Xor: | ||||||
| 2542 | return Builder.CreateXor(Src1, Src2); | ||||||
| 2543 | case AtomicRMWInst::Xchg: | ||||||
| 2544 | case AtomicRMWInst::FAdd: | ||||||
| 2545 | case AtomicRMWInst::FSub: | ||||||
| 2546 | case AtomicRMWInst::BAD_BINOP: | ||||||
| 2547 | case AtomicRMWInst::Max: | ||||||
| 2548 | case AtomicRMWInst::Min: | ||||||
| 2549 | case AtomicRMWInst::UMax: | ||||||
| 2550 | case AtomicRMWInst::UMin: | ||||||
| 2551 | llvm_unreachable("Unsupported atomic update operation")__builtin_unreachable(); | ||||||
| 2552 | } | ||||||
| 2553 | llvm_unreachable("Unsupported atomic update operation")__builtin_unreachable(); | ||||||
| 2554 | } | ||||||
| 2555 | |||||||
| 2556 | std::pair<Value *, Value *> | ||||||
| 2557 | OpenMPIRBuilder::emitAtomicUpdate(Instruction *AllocIP, Value *X, Value *Expr, | ||||||
| 2558 | AtomicOrdering AO, AtomicRMWInst::BinOp RMWOp, | ||||||
| 2559 | AtomicUpdateCallbackTy &UpdateOp, | ||||||
| 2560 | bool VolatileX, bool IsXLHSInRHSPart) { | ||||||
| 2561 | Type *XElemTy = X->getType()->getPointerElementType(); | ||||||
| 2562 | |||||||
| 2563 | bool DoCmpExch = | ||||||
| 2564 | ((RMWOp == AtomicRMWInst::BAD_BINOP) || (RMWOp == AtomicRMWInst::FAdd)) || | ||||||
| 2565 | (RMWOp == AtomicRMWInst::FSub) || | ||||||
| 2566 | (RMWOp == AtomicRMWInst::Sub && !IsXLHSInRHSPart); | ||||||
| 2567 | |||||||
| 2568 | std::pair<Value *, Value *> Res; | ||||||
| 2569 | if (XElemTy->isIntegerTy() && !DoCmpExch) { | ||||||
| 2570 | Res.first = Builder.CreateAtomicRMW(RMWOp, X, Expr, llvm::MaybeAlign(), AO); | ||||||
| 2571 | // not needed except in case of postfix captures. Generate anyway for | ||||||
| 2572 | // consistency with the else part. Will be removed with any DCE pass. | ||||||
| 2573 | Res.second = emitRMWOpAsInstruction(Res.first, Expr, RMWOp); | ||||||
| 2574 | } else { | ||||||
| 2575 | unsigned Addrspace = cast<PointerType>(X->getType())->getAddressSpace(); | ||||||
| 2576 | IntegerType *IntCastTy = | ||||||
| 2577 | IntegerType::get(M.getContext(), XElemTy->getScalarSizeInBits()); | ||||||
| 2578 | Value *XBCast = | ||||||
| 2579 | Builder.CreateBitCast(X, IntCastTy->getPointerTo(Addrspace)); | ||||||
| 2580 | LoadInst *OldVal = | ||||||
| 2581 | Builder.CreateLoad(IntCastTy, XBCast, X->getName() + ".atomic.load"); | ||||||
| 2582 | OldVal->setAtomic(AO); | ||||||
| 2583 | // CurBB | ||||||
| 2584 | // | /---\ | ||||||
| 2585 | // ContBB | | ||||||
| 2586 | // | \---/ | ||||||
| 2587 | // ExitBB | ||||||
| 2588 | BasicBlock *CurBB = Builder.GetInsertBlock(); | ||||||
| 2589 | Instruction *CurBBTI = CurBB->getTerminator(); | ||||||
| 2590 | CurBBTI = CurBBTI ? CurBBTI : Builder.CreateUnreachable(); | ||||||
| 2591 | BasicBlock *ExitBB = | ||||||
| 2592 | CurBB->splitBasicBlock(CurBBTI, X->getName() + ".atomic.exit"); | ||||||
| 2593 | BasicBlock *ContBB = CurBB->splitBasicBlock(CurBB->getTerminator(), | ||||||
| 2594 | X->getName() + ".atomic.cont"); | ||||||
| 2595 | ContBB->getTerminator()->eraseFromParent(); | ||||||
| 2596 | Builder.SetInsertPoint(ContBB); | ||||||
| 2597 | llvm::PHINode *PHI = Builder.CreatePHI(OldVal->getType(), 2); | ||||||
| 2598 | PHI->addIncoming(OldVal, CurBB); | ||||||
| 2599 | AllocaInst *NewAtomicAddr = Builder.CreateAlloca(XElemTy); | ||||||
| 2600 | NewAtomicAddr->setName(X->getName() + "x.new.val"); | ||||||
| 2601 | NewAtomicAddr->moveBefore(AllocIP); | ||||||
| 2602 | IntegerType *NewAtomicCastTy = | ||||||
| 2603 | IntegerType::get(M.getContext(), XElemTy->getScalarSizeInBits()); | ||||||
| 2604 | bool IsIntTy = XElemTy->isIntegerTy(); | ||||||
| 2605 | Value *NewAtomicIntAddr = | ||||||
| 2606 | (IsIntTy
| ||||||
| 2607 | ? NewAtomicAddr | ||||||
| 2608 | : Builder.CreateBitCast(NewAtomicAddr, | ||||||
| 2609 | NewAtomicCastTy->getPointerTo(Addrspace)); | ||||||
| 2610 | Value *OldExprVal = PHI; | ||||||
| 2611 | if (!IsIntTy
| ||||||
| 2612 | if (XElemTy->isFloatingPointTy()) { | ||||||
| 2613 | OldExprVal = Builder.CreateBitCast(PHI, XElemTy, | ||||||
| 2614 | X->getName() + ".atomic.fltCast"); | ||||||
| 2615 | } else { | ||||||
| 2616 | OldExprVal = Builder.CreateIntToPtr(PHI, XElemTy, | ||||||
| 2617 | X->getName() + ".atomic.ptrCast"); | ||||||
| 2618 | } | ||||||
| 2619 | } | ||||||
| 2620 | |||||||
| 2621 | Value *Upd = UpdateOp(OldExprVal, Builder); | ||||||
| 2622 | Builder.CreateStore(Upd, NewAtomicAddr); | ||||||
| 2623 | LoadInst *DesiredVal = Builder.CreateLoad(XElemTy, NewAtomicIntAddr); | ||||||
| 2624 | Value *XAddr = | ||||||
| 2625 | (IsIntTy
| ||||||
| 2626 | ? X | ||||||
| 2627 | : Builder.CreateBitCast(X, IntCastTy->getPointerTo(Addrspace)); | ||||||
| 2628 | AtomicOrdering Failure = | ||||||
| 2629 | llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO); | ||||||
| 2630 | AtomicCmpXchgInst *Result = Builder.CreateAtomicCmpXchg( | ||||||
| 2631 | XAddr, OldExprVal, DesiredVal, llvm::MaybeAlign(), AO, Failure); | ||||||
| 2632 | Result->setVolatile(VolatileX); | ||||||
| 2633 | Value *PreviousVal = Builder.CreateExtractValue(Result, /*Idxs=*/0); | ||||||
| 2634 | Value *SuccessFailureVal = Builder.CreateExtractValue(Result, /*Idxs=*/1); | ||||||
| 2635 | PHI->addIncoming(PreviousVal, Builder.GetInsertBlock()); | ||||||
| 2636 | Builder.CreateCondBr(SuccessFailureVal, ExitBB, ContBB); | ||||||
| 2637 | |||||||
| 2638 | Res.first = OldExprVal; | ||||||
| 2639 | Res.second = Upd; | ||||||
| 2640 | |||||||
| 2641 | // set Insertion point in exit block | ||||||
| 2642 | if (UnreachableInst *ExitTI
| ||||||
| 2643 | dyn_cast<UnreachableInst>(ExitBB->getTerminator())) { | ||||||
| 2644 | CurBBTI->eraseFromParent(); | ||||||
| 2645 | Builder.SetInsertPoint(ExitBB); | ||||||
| 2646 | } else { | ||||||
| 2647 | Builder.SetInsertPoint(ExitTI); | ||||||
| 2648 | } | ||||||
| 2649 | } | ||||||
| 2650 | |||||||
| 2651 | return Res; | ||||||
| 2652 | } | ||||||
| 2653 | |||||||
| 2654 | OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createAtomicCapture( | ||||||
| 2655 | const LocationDescription &Loc, Instruction *AllocIP, AtomicOpValue &X, | ||||||
| 2656 | AtomicOpValue &V, Value *Expr, AtomicOrdering AO, | ||||||
| 2657 | AtomicRMWInst::BinOp RMWOp, AtomicUpdateCallbackTy &UpdateOp, | ||||||
| 2658 | bool UpdateExpr, bool IsPostfixUpdate, bool IsXLHSInRHSPart) { | ||||||
| 2659 | if (!updateToLocation(Loc)) | ||||||
| 2660 | return Loc.IP; | ||||||
| 2661 | |||||||
| 2662 | LLVM_DEBUG({do { } while (false) | ||||||
| 2663 | Type *XTy = X.Var->getType();do { } while (false) | ||||||
| 2664 | assert(XTy->isPointerTy() &&do { } while (false) | ||||||
| 2665 | "OMP Atomic expects a pointer to target memory");do { } while (false) | ||||||
| 2666 | Type *XElemTy = XTy->getPointerElementType();do { } while (false) | ||||||
| 2667 | assert((XElemTy->isFloatingPointTy() || XElemTy->isIntegerTy() ||do { } while (false) | ||||||
| 2668 | XElemTy->isPointerTy()) &&do { } while (false) | ||||||
| 2669 | "OMP atomic capture expected a scalar type");do { } while (false) | ||||||
| 2670 | assert((RMWOp != AtomicRMWInst::Max) && (RMWOp != AtomicRMWInst::Min) &&do { } while (false) | ||||||
| 2671 | "OpenMP atomic does not support LT or GT operations");do { } while (false) | ||||||
| 2672 | })do { } while (false); | ||||||
| 2673 | |||||||
| 2674 | // If UpdateExpr is 'x' updated with some `expr` not based on 'x', | ||||||
| 2675 | // 'x' is simply atomically rewritten with 'expr'. | ||||||
| 2676 | AtomicRMWInst::BinOp AtomicOp = (UpdateExpr ? RMWOp : AtomicRMWInst::Xchg); | ||||||
| 2677 | std::pair<Value *, Value *> Result = | ||||||
| 2678 | emitAtomicUpdate(AllocIP, X.Var, Expr, AO, AtomicOp, UpdateOp, | ||||||
| 2679 | X.IsVolatile, IsXLHSInRHSPart); | ||||||
| 2680 | |||||||
| 2681 | Value *CapturedVal = (IsPostfixUpdate ? Result.first : Result.second); | ||||||
| 2682 | Builder.CreateStore(CapturedVal, V.Var, V.IsVolatile); | ||||||
| 2683 | |||||||
| 2684 | checkAndEmitFlushAfterAtomic(Loc, AO, AtomicKind::Capture); | ||||||
| 2685 | return Builder.saveIP(); | ||||||
| 2686 | } | ||||||
| 2687 | |||||||
| 2688 | GlobalVariable * | ||||||
| 2689 | OpenMPIRBuilder::createOffloadMapnames(SmallVectorImpl<llvm::Constant *> &Names, | ||||||
| 2690 | std::string VarName) { | ||||||
| 2691 | llvm::Constant *MapNamesArrayInit = llvm::ConstantArray::get( | ||||||
| 2692 | llvm::ArrayType::get( | ||||||
| 2693 | llvm::Type::getInt8Ty(M.getContext())->getPointerTo(), Names.size()), | ||||||
| 2694 | Names); | ||||||
| 2695 | auto *MapNamesArrayGlobal = new llvm::GlobalVariable( | ||||||
| 2696 | M, MapNamesArrayInit->getType(), | ||||||
| 2697 | /*isConstant=*/true, llvm::GlobalValue::PrivateLinkage, MapNamesArrayInit, | ||||||
| 2698 | VarName); | ||||||
| 2699 | return MapNamesArrayGlobal; | ||||||
| 2700 | } | ||||||
| 2701 | |||||||
| 2702 | // Create all simple and struct types exposed by the runtime and remember | ||||||
| 2703 | // the llvm::PointerTypes of them for easy access later. | ||||||
| 2704 | void OpenMPIRBuilder::initializeTypes(Module &M) { | ||||||
| 2705 | LLVMContext &Ctx = M.getContext(); | ||||||
| 2706 | StructType *T; | ||||||
| 2707 | #define OMP_TYPE(VarName, InitValue) VarName = InitValue; | ||||||
| 2708 | #define OMP_ARRAY_TYPE(VarName, ElemTy, ArraySize) \ | ||||||
| 2709 | VarName##Ty = ArrayType::get(ElemTy, ArraySize); \ | ||||||
| 2710 | VarName##PtrTy = PointerType::getUnqual(VarName##Ty); | ||||||
| 2711 | #define OMP_FUNCTION_TYPE(VarName, IsVarArg, ReturnType, ...) \ | ||||||
| 2712 | VarName = FunctionType::get(ReturnType, {__VA_ARGS__}, IsVarArg); \ | ||||||
| 2713 | VarName##Ptr = PointerType::getUnqual(VarName); | ||||||
| 2714 | #define OMP_STRUCT_TYPE(VarName, StructName, ...) \ | ||||||
| 2715 | T = StructType::getTypeByName(Ctx, StructName); \ | ||||||
| 2716 | if (!T) \ | ||||||
| 2717 | T = StructType::create(Ctx, {__VA_ARGS__}, StructName); \ | ||||||
| 2718 | VarName = T; \ | ||||||
| 2719 | VarName##Ptr = PointerType::getUnqual(T); | ||||||
| 2720 | #include "llvm/Frontend/OpenMP/OMPKinds.def" | ||||||
| 2721 | } | ||||||
| 2722 | |||||||
| 2723 | void OpenMPIRBuilder::OutlineInfo::collectBlocks( | ||||||
| 2724 | SmallPtrSetImpl<BasicBlock *> &BlockSet, | ||||||
| 2725 | SmallVectorImpl<BasicBlock *> &BlockVector) { | ||||||
| 2726 | SmallVector<BasicBlock *, 32> Worklist; | ||||||
| 2727 | BlockSet.insert(EntryBB); | ||||||
| 2728 | BlockSet.insert(ExitBB); | ||||||
| 2729 | |||||||
| 2730 | Worklist.push_back(EntryBB); | ||||||
| 2731 | while (!Worklist.empty()) { | ||||||
| 2732 | BasicBlock *BB = Worklist.pop_back_val(); | ||||||
| 2733 | BlockVector.push_back(BB); | ||||||
| 2734 | for (BasicBlock *SuccBB : successors(BB)) | ||||||
| 2735 | if (BlockSet.insert(SuccBB).second) | ||||||
| 2736 | Worklist.push_back(SuccBB); | ||||||
| 2737 | } | ||||||
| 2738 | } | ||||||
| 2739 | |||||||
| 2740 | void CanonicalLoopInfo::collectControlBlocks( | ||||||
| 2741 | SmallVectorImpl<BasicBlock *> &BBs) { | ||||||
| 2742 | // We only count those BBs as control block for which we do not need to | ||||||
| 2743 | // reverse the CFG, i.e. not the loop body which can contain arbitrary control | ||||||
| 2744 | // flow. For consistency, this also means we do not add the Body block, which | ||||||
| 2745 | // is just the entry to the body code. | ||||||
| 2746 | BBs.reserve(BBs.size() + 6); | ||||||
| 2747 | BBs.append({Preheader, Header, Cond, Latch, Exit, After}); | ||||||
| 2748 | } | ||||||
| 2749 | |||||||
| 2750 | void CanonicalLoopInfo::assertOK() const { | ||||||
| 2751 | #ifndef NDEBUG1 | ||||||
| 2752 | if (!IsValid) | ||||||
| 2753 | return; | ||||||
| 2754 | |||||||
| 2755 | // Verify standard control-flow we use for OpenMP loops. | ||||||
| 2756 | assert(Preheader)((void)0); | ||||||
| 2757 | assert(isa<BranchInst>(Preheader->getTerminator()) &&((void)0) | ||||||
| 2758 | "Preheader must terminate with unconditional branch")((void)0); | ||||||
| 2759 | assert(Preheader->getSingleSuccessor() == Header &&((void)0) | ||||||
| 2760 | "Preheader must jump to header")((void)0); | ||||||
| 2761 | |||||||
| 2762 | assert(Header)((void)0); | ||||||
| 2763 | assert(isa<BranchInst>(Header->getTerminator()) &&((void)0) | ||||||
| 2764 | "Header must terminate with unconditional branch")((void)0); | ||||||
| 2765 | assert(Header->getSingleSuccessor() == Cond &&((void)0) | ||||||
| 2766 | "Header must jump to exiting block")((void)0); | ||||||
| 2767 | |||||||
| 2768 | assert(Cond)((void)0); | ||||||
| 2769 | assert(Cond->getSinglePredecessor() == Header &&((void)0) | ||||||
| 2770 | "Exiting block only reachable from header")((void)0); | ||||||
| 2771 | |||||||
| 2772 | assert(isa<BranchInst>(Cond->getTerminator()) &&((void)0) | ||||||
| 2773 | "Exiting block must terminate with conditional branch")((void)0); | ||||||
| 2774 | assert(size(successors(Cond)) == 2 &&((void)0) | ||||||
| 2775 | "Exiting block must have two successors")((void)0); | ||||||
| 2776 | assert(cast<BranchInst>(Cond->getTerminator())->getSuccessor(0) == Body &&((void)0) | ||||||
| 2777 | "Exiting block's first successor jump to the body")((void)0); | ||||||
| 2778 | assert(cast<BranchInst>(Cond->getTerminator())->getSuccessor(1) == Exit &&((void)0) | ||||||
| 2779 | "Exiting block's second successor must exit the loop")((void)0); | ||||||
| 2780 | |||||||
| 2781 | assert(Body)((void)0); | ||||||
| 2782 | assert(Body->getSinglePredecessor() == Cond &&((void)0) | ||||||
| 2783 | "Body only reachable from exiting block")((void)0); | ||||||
| 2784 | assert(!isa<PHINode>(Body->front()))((void)0); | ||||||
| 2785 | |||||||
| 2786 | assert(Latch)((void)0); | ||||||
| 2787 | assert(isa<BranchInst>(Latch->getTerminator()) &&((void)0) | ||||||
| 2788 | "Latch must terminate with unconditional branch")((void)0); | ||||||
| 2789 | assert(Latch->getSingleSuccessor() == Header && "Latch must jump to header")((void)0); | ||||||
| 2790 | // TODO: To support simple redirecting of the end of the body code that has | ||||||
| 2791 | // multiple; introduce another auxiliary basic block like preheader and after. | ||||||
| 2792 | assert(Latch->getSinglePredecessor() != nullptr)((void)0); | ||||||
| 2793 | assert(!isa<PHINode>(Latch->front()))((void)0); | ||||||
| 2794 | |||||||
| 2795 | assert(Exit)((void)0); | ||||||
| 2796 | assert(isa<BranchInst>(Exit->getTerminator()) &&((void)0) | ||||||
| 2797 | "Exit block must terminate with unconditional branch")((void)0); | ||||||
| 2798 | assert(Exit->getSingleSuccessor() == After &&((void)0) | ||||||
| 2799 | "Exit block must jump to after block")((void)0); | ||||||
| 2800 | |||||||
| 2801 | assert(After)((void)0); | ||||||
| 2802 | assert(After->getSinglePredecessor() == Exit &&((void)0) | ||||||
| 2803 | "After block only reachable from exit block")((void)0); | ||||||
| 2804 | assert(After->empty() || !isa<PHINode>(After->front()))((void)0); | ||||||
| 2805 | |||||||
| 2806 | Instruction *IndVar = getIndVar(); | ||||||
| 2807 | assert(IndVar && "Canonical induction variable not found?")((void)0); | ||||||
| 2808 | assert(isa<IntegerType>(IndVar->getType()) &&((void)0) | ||||||
| 2809 | "Induction variable must be an integer")((void)0); | ||||||
| 2810 | assert(cast<PHINode>(IndVar)->getParent() == Header &&((void)0) | ||||||
| 2811 | "Induction variable must be a PHI in the loop header")((void)0); | ||||||
| 2812 | assert(cast<PHINode>(IndVar)->getIncomingBlock(0) == Preheader)((void)0); | ||||||
| 2813 | assert(((void)0) | ||||||
| 2814 | cast<ConstantInt>(cast<PHINode>(IndVar)->getIncomingValue(0))->isZero())((void)0); | ||||||
| 2815 | assert(cast<PHINode>(IndVar)->getIncomingBlock(1) == Latch)((void)0); | ||||||
| 2816 | |||||||
| 2817 | auto *NextIndVar = cast<PHINode>(IndVar)->getIncomingValue(1); | ||||||
| 2818 | assert(cast<Instruction>(NextIndVar)->getParent() == Latch)((void)0); | ||||||
| 2819 | assert(cast<BinaryOperator>(NextIndVar)->getOpcode() == BinaryOperator::Add)((void)0); | ||||||
| 2820 | assert(cast<BinaryOperator>(NextIndVar)->getOperand(0) == IndVar)((void)0); | ||||||
| 2821 | assert(cast<ConstantInt>(cast<BinaryOperator>(NextIndVar)->getOperand(1))((void)0) | ||||||
| 2822 | ->isOne())((void)0); | ||||||
| 2823 | |||||||
| 2824 | Value *TripCount = getTripCount(); | ||||||
| 2825 | assert(TripCount && "Loop trip count not found?")((void)0); | ||||||
| 2826 | assert(IndVar->getType() == TripCount->getType() &&((void)0) | ||||||
| 2827 | "Trip count and induction variable must have the same type")((void)0); | ||||||
| 2828 | |||||||
| 2829 | auto *CmpI = cast<CmpInst>(&Cond->front()); | ||||||
| 2830 | assert(CmpI->getPredicate() == CmpInst::ICMP_ULT &&((void)0) | ||||||
| 2831 | "Exit condition must be a signed less-than comparison")((void)0); | ||||||
| 2832 | assert(CmpI->getOperand(0) == IndVar &&((void)0) | ||||||
| 2833 | "Exit condition must compare the induction variable")((void)0); | ||||||
| 2834 | assert(CmpI->getOperand(1) == TripCount &&((void)0) | ||||||
| 2835 | "Exit condition must compare with the trip count")((void)0); | ||||||
| 2836 | #endif | ||||||
| 2837 | } |
| 1 | //===- llvm/Type.h - Classes for handling data types ------------*- C++ -*-===// |
| 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 file contains the declaration of the Type class. For more "Type" |
| 10 | // stuff, look in DerivedTypes.h. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #ifndef LLVM_IR_TYPE_H |
| 15 | #define LLVM_IR_TYPE_H |
| 16 | |
| 17 | #include "llvm/ADT/APFloat.h" |
| 18 | #include "llvm/ADT/ArrayRef.h" |
| 19 | #include "llvm/ADT/SmallPtrSet.h" |
| 20 | #include "llvm/Support/CBindingWrapping.h" |
| 21 | #include "llvm/Support/Casting.h" |
| 22 | #include "llvm/Support/Compiler.h" |
| 23 | #include "llvm/Support/ErrorHandling.h" |
| 24 | #include "llvm/Support/TypeSize.h" |
| 25 | #include <cassert> |
| 26 | #include <cstdint> |
| 27 | #include <iterator> |
| 28 | |
| 29 | namespace llvm { |
| 30 | |
| 31 | class IntegerType; |
| 32 | class LLVMContext; |
| 33 | class PointerType; |
| 34 | class raw_ostream; |
| 35 | class StringRef; |
| 36 | |
| 37 | /// The instances of the Type class are immutable: once they are created, |
| 38 | /// they are never changed. Also note that only one instance of a particular |
| 39 | /// type is ever created. Thus seeing if two types are equal is a matter of |
| 40 | /// doing a trivial pointer comparison. To enforce that no two equal instances |
| 41 | /// are created, Type instances can only be created via static factory methods |
| 42 | /// in class Type and in derived classes. Once allocated, Types are never |
| 43 | /// free'd. |
| 44 | /// |
| 45 | class Type { |
| 46 | public: |
| 47 | //===--------------------------------------------------------------------===// |
| 48 | /// Definitions of all of the base types for the Type system. Based on this |
| 49 | /// value, you can cast to a class defined in DerivedTypes.h. |
| 50 | /// Note: If you add an element to this, you need to add an element to the |
| 51 | /// Type::getPrimitiveType function, or else things will break! |
| 52 | /// Also update LLVMTypeKind and LLVMGetTypeKind () in the C binding. |
| 53 | /// |
| 54 | enum TypeID { |
| 55 | // PrimitiveTypes |
| 56 | HalfTyID = 0, ///< 16-bit floating point type |
| 57 | BFloatTyID, ///< 16-bit floating point type (7-bit significand) |
| 58 | FloatTyID, ///< 32-bit floating point type |
| 59 | DoubleTyID, ///< 64-bit floating point type |
| 60 | X86_FP80TyID, ///< 80-bit floating point type (X87) |
| 61 | FP128TyID, ///< 128-bit floating point type (112-bit significand) |
| 62 | PPC_FP128TyID, ///< 128-bit floating point type (two 64-bits, PowerPC) |
| 63 | VoidTyID, ///< type with no size |
| 64 | LabelTyID, ///< Labels |
| 65 | MetadataTyID, ///< Metadata |
| 66 | X86_MMXTyID, ///< MMX vectors (64 bits, X86 specific) |
| 67 | X86_AMXTyID, ///< AMX vectors (8192 bits, X86 specific) |
| 68 | TokenTyID, ///< Tokens |
| 69 | |
| 70 | // Derived types... see DerivedTypes.h file. |
| 71 | IntegerTyID, ///< Arbitrary bit width integers |
| 72 | FunctionTyID, ///< Functions |
| 73 | PointerTyID, ///< Pointers |
| 74 | StructTyID, ///< Structures |
| 75 | ArrayTyID, ///< Arrays |
| 76 | FixedVectorTyID, ///< Fixed width SIMD vector type |
| 77 | ScalableVectorTyID ///< Scalable SIMD vector type |
| 78 | }; |
| 79 | |
| 80 | private: |
| 81 | /// This refers to the LLVMContext in which this type was uniqued. |
| 82 | LLVMContext &Context; |
| 83 | |
| 84 | TypeID ID : 8; // The current base type of this type. |
| 85 | unsigned SubclassData : 24; // Space for subclasses to store data. |
| 86 | // Note that this should be synchronized with |
| 87 | // MAX_INT_BITS value in IntegerType class. |
| 88 | |
| 89 | protected: |
| 90 | friend class LLVMContextImpl; |
| 91 | |
| 92 | explicit Type(LLVMContext &C, TypeID tid) |
| 93 | : Context(C), ID(tid), SubclassData(0) {} |
| 94 | ~Type() = default; |
| 95 | |
| 96 | unsigned getSubclassData() const { return SubclassData; } |
| 97 | |
| 98 | void setSubclassData(unsigned val) { |
| 99 | SubclassData = val; |
| 100 | // Ensure we don't have any accidental truncation. |
| 101 | assert(getSubclassData() == val && "Subclass data too large for field")((void)0); |
| 102 | } |
| 103 | |
| 104 | /// Keeps track of how many Type*'s there are in the ContainedTys list. |
| 105 | unsigned NumContainedTys = 0; |
| 106 | |
| 107 | /// A pointer to the array of Types contained by this Type. For example, this |
| 108 | /// includes the arguments of a function type, the elements of a structure, |
| 109 | /// the pointee of a pointer, the element type of an array, etc. This pointer |
| 110 | /// may be 0 for types that don't contain other types (Integer, Double, |
| 111 | /// Float). |
| 112 | Type * const *ContainedTys = nullptr; |
| 113 | |
| 114 | public: |
| 115 | /// Print the current type. |
| 116 | /// Omit the type details if \p NoDetails == true. |
| 117 | /// E.g., let %st = type { i32, i16 } |
| 118 | /// When \p NoDetails is true, we only print %st. |
| 119 | /// Put differently, \p NoDetails prints the type as if |
| 120 | /// inlined with the operands when printing an instruction. |
| 121 | void print(raw_ostream &O, bool IsForDebug = false, |
| 122 | bool NoDetails = false) const; |
| 123 | |
| 124 | void dump() const; |
| 125 | |
| 126 | /// Return the LLVMContext in which this type was uniqued. |
| 127 | LLVMContext &getContext() const { return Context; } |
| 128 | |
| 129 | //===--------------------------------------------------------------------===// |
| 130 | // Accessors for working with types. |
| 131 | // |
| 132 | |
| 133 | /// Return the type id for the type. This will return one of the TypeID enum |
| 134 | /// elements defined above. |
| 135 | TypeID getTypeID() const { return ID; } |
| 136 | |
| 137 | /// Return true if this is 'void'. |
| 138 | bool isVoidTy() const { return getTypeID() == VoidTyID; } |
| 139 | |
| 140 | /// Return true if this is 'half', a 16-bit IEEE fp type. |
| 141 | bool isHalfTy() const { return getTypeID() == HalfTyID; } |
| 142 | |
| 143 | /// Return true if this is 'bfloat', a 16-bit bfloat type. |
| 144 | bool isBFloatTy() const { return getTypeID() == BFloatTyID; } |
| 145 | |
| 146 | /// Return true if this is 'float', a 32-bit IEEE fp type. |
| 147 | bool isFloatTy() const { return getTypeID() == FloatTyID; } |
| 148 | |
| 149 | /// Return true if this is 'double', a 64-bit IEEE fp type. |
| 150 | bool isDoubleTy() const { return getTypeID() == DoubleTyID; } |
| 151 | |
| 152 | /// Return true if this is x86 long double. |
| 153 | bool isX86_FP80Ty() const { return getTypeID() == X86_FP80TyID; } |
| 154 | |
| 155 | /// Return true if this is 'fp128'. |
| 156 | bool isFP128Ty() const { return getTypeID() == FP128TyID; } |
| 157 | |
| 158 | /// Return true if this is powerpc long double. |
| 159 | bool isPPC_FP128Ty() const { return getTypeID() == PPC_FP128TyID; } |
| 160 | |
| 161 | /// Return true if this is one of the six floating-point types |
| 162 | bool isFloatingPointTy() const { |
| 163 | return getTypeID() == HalfTyID || getTypeID() == BFloatTyID || |
| 164 | getTypeID() == FloatTyID || getTypeID() == DoubleTyID || |
| 165 | getTypeID() == X86_FP80TyID || getTypeID() == FP128TyID || |
| 166 | getTypeID() == PPC_FP128TyID; |
| 167 | } |
| 168 | |
| 169 | const fltSemantics &getFltSemantics() const { |
| 170 | switch (getTypeID()) { |
| 171 | case HalfTyID: return APFloat::IEEEhalf(); |
| 172 | case BFloatTyID: return APFloat::BFloat(); |
| 173 | case FloatTyID: return APFloat::IEEEsingle(); |
| 174 | case DoubleTyID: return APFloat::IEEEdouble(); |
| 175 | case X86_FP80TyID: return APFloat::x87DoubleExtended(); |
| 176 | case FP128TyID: return APFloat::IEEEquad(); |
| 177 | case PPC_FP128TyID: return APFloat::PPCDoubleDouble(); |
| 178 | default: llvm_unreachable("Invalid floating type")__builtin_unreachable(); |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | /// Return true if this is X86 MMX. |
| 183 | bool isX86_MMXTy() const { return getTypeID() == X86_MMXTyID; } |
| 184 | |
| 185 | /// Return true if this is X86 AMX. |
| 186 | bool isX86_AMXTy() const { return getTypeID() == X86_AMXTyID; } |
| 187 | |
| 188 | /// Return true if this is a FP type or a vector of FP. |
| 189 | bool isFPOrFPVectorTy() const { return getScalarType()->isFloatingPointTy(); } |
| 190 | |
| 191 | /// Return true if this is 'label'. |
| 192 | bool isLabelTy() const { return getTypeID() == LabelTyID; } |
| 193 | |
| 194 | /// Return true if this is 'metadata'. |
| 195 | bool isMetadataTy() const { return getTypeID() == MetadataTyID; } |
| 196 | |
| 197 | /// Return true if this is 'token'. |
| 198 | bool isTokenTy() const { return getTypeID() == TokenTyID; } |
| 199 | |
| 200 | /// True if this is an instance of IntegerType. |
| 201 | bool isIntegerTy() const { return getTypeID() == IntegerTyID; } |
| 202 | |
| 203 | /// Return true if this is an IntegerType of the given width. |
| 204 | bool isIntegerTy(unsigned Bitwidth) const; |
| 205 | |
| 206 | /// Return true if this is an integer type or a vector of integer types. |
| 207 | bool isIntOrIntVectorTy() const { return getScalarType()->isIntegerTy(); } |
| 208 | |
| 209 | /// Return true if this is an integer type or a vector of integer types of |
| 210 | /// the given width. |
| 211 | bool isIntOrIntVectorTy(unsigned BitWidth) const { |
| 212 | return getScalarType()->isIntegerTy(BitWidth); |
| 213 | } |
| 214 | |
| 215 | /// Return true if this is an integer type or a pointer type. |
| 216 | bool isIntOrPtrTy() const { return isIntegerTy() || isPointerTy(); } |
| 217 | |
| 218 | /// True if this is an instance of FunctionType. |
| 219 | bool isFunctionTy() const { return getTypeID() == FunctionTyID; } |
| 220 | |
| 221 | /// True if this is an instance of StructType. |
| 222 | bool isStructTy() const { return getTypeID() == StructTyID; } |
| 223 | |
| 224 | /// True if this is an instance of ArrayType. |
| 225 | bool isArrayTy() const { return getTypeID() == ArrayTyID; } |
| 226 | |
| 227 | /// True if this is an instance of PointerType. |
| 228 | bool isPointerTy() const { return getTypeID() == PointerTyID; } |
| 229 | |
| 230 | /// True if this is an instance of an opaque PointerType. |
| 231 | bool isOpaquePointerTy() const; |
| 232 | |
| 233 | /// Return true if this is a pointer type or a vector of pointer types. |
| 234 | bool isPtrOrPtrVectorTy() const { return getScalarType()->isPointerTy(); } |
| 235 | |
| 236 | /// True if this is an instance of VectorType. |
| 237 | inline bool isVectorTy() const { |
| 238 | return getTypeID() == ScalableVectorTyID || getTypeID() == FixedVectorTyID; |
| 239 | } |
| 240 | |
| 241 | /// Return true if this type could be converted with a lossless BitCast to |
| 242 | /// type 'Ty'. For example, i8* to i32*. BitCasts are valid for types of the |
| 243 | /// same size only where no re-interpretation of the bits is done. |
| 244 | /// Determine if this type could be losslessly bitcast to Ty |
| 245 | bool canLosslesslyBitCastTo(Type *Ty) const; |
| 246 | |
| 247 | /// Return true if this type is empty, that is, it has no elements or all of |
| 248 | /// its elements are empty. |
| 249 | bool isEmptyTy() const; |
| 250 | |
| 251 | /// Return true if the type is "first class", meaning it is a valid type for a |
| 252 | /// Value. |
| 253 | bool isFirstClassType() const { |
| 254 | return getTypeID() != FunctionTyID && getTypeID() != VoidTyID; |
| 255 | } |
| 256 | |
| 257 | /// Return true if the type is a valid type for a register in codegen. This |
| 258 | /// includes all first-class types except struct and array types. |
| 259 | bool isSingleValueType() const { |
| 260 | return isFloatingPointTy() || isX86_MMXTy() || isIntegerTy() || |
| 261 | isPointerTy() || isVectorTy() || isX86_AMXTy(); |
| 262 | } |
| 263 | |
| 264 | /// Return true if the type is an aggregate type. This means it is valid as |
| 265 | /// the first operand of an insertvalue or extractvalue instruction. This |
| 266 | /// includes struct and array types, but does not include vector types. |
| 267 | bool isAggregateType() const { |
| 268 | return getTypeID() == StructTyID || getTypeID() == ArrayTyID; |
| 269 | } |
| 270 | |
| 271 | /// Return true if it makes sense to take the size of this type. To get the |
| 272 | /// actual size for a particular target, it is reasonable to use the |
| 273 | /// DataLayout subsystem to do this. |
| 274 | bool isSized(SmallPtrSetImpl<Type*> *Visited = nullptr) const { |
| 275 | // If it's a primitive, it is always sized. |
| 276 | if (getTypeID() == IntegerTyID || isFloatingPointTy() || |
| 277 | getTypeID() == PointerTyID || getTypeID() == X86_MMXTyID || |
| 278 | getTypeID() == X86_AMXTyID) |
| 279 | return true; |
| 280 | // If it is not something that can have a size (e.g. a function or label), |
| 281 | // it doesn't have a size. |
| 282 | if (getTypeID() != StructTyID && getTypeID() != ArrayTyID && !isVectorTy()) |
| 283 | return false; |
| 284 | // Otherwise we have to try harder to decide. |
| 285 | return isSizedDerivedType(Visited); |
| 286 | } |
| 287 | |
| 288 | /// Return the basic size of this type if it is a primitive type. These are |
| 289 | /// fixed by LLVM and are not target-dependent. |
| 290 | /// This will return zero if the type does not have a size or is not a |
| 291 | /// primitive type. |
| 292 | /// |
| 293 | /// If this is a scalable vector type, the scalable property will be set and |
| 294 | /// the runtime size will be a positive integer multiple of the base size. |
| 295 | /// |
| 296 | /// Note that this may not reflect the size of memory allocated for an |
| 297 | /// instance of the type or the number of bytes that are written when an |
| 298 | /// instance of the type is stored to memory. The DataLayout class provides |
| 299 | /// additional query functions to provide this information. |
| 300 | /// |
| 301 | TypeSize getPrimitiveSizeInBits() const LLVM_READONLY__attribute__((__pure__)); |
| 302 | |
| 303 | /// If this is a vector type, return the getPrimitiveSizeInBits value for the |
| 304 | /// element type. Otherwise return the getPrimitiveSizeInBits value for this |
| 305 | /// type. |
| 306 | unsigned getScalarSizeInBits() const LLVM_READONLY__attribute__((__pure__)); |
| 307 | |
| 308 | /// Return the width of the mantissa of this type. This is only valid on |
| 309 | /// floating-point types. If the FP type does not have a stable mantissa (e.g. |
| 310 | /// ppc long double), this method returns -1. |
| 311 | int getFPMantissaWidth() const; |
| 312 | |
| 313 | /// Return whether the type is IEEE compatible, as defined by the eponymous |
| 314 | /// method in APFloat. |
| 315 | bool isIEEE() const { return APFloat::getZero(getFltSemantics()).isIEEE(); } |
| 316 | |
| 317 | /// If this is a vector type, return the element type, otherwise return |
| 318 | /// 'this'. |
| 319 | inline Type *getScalarType() const { |
| 320 | if (isVectorTy()) |
| 321 | return getContainedType(0); |
| 322 | return const_cast<Type *>(this); |
| 323 | } |
| 324 | |
| 325 | //===--------------------------------------------------------------------===// |
| 326 | // Type Iteration support. |
| 327 | // |
| 328 | using subtype_iterator = Type * const *; |
| 329 | |
| 330 | subtype_iterator subtype_begin() const { return ContainedTys; } |
| 331 | subtype_iterator subtype_end() const { return &ContainedTys[NumContainedTys];} |
| 332 | ArrayRef<Type*> subtypes() const { |
| 333 | return makeArrayRef(subtype_begin(), subtype_end()); |
| 334 | } |
| 335 | |
| 336 | using subtype_reverse_iterator = std::reverse_iterator<subtype_iterator>; |
| 337 | |
| 338 | subtype_reverse_iterator subtype_rbegin() const { |
| 339 | return subtype_reverse_iterator(subtype_end()); |
| 340 | } |
| 341 | subtype_reverse_iterator subtype_rend() const { |
| 342 | return subtype_reverse_iterator(subtype_begin()); |
| 343 | } |
| 344 | |
| 345 | /// This method is used to implement the type iterator (defined at the end of |
| 346 | /// the file). For derived types, this returns the types 'contained' in the |
| 347 | /// derived type. |
| 348 | Type *getContainedType(unsigned i) const { |
| 349 | assert(i < NumContainedTys && "Index out of range!")((void)0); |
| 350 | return ContainedTys[i]; |
| 351 | } |
| 352 | |
| 353 | /// Return the number of types in the derived type. |
| 354 | unsigned getNumContainedTypes() const { return NumContainedTys; } |
| 355 | |
| 356 | //===--------------------------------------------------------------------===// |
| 357 | // Helper methods corresponding to subclass methods. This forces a cast to |
| 358 | // the specified subclass and calls its accessor. "getArrayNumElements" (for |
| 359 | // example) is shorthand for cast<ArrayType>(Ty)->getNumElements(). This is |
| 360 | // only intended to cover the core methods that are frequently used, helper |
| 361 | // methods should not be added here. |
| 362 | |
| 363 | inline unsigned getIntegerBitWidth() const; |
| 364 | |
| 365 | inline Type *getFunctionParamType(unsigned i) const; |
| 366 | inline unsigned getFunctionNumParams() const; |
| 367 | inline bool isFunctionVarArg() const; |
| 368 | |
| 369 | inline StringRef getStructName() const; |
| 370 | inline unsigned getStructNumElements() const; |
| 371 | inline Type *getStructElementType(unsigned N) const; |
| 372 | |
| 373 | inline uint64_t getArrayNumElements() const; |
| 374 | |
| 375 | Type *getArrayElementType() const { |
| 376 | assert(getTypeID() == ArrayTyID)((void)0); |
| 377 | return ContainedTys[0]; |
| 378 | } |
| 379 | |
| 380 | Type *getPointerElementType() const { |
| 381 | assert(getTypeID() == PointerTyID)((void)0); |
| 382 | return ContainedTys[0]; |
| 383 | } |
| 384 | |
| 385 | /// Given vector type, change the element type, |
| 386 | /// whilst keeping the old number of elements. |
| 387 | /// For non-vectors simply returns \p EltTy. |
| 388 | inline Type *getWithNewType(Type *EltTy) const; |
| 389 | |
| 390 | /// Given an integer or vector type, change the lane bitwidth to NewBitwidth, |
| 391 | /// whilst keeping the old number of lanes. |
| 392 | inline Type *getWithNewBitWidth(unsigned NewBitWidth) const; |
| 393 | |
| 394 | /// Given scalar/vector integer type, returns a type with elements twice as |
| 395 | /// wide as in the original type. For vectors, preserves element count. |
| 396 | inline Type *getExtendedType() const; |
| 397 | |
| 398 | /// Get the address space of this pointer or pointer vector type. |
| 399 | inline unsigned getPointerAddressSpace() const; |
| 400 | |
| 401 | //===--------------------------------------------------------------------===// |
| 402 | // Static members exported by the Type class itself. Useful for getting |
| 403 | // instances of Type. |
| 404 | // |
| 405 | |
| 406 | /// Return a type based on an identifier. |
| 407 | static Type *getPrimitiveType(LLVMContext &C, TypeID IDNumber); |
| 408 | |
| 409 | //===--------------------------------------------------------------------===// |
| 410 | // These are the builtin types that are always available. |
| 411 | // |
| 412 | static Type *getVoidTy(LLVMContext &C); |
| 413 | static Type *getLabelTy(LLVMContext &C); |
| 414 | static Type *getHalfTy(LLVMContext &C); |
| 415 | static Type *getBFloatTy(LLVMContext &C); |
| 416 | static Type *getFloatTy(LLVMContext &C); |
| 417 | static Type *getDoubleTy(LLVMContext &C); |
| 418 | static Type *getMetadataTy(LLVMContext &C); |
| 419 | static Type *getX86_FP80Ty(LLVMContext &C); |
| 420 | static Type *getFP128Ty(LLVMContext &C); |
| 421 | static Type *getPPC_FP128Ty(LLVMContext &C); |
| 422 | static Type *getX86_MMXTy(LLVMContext &C); |
| 423 | static Type *getX86_AMXTy(LLVMContext &C); |
| 424 | static Type *getTokenTy(LLVMContext &C); |
| 425 | static IntegerType *getIntNTy(LLVMContext &C, unsigned N); |
| 426 | static IntegerType *getInt1Ty(LLVMContext &C); |
| 427 | static IntegerType *getInt8Ty(LLVMContext &C); |
| 428 | static IntegerType *getInt16Ty(LLVMContext &C); |
| 429 | static IntegerType *getInt32Ty(LLVMContext &C); |
| 430 | static IntegerType *getInt64Ty(LLVMContext &C); |
| 431 | static IntegerType *getInt128Ty(LLVMContext &C); |
| 432 | template <typename ScalarTy> static Type *getScalarTy(LLVMContext &C) { |
| 433 | int noOfBits = sizeof(ScalarTy) * CHAR_BIT8; |
| 434 | if (std::is_integral<ScalarTy>::value) { |
| 435 | return (Type*) Type::getIntNTy(C, noOfBits); |
| 436 | } else if (std::is_floating_point<ScalarTy>::value) { |
| 437 | switch (noOfBits) { |
| 438 | case 32: |
| 439 | return Type::getFloatTy(C); |
| 440 | case 64: |
| 441 | return Type::getDoubleTy(C); |
| 442 | } |
| 443 | } |
| 444 | llvm_unreachable("Unsupported type in Type::getScalarTy")__builtin_unreachable(); |
| 445 | } |
| 446 | static Type *getFloatingPointTy(LLVMContext &C, const fltSemantics &S) { |
| 447 | Type *Ty; |
| 448 | if (&S == &APFloat::IEEEhalf()) |
| 449 | Ty = Type::getHalfTy(C); |
| 450 | else if (&S == &APFloat::BFloat()) |
| 451 | Ty = Type::getBFloatTy(C); |
| 452 | else if (&S == &APFloat::IEEEsingle()) |
| 453 | Ty = Type::getFloatTy(C); |
| 454 | else if (&S == &APFloat::IEEEdouble()) |
| 455 | Ty = Type::getDoubleTy(C); |
| 456 | else if (&S == &APFloat::x87DoubleExtended()) |
| 457 | Ty = Type::getX86_FP80Ty(C); |
| 458 | else if (&S == &APFloat::IEEEquad()) |
| 459 | Ty = Type::getFP128Ty(C); |
| 460 | else { |
| 461 | assert(&S == &APFloat::PPCDoubleDouble() && "Unknown FP format")((void)0); |
| 462 | Ty = Type::getPPC_FP128Ty(C); |
| 463 | } |
| 464 | return Ty; |
| 465 | } |
| 466 | |
| 467 | //===--------------------------------------------------------------------===// |
| 468 | // Convenience methods for getting pointer types with one of the above builtin |
| 469 | // types as pointee. |
| 470 | // |
| 471 | static PointerType *getHalfPtrTy(LLVMContext &C, unsigned AS = 0); |
| 472 | static PointerType *getBFloatPtrTy(LLVMContext &C, unsigned AS = 0); |
| 473 | static PointerType *getFloatPtrTy(LLVMContext &C, unsigned AS = 0); |
| 474 | static PointerType *getDoublePtrTy(LLVMContext &C, unsigned AS = 0); |
| 475 | static PointerType *getX86_FP80PtrTy(LLVMContext &C, unsigned AS = 0); |
| 476 | static PointerType *getFP128PtrTy(LLVMContext &C, unsigned AS = 0); |
| 477 | static PointerType *getPPC_FP128PtrTy(LLVMContext &C, unsigned AS = 0); |
| 478 | static PointerType *getX86_MMXPtrTy(LLVMContext &C, unsigned AS = 0); |
| 479 | static PointerType *getX86_AMXPtrTy(LLVMContext &C, unsigned AS = 0); |
| 480 | static PointerType *getIntNPtrTy(LLVMContext &C, unsigned N, unsigned AS = 0); |
| 481 | static PointerType *getInt1PtrTy(LLVMContext &C, unsigned AS = 0); |
| 482 | static PointerType *getInt8PtrTy(LLVMContext &C, unsigned AS = 0); |
| 483 | static PointerType *getInt16PtrTy(LLVMContext &C, unsigned AS = 0); |
| 484 | static PointerType *getInt32PtrTy(LLVMContext &C, unsigned AS = 0); |
| 485 | static PointerType *getInt64PtrTy(LLVMContext &C, unsigned AS = 0); |
| 486 | |
| 487 | /// Return a pointer to the current type. This is equivalent to |
| 488 | /// PointerType::get(Foo, AddrSpace). |
| 489 | /// TODO: Remove this after opaque pointer transition is complete. |
| 490 | PointerType *getPointerTo(unsigned AddrSpace = 0) const; |
| 491 | |
| 492 | private: |
| 493 | /// Derived types like structures and arrays are sized iff all of the members |
| 494 | /// of the type are sized as well. Since asking for their size is relatively |
| 495 | /// uncommon, move this operation out-of-line. |
| 496 | bool isSizedDerivedType(SmallPtrSetImpl<Type*> *Visited = nullptr) const; |
| 497 | }; |
| 498 | |
| 499 | // Printing of types. |
| 500 | inline raw_ostream &operator<<(raw_ostream &OS, const Type &T) { |
| 501 | T.print(OS); |
| 502 | return OS; |
| 503 | } |
| 504 | |
| 505 | // allow isa<PointerType>(x) to work without DerivedTypes.h included. |
| 506 | template <> struct isa_impl<PointerType, Type> { |
| 507 | static inline bool doit(const Type &Ty) { |
| 508 | return Ty.getTypeID() == Type::PointerTyID; |
| 509 | } |
| 510 | }; |
| 511 | |
| 512 | // Create wrappers for C Binding types (see CBindingWrapping.h). |
| 513 | DEFINE_ISA_CONVERSION_FUNCTIONS(Type, LLVMTypeRef)inline Type *unwrap(LLVMTypeRef P) { return reinterpret_cast< Type*>(P); } inline LLVMTypeRef wrap(const Type *P) { return reinterpret_cast<LLVMTypeRef>(const_cast<Type*>( P)); } template<typename T> inline T *unwrap(LLVMTypeRef P) { return cast<T>(unwrap(P)); } |
| 514 | |
| 515 | /* Specialized opaque type conversions. |
| 516 | */ |
| 517 | inline Type **unwrap(LLVMTypeRef* Tys) { |
| 518 | return reinterpret_cast<Type**>(Tys); |
| 519 | } |
| 520 | |
| 521 | inline LLVMTypeRef *wrap(Type **Tys) { |
| 522 | return reinterpret_cast<LLVMTypeRef*>(const_cast<Type**>(Tys)); |
| 523 | } |
| 524 | |
| 525 | } // end namespace llvm |
| 526 | |
| 527 | #endif // LLVM_IR_TYPE_H |
| 1 | //===- llvm/IRBuilder.h - Builder for LLVM Instructions ---------*- C++ -*-===// | |||
| 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 file defines the IRBuilder class, which is used as a convenient way | |||
| 10 | // to create LLVM instructions with a consistent and simplified interface. | |||
| 11 | // | |||
| 12 | //===----------------------------------------------------------------------===// | |||
| 13 | ||||
| 14 | #ifndef LLVM_IR_IRBUILDER_H | |||
| 15 | #define LLVM_IR_IRBUILDER_H | |||
| 16 | ||||
| 17 | #include "llvm-c/Types.h" | |||
| 18 | #include "llvm/ADT/ArrayRef.h" | |||
| 19 | #include "llvm/ADT/None.h" | |||
| 20 | #include "llvm/ADT/STLExtras.h" | |||
| 21 | #include "llvm/ADT/StringRef.h" | |||
| 22 | #include "llvm/ADT/Twine.h" | |||
| 23 | #include "llvm/IR/BasicBlock.h" | |||
| 24 | #include "llvm/IR/Constant.h" | |||
| 25 | #include "llvm/IR/ConstantFolder.h" | |||
| 26 | #include "llvm/IR/Constants.h" | |||
| 27 | #include "llvm/IR/DataLayout.h" | |||
| 28 | #include "llvm/IR/DebugInfoMetadata.h" | |||
| 29 | #include "llvm/IR/DebugLoc.h" | |||
| 30 | #include "llvm/IR/DerivedTypes.h" | |||
| 31 | #include "llvm/IR/Function.h" | |||
| 32 | #include "llvm/IR/GlobalVariable.h" | |||
| 33 | #include "llvm/IR/InstrTypes.h" | |||
| 34 | #include "llvm/IR/Instruction.h" | |||
| 35 | #include "llvm/IR/Instructions.h" | |||
| 36 | #include "llvm/IR/IntrinsicInst.h" | |||
| 37 | #include "llvm/IR/LLVMContext.h" | |||
| 38 | #include "llvm/IR/Module.h" | |||
| 39 | #include "llvm/IR/Operator.h" | |||
| 40 | #include "llvm/IR/Type.h" | |||
| 41 | #include "llvm/IR/Value.h" | |||
| 42 | #include "llvm/IR/ValueHandle.h" | |||
| 43 | #include "llvm/Support/AtomicOrdering.h" | |||
| 44 | #include "llvm/Support/CBindingWrapping.h" | |||
| 45 | #include "llvm/Support/Casting.h" | |||
| 46 | #include <cassert> | |||
| 47 | #include <cstddef> | |||
| 48 | #include <cstdint> | |||
| 49 | #include <functional> | |||
| 50 | #include <utility> | |||
| 51 | ||||
| 52 | namespace llvm { | |||
| 53 | ||||
| 54 | class APInt; | |||
| 55 | class MDNode; | |||
| 56 | class Use; | |||
| 57 | ||||
| 58 | /// This provides the default implementation of the IRBuilder | |||
| 59 | /// 'InsertHelper' method that is called whenever an instruction is created by | |||
| 60 | /// IRBuilder and needs to be inserted. | |||
| 61 | /// | |||
| 62 | /// By default, this inserts the instruction at the insertion point. | |||
| 63 | class IRBuilderDefaultInserter { | |||
| 64 | public: | |||
| 65 | virtual ~IRBuilderDefaultInserter(); | |||
| 66 | ||||
| 67 | virtual void InsertHelper(Instruction *I, const Twine &Name, | |||
| 68 | BasicBlock *BB, | |||
| 69 | BasicBlock::iterator InsertPt) const { | |||
| 70 | if (BB) BB->getInstList().insert(InsertPt, I); | |||
| 71 | I->setName(Name); | |||
| 72 | } | |||
| 73 | }; | |||
| 74 | ||||
| 75 | /// Provides an 'InsertHelper' that calls a user-provided callback after | |||
| 76 | /// performing the default insertion. | |||
| 77 | class IRBuilderCallbackInserter : public IRBuilderDefaultInserter { | |||
| 78 | std::function<void(Instruction *)> Callback; | |||
| 79 | ||||
| 80 | public: | |||
| 81 | virtual ~IRBuilderCallbackInserter(); | |||
| 82 | ||||
| 83 | IRBuilderCallbackInserter(std::function<void(Instruction *)> Callback) | |||
| 84 | : Callback(std::move(Callback)) {} | |||
| 85 | ||||
| 86 | void InsertHelper(Instruction *I, const Twine &Name, | |||
| 87 | BasicBlock *BB, | |||
| 88 | BasicBlock::iterator InsertPt) const override { | |||
| 89 | IRBuilderDefaultInserter::InsertHelper(I, Name, BB, InsertPt); | |||
| 90 | Callback(I); | |||
| 91 | } | |||
| 92 | }; | |||
| 93 | ||||
| 94 | /// Common base class shared among various IRBuilders. | |||
| 95 | class IRBuilderBase { | |||
| 96 | /// Pairs of (metadata kind, MDNode *) that should be added to all newly | |||
| 97 | /// created instructions, like !dbg metadata. | |||
| 98 | SmallVector<std::pair<unsigned, MDNode *>, 2> MetadataToCopy; | |||
| 99 | ||||
| 100 | /// Add or update the an entry (Kind, MD) to MetadataToCopy, if \p MD is not | |||
| 101 | /// null. If \p MD is null, remove the entry with \p Kind. | |||
| 102 | void AddOrRemoveMetadataToCopy(unsigned Kind, MDNode *MD) { | |||
| 103 | if (!MD) { | |||
| 104 | erase_if(MetadataToCopy, [Kind](const std::pair<unsigned, MDNode *> &KV) { | |||
| 105 | return KV.first == Kind; | |||
| 106 | }); | |||
| 107 | return; | |||
| 108 | } | |||
| 109 | ||||
| 110 | for (auto &KV : MetadataToCopy) | |||
| 111 | if (KV.first == Kind) { | |||
| 112 | KV.second = MD; | |||
| 113 | return; | |||
| 114 | } | |||
| 115 | ||||
| 116 | MetadataToCopy.emplace_back(Kind, MD); | |||
| 117 | } | |||
| 118 | ||||
| 119 | protected: | |||
| 120 | BasicBlock *BB; | |||
| 121 | BasicBlock::iterator InsertPt; | |||
| 122 | LLVMContext &Context; | |||
| 123 | const IRBuilderFolder &Folder; | |||
| 124 | const IRBuilderDefaultInserter &Inserter; | |||
| 125 | ||||
| 126 | MDNode *DefaultFPMathTag; | |||
| 127 | FastMathFlags FMF; | |||
| 128 | ||||
| 129 | bool IsFPConstrained; | |||
| 130 | fp::ExceptionBehavior DefaultConstrainedExcept; | |||
| 131 | RoundingMode DefaultConstrainedRounding; | |||
| 132 | ||||
| 133 | ArrayRef<OperandBundleDef> DefaultOperandBundles; | |||
| 134 | ||||
| 135 | public: | |||
| 136 | IRBuilderBase(LLVMContext &context, const IRBuilderFolder &Folder, | |||
| 137 | const IRBuilderDefaultInserter &Inserter, | |||
| 138 | MDNode *FPMathTag, ArrayRef<OperandBundleDef> OpBundles) | |||
| 139 | : Context(context), Folder(Folder), Inserter(Inserter), | |||
| 140 | DefaultFPMathTag(FPMathTag), IsFPConstrained(false), | |||
| 141 | DefaultConstrainedExcept(fp::ebStrict), | |||
| 142 | DefaultConstrainedRounding(RoundingMode::Dynamic), | |||
| 143 | DefaultOperandBundles(OpBundles) { | |||
| 144 | ClearInsertionPoint(); | |||
| 145 | } | |||
| 146 | ||||
| 147 | /// Insert and return the specified instruction. | |||
| 148 | template<typename InstTy> | |||
| 149 | InstTy *Insert(InstTy *I, const Twine &Name = "") const { | |||
| 150 | Inserter.InsertHelper(I, Name, BB, InsertPt); | |||
| 151 | AddMetadataToInst(I); | |||
| 152 | return I; | |||
| 153 | } | |||
| 154 | ||||
| 155 | /// No-op overload to handle constants. | |||
| 156 | Constant *Insert(Constant *C, const Twine& = "") const { | |||
| 157 | return C; | |||
| 158 | } | |||
| 159 | ||||
| 160 | Value *Insert(Value *V, const Twine &Name = "") const { | |||
| 161 | if (Instruction *I = dyn_cast<Instruction>(V)) | |||
| 162 | return Insert(I, Name); | |||
| 163 | assert(isa<Constant>(V))((void)0); | |||
| 164 | return V; | |||
| 165 | } | |||
| 166 | ||||
| 167 | //===--------------------------------------------------------------------===// | |||
| 168 | // Builder configuration methods | |||
| 169 | //===--------------------------------------------------------------------===// | |||
| 170 | ||||
| 171 | /// Clear the insertion point: created instructions will not be | |||
| 172 | /// inserted into a block. | |||
| 173 | void ClearInsertionPoint() { | |||
| 174 | BB = nullptr; | |||
| 175 | InsertPt = BasicBlock::iterator(); | |||
| 176 | } | |||
| 177 | ||||
| 178 | BasicBlock *GetInsertBlock() const { return BB; } | |||
| 179 | BasicBlock::iterator GetInsertPoint() const { return InsertPt; } | |||
| 180 | LLVMContext &getContext() const { return Context; } | |||
| 181 | ||||
| 182 | /// This specifies that created instructions should be appended to the | |||
| 183 | /// end of the specified block. | |||
| 184 | void SetInsertPoint(BasicBlock *TheBB) { | |||
| 185 | BB = TheBB; | |||
| 186 | InsertPt = BB->end(); | |||
| 187 | } | |||
| 188 | ||||
| 189 | /// This specifies that created instructions should be inserted before | |||
| 190 | /// the specified instruction. | |||
| 191 | void SetInsertPoint(Instruction *I) { | |||
| 192 | BB = I->getParent(); | |||
| ||||
| 193 | InsertPt = I->getIterator(); | |||
| 194 | assert(InsertPt != BB->end() && "Can't read debug loc from end()")((void)0); | |||
| 195 | SetCurrentDebugLocation(I->getDebugLoc()); | |||
| 196 | } | |||
| 197 | ||||
| 198 | /// This specifies that created instructions should be inserted at the | |||
| 199 | /// specified point. | |||
| 200 | void SetInsertPoint(BasicBlock *TheBB, BasicBlock::iterator IP) { | |||
| 201 | BB = TheBB; | |||
| 202 | InsertPt = IP; | |||
| 203 | if (IP != TheBB->end()) | |||
| 204 | SetCurrentDebugLocation(IP->getDebugLoc()); | |||
| 205 | } | |||
| 206 | ||||
| 207 | /// Set location information used by debugging information. | |||
| 208 | void SetCurrentDebugLocation(DebugLoc L) { | |||
| 209 | AddOrRemoveMetadataToCopy(LLVMContext::MD_dbg, L.getAsMDNode()); | |||
| 210 | } | |||
| 211 | ||||
| 212 | /// Collect metadata with IDs \p MetadataKinds from \p Src which should be | |||
| 213 | /// added to all created instructions. Entries present in MedataDataToCopy but | |||
| 214 | /// not on \p Src will be dropped from MetadataToCopy. | |||
| 215 | void CollectMetadataToCopy(Instruction *Src, | |||
| 216 | ArrayRef<unsigned> MetadataKinds) { | |||
| 217 | for (unsigned K : MetadataKinds) | |||
| 218 | AddOrRemoveMetadataToCopy(K, Src->getMetadata(K)); | |||
| 219 | } | |||
| 220 | ||||
| 221 | /// Get location information used by debugging information. | |||
| 222 | DebugLoc getCurrentDebugLocation() const { | |||
| 223 | for (auto &KV : MetadataToCopy) | |||
| 224 | if (KV.first == LLVMContext::MD_dbg) | |||
| 225 | return {cast<DILocation>(KV.second)}; | |||
| 226 | ||||
| 227 | return {}; | |||
| 228 | } | |||
| 229 | ||||
| 230 | /// If this builder has a current debug location, set it on the | |||
| 231 | /// specified instruction. | |||
| 232 | void SetInstDebugLocation(Instruction *I) const { | |||
| 233 | for (const auto &KV : MetadataToCopy) | |||
| 234 | if (KV.first == LLVMContext::MD_dbg) { | |||
| 235 | I->setDebugLoc(DebugLoc(KV.second)); | |||
| 236 | return; | |||
| 237 | } | |||
| 238 | } | |||
| 239 | ||||
| 240 | /// Add all entries in MetadataToCopy to \p I. | |||
| 241 | void AddMetadataToInst(Instruction *I) const { | |||
| 242 | for (auto &KV : MetadataToCopy) | |||
| 243 | I->setMetadata(KV.first, KV.second); | |||
| 244 | } | |||
| 245 | ||||
| 246 | /// Get the return type of the current function that we're emitting | |||
| 247 | /// into. | |||
| 248 | Type *getCurrentFunctionReturnType() const; | |||
| 249 | ||||
| 250 | /// InsertPoint - A saved insertion point. | |||
| 251 | class InsertPoint { | |||
| 252 | BasicBlock *Block = nullptr; | |||
| 253 | BasicBlock::iterator Point; | |||
| 254 | ||||
| 255 | public: | |||
| 256 | /// Creates a new insertion point which doesn't point to anything. | |||
| 257 | InsertPoint() = default; | |||
| 258 | ||||
| 259 | /// Creates a new insertion point at the given location. | |||
| 260 | InsertPoint(BasicBlock *InsertBlock, BasicBlock::iterator InsertPoint) | |||
| 261 | : Block(InsertBlock), Point(InsertPoint) {} | |||
| 262 | ||||
| 263 | /// Returns true if this insert point is set. | |||
| 264 | bool isSet() const { return (Block != nullptr); } | |||
| 265 | ||||
| 266 | BasicBlock *getBlock() const { return Block; } | |||
| 267 | BasicBlock::iterator getPoint() const { return Point; } | |||
| 268 | }; | |||
| 269 | ||||
| 270 | /// Returns the current insert point. | |||
| 271 | InsertPoint saveIP() const { | |||
| 272 | return InsertPoint(GetInsertBlock(), GetInsertPoint()); | |||
| 273 | } | |||
| 274 | ||||
| 275 | /// Returns the current insert point, clearing it in the process. | |||
| 276 | InsertPoint saveAndClearIP() { | |||
| 277 | InsertPoint IP(GetInsertBlock(), GetInsertPoint()); | |||
| 278 | ClearInsertionPoint(); | |||
| 279 | return IP; | |||
| 280 | } | |||
| 281 | ||||
| 282 | /// Sets the current insert point to a previously-saved location. | |||
| 283 | void restoreIP(InsertPoint IP) { | |||
| 284 | if (IP.isSet()) | |||
| 285 | SetInsertPoint(IP.getBlock(), IP.getPoint()); | |||
| 286 | else | |||
| 287 | ClearInsertionPoint(); | |||
| 288 | } | |||
| 289 | ||||
| 290 | /// Get the floating point math metadata being used. | |||
| 291 | MDNode *getDefaultFPMathTag() const { return DefaultFPMathTag; } | |||
| 292 | ||||
| 293 | /// Get the flags to be applied to created floating point ops | |||
| 294 | FastMathFlags getFastMathFlags() const { return FMF; } | |||
| 295 | ||||
| 296 | FastMathFlags &getFastMathFlags() { return FMF; } | |||
| 297 | ||||
| 298 | /// Clear the fast-math flags. | |||
| 299 | void clearFastMathFlags() { FMF.clear(); } | |||
| 300 | ||||
| 301 | /// Set the floating point math metadata to be used. | |||
| 302 | void setDefaultFPMathTag(MDNode *FPMathTag) { DefaultFPMathTag = FPMathTag; } | |||
| 303 | ||||
| 304 | /// Set the fast-math flags to be used with generated fp-math operators | |||
| 305 | void setFastMathFlags(FastMathFlags NewFMF) { FMF = NewFMF; } | |||
| 306 | ||||
| 307 | /// Enable/Disable use of constrained floating point math. When | |||
| 308 | /// enabled the CreateF<op>() calls instead create constrained | |||
| 309 | /// floating point intrinsic calls. Fast math flags are unaffected | |||
| 310 | /// by this setting. | |||
| 311 | void setIsFPConstrained(bool IsCon) { IsFPConstrained = IsCon; } | |||
| 312 | ||||
| 313 | /// Query for the use of constrained floating point math | |||
| 314 | bool getIsFPConstrained() { return IsFPConstrained; } | |||
| 315 | ||||
| 316 | /// Set the exception handling to be used with constrained floating point | |||
| 317 | void setDefaultConstrainedExcept(fp::ExceptionBehavior NewExcept) { | |||
| 318 | #ifndef NDEBUG1 | |||
| 319 | Optional<StringRef> ExceptStr = ExceptionBehaviorToStr(NewExcept); | |||
| 320 | assert(ExceptStr.hasValue() && "Garbage strict exception behavior!")((void)0); | |||
| 321 | #endif | |||
| 322 | DefaultConstrainedExcept = NewExcept; | |||
| 323 | } | |||
| 324 | ||||
| 325 | /// Set the rounding mode handling to be used with constrained floating point | |||
| 326 | void setDefaultConstrainedRounding(RoundingMode NewRounding) { | |||
| 327 | #ifndef NDEBUG1 | |||
| 328 | Optional<StringRef> RoundingStr = RoundingModeToStr(NewRounding); | |||
| 329 | assert(RoundingStr.hasValue() && "Garbage strict rounding mode!")((void)0); | |||
| 330 | #endif | |||
| 331 | DefaultConstrainedRounding = NewRounding; | |||
| 332 | } | |||
| 333 | ||||
| 334 | /// Get the exception handling used with constrained floating point | |||
| 335 | fp::ExceptionBehavior getDefaultConstrainedExcept() { | |||
| 336 | return DefaultConstrainedExcept; | |||
| 337 | } | |||
| 338 | ||||
| 339 | /// Get the rounding mode handling used with constrained floating point | |||
| 340 | RoundingMode getDefaultConstrainedRounding() { | |||
| 341 | return DefaultConstrainedRounding; | |||
| 342 | } | |||
| 343 | ||||
| 344 | void setConstrainedFPFunctionAttr() { | |||
| 345 | assert(BB && "Must have a basic block to set any function attributes!")((void)0); | |||
| 346 | ||||
| 347 | Function *F = BB->getParent(); | |||
| 348 | if (!F->hasFnAttribute(Attribute::StrictFP)) { | |||
| 349 | F->addFnAttr(Attribute::StrictFP); | |||
| 350 | } | |||
| 351 | } | |||
| 352 | ||||
| 353 | void setConstrainedFPCallAttr(CallBase *I) { | |||
| 354 | I->addAttribute(AttributeList::FunctionIndex, Attribute::StrictFP); | |||
| 355 | } | |||
| 356 | ||||
| 357 | void setDefaultOperandBundles(ArrayRef<OperandBundleDef> OpBundles) { | |||
| 358 | DefaultOperandBundles = OpBundles; | |||
| 359 | } | |||
| 360 | ||||
| 361 | //===--------------------------------------------------------------------===// | |||
| 362 | // RAII helpers. | |||
| 363 | //===--------------------------------------------------------------------===// | |||
| 364 | ||||
| 365 | // RAII object that stores the current insertion point and restores it | |||
| 366 | // when the object is destroyed. This includes the debug location. | |||
| 367 | class InsertPointGuard { | |||
| 368 | IRBuilderBase &Builder; | |||
| 369 | AssertingVH<BasicBlock> Block; | |||
| 370 | BasicBlock::iterator Point; | |||
| 371 | DebugLoc DbgLoc; | |||
| 372 | ||||
| 373 | public: | |||
| 374 | InsertPointGuard(IRBuilderBase &B) | |||
| 375 | : Builder(B), Block(B.GetInsertBlock()), Point(B.GetInsertPoint()), | |||
| 376 | DbgLoc(B.getCurrentDebugLocation()) {} | |||
| 377 | ||||
| 378 | InsertPointGuard(const InsertPointGuard &) = delete; | |||
| 379 | InsertPointGuard &operator=(const InsertPointGuard &) = delete; | |||
| 380 | ||||
| 381 | ~InsertPointGuard() { | |||
| 382 | Builder.restoreIP(InsertPoint(Block, Point)); | |||
| 383 | Builder.SetCurrentDebugLocation(DbgLoc); | |||
| 384 | } | |||
| 385 | }; | |||
| 386 | ||||
| 387 | // RAII object that stores the current fast math settings and restores | |||
| 388 | // them when the object is destroyed. | |||
| 389 | class FastMathFlagGuard { | |||
| 390 | IRBuilderBase &Builder; | |||
| 391 | FastMathFlags FMF; | |||
| 392 | MDNode *FPMathTag; | |||
| 393 | bool IsFPConstrained; | |||
| 394 | fp::ExceptionBehavior DefaultConstrainedExcept; | |||
| 395 | RoundingMode DefaultConstrainedRounding; | |||
| 396 | ||||
| 397 | public: | |||
| 398 | FastMathFlagGuard(IRBuilderBase &B) | |||
| 399 | : Builder(B), FMF(B.FMF), FPMathTag(B.DefaultFPMathTag), | |||
| 400 | IsFPConstrained(B.IsFPConstrained), | |||
| 401 | DefaultConstrainedExcept(B.DefaultConstrainedExcept), | |||
| 402 | DefaultConstrainedRounding(B.DefaultConstrainedRounding) {} | |||
| 403 | ||||
| 404 | FastMathFlagGuard(const FastMathFlagGuard &) = delete; | |||
| 405 | FastMathFlagGuard &operator=(const FastMathFlagGuard &) = delete; | |||
| 406 | ||||
| 407 | ~FastMathFlagGuard() { | |||
| 408 | Builder.FMF = FMF; | |||
| 409 | Builder.DefaultFPMathTag = FPMathTag; | |||
| 410 | Builder.IsFPConstrained = IsFPConstrained; | |||
| 411 | Builder.DefaultConstrainedExcept = DefaultConstrainedExcept; | |||
| 412 | Builder.DefaultConstrainedRounding = DefaultConstrainedRounding; | |||
| 413 | } | |||
| 414 | }; | |||
| 415 | ||||
| 416 | // RAII object that stores the current default operand bundles and restores | |||
| 417 | // them when the object is destroyed. | |||
| 418 | class OperandBundlesGuard { | |||
| 419 | IRBuilderBase &Builder; | |||
| 420 | ArrayRef<OperandBundleDef> DefaultOperandBundles; | |||
| 421 | ||||
| 422 | public: | |||
| 423 | OperandBundlesGuard(IRBuilderBase &B) | |||
| 424 | : Builder(B), DefaultOperandBundles(B.DefaultOperandBundles) {} | |||
| 425 | ||||
| 426 | OperandBundlesGuard(const OperandBundlesGuard &) = delete; | |||
| 427 | OperandBundlesGuard &operator=(const OperandBundlesGuard &) = delete; | |||
| 428 | ||||
| 429 | ~OperandBundlesGuard() { | |||
| 430 | Builder.DefaultOperandBundles = DefaultOperandBundles; | |||
| 431 | } | |||
| 432 | }; | |||
| 433 | ||||
| 434 | ||||
| 435 | //===--------------------------------------------------------------------===// | |||
| 436 | // Miscellaneous creation methods. | |||
| 437 | //===--------------------------------------------------------------------===// | |||
| 438 | ||||
| 439 | /// Make a new global variable with initializer type i8* | |||
| 440 | /// | |||
| 441 | /// Make a new global variable with an initializer that has array of i8 type | |||
| 442 | /// filled in with the null terminated string value specified. The new global | |||
| 443 | /// variable will be marked mergable with any others of the same contents. If | |||
| 444 | /// Name is specified, it is the name of the global variable created. | |||
| 445 | /// | |||
| 446 | /// If no module is given via \p M, it is take from the insertion point basic | |||
| 447 | /// block. | |||
| 448 | GlobalVariable *CreateGlobalString(StringRef Str, const Twine &Name = "", | |||
| 449 | unsigned AddressSpace = 0, | |||
| 450 | Module *M = nullptr); | |||
| 451 | ||||
| 452 | /// Get a constant value representing either true or false. | |||
| 453 | ConstantInt *getInt1(bool V) { | |||
| 454 | return ConstantInt::get(getInt1Ty(), V); | |||
| 455 | } | |||
| 456 | ||||
| 457 | /// Get the constant value for i1 true. | |||
| 458 | ConstantInt *getTrue() { | |||
| 459 | return ConstantInt::getTrue(Context); | |||
| 460 | } | |||
| 461 | ||||
| 462 | /// Get the constant value for i1 false. | |||
| 463 | ConstantInt *getFalse() { | |||
| 464 | return ConstantInt::getFalse(Context); | |||
| 465 | } | |||
| 466 | ||||
| 467 | /// Get a constant 8-bit value. | |||
| 468 | ConstantInt *getInt8(uint8_t C) { | |||
| 469 | return ConstantInt::get(getInt8Ty(), C); | |||
| 470 | } | |||
| 471 | ||||
| 472 | /// Get a constant 16-bit value. | |||
| 473 | ConstantInt *getInt16(uint16_t C) { | |||
| 474 | return ConstantInt::get(getInt16Ty(), C); | |||
| 475 | } | |||
| 476 | ||||
| 477 | /// Get a constant 32-bit value. | |||
| 478 | ConstantInt *getInt32(uint32_t C) { | |||
| 479 | return ConstantInt::get(getInt32Ty(), C); | |||
| 480 | } | |||
| 481 | ||||
| 482 | /// Get a constant 64-bit value. | |||
| 483 | ConstantInt *getInt64(uint64_t C) { | |||
| 484 | return ConstantInt::get(getInt64Ty(), C); | |||
| 485 | } | |||
| 486 | ||||
| 487 | /// Get a constant N-bit value, zero extended or truncated from | |||
| 488 | /// a 64-bit value. | |||
| 489 | ConstantInt *getIntN(unsigned N, uint64_t C) { | |||
| 490 | return ConstantInt::get(getIntNTy(N), C); | |||
| 491 | } | |||
| 492 | ||||
| 493 | /// Get a constant integer value. | |||
| 494 | ConstantInt *getInt(const APInt &AI) { | |||
| 495 | return ConstantInt::get(Context, AI); | |||
| 496 | } | |||
| 497 | ||||
| 498 | //===--------------------------------------------------------------------===// | |||
| 499 | // Type creation methods | |||
| 500 | //===--------------------------------------------------------------------===// | |||
| 501 | ||||
| 502 | /// Fetch the type representing a single bit | |||
| 503 | IntegerType *getInt1Ty() { | |||
| 504 | return Type::getInt1Ty(Context); | |||
| 505 | } | |||
| 506 | ||||
| 507 | /// Fetch the type representing an 8-bit integer. | |||
| 508 | IntegerType *getInt8Ty() { | |||
| 509 | return Type::getInt8Ty(Context); | |||
| 510 | } | |||
| 511 | ||||
| 512 | /// Fetch the type representing a 16-bit integer. | |||
| 513 | IntegerType *getInt16Ty() { | |||
| 514 | return Type::getInt16Ty(Context); | |||
| 515 | } | |||
| 516 | ||||
| 517 | /// Fetch the type representing a 32-bit integer. | |||
| 518 | IntegerType *getInt32Ty() { | |||
| 519 | return Type::getInt32Ty(Context); | |||
| 520 | } | |||
| 521 | ||||
| 522 | /// Fetch the type representing a 64-bit integer. | |||
| 523 | IntegerType *getInt64Ty() { | |||
| 524 | return Type::getInt64Ty(Context); | |||
| 525 | } | |||
| 526 | ||||
| 527 | /// Fetch the type representing a 128-bit integer. | |||
| 528 | IntegerType *getInt128Ty() { return Type::getInt128Ty(Context); } | |||
| 529 | ||||
| 530 | /// Fetch the type representing an N-bit integer. | |||
| 531 | IntegerType *getIntNTy(unsigned N) { | |||
| 532 | return Type::getIntNTy(Context, N); | |||
| 533 | } | |||
| 534 | ||||
| 535 | /// Fetch the type representing a 16-bit floating point value. | |||
| 536 | Type *getHalfTy() { | |||
| 537 | return Type::getHalfTy(Context); | |||
| 538 | } | |||
| 539 | ||||
| 540 | /// Fetch the type representing a 16-bit brain floating point value. | |||
| 541 | Type *getBFloatTy() { | |||
| 542 | return Type::getBFloatTy(Context); | |||
| 543 | } | |||
| 544 | ||||
| 545 | /// Fetch the type representing a 32-bit floating point value. | |||
| 546 | Type *getFloatTy() { | |||
| 547 | return Type::getFloatTy(Context); | |||
| 548 | } | |||
| 549 | ||||
| 550 | /// Fetch the type representing a 64-bit floating point value. | |||
| 551 | Type *getDoubleTy() { | |||
| 552 | return Type::getDoubleTy(Context); | |||
| 553 | } | |||
| 554 | ||||
| 555 | /// Fetch the type representing void. | |||
| 556 | Type *getVoidTy() { | |||
| 557 | return Type::getVoidTy(Context); | |||
| 558 | } | |||
| 559 | ||||
| 560 | /// Fetch the type representing a pointer to an 8-bit integer value. | |||
| 561 | PointerType *getInt8PtrTy(unsigned AddrSpace = 0) { | |||
| 562 | return Type::getInt8PtrTy(Context, AddrSpace); | |||
| 563 | } | |||
| 564 | ||||
| 565 | /// Fetch the type representing a pointer to an integer value. | |||
| 566 | IntegerType *getIntPtrTy(const DataLayout &DL, unsigned AddrSpace = 0) { | |||
| 567 | return DL.getIntPtrType(Context, AddrSpace); | |||
| 568 | } | |||
| 569 | ||||
| 570 | //===--------------------------------------------------------------------===// | |||
| 571 | // Intrinsic creation methods | |||
| 572 | //===--------------------------------------------------------------------===// | |||
| 573 | ||||
| 574 | /// Create and insert a memset to the specified pointer and the | |||
| 575 | /// specified value. | |||
| 576 | /// | |||
| 577 | /// If the pointer isn't an i8*, it will be converted. If a TBAA tag is | |||
| 578 | /// specified, it will be added to the instruction. Likewise with alias.scope | |||
| 579 | /// and noalias tags. | |||
| 580 | CallInst *CreateMemSet(Value *Ptr, Value *Val, uint64_t Size, | |||
| 581 | MaybeAlign Align, bool isVolatile = false, | |||
| 582 | MDNode *TBAATag = nullptr, MDNode *ScopeTag = nullptr, | |||
| 583 | MDNode *NoAliasTag = nullptr) { | |||
| 584 | return CreateMemSet(Ptr, Val, getInt64(Size), Align, isVolatile, | |||
| 585 | TBAATag, ScopeTag, NoAliasTag); | |||
| 586 | } | |||
| 587 | ||||
| 588 | CallInst *CreateMemSet(Value *Ptr, Value *Val, Value *Size, MaybeAlign Align, | |||
| 589 | bool isVolatile = false, MDNode *TBAATag = nullptr, | |||
| 590 | MDNode *ScopeTag = nullptr, | |||
| 591 | MDNode *NoAliasTag = nullptr); | |||
| 592 | ||||
| 593 | /// Create and insert an element unordered-atomic memset of the region of | |||
| 594 | /// memory starting at the given pointer to the given value. | |||
| 595 | /// | |||
| 596 | /// If the pointer isn't an i8*, it will be converted. If a TBAA tag is | |||
| 597 | /// specified, it will be added to the instruction. Likewise with alias.scope | |||
| 598 | /// and noalias tags. | |||
| 599 | CallInst *CreateElementUnorderedAtomicMemSet(Value *Ptr, Value *Val, | |||
| 600 | uint64_t Size, Align Alignment, | |||
| 601 | uint32_t ElementSize, | |||
| 602 | MDNode *TBAATag = nullptr, | |||
| 603 | MDNode *ScopeTag = nullptr, | |||
| 604 | MDNode *NoAliasTag = nullptr) { | |||
| 605 | return CreateElementUnorderedAtomicMemSet(Ptr, Val, getInt64(Size), | |||
| 606 | Align(Alignment), ElementSize, | |||
| 607 | TBAATag, ScopeTag, NoAliasTag); | |||
| 608 | } | |||
| 609 | ||||
| 610 | CallInst *CreateElementUnorderedAtomicMemSet(Value *Ptr, Value *Val, | |||
| 611 | Value *Size, Align Alignment, | |||
| 612 | uint32_t ElementSize, | |||
| 613 | MDNode *TBAATag = nullptr, | |||
| 614 | MDNode *ScopeTag = nullptr, | |||
| 615 | MDNode *NoAliasTag = nullptr); | |||
| 616 | ||||
| 617 | /// Create and insert a memcpy between the specified pointers. | |||
| 618 | /// | |||
| 619 | /// If the pointers aren't i8*, they will be converted. If a TBAA tag is | |||
| 620 | /// specified, it will be added to the instruction. Likewise with alias.scope | |||
| 621 | /// and noalias tags. | |||
| 622 | CallInst *CreateMemCpy(Value *Dst, MaybeAlign DstAlign, Value *Src, | |||
| 623 | MaybeAlign SrcAlign, uint64_t Size, | |||
| 624 | bool isVolatile = false, MDNode *TBAATag = nullptr, | |||
| 625 | MDNode *TBAAStructTag = nullptr, | |||
| 626 | MDNode *ScopeTag = nullptr, | |||
| 627 | MDNode *NoAliasTag = nullptr) { | |||
| 628 | return CreateMemCpy(Dst, DstAlign, Src, SrcAlign, getInt64(Size), | |||
| 629 | isVolatile, TBAATag, TBAAStructTag, ScopeTag, | |||
| 630 | NoAliasTag); | |||
| 631 | } | |||
| 632 | ||||
| 633 | CallInst *CreateMemTransferInst( | |||
| 634 | Intrinsic::ID IntrID, Value *Dst, MaybeAlign DstAlign, Value *Src, | |||
| 635 | MaybeAlign SrcAlign, Value *Size, bool isVolatile = false, | |||
| 636 | MDNode *TBAATag = nullptr, MDNode *TBAAStructTag = nullptr, | |||
| 637 | MDNode *ScopeTag = nullptr, MDNode *NoAliasTag = nullptr); | |||
| 638 | ||||
| 639 | CallInst *CreateMemCpy(Value *Dst, MaybeAlign DstAlign, Value *Src, | |||
| 640 | MaybeAlign SrcAlign, Value *Size, | |||
| 641 | bool isVolatile = false, MDNode *TBAATag = nullptr, | |||
| 642 | MDNode *TBAAStructTag = nullptr, | |||
| 643 | MDNode *ScopeTag = nullptr, | |||
| 644 | MDNode *NoAliasTag = nullptr) { | |||
| 645 | return CreateMemTransferInst(Intrinsic::memcpy, Dst, DstAlign, Src, | |||
| 646 | SrcAlign, Size, isVolatile, TBAATag, | |||
| 647 | TBAAStructTag, ScopeTag, NoAliasTag); | |||
| 648 | } | |||
| 649 | ||||
| 650 | CallInst * | |||
| 651 | CreateMemCpyInline(Value *Dst, MaybeAlign DstAlign, Value *Src, | |||
| 652 | MaybeAlign SrcAlign, Value *Size, bool IsVolatile = false, | |||
| 653 | MDNode *TBAATag = nullptr, MDNode *TBAAStructTag = nullptr, | |||
| 654 | MDNode *ScopeTag = nullptr, MDNode *NoAliasTag = nullptr); | |||
| 655 | ||||
| 656 | /// Create and insert an element unordered-atomic memcpy between the | |||
| 657 | /// specified pointers. | |||
| 658 | /// | |||
| 659 | /// DstAlign/SrcAlign are the alignments of the Dst/Src pointers, respectively. | |||
| 660 | /// | |||
| 661 | /// If the pointers aren't i8*, they will be converted. If a TBAA tag is | |||
| 662 | /// specified, it will be added to the instruction. Likewise with alias.scope | |||
| 663 | /// and noalias tags. | |||
| 664 | CallInst *CreateElementUnorderedAtomicMemCpy( | |||
| 665 | Value *Dst, Align DstAlign, Value *Src, Align SrcAlign, Value *Size, | |||
| 666 | uint32_t ElementSize, MDNode *TBAATag = nullptr, | |||
| 667 | MDNode *TBAAStructTag = nullptr, MDNode *ScopeTag = nullptr, | |||
| 668 | MDNode *NoAliasTag = nullptr); | |||
| 669 | ||||
| 670 | CallInst *CreateMemMove(Value *Dst, MaybeAlign DstAlign, Value *Src, | |||
| 671 | MaybeAlign SrcAlign, uint64_t Size, | |||
| 672 | bool isVolatile = false, MDNode *TBAATag = nullptr, | |||
| 673 | MDNode *ScopeTag = nullptr, | |||
| 674 | MDNode *NoAliasTag = nullptr) { | |||
| 675 | return CreateMemMove(Dst, DstAlign, Src, SrcAlign, getInt64(Size), | |||
| 676 | isVolatile, TBAATag, ScopeTag, NoAliasTag); | |||
| 677 | } | |||
| 678 | ||||
| 679 | CallInst *CreateMemMove(Value *Dst, MaybeAlign DstAlign, Value *Src, | |||
| 680 | MaybeAlign SrcAlign, Value *Size, | |||
| 681 | bool isVolatile = false, MDNode *TBAATag = nullptr, | |||
| 682 | MDNode *ScopeTag = nullptr, | |||
| 683 | MDNode *NoAliasTag = nullptr); | |||
| 684 | ||||
| 685 | /// \brief Create and insert an element unordered-atomic memmove between the | |||
| 686 | /// specified pointers. | |||
| 687 | /// | |||
| 688 | /// DstAlign/SrcAlign are the alignments of the Dst/Src pointers, | |||
| 689 | /// respectively. | |||
| 690 | /// | |||
| 691 | /// If the pointers aren't i8*, they will be converted. If a TBAA tag is | |||
| 692 | /// specified, it will be added to the instruction. Likewise with alias.scope | |||
| 693 | /// and noalias tags. | |||
| 694 | CallInst *CreateElementUnorderedAtomicMemMove( | |||
| 695 | Value *Dst, Align DstAlign, Value *Src, Align SrcAlign, Value *Size, | |||
| 696 | uint32_t ElementSize, MDNode *TBAATag = nullptr, | |||
| 697 | MDNode *TBAAStructTag = nullptr, MDNode *ScopeTag = nullptr, | |||
| 698 | MDNode *NoAliasTag = nullptr); | |||
| 699 | ||||
| 700 | /// Create a vector fadd reduction intrinsic of the source vector. | |||
| 701 | /// The first parameter is a scalar accumulator value for ordered reductions. | |||
| 702 | CallInst *CreateFAddReduce(Value *Acc, Value *Src); | |||
| 703 | ||||
| 704 | /// Create a vector fmul reduction intrinsic of the source vector. | |||
| 705 | /// The first parameter is a scalar accumulator value for ordered reductions. | |||
| 706 | CallInst *CreateFMulReduce(Value *Acc, Value *Src); | |||
| 707 | ||||
| 708 | /// Create a vector int add reduction intrinsic of the source vector. | |||
| 709 | CallInst *CreateAddReduce(Value *Src); | |||
| 710 | ||||
| 711 | /// Create a vector int mul reduction intrinsic of the source vector. | |||
| 712 | CallInst *CreateMulReduce(Value *Src); | |||
| 713 | ||||
| 714 | /// Create a vector int AND reduction intrinsic of the source vector. | |||
| 715 | CallInst *CreateAndReduce(Value *Src); | |||
| 716 | ||||
| 717 | /// Create a vector int OR reduction intrinsic of the source vector. | |||
| 718 | CallInst *CreateOrReduce(Value *Src); | |||
| 719 | ||||
| 720 | /// Create a vector int XOR reduction intrinsic of the source vector. | |||
| 721 | CallInst *CreateXorReduce(Value *Src); | |||
| 722 | ||||
| 723 | /// Create a vector integer max reduction intrinsic of the source | |||
| 724 | /// vector. | |||
| 725 | CallInst *CreateIntMaxReduce(Value *Src, bool IsSigned = false); | |||
| 726 | ||||
| 727 | /// Create a vector integer min reduction intrinsic of the source | |||
| 728 | /// vector. | |||
| 729 | CallInst *CreateIntMinReduce(Value *Src, bool IsSigned = false); | |||
| 730 | ||||
| 731 | /// Create a vector float max reduction intrinsic of the source | |||
| 732 | /// vector. | |||
| 733 | CallInst *CreateFPMaxReduce(Value *Src); | |||
| 734 | ||||
| 735 | /// Create a vector float min reduction intrinsic of the source | |||
| 736 | /// vector. | |||
| 737 | CallInst *CreateFPMinReduce(Value *Src); | |||
| 738 | ||||
| 739 | /// Create a lifetime.start intrinsic. | |||
| 740 | /// | |||
| 741 | /// If the pointer isn't i8* it will be converted. | |||
| 742 | CallInst *CreateLifetimeStart(Value *Ptr, ConstantInt *Size = nullptr); | |||
| 743 | ||||
| 744 | /// Create a lifetime.end intrinsic. | |||
| 745 | /// | |||
| 746 | /// If the pointer isn't i8* it will be converted. | |||
| 747 | CallInst *CreateLifetimeEnd(Value *Ptr, ConstantInt *Size = nullptr); | |||
| 748 | ||||
| 749 | /// Create a call to invariant.start intrinsic. | |||
| 750 | /// | |||
| 751 | /// If the pointer isn't i8* it will be converted. | |||
| 752 | CallInst *CreateInvariantStart(Value *Ptr, ConstantInt *Size = nullptr); | |||
| 753 | ||||
| 754 | /// Create a call to Masked Load intrinsic | |||
| 755 | CallInst *CreateMaskedLoad(Type *Ty, Value *Ptr, Align Alignment, Value *Mask, | |||
| 756 | Value *PassThru = nullptr, const Twine &Name = ""); | |||
| 757 | ||||
| 758 | /// Create a call to Masked Store intrinsic | |||
| 759 | CallInst *CreateMaskedStore(Value *Val, Value *Ptr, Align Alignment, | |||
| 760 | Value *Mask); | |||
| 761 | ||||
| 762 | /// Create a call to Masked Gather intrinsic | |||
| 763 | CallInst *CreateMaskedGather(Type *Ty, Value *Ptrs, Align Alignment, | |||
| 764 | Value *Mask = nullptr, Value *PassThru = nullptr, | |||
| 765 | const Twine &Name = ""); | |||
| 766 | ||||
| 767 | /// Create a call to Masked Scatter intrinsic | |||
| 768 | CallInst *CreateMaskedScatter(Value *Val, Value *Ptrs, Align Alignment, | |||
| 769 | Value *Mask = nullptr); | |||
| 770 | ||||
| 771 | /// Create an assume intrinsic call that allows the optimizer to | |||
| 772 | /// assume that the provided condition will be true. | |||
| 773 | /// | |||
| 774 | /// The optional argument \p OpBundles specifies operand bundles that are | |||
| 775 | /// added to the call instruction. | |||
| 776 | CallInst *CreateAssumption(Value *Cond, | |||
| 777 | ArrayRef<OperandBundleDef> OpBundles = llvm::None); | |||
| 778 | ||||
| 779 | /// Create a llvm.experimental.noalias.scope.decl intrinsic call. | |||
| 780 | Instruction *CreateNoAliasScopeDeclaration(Value *Scope); | |||
| 781 | Instruction *CreateNoAliasScopeDeclaration(MDNode *ScopeTag) { | |||
| 782 | return CreateNoAliasScopeDeclaration( | |||
| 783 | MetadataAsValue::get(Context, ScopeTag)); | |||
| 784 | } | |||
| 785 | ||||
| 786 | /// Create a call to the experimental.gc.statepoint intrinsic to | |||
| 787 | /// start a new statepoint sequence. | |||
| 788 | CallInst *CreateGCStatepointCall(uint64_t ID, uint32_t NumPatchBytes, | |||
| 789 | Value *ActualCallee, | |||
| 790 | ArrayRef<Value *> CallArgs, | |||
| 791 | Optional<ArrayRef<Value *>> DeoptArgs, | |||
| 792 | ArrayRef<Value *> GCArgs, | |||
| 793 | const Twine &Name = ""); | |||
| 794 | ||||
| 795 | /// Create a call to the experimental.gc.statepoint intrinsic to | |||
| 796 | /// start a new statepoint sequence. | |||
| 797 | CallInst *CreateGCStatepointCall(uint64_t ID, uint32_t NumPatchBytes, | |||
| 798 | Value *ActualCallee, uint32_t Flags, | |||
| 799 | ArrayRef<Value *> CallArgs, | |||
| 800 | Optional<ArrayRef<Use>> TransitionArgs, | |||
| 801 | Optional<ArrayRef<Use>> DeoptArgs, | |||
| 802 | ArrayRef<Value *> GCArgs, | |||
| 803 | const Twine &Name = ""); | |||
| 804 | ||||
| 805 | /// Conveninence function for the common case when CallArgs are filled | |||
| 806 | /// in using makeArrayRef(CS.arg_begin(), CS.arg_end()); Use needs to be | |||
| 807 | /// .get()'ed to get the Value pointer. | |||
| 808 | CallInst *CreateGCStatepointCall(uint64_t ID, uint32_t NumPatchBytes, | |||
| 809 | Value *ActualCallee, ArrayRef<Use> CallArgs, | |||
| 810 | Optional<ArrayRef<Value *>> DeoptArgs, | |||
| 811 | ArrayRef<Value *> GCArgs, | |||
| 812 | const Twine &Name = ""); | |||
| 813 | ||||
| 814 | /// Create an invoke to the experimental.gc.statepoint intrinsic to | |||
| 815 | /// start a new statepoint sequence. | |||
| 816 | InvokeInst * | |||
| 817 | CreateGCStatepointInvoke(uint64_t ID, uint32_t NumPatchBytes, | |||
| 818 | Value *ActualInvokee, BasicBlock *NormalDest, | |||
| 819 | BasicBlock *UnwindDest, ArrayRef<Value *> InvokeArgs, | |||
| 820 | Optional<ArrayRef<Value *>> DeoptArgs, | |||
| 821 | ArrayRef<Value *> GCArgs, const Twine &Name = ""); | |||
| 822 | ||||
| 823 | /// Create an invoke to the experimental.gc.statepoint intrinsic to | |||
| 824 | /// start a new statepoint sequence. | |||
| 825 | InvokeInst *CreateGCStatepointInvoke( | |||
| 826 | uint64_t ID, uint32_t NumPatchBytes, Value *ActualInvokee, | |||
| 827 | BasicBlock *NormalDest, BasicBlock *UnwindDest, uint32_t Flags, | |||
| 828 | ArrayRef<Value *> InvokeArgs, Optional<ArrayRef<Use>> TransitionArgs, | |||
| 829 | Optional<ArrayRef<Use>> DeoptArgs, ArrayRef<Value *> GCArgs, | |||
| 830 | const Twine &Name = ""); | |||
| 831 | ||||
| 832 | // Convenience function for the common case when CallArgs are filled in using | |||
| 833 | // makeArrayRef(CS.arg_begin(), CS.arg_end()); Use needs to be .get()'ed to | |||
| 834 | // get the Value *. | |||
| 835 | InvokeInst * | |||
| 836 | CreateGCStatepointInvoke(uint64_t ID, uint32_t NumPatchBytes, | |||
| 837 | Value *ActualInvokee, BasicBlock *NormalDest, | |||
| 838 | BasicBlock *UnwindDest, ArrayRef<Use> InvokeArgs, | |||
| 839 | Optional<ArrayRef<Value *>> DeoptArgs, | |||
| 840 | ArrayRef<Value *> GCArgs, const Twine &Name = ""); | |||
| 841 | ||||
| 842 | /// Create a call to the experimental.gc.result intrinsic to extract | |||
| 843 | /// the result from a call wrapped in a statepoint. | |||
| 844 | CallInst *CreateGCResult(Instruction *Statepoint, | |||
| 845 | Type *ResultType, | |||
| 846 | const Twine &Name = ""); | |||
| 847 | ||||
| 848 | /// Create a call to the experimental.gc.relocate intrinsics to | |||
| 849 | /// project the relocated value of one pointer from the statepoint. | |||
| 850 | CallInst *CreateGCRelocate(Instruction *Statepoint, | |||
| 851 | int BaseOffset, | |||
| 852 | int DerivedOffset, | |||
| 853 | Type *ResultType, | |||
| 854 | const Twine &Name = ""); | |||
| 855 | ||||
| 856 | /// Create a call to the experimental.gc.pointer.base intrinsic to get the | |||
| 857 | /// base pointer for the specified derived pointer. | |||
| 858 | CallInst *CreateGCGetPointerBase(Value *DerivedPtr, const Twine &Name = ""); | |||
| 859 | ||||
| 860 | /// Create a call to the experimental.gc.get.pointer.offset intrinsic to get | |||
| 861 | /// the offset of the specified derived pointer from its base. | |||
| 862 | CallInst *CreateGCGetPointerOffset(Value *DerivedPtr, const Twine &Name = ""); | |||
| 863 | ||||
| 864 | /// Create a call to llvm.vscale, multiplied by \p Scaling. The type of VScale | |||
| 865 | /// will be the same type as that of \p Scaling. | |||
| 866 | Value *CreateVScale(Constant *Scaling, const Twine &Name = ""); | |||
| 867 | ||||
| 868 | /// Creates a vector of type \p DstType with the linear sequence <0, 1, ...> | |||
| 869 | Value *CreateStepVector(Type *DstType, const Twine &Name = ""); | |||
| 870 | ||||
| 871 | /// Create a call to intrinsic \p ID with 1 operand which is mangled on its | |||
| 872 | /// type. | |||
| 873 | CallInst *CreateUnaryIntrinsic(Intrinsic::ID ID, Value *V, | |||
| 874 | Instruction *FMFSource = nullptr, | |||
| 875 | const Twine &Name = ""); | |||
| 876 | ||||
| 877 | /// Create a call to intrinsic \p ID with 2 operands which is mangled on the | |||
| 878 | /// first type. | |||
| 879 | CallInst *CreateBinaryIntrinsic(Intrinsic::ID ID, Value *LHS, Value *RHS, | |||
| 880 | Instruction *FMFSource = nullptr, | |||
| 881 | const Twine &Name = ""); | |||
| 882 | ||||
| 883 | /// Create a call to intrinsic \p ID with \p args, mangled using \p Types. If | |||
| 884 | /// \p FMFSource is provided, copy fast-math-flags from that instruction to | |||
| 885 | /// the intrinsic. | |||
| 886 | CallInst *CreateIntrinsic(Intrinsic::ID ID, ArrayRef<Type *> Types, | |||
| 887 | ArrayRef<Value *> Args, | |||
| 888 | Instruction *FMFSource = nullptr, | |||
| 889 | const Twine &Name = ""); | |||
| 890 | ||||
| 891 | /// Create call to the minnum intrinsic. | |||
| 892 | CallInst *CreateMinNum(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 893 | return CreateBinaryIntrinsic(Intrinsic::minnum, LHS, RHS, nullptr, Name); | |||
| 894 | } | |||
| 895 | ||||
| 896 | /// Create call to the maxnum intrinsic. | |||
| 897 | CallInst *CreateMaxNum(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 898 | return CreateBinaryIntrinsic(Intrinsic::maxnum, LHS, RHS, nullptr, Name); | |||
| 899 | } | |||
| 900 | ||||
| 901 | /// Create call to the minimum intrinsic. | |||
| 902 | CallInst *CreateMinimum(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 903 | return CreateBinaryIntrinsic(Intrinsic::minimum, LHS, RHS, nullptr, Name); | |||
| 904 | } | |||
| 905 | ||||
| 906 | /// Create call to the maximum intrinsic. | |||
| 907 | CallInst *CreateMaximum(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 908 | return CreateBinaryIntrinsic(Intrinsic::maximum, LHS, RHS, nullptr, Name); | |||
| 909 | } | |||
| 910 | ||||
| 911 | /// Create a call to the arithmetic_fence intrinsic. | |||
| 912 | CallInst *CreateArithmeticFence(Value *Val, Type *DstType, | |||
| 913 | const Twine &Name = "") { | |||
| 914 | return CreateIntrinsic(Intrinsic::arithmetic_fence, DstType, Val, nullptr, | |||
| 915 | Name); | |||
| 916 | } | |||
| 917 | ||||
| 918 | /// Create a call to the experimental.vector.extract intrinsic. | |||
| 919 | CallInst *CreateExtractVector(Type *DstType, Value *SrcVec, Value *Idx, | |||
| 920 | const Twine &Name = "") { | |||
| 921 | return CreateIntrinsic(Intrinsic::experimental_vector_extract, | |||
| 922 | {DstType, SrcVec->getType()}, {SrcVec, Idx}, nullptr, | |||
| 923 | Name); | |||
| 924 | } | |||
| 925 | ||||
| 926 | /// Create a call to the experimental.vector.insert intrinsic. | |||
| 927 | CallInst *CreateInsertVector(Type *DstType, Value *SrcVec, Value *SubVec, | |||
| 928 | Value *Idx, const Twine &Name = "") { | |||
| 929 | return CreateIntrinsic(Intrinsic::experimental_vector_insert, | |||
| 930 | {DstType, SubVec->getType()}, {SrcVec, SubVec, Idx}, | |||
| 931 | nullptr, Name); | |||
| 932 | } | |||
| 933 | ||||
| 934 | private: | |||
| 935 | /// Create a call to a masked intrinsic with given Id. | |||
| 936 | CallInst *CreateMaskedIntrinsic(Intrinsic::ID Id, ArrayRef<Value *> Ops, | |||
| 937 | ArrayRef<Type *> OverloadedTypes, | |||
| 938 | const Twine &Name = ""); | |||
| 939 | ||||
| 940 | Value *getCastedInt8PtrValue(Value *Ptr); | |||
| 941 | ||||
| 942 | //===--------------------------------------------------------------------===// | |||
| 943 | // Instruction creation methods: Terminators | |||
| 944 | //===--------------------------------------------------------------------===// | |||
| 945 | ||||
| 946 | private: | |||
| 947 | /// Helper to add branch weight and unpredictable metadata onto an | |||
| 948 | /// instruction. | |||
| 949 | /// \returns The annotated instruction. | |||
| 950 | template <typename InstTy> | |||
| 951 | InstTy *addBranchMetadata(InstTy *I, MDNode *Weights, MDNode *Unpredictable) { | |||
| 952 | if (Weights) | |||
| 953 | I->setMetadata(LLVMContext::MD_prof, Weights); | |||
| 954 | if (Unpredictable) | |||
| 955 | I->setMetadata(LLVMContext::MD_unpredictable, Unpredictable); | |||
| 956 | return I; | |||
| 957 | } | |||
| 958 | ||||
| 959 | public: | |||
| 960 | /// Create a 'ret void' instruction. | |||
| 961 | ReturnInst *CreateRetVoid() { | |||
| 962 | return Insert(ReturnInst::Create(Context)); | |||
| 963 | } | |||
| 964 | ||||
| 965 | /// Create a 'ret <val>' instruction. | |||
| 966 | ReturnInst *CreateRet(Value *V) { | |||
| 967 | return Insert(ReturnInst::Create(Context, V)); | |||
| 968 | } | |||
| 969 | ||||
| 970 | /// Create a sequence of N insertvalue instructions, | |||
| 971 | /// with one Value from the retVals array each, that build a aggregate | |||
| 972 | /// return value one value at a time, and a ret instruction to return | |||
| 973 | /// the resulting aggregate value. | |||
| 974 | /// | |||
| 975 | /// This is a convenience function for code that uses aggregate return values | |||
| 976 | /// as a vehicle for having multiple return values. | |||
| 977 | ReturnInst *CreateAggregateRet(Value *const *retVals, unsigned N) { | |||
| 978 | Value *V = UndefValue::get(getCurrentFunctionReturnType()); | |||
| 979 | for (unsigned i = 0; i != N; ++i) | |||
| 980 | V = CreateInsertValue(V, retVals[i], i, "mrv"); | |||
| 981 | return Insert(ReturnInst::Create(Context, V)); | |||
| 982 | } | |||
| 983 | ||||
| 984 | /// Create an unconditional 'br label X' instruction. | |||
| 985 | BranchInst *CreateBr(BasicBlock *Dest) { | |||
| 986 | return Insert(BranchInst::Create(Dest)); | |||
| 987 | } | |||
| 988 | ||||
| 989 | /// Create a conditional 'br Cond, TrueDest, FalseDest' | |||
| 990 | /// instruction. | |||
| 991 | BranchInst *CreateCondBr(Value *Cond, BasicBlock *True, BasicBlock *False, | |||
| 992 | MDNode *BranchWeights = nullptr, | |||
| 993 | MDNode *Unpredictable = nullptr) { | |||
| 994 | return Insert(addBranchMetadata(BranchInst::Create(True, False, Cond), | |||
| 995 | BranchWeights, Unpredictable)); | |||
| 996 | } | |||
| 997 | ||||
| 998 | /// Create a conditional 'br Cond, TrueDest, FalseDest' | |||
| 999 | /// instruction. Copy branch meta data if available. | |||
| 1000 | BranchInst *CreateCondBr(Value *Cond, BasicBlock *True, BasicBlock *False, | |||
| 1001 | Instruction *MDSrc) { | |||
| 1002 | BranchInst *Br = BranchInst::Create(True, False, Cond); | |||
| 1003 | if (MDSrc) { | |||
| 1004 | unsigned WL[4] = {LLVMContext::MD_prof, LLVMContext::MD_unpredictable, | |||
| 1005 | LLVMContext::MD_make_implicit, LLVMContext::MD_dbg}; | |||
| 1006 | Br->copyMetadata(*MDSrc, makeArrayRef(&WL[0], 4)); | |||
| 1007 | } | |||
| 1008 | return Insert(Br); | |||
| 1009 | } | |||
| 1010 | ||||
| 1011 | /// Create a switch instruction with the specified value, default dest, | |||
| 1012 | /// and with a hint for the number of cases that will be added (for efficient | |||
| 1013 | /// allocation). | |||
| 1014 | SwitchInst *CreateSwitch(Value *V, BasicBlock *Dest, unsigned NumCases = 10, | |||
| 1015 | MDNode *BranchWeights = nullptr, | |||
| 1016 | MDNode *Unpredictable = nullptr) { | |||
| 1017 | return Insert(addBranchMetadata(SwitchInst::Create(V, Dest, NumCases), | |||
| 1018 | BranchWeights, Unpredictable)); | |||
| 1019 | } | |||
| 1020 | ||||
| 1021 | /// Create an indirect branch instruction with the specified address | |||
| 1022 | /// operand, with an optional hint for the number of destinations that will be | |||
| 1023 | /// added (for efficient allocation). | |||
| 1024 | IndirectBrInst *CreateIndirectBr(Value *Addr, unsigned NumDests = 10) { | |||
| 1025 | return Insert(IndirectBrInst::Create(Addr, NumDests)); | |||
| 1026 | } | |||
| 1027 | ||||
| 1028 | /// Create an invoke instruction. | |||
| 1029 | InvokeInst *CreateInvoke(FunctionType *Ty, Value *Callee, | |||
| 1030 | BasicBlock *NormalDest, BasicBlock *UnwindDest, | |||
| 1031 | ArrayRef<Value *> Args, | |||
| 1032 | ArrayRef<OperandBundleDef> OpBundles, | |||
| 1033 | const Twine &Name = "") { | |||
| 1034 | InvokeInst *II = | |||
| 1035 | InvokeInst::Create(Ty, Callee, NormalDest, UnwindDest, Args, OpBundles); | |||
| 1036 | if (IsFPConstrained) | |||
| 1037 | setConstrainedFPCallAttr(II); | |||
| 1038 | return Insert(II, Name); | |||
| 1039 | } | |||
| 1040 | InvokeInst *CreateInvoke(FunctionType *Ty, Value *Callee, | |||
| 1041 | BasicBlock *NormalDest, BasicBlock *UnwindDest, | |||
| 1042 | ArrayRef<Value *> Args = None, | |||
| 1043 | const Twine &Name = "") { | |||
| 1044 | InvokeInst *II = | |||
| 1045 | InvokeInst::Create(Ty, Callee, NormalDest, UnwindDest, Args); | |||
| 1046 | if (IsFPConstrained) | |||
| 1047 | setConstrainedFPCallAttr(II); | |||
| 1048 | return Insert(II, Name); | |||
| 1049 | } | |||
| 1050 | ||||
| 1051 | InvokeInst *CreateInvoke(FunctionCallee Callee, BasicBlock *NormalDest, | |||
| 1052 | BasicBlock *UnwindDest, ArrayRef<Value *> Args, | |||
| 1053 | ArrayRef<OperandBundleDef> OpBundles, | |||
| 1054 | const Twine &Name = "") { | |||
| 1055 | return CreateInvoke(Callee.getFunctionType(), Callee.getCallee(), | |||
| 1056 | NormalDest, UnwindDest, Args, OpBundles, Name); | |||
| 1057 | } | |||
| 1058 | ||||
| 1059 | InvokeInst *CreateInvoke(FunctionCallee Callee, BasicBlock *NormalDest, | |||
| 1060 | BasicBlock *UnwindDest, | |||
| 1061 | ArrayRef<Value *> Args = None, | |||
| 1062 | const Twine &Name = "") { | |||
| 1063 | return CreateInvoke(Callee.getFunctionType(), Callee.getCallee(), | |||
| 1064 | NormalDest, UnwindDest, Args, Name); | |||
| 1065 | } | |||
| 1066 | ||||
| 1067 | /// \brief Create a callbr instruction. | |||
| 1068 | CallBrInst *CreateCallBr(FunctionType *Ty, Value *Callee, | |||
| 1069 | BasicBlock *DefaultDest, | |||
| 1070 | ArrayRef<BasicBlock *> IndirectDests, | |||
| 1071 | ArrayRef<Value *> Args = None, | |||
| 1072 | const Twine &Name = "") { | |||
| 1073 | return Insert(CallBrInst::Create(Ty, Callee, DefaultDest, IndirectDests, | |||
| 1074 | Args), Name); | |||
| 1075 | } | |||
| 1076 | CallBrInst *CreateCallBr(FunctionType *Ty, Value *Callee, | |||
| 1077 | BasicBlock *DefaultDest, | |||
| 1078 | ArrayRef<BasicBlock *> IndirectDests, | |||
| 1079 | ArrayRef<Value *> Args, | |||
| 1080 | ArrayRef<OperandBundleDef> OpBundles, | |||
| 1081 | const Twine &Name = "") { | |||
| 1082 | return Insert( | |||
| 1083 | CallBrInst::Create(Ty, Callee, DefaultDest, IndirectDests, Args, | |||
| 1084 | OpBundles), Name); | |||
| 1085 | } | |||
| 1086 | ||||
| 1087 | CallBrInst *CreateCallBr(FunctionCallee Callee, BasicBlock *DefaultDest, | |||
| 1088 | ArrayRef<BasicBlock *> IndirectDests, | |||
| 1089 | ArrayRef<Value *> Args = None, | |||
| 1090 | const Twine &Name = "") { | |||
| 1091 | return CreateCallBr(Callee.getFunctionType(), Callee.getCallee(), | |||
| 1092 | DefaultDest, IndirectDests, Args, Name); | |||
| 1093 | } | |||
| 1094 | CallBrInst *CreateCallBr(FunctionCallee Callee, BasicBlock *DefaultDest, | |||
| 1095 | ArrayRef<BasicBlock *> IndirectDests, | |||
| 1096 | ArrayRef<Value *> Args, | |||
| 1097 | ArrayRef<OperandBundleDef> OpBundles, | |||
| 1098 | const Twine &Name = "") { | |||
| 1099 | return CreateCallBr(Callee.getFunctionType(), Callee.getCallee(), | |||
| 1100 | DefaultDest, IndirectDests, Args, Name); | |||
| 1101 | } | |||
| 1102 | ||||
| 1103 | ResumeInst *CreateResume(Value *Exn) { | |||
| 1104 | return Insert(ResumeInst::Create(Exn)); | |||
| 1105 | } | |||
| 1106 | ||||
| 1107 | CleanupReturnInst *CreateCleanupRet(CleanupPadInst *CleanupPad, | |||
| 1108 | BasicBlock *UnwindBB = nullptr) { | |||
| 1109 | return Insert(CleanupReturnInst::Create(CleanupPad, UnwindBB)); | |||
| 1110 | } | |||
| 1111 | ||||
| 1112 | CatchSwitchInst *CreateCatchSwitch(Value *ParentPad, BasicBlock *UnwindBB, | |||
| 1113 | unsigned NumHandlers, | |||
| 1114 | const Twine &Name = "") { | |||
| 1115 | return Insert(CatchSwitchInst::Create(ParentPad, UnwindBB, NumHandlers), | |||
| 1116 | Name); | |||
| 1117 | } | |||
| 1118 | ||||
| 1119 | CatchPadInst *CreateCatchPad(Value *ParentPad, ArrayRef<Value *> Args, | |||
| 1120 | const Twine &Name = "") { | |||
| 1121 | return Insert(CatchPadInst::Create(ParentPad, Args), Name); | |||
| 1122 | } | |||
| 1123 | ||||
| 1124 | CleanupPadInst *CreateCleanupPad(Value *ParentPad, | |||
| 1125 | ArrayRef<Value *> Args = None, | |||
| 1126 | const Twine &Name = "") { | |||
| 1127 | return Insert(CleanupPadInst::Create(ParentPad, Args), Name); | |||
| 1128 | } | |||
| 1129 | ||||
| 1130 | CatchReturnInst *CreateCatchRet(CatchPadInst *CatchPad, BasicBlock *BB) { | |||
| 1131 | return Insert(CatchReturnInst::Create(CatchPad, BB)); | |||
| 1132 | } | |||
| 1133 | ||||
| 1134 | UnreachableInst *CreateUnreachable() { | |||
| 1135 | return Insert(new UnreachableInst(Context)); | |||
| 1136 | } | |||
| 1137 | ||||
| 1138 | //===--------------------------------------------------------------------===// | |||
| 1139 | // Instruction creation methods: Binary Operators | |||
| 1140 | //===--------------------------------------------------------------------===// | |||
| 1141 | private: | |||
| 1142 | BinaryOperator *CreateInsertNUWNSWBinOp(BinaryOperator::BinaryOps Opc, | |||
| 1143 | Value *LHS, Value *RHS, | |||
| 1144 | const Twine &Name, | |||
| 1145 | bool HasNUW, bool HasNSW) { | |||
| 1146 | BinaryOperator *BO = Insert(BinaryOperator::Create(Opc, LHS, RHS), Name); | |||
| 1147 | if (HasNUW) BO->setHasNoUnsignedWrap(); | |||
| 1148 | if (HasNSW) BO->setHasNoSignedWrap(); | |||
| 1149 | return BO; | |||
| 1150 | } | |||
| 1151 | ||||
| 1152 | Instruction *setFPAttrs(Instruction *I, MDNode *FPMD, | |||
| 1153 | FastMathFlags FMF) const { | |||
| 1154 | if (!FPMD) | |||
| 1155 | FPMD = DefaultFPMathTag; | |||
| 1156 | if (FPMD) | |||
| 1157 | I->setMetadata(LLVMContext::MD_fpmath, FPMD); | |||
| 1158 | I->setFastMathFlags(FMF); | |||
| 1159 | return I; | |||
| 1160 | } | |||
| 1161 | ||||
| 1162 | Value *foldConstant(Instruction::BinaryOps Opc, Value *L, | |||
| 1163 | Value *R, const Twine &Name) const { | |||
| 1164 | auto *LC = dyn_cast<Constant>(L); | |||
| 1165 | auto *RC = dyn_cast<Constant>(R); | |||
| 1166 | return (LC && RC) ? Insert(Folder.CreateBinOp(Opc, LC, RC), Name) : nullptr; | |||
| 1167 | } | |||
| 1168 | ||||
| 1169 | Value *getConstrainedFPRounding(Optional<RoundingMode> Rounding) { | |||
| 1170 | RoundingMode UseRounding = DefaultConstrainedRounding; | |||
| 1171 | ||||
| 1172 | if (Rounding.hasValue()) | |||
| 1173 | UseRounding = Rounding.getValue(); | |||
| 1174 | ||||
| 1175 | Optional<StringRef> RoundingStr = RoundingModeToStr(UseRounding); | |||
| 1176 | assert(RoundingStr.hasValue() && "Garbage strict rounding mode!")((void)0); | |||
| 1177 | auto *RoundingMDS = MDString::get(Context, RoundingStr.getValue()); | |||
| 1178 | ||||
| 1179 | return MetadataAsValue::get(Context, RoundingMDS); | |||
| 1180 | } | |||
| 1181 | ||||
| 1182 | Value *getConstrainedFPExcept(Optional<fp::ExceptionBehavior> Except) { | |||
| 1183 | fp::ExceptionBehavior UseExcept = DefaultConstrainedExcept; | |||
| 1184 | ||||
| 1185 | if (Except.hasValue()) | |||
| 1186 | UseExcept = Except.getValue(); | |||
| 1187 | ||||
| 1188 | Optional<StringRef> ExceptStr = ExceptionBehaviorToStr(UseExcept); | |||
| 1189 | assert(ExceptStr.hasValue() && "Garbage strict exception behavior!")((void)0); | |||
| 1190 | auto *ExceptMDS = MDString::get(Context, ExceptStr.getValue()); | |||
| 1191 | ||||
| 1192 | return MetadataAsValue::get(Context, ExceptMDS); | |||
| 1193 | } | |||
| 1194 | ||||
| 1195 | Value *getConstrainedFPPredicate(CmpInst::Predicate Predicate) { | |||
| 1196 | assert(CmpInst::isFPPredicate(Predicate) &&((void)0) | |||
| 1197 | Predicate != CmpInst::FCMP_FALSE &&((void)0) | |||
| 1198 | Predicate != CmpInst::FCMP_TRUE &&((void)0) | |||
| 1199 | "Invalid constrained FP comparison predicate!")((void)0); | |||
| 1200 | ||||
| 1201 | StringRef PredicateStr = CmpInst::getPredicateName(Predicate); | |||
| 1202 | auto *PredicateMDS = MDString::get(Context, PredicateStr); | |||
| 1203 | ||||
| 1204 | return MetadataAsValue::get(Context, PredicateMDS); | |||
| 1205 | } | |||
| 1206 | ||||
| 1207 | public: | |||
| 1208 | Value *CreateAdd(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1209 | bool HasNUW = false, bool HasNSW = false) { | |||
| 1210 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1211 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1212 | return Insert(Folder.CreateAdd(LC, RC, HasNUW, HasNSW), Name); | |||
| 1213 | return CreateInsertNUWNSWBinOp(Instruction::Add, LHS, RHS, Name, | |||
| 1214 | HasNUW, HasNSW); | |||
| 1215 | } | |||
| 1216 | ||||
| 1217 | Value *CreateNSWAdd(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1218 | return CreateAdd(LHS, RHS, Name, false, true); | |||
| 1219 | } | |||
| 1220 | ||||
| 1221 | Value *CreateNUWAdd(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1222 | return CreateAdd(LHS, RHS, Name, true, false); | |||
| 1223 | } | |||
| 1224 | ||||
| 1225 | Value *CreateSub(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1226 | bool HasNUW = false, bool HasNSW = false) { | |||
| 1227 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1228 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1229 | return Insert(Folder.CreateSub(LC, RC, HasNUW, HasNSW), Name); | |||
| 1230 | return CreateInsertNUWNSWBinOp(Instruction::Sub, LHS, RHS, Name, | |||
| 1231 | HasNUW, HasNSW); | |||
| 1232 | } | |||
| 1233 | ||||
| 1234 | Value *CreateNSWSub(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1235 | return CreateSub(LHS, RHS, Name, false, true); | |||
| 1236 | } | |||
| 1237 | ||||
| 1238 | Value *CreateNUWSub(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1239 | return CreateSub(LHS, RHS, Name, true, false); | |||
| 1240 | } | |||
| 1241 | ||||
| 1242 | Value *CreateMul(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1243 | bool HasNUW = false, bool HasNSW = false) { | |||
| 1244 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1245 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1246 | return Insert(Folder.CreateMul(LC, RC, HasNUW, HasNSW), Name); | |||
| 1247 | return CreateInsertNUWNSWBinOp(Instruction::Mul, LHS, RHS, Name, | |||
| 1248 | HasNUW, HasNSW); | |||
| 1249 | } | |||
| 1250 | ||||
| 1251 | Value *CreateNSWMul(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1252 | return CreateMul(LHS, RHS, Name, false, true); | |||
| 1253 | } | |||
| 1254 | ||||
| 1255 | Value *CreateNUWMul(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1256 | return CreateMul(LHS, RHS, Name, true, false); | |||
| 1257 | } | |||
| 1258 | ||||
| 1259 | Value *CreateUDiv(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1260 | bool isExact = false) { | |||
| 1261 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1262 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1263 | return Insert(Folder.CreateUDiv(LC, RC, isExact), Name); | |||
| 1264 | if (!isExact) | |||
| 1265 | return Insert(BinaryOperator::CreateUDiv(LHS, RHS), Name); | |||
| 1266 | return Insert(BinaryOperator::CreateExactUDiv(LHS, RHS), Name); | |||
| 1267 | } | |||
| 1268 | ||||
| 1269 | Value *CreateExactUDiv(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1270 | return CreateUDiv(LHS, RHS, Name, true); | |||
| 1271 | } | |||
| 1272 | ||||
| 1273 | Value *CreateSDiv(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1274 | bool isExact = false) { | |||
| 1275 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1276 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1277 | return Insert(Folder.CreateSDiv(LC, RC, isExact), Name); | |||
| 1278 | if (!isExact) | |||
| 1279 | return Insert(BinaryOperator::CreateSDiv(LHS, RHS), Name); | |||
| 1280 | return Insert(BinaryOperator::CreateExactSDiv(LHS, RHS), Name); | |||
| 1281 | } | |||
| 1282 | ||||
| 1283 | Value *CreateExactSDiv(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1284 | return CreateSDiv(LHS, RHS, Name, true); | |||
| 1285 | } | |||
| 1286 | ||||
| 1287 | Value *CreateURem(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1288 | if (Value *V = foldConstant(Instruction::URem, LHS, RHS, Name)) return V; | |||
| 1289 | return Insert(BinaryOperator::CreateURem(LHS, RHS), Name); | |||
| 1290 | } | |||
| 1291 | ||||
| 1292 | Value *CreateSRem(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1293 | if (Value *V = foldConstant(Instruction::SRem, LHS, RHS, Name)) return V; | |||
| 1294 | return Insert(BinaryOperator::CreateSRem(LHS, RHS), Name); | |||
| 1295 | } | |||
| 1296 | ||||
| 1297 | Value *CreateShl(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1298 | bool HasNUW = false, bool HasNSW = false) { | |||
| 1299 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1300 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1301 | return Insert(Folder.CreateShl(LC, RC, HasNUW, HasNSW), Name); | |||
| 1302 | return CreateInsertNUWNSWBinOp(Instruction::Shl, LHS, RHS, Name, | |||
| 1303 | HasNUW, HasNSW); | |||
| 1304 | } | |||
| 1305 | ||||
| 1306 | Value *CreateShl(Value *LHS, const APInt &RHS, const Twine &Name = "", | |||
| 1307 | bool HasNUW = false, bool HasNSW = false) { | |||
| 1308 | return CreateShl(LHS, ConstantInt::get(LHS->getType(), RHS), Name, | |||
| 1309 | HasNUW, HasNSW); | |||
| 1310 | } | |||
| 1311 | ||||
| 1312 | Value *CreateShl(Value *LHS, uint64_t RHS, const Twine &Name = "", | |||
| 1313 | bool HasNUW = false, bool HasNSW = false) { | |||
| 1314 | return CreateShl(LHS, ConstantInt::get(LHS->getType(), RHS), Name, | |||
| 1315 | HasNUW, HasNSW); | |||
| 1316 | } | |||
| 1317 | ||||
| 1318 | Value *CreateLShr(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1319 | bool isExact = false) { | |||
| 1320 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1321 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1322 | return Insert(Folder.CreateLShr(LC, RC, isExact), Name); | |||
| 1323 | if (!isExact) | |||
| 1324 | return Insert(BinaryOperator::CreateLShr(LHS, RHS), Name); | |||
| 1325 | return Insert(BinaryOperator::CreateExactLShr(LHS, RHS), Name); | |||
| 1326 | } | |||
| 1327 | ||||
| 1328 | Value *CreateLShr(Value *LHS, const APInt &RHS, const Twine &Name = "", | |||
| 1329 | bool isExact = false) { | |||
| 1330 | return CreateLShr(LHS, ConstantInt::get(LHS->getType(), RHS), Name,isExact); | |||
| 1331 | } | |||
| 1332 | ||||
| 1333 | Value *CreateLShr(Value *LHS, uint64_t RHS, const Twine &Name = "", | |||
| 1334 | bool isExact = false) { | |||
| 1335 | return CreateLShr(LHS, ConstantInt::get(LHS->getType(), RHS), Name,isExact); | |||
| 1336 | } | |||
| 1337 | ||||
| 1338 | Value *CreateAShr(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1339 | bool isExact = false) { | |||
| 1340 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1341 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 1342 | return Insert(Folder.CreateAShr(LC, RC, isExact), Name); | |||
| 1343 | if (!isExact) | |||
| 1344 | return Insert(BinaryOperator::CreateAShr(LHS, RHS), Name); | |||
| 1345 | return Insert(BinaryOperator::CreateExactAShr(LHS, RHS), Name); | |||
| 1346 | } | |||
| 1347 | ||||
| 1348 | Value *CreateAShr(Value *LHS, const APInt &RHS, const Twine &Name = "", | |||
| 1349 | bool isExact = false) { | |||
| 1350 | return CreateAShr(LHS, ConstantInt::get(LHS->getType(), RHS), Name,isExact); | |||
| 1351 | } | |||
| 1352 | ||||
| 1353 | Value *CreateAShr(Value *LHS, uint64_t RHS, const Twine &Name = "", | |||
| 1354 | bool isExact = false) { | |||
| 1355 | return CreateAShr(LHS, ConstantInt::get(LHS->getType(), RHS), Name,isExact); | |||
| 1356 | } | |||
| 1357 | ||||
| 1358 | Value *CreateAnd(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1359 | if (auto *RC = dyn_cast<Constant>(RHS)) { | |||
| 1360 | if (isa<ConstantInt>(RC) && cast<ConstantInt>(RC)->isMinusOne()) | |||
| 1361 | return LHS; // LHS & -1 -> LHS | |||
| 1362 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1363 | return Insert(Folder.CreateAnd(LC, RC), Name); | |||
| 1364 | } | |||
| 1365 | return Insert(BinaryOperator::CreateAnd(LHS, RHS), Name); | |||
| 1366 | } | |||
| 1367 | ||||
| 1368 | Value *CreateAnd(Value *LHS, const APInt &RHS, const Twine &Name = "") { | |||
| 1369 | return CreateAnd(LHS, ConstantInt::get(LHS->getType(), RHS), Name); | |||
| 1370 | } | |||
| 1371 | ||||
| 1372 | Value *CreateAnd(Value *LHS, uint64_t RHS, const Twine &Name = "") { | |||
| 1373 | return CreateAnd(LHS, ConstantInt::get(LHS->getType(), RHS), Name); | |||
| 1374 | } | |||
| 1375 | ||||
| 1376 | Value *CreateAnd(ArrayRef<Value*> Ops) { | |||
| 1377 | assert(!Ops.empty())((void)0); | |||
| 1378 | Value *Accum = Ops[0]; | |||
| 1379 | for (unsigned i = 1; i < Ops.size(); i++) | |||
| 1380 | Accum = CreateAnd(Accum, Ops[i]); | |||
| 1381 | return Accum; | |||
| 1382 | } | |||
| 1383 | ||||
| 1384 | Value *CreateOr(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1385 | if (auto *RC = dyn_cast<Constant>(RHS)) { | |||
| 1386 | if (RC->isNullValue()) | |||
| 1387 | return LHS; // LHS | 0 -> LHS | |||
| 1388 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 1389 | return Insert(Folder.CreateOr(LC, RC), Name); | |||
| 1390 | } | |||
| 1391 | return Insert(BinaryOperator::CreateOr(LHS, RHS), Name); | |||
| 1392 | } | |||
| 1393 | ||||
| 1394 | Value *CreateOr(Value *LHS, const APInt &RHS, const Twine &Name = "") { | |||
| 1395 | return CreateOr(LHS, ConstantInt::get(LHS->getType(), RHS), Name); | |||
| 1396 | } | |||
| 1397 | ||||
| 1398 | Value *CreateOr(Value *LHS, uint64_t RHS, const Twine &Name = "") { | |||
| 1399 | return CreateOr(LHS, ConstantInt::get(LHS->getType(), RHS), Name); | |||
| 1400 | } | |||
| 1401 | ||||
| 1402 | Value *CreateOr(ArrayRef<Value*> Ops) { | |||
| 1403 | assert(!Ops.empty())((void)0); | |||
| 1404 | Value *Accum = Ops[0]; | |||
| 1405 | for (unsigned i = 1; i < Ops.size(); i++) | |||
| 1406 | Accum = CreateOr(Accum, Ops[i]); | |||
| 1407 | return Accum; | |||
| 1408 | } | |||
| 1409 | ||||
| 1410 | Value *CreateXor(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 1411 | if (Value *V = foldConstant(Instruction::Xor, LHS, RHS, Name)) return V; | |||
| 1412 | return Insert(BinaryOperator::CreateXor(LHS, RHS), Name); | |||
| 1413 | } | |||
| 1414 | ||||
| 1415 | Value *CreateXor(Value *LHS, const APInt &RHS, const Twine &Name = "") { | |||
| 1416 | return CreateXor(LHS, ConstantInt::get(LHS->getType(), RHS), Name); | |||
| 1417 | } | |||
| 1418 | ||||
| 1419 | Value *CreateXor(Value *LHS, uint64_t RHS, const Twine &Name = "") { | |||
| 1420 | return CreateXor(LHS, ConstantInt::get(LHS->getType(), RHS), Name); | |||
| 1421 | } | |||
| 1422 | ||||
| 1423 | Value *CreateFAdd(Value *L, Value *R, const Twine &Name = "", | |||
| 1424 | MDNode *FPMD = nullptr) { | |||
| 1425 | if (IsFPConstrained) | |||
| 1426 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fadd, | |||
| 1427 | L, R, nullptr, Name, FPMD); | |||
| 1428 | ||||
| 1429 | if (Value *V = foldConstant(Instruction::FAdd, L, R, Name)) return V; | |||
| 1430 | Instruction *I = setFPAttrs(BinaryOperator::CreateFAdd(L, R), FPMD, FMF); | |||
| 1431 | return Insert(I, Name); | |||
| 1432 | } | |||
| 1433 | ||||
| 1434 | /// Copy fast-math-flags from an instruction rather than using the builder's | |||
| 1435 | /// default FMF. | |||
| 1436 | Value *CreateFAddFMF(Value *L, Value *R, Instruction *FMFSource, | |||
| 1437 | const Twine &Name = "") { | |||
| 1438 | if (IsFPConstrained) | |||
| 1439 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fadd, | |||
| 1440 | L, R, FMFSource, Name); | |||
| 1441 | ||||
| 1442 | if (Value *V = foldConstant(Instruction::FAdd, L, R, Name)) return V; | |||
| 1443 | Instruction *I = setFPAttrs(BinaryOperator::CreateFAdd(L, R), nullptr, | |||
| 1444 | FMFSource->getFastMathFlags()); | |||
| 1445 | return Insert(I, Name); | |||
| 1446 | } | |||
| 1447 | ||||
| 1448 | Value *CreateFSub(Value *L, Value *R, const Twine &Name = "", | |||
| 1449 | MDNode *FPMD = nullptr) { | |||
| 1450 | if (IsFPConstrained) | |||
| 1451 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fsub, | |||
| 1452 | L, R, nullptr, Name, FPMD); | |||
| 1453 | ||||
| 1454 | if (Value *V = foldConstant(Instruction::FSub, L, R, Name)) return V; | |||
| 1455 | Instruction *I = setFPAttrs(BinaryOperator::CreateFSub(L, R), FPMD, FMF); | |||
| 1456 | return Insert(I, Name); | |||
| 1457 | } | |||
| 1458 | ||||
| 1459 | /// Copy fast-math-flags from an instruction rather than using the builder's | |||
| 1460 | /// default FMF. | |||
| 1461 | Value *CreateFSubFMF(Value *L, Value *R, Instruction *FMFSource, | |||
| 1462 | const Twine &Name = "") { | |||
| 1463 | if (IsFPConstrained) | |||
| 1464 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fsub, | |||
| 1465 | L, R, FMFSource, Name); | |||
| 1466 | ||||
| 1467 | if (Value *V = foldConstant(Instruction::FSub, L, R, Name)) return V; | |||
| 1468 | Instruction *I = setFPAttrs(BinaryOperator::CreateFSub(L, R), nullptr, | |||
| 1469 | FMFSource->getFastMathFlags()); | |||
| 1470 | return Insert(I, Name); | |||
| 1471 | } | |||
| 1472 | ||||
| 1473 | Value *CreateFMul(Value *L, Value *R, const Twine &Name = "", | |||
| 1474 | MDNode *FPMD = nullptr) { | |||
| 1475 | if (IsFPConstrained) | |||
| 1476 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fmul, | |||
| 1477 | L, R, nullptr, Name, FPMD); | |||
| 1478 | ||||
| 1479 | if (Value *V = foldConstant(Instruction::FMul, L, R, Name)) return V; | |||
| 1480 | Instruction *I = setFPAttrs(BinaryOperator::CreateFMul(L, R), FPMD, FMF); | |||
| 1481 | return Insert(I, Name); | |||
| 1482 | } | |||
| 1483 | ||||
| 1484 | /// Copy fast-math-flags from an instruction rather than using the builder's | |||
| 1485 | /// default FMF. | |||
| 1486 | Value *CreateFMulFMF(Value *L, Value *R, Instruction *FMFSource, | |||
| 1487 | const Twine &Name = "") { | |||
| 1488 | if (IsFPConstrained) | |||
| 1489 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fmul, | |||
| 1490 | L, R, FMFSource, Name); | |||
| 1491 | ||||
| 1492 | if (Value *V = foldConstant(Instruction::FMul, L, R, Name)) return V; | |||
| 1493 | Instruction *I = setFPAttrs(BinaryOperator::CreateFMul(L, R), nullptr, | |||
| 1494 | FMFSource->getFastMathFlags()); | |||
| 1495 | return Insert(I, Name); | |||
| 1496 | } | |||
| 1497 | ||||
| 1498 | Value *CreateFDiv(Value *L, Value *R, const Twine &Name = "", | |||
| 1499 | MDNode *FPMD = nullptr) { | |||
| 1500 | if (IsFPConstrained) | |||
| 1501 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fdiv, | |||
| 1502 | L, R, nullptr, Name, FPMD); | |||
| 1503 | ||||
| 1504 | if (Value *V = foldConstant(Instruction::FDiv, L, R, Name)) return V; | |||
| 1505 | Instruction *I = setFPAttrs(BinaryOperator::CreateFDiv(L, R), FPMD, FMF); | |||
| 1506 | return Insert(I, Name); | |||
| 1507 | } | |||
| 1508 | ||||
| 1509 | /// Copy fast-math-flags from an instruction rather than using the builder's | |||
| 1510 | /// default FMF. | |||
| 1511 | Value *CreateFDivFMF(Value *L, Value *R, Instruction *FMFSource, | |||
| 1512 | const Twine &Name = "") { | |||
| 1513 | if (IsFPConstrained) | |||
| 1514 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fdiv, | |||
| 1515 | L, R, FMFSource, Name); | |||
| 1516 | ||||
| 1517 | if (Value *V = foldConstant(Instruction::FDiv, L, R, Name)) return V; | |||
| 1518 | Instruction *I = setFPAttrs(BinaryOperator::CreateFDiv(L, R), nullptr, | |||
| 1519 | FMFSource->getFastMathFlags()); | |||
| 1520 | return Insert(I, Name); | |||
| 1521 | } | |||
| 1522 | ||||
| 1523 | Value *CreateFRem(Value *L, Value *R, const Twine &Name = "", | |||
| 1524 | MDNode *FPMD = nullptr) { | |||
| 1525 | if (IsFPConstrained) | |||
| 1526 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_frem, | |||
| 1527 | L, R, nullptr, Name, FPMD); | |||
| 1528 | ||||
| 1529 | if (Value *V = foldConstant(Instruction::FRem, L, R, Name)) return V; | |||
| 1530 | Instruction *I = setFPAttrs(BinaryOperator::CreateFRem(L, R), FPMD, FMF); | |||
| 1531 | return Insert(I, Name); | |||
| 1532 | } | |||
| 1533 | ||||
| 1534 | /// Copy fast-math-flags from an instruction rather than using the builder's | |||
| 1535 | /// default FMF. | |||
| 1536 | Value *CreateFRemFMF(Value *L, Value *R, Instruction *FMFSource, | |||
| 1537 | const Twine &Name = "") { | |||
| 1538 | if (IsFPConstrained) | |||
| 1539 | return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_frem, | |||
| 1540 | L, R, FMFSource, Name); | |||
| 1541 | ||||
| 1542 | if (Value *V = foldConstant(Instruction::FRem, L, R, Name)) return V; | |||
| 1543 | Instruction *I = setFPAttrs(BinaryOperator::CreateFRem(L, R), nullptr, | |||
| 1544 | FMFSource->getFastMathFlags()); | |||
| 1545 | return Insert(I, Name); | |||
| 1546 | } | |||
| 1547 | ||||
| 1548 | Value *CreateBinOp(Instruction::BinaryOps Opc, | |||
| 1549 | Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 1550 | MDNode *FPMathTag = nullptr) { | |||
| 1551 | if (Value *V = foldConstant(Opc, LHS, RHS, Name)) return V; | |||
| 1552 | Instruction *BinOp = BinaryOperator::Create(Opc, LHS, RHS); | |||
| 1553 | if (isa<FPMathOperator>(BinOp)) | |||
| 1554 | setFPAttrs(BinOp, FPMathTag, FMF); | |||
| 1555 | return Insert(BinOp, Name); | |||
| 1556 | } | |||
| 1557 | ||||
| 1558 | Value *CreateLogicalAnd(Value *Cond1, Value *Cond2, const Twine &Name = "") { | |||
| 1559 | assert(Cond2->getType()->isIntOrIntVectorTy(1))((void)0); | |||
| 1560 | return CreateSelect(Cond1, Cond2, | |||
| 1561 | ConstantInt::getNullValue(Cond2->getType()), Name); | |||
| 1562 | } | |||
| 1563 | ||||
| 1564 | Value *CreateLogicalOr(Value *Cond1, Value *Cond2, const Twine &Name = "") { | |||
| 1565 | assert(Cond2->getType()->isIntOrIntVectorTy(1))((void)0); | |||
| 1566 | return CreateSelect(Cond1, ConstantInt::getAllOnesValue(Cond2->getType()), | |||
| 1567 | Cond2, Name); | |||
| 1568 | } | |||
| 1569 | ||||
| 1570 | CallInst *CreateConstrainedFPBinOp( | |||
| 1571 | Intrinsic::ID ID, Value *L, Value *R, Instruction *FMFSource = nullptr, | |||
| 1572 | const Twine &Name = "", MDNode *FPMathTag = nullptr, | |||
| 1573 | Optional<RoundingMode> Rounding = None, | |||
| 1574 | Optional<fp::ExceptionBehavior> Except = None); | |||
| 1575 | ||||
| 1576 | Value *CreateNeg(Value *V, const Twine &Name = "", | |||
| 1577 | bool HasNUW = false, bool HasNSW = false) { | |||
| 1578 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 1579 | return Insert(Folder.CreateNeg(VC, HasNUW, HasNSW), Name); | |||
| 1580 | BinaryOperator *BO = Insert(BinaryOperator::CreateNeg(V), Name); | |||
| 1581 | if (HasNUW) BO->setHasNoUnsignedWrap(); | |||
| 1582 | if (HasNSW) BO->setHasNoSignedWrap(); | |||
| 1583 | return BO; | |||
| 1584 | } | |||
| 1585 | ||||
| 1586 | Value *CreateNSWNeg(Value *V, const Twine &Name = "") { | |||
| 1587 | return CreateNeg(V, Name, false, true); | |||
| 1588 | } | |||
| 1589 | ||||
| 1590 | Value *CreateNUWNeg(Value *V, const Twine &Name = "") { | |||
| 1591 | return CreateNeg(V, Name, true, false); | |||
| 1592 | } | |||
| 1593 | ||||
| 1594 | Value *CreateFNeg(Value *V, const Twine &Name = "", | |||
| 1595 | MDNode *FPMathTag = nullptr) { | |||
| 1596 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 1597 | return Insert(Folder.CreateFNeg(VC), Name); | |||
| 1598 | return Insert(setFPAttrs(UnaryOperator::CreateFNeg(V), FPMathTag, FMF), | |||
| 1599 | Name); | |||
| 1600 | } | |||
| 1601 | ||||
| 1602 | /// Copy fast-math-flags from an instruction rather than using the builder's | |||
| 1603 | /// default FMF. | |||
| 1604 | Value *CreateFNegFMF(Value *V, Instruction *FMFSource, | |||
| 1605 | const Twine &Name = "") { | |||
| 1606 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 1607 | return Insert(Folder.CreateFNeg(VC), Name); | |||
| 1608 | return Insert(setFPAttrs(UnaryOperator::CreateFNeg(V), nullptr, | |||
| 1609 | FMFSource->getFastMathFlags()), | |||
| 1610 | Name); | |||
| 1611 | } | |||
| 1612 | ||||
| 1613 | Value *CreateNot(Value *V, const Twine &Name = "") { | |||
| 1614 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 1615 | return Insert(Folder.CreateNot(VC), Name); | |||
| 1616 | return Insert(BinaryOperator::CreateNot(V), Name); | |||
| 1617 | } | |||
| 1618 | ||||
| 1619 | Value *CreateUnOp(Instruction::UnaryOps Opc, | |||
| 1620 | Value *V, const Twine &Name = "", | |||
| 1621 | MDNode *FPMathTag = nullptr) { | |||
| 1622 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 1623 | return Insert(Folder.CreateUnOp(Opc, VC), Name); | |||
| 1624 | Instruction *UnOp = UnaryOperator::Create(Opc, V); | |||
| 1625 | if (isa<FPMathOperator>(UnOp)) | |||
| 1626 | setFPAttrs(UnOp, FPMathTag, FMF); | |||
| 1627 | return Insert(UnOp, Name); | |||
| 1628 | } | |||
| 1629 | ||||
| 1630 | /// Create either a UnaryOperator or BinaryOperator depending on \p Opc. | |||
| 1631 | /// Correct number of operands must be passed accordingly. | |||
| 1632 | Value *CreateNAryOp(unsigned Opc, ArrayRef<Value *> Ops, | |||
| 1633 | const Twine &Name = "", MDNode *FPMathTag = nullptr); | |||
| 1634 | ||||
| 1635 | //===--------------------------------------------------------------------===// | |||
| 1636 | // Instruction creation methods: Memory Instructions | |||
| 1637 | //===--------------------------------------------------------------------===// | |||
| 1638 | ||||
| 1639 | AllocaInst *CreateAlloca(Type *Ty, unsigned AddrSpace, | |||
| 1640 | Value *ArraySize = nullptr, const Twine &Name = "") { | |||
| 1641 | const DataLayout &DL = BB->getModule()->getDataLayout(); | |||
| 1642 | Align AllocaAlign = DL.getPrefTypeAlign(Ty); | |||
| 1643 | return Insert(new AllocaInst(Ty, AddrSpace, ArraySize, AllocaAlign), Name); | |||
| 1644 | } | |||
| 1645 | ||||
| 1646 | AllocaInst *CreateAlloca(Type *Ty, Value *ArraySize = nullptr, | |||
| 1647 | const Twine &Name = "") { | |||
| 1648 | const DataLayout &DL = BB->getModule()->getDataLayout(); | |||
| 1649 | Align AllocaAlign = DL.getPrefTypeAlign(Ty); | |||
| 1650 | unsigned AddrSpace = DL.getAllocaAddrSpace(); | |||
| 1651 | return Insert(new AllocaInst(Ty, AddrSpace, ArraySize, AllocaAlign), Name); | |||
| 1652 | } | |||
| 1653 | ||||
| 1654 | /// Provided to resolve 'CreateLoad(Ty, Ptr, "...")' correctly, instead of | |||
| 1655 | /// converting the string to 'bool' for the isVolatile parameter. | |||
| 1656 | LoadInst *CreateLoad(Type *Ty, Value *Ptr, const char *Name) { | |||
| 1657 | return CreateAlignedLoad(Ty, Ptr, MaybeAlign(), Name); | |||
| 1658 | } | |||
| 1659 | ||||
| 1660 | LoadInst *CreateLoad(Type *Ty, Value *Ptr, const Twine &Name = "") { | |||
| 1661 | return CreateAlignedLoad(Ty, Ptr, MaybeAlign(), Name); | |||
| 1662 | } | |||
| 1663 | ||||
| 1664 | LoadInst *CreateLoad(Type *Ty, Value *Ptr, bool isVolatile, | |||
| 1665 | const Twine &Name = "") { | |||
| 1666 | return CreateAlignedLoad(Ty, Ptr, MaybeAlign(), isVolatile, Name); | |||
| 1667 | } | |||
| 1668 | ||||
| 1669 | // Deprecated [opaque pointer types] | |||
| 1670 | LLVM_ATTRIBUTE_DEPRECATED(LoadInst *CreateLoad(Value *Ptr,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const char *Name) | |||
| 1671 | const char *Name),[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const char *Name) | |||
| 1672 | "Use the version that explicitly specifies the "[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const char *Name) | |||
| 1673 | "loaded type instead")[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const char *Name) { | |||
| 1674 | return CreateLoad(Ptr->getType()->getPointerElementType(), Ptr, Name); | |||
| 1675 | } | |||
| 1676 | ||||
| 1677 | // Deprecated [opaque pointer types] | |||
| 1678 | LLVM_ATTRIBUTE_DEPRECATED(LoadInst *CreateLoad(Value *Ptr,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const Twine &Name = "") | |||
| 1679 | const Twine &Name = ""),[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const Twine &Name = "") | |||
| 1680 | "Use the version that explicitly specifies the "[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const Twine &Name = "") | |||
| 1681 | "loaded type instead")[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, const Twine &Name = "") { | |||
| 1682 | return CreateLoad(Ptr->getType()->getPointerElementType(), Ptr, Name); | |||
| 1683 | } | |||
| 1684 | ||||
| 1685 | // Deprecated [opaque pointer types] | |||
| 1686 | LLVM_ATTRIBUTE_DEPRECATED(LoadInst *CreateLoad(Value *Ptr,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, bool isVolatile, const Twine &Name = "") | |||
| 1687 | bool isVolatile,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, bool isVolatile, const Twine &Name = "") | |||
| 1688 | const Twine &Name = ""),[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, bool isVolatile, const Twine &Name = "") | |||
| 1689 | "Use the version that explicitly specifies the "[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, bool isVolatile, const Twine &Name = "") | |||
| 1690 | "loaded type instead")[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateLoad(Value *Ptr, bool isVolatile, const Twine &Name = "") { | |||
| 1691 | return CreateLoad(Ptr->getType()->getPointerElementType(), Ptr, isVolatile, | |||
| 1692 | Name); | |||
| 1693 | } | |||
| 1694 | ||||
| 1695 | StoreInst *CreateStore(Value *Val, Value *Ptr, bool isVolatile = false) { | |||
| 1696 | return CreateAlignedStore(Val, Ptr, MaybeAlign(), isVolatile); | |||
| 1697 | } | |||
| 1698 | ||||
| 1699 | LoadInst *CreateAlignedLoad(Type *Ty, Value *Ptr, MaybeAlign Align, | |||
| 1700 | const char *Name) { | |||
| 1701 | return CreateAlignedLoad(Ty, Ptr, Align, /*isVolatile*/false, Name); | |||
| 1702 | } | |||
| 1703 | ||||
| 1704 | LoadInst *CreateAlignedLoad(Type *Ty, Value *Ptr, MaybeAlign Align, | |||
| 1705 | const Twine &Name = "") { | |||
| 1706 | return CreateAlignedLoad(Ty, Ptr, Align, /*isVolatile*/false, Name); | |||
| 1707 | } | |||
| 1708 | ||||
| 1709 | LoadInst *CreateAlignedLoad(Type *Ty, Value *Ptr, MaybeAlign Align, | |||
| 1710 | bool isVolatile, const Twine &Name = "") { | |||
| 1711 | if (!Align) { | |||
| 1712 | const DataLayout &DL = BB->getModule()->getDataLayout(); | |||
| 1713 | Align = DL.getABITypeAlign(Ty); | |||
| 1714 | } | |||
| 1715 | return Insert(new LoadInst(Ty, Ptr, Twine(), isVolatile, *Align), Name); | |||
| 1716 | } | |||
| 1717 | ||||
| 1718 | // Deprecated [opaque pointer types] | |||
| 1719 | LLVM_ATTRIBUTE_DEPRECATED(LoadInst *CreateAlignedLoad(Value *Ptr,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const char *Name) | |||
| 1720 | MaybeAlign Align,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const char *Name) | |||
| 1721 | const char *Name),[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const char *Name) | |||
| 1722 | "Use the version that explicitly specifies the "[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const char *Name) | |||
| 1723 | "loaded type instead")[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const char *Name) { | |||
| 1724 | return CreateAlignedLoad(Ptr->getType()->getPointerElementType(), Ptr, | |||
| 1725 | Align, Name); | |||
| 1726 | } | |||
| 1727 | // Deprecated [opaque pointer types] | |||
| 1728 | LLVM_ATTRIBUTE_DEPRECATED(LoadInst *CreateAlignedLoad(Value *Ptr,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const Twine &Name = "") | |||
| 1729 | MaybeAlign Align,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const Twine &Name = "") | |||
| 1730 | const Twine &Name = ""),[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const Twine &Name = "") | |||
| 1731 | "Use the version that explicitly specifies the "[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const Twine &Name = "") | |||
| 1732 | "loaded type instead")[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, const Twine &Name = "") { | |||
| 1733 | return CreateAlignedLoad(Ptr->getType()->getPointerElementType(), Ptr, | |||
| 1734 | Align, Name); | |||
| 1735 | } | |||
| 1736 | // Deprecated [opaque pointer types] | |||
| 1737 | LLVM_ATTRIBUTE_DEPRECATED(LoadInst *CreateAlignedLoad(Value *Ptr,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, bool isVolatile, const Twine &Name = "") | |||
| 1738 | MaybeAlign Align,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, bool isVolatile, const Twine &Name = "") | |||
| 1739 | bool isVolatile,[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, bool isVolatile, const Twine &Name = "") | |||
| 1740 | const Twine &Name = ""),[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, bool isVolatile, const Twine &Name = "") | |||
| 1741 | "Use the version that explicitly specifies the "[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, bool isVolatile, const Twine &Name = "") | |||
| 1742 | "loaded type instead")[[deprecated("Use the version that explicitly specifies the " "loaded type instead")]] LoadInst *CreateAlignedLoad(Value * Ptr, MaybeAlign Align, bool isVolatile, const Twine &Name = "") { | |||
| 1743 | return CreateAlignedLoad(Ptr->getType()->getPointerElementType(), Ptr, | |||
| 1744 | Align, isVolatile, Name); | |||
| 1745 | } | |||
| 1746 | ||||
| 1747 | StoreInst *CreateAlignedStore(Value *Val, Value *Ptr, MaybeAlign Align, | |||
| 1748 | bool isVolatile = false) { | |||
| 1749 | if (!Align) { | |||
| 1750 | const DataLayout &DL = BB->getModule()->getDataLayout(); | |||
| 1751 | Align = DL.getABITypeAlign(Val->getType()); | |||
| 1752 | } | |||
| 1753 | return Insert(new StoreInst(Val, Ptr, isVolatile, *Align)); | |||
| 1754 | } | |||
| 1755 | FenceInst *CreateFence(AtomicOrdering Ordering, | |||
| 1756 | SyncScope::ID SSID = SyncScope::System, | |||
| 1757 | const Twine &Name = "") { | |||
| 1758 | return Insert(new FenceInst(Context, Ordering, SSID), Name); | |||
| 1759 | } | |||
| 1760 | ||||
| 1761 | AtomicCmpXchgInst * | |||
| 1762 | CreateAtomicCmpXchg(Value *Ptr, Value *Cmp, Value *New, MaybeAlign Align, | |||
| 1763 | AtomicOrdering SuccessOrdering, | |||
| 1764 | AtomicOrdering FailureOrdering, | |||
| 1765 | SyncScope::ID SSID = SyncScope::System) { | |||
| 1766 | if (!Align) { | |||
| 1767 | const DataLayout &DL = BB->getModule()->getDataLayout(); | |||
| 1768 | Align = llvm::Align(DL.getTypeStoreSize(New->getType())); | |||
| 1769 | } | |||
| 1770 | ||||
| 1771 | return Insert(new AtomicCmpXchgInst(Ptr, Cmp, New, *Align, SuccessOrdering, | |||
| 1772 | FailureOrdering, SSID)); | |||
| 1773 | } | |||
| 1774 | ||||
| 1775 | AtomicRMWInst *CreateAtomicRMW(AtomicRMWInst::BinOp Op, Value *Ptr, | |||
| 1776 | Value *Val, MaybeAlign Align, | |||
| 1777 | AtomicOrdering Ordering, | |||
| 1778 | SyncScope::ID SSID = SyncScope::System) { | |||
| 1779 | if (!Align) { | |||
| 1780 | const DataLayout &DL = BB->getModule()->getDataLayout(); | |||
| 1781 | Align = llvm::Align(DL.getTypeStoreSize(Val->getType())); | |||
| 1782 | } | |||
| 1783 | ||||
| 1784 | return Insert(new AtomicRMWInst(Op, Ptr, Val, *Align, Ordering, SSID)); | |||
| 1785 | } | |||
| 1786 | ||||
| 1787 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateGEP(Value *Ptr, ArrayRef<Value *> IdxList , const Twine &Name = "") | |||
| 1788 | Value *CreateGEP(Value *Ptr, ArrayRef<Value *> IdxList,[[deprecated("Use the version with explicit element type instead" )]] Value *CreateGEP(Value *Ptr, ArrayRef<Value *> IdxList , const Twine &Name = "") | |||
| 1789 | const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateGEP(Value *Ptr, ArrayRef<Value *> IdxList , const Twine &Name = "") | |||
| 1790 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateGEP(Value *Ptr, ArrayRef<Value *> IdxList , const Twine &Name = "") { | |||
| 1791 | return CreateGEP(Ptr->getType()->getScalarType()->getPointerElementType(), | |||
| 1792 | Ptr, IdxList, Name); | |||
| 1793 | } | |||
| 1794 | ||||
| 1795 | Value *CreateGEP(Type *Ty, Value *Ptr, ArrayRef<Value *> IdxList, | |||
| 1796 | const Twine &Name = "") { | |||
| 1797 | if (auto *PC = dyn_cast<Constant>(Ptr)) { | |||
| 1798 | // Every index must be constant. | |||
| 1799 | size_t i, e; | |||
| 1800 | for (i = 0, e = IdxList.size(); i != e; ++i) | |||
| 1801 | if (!isa<Constant>(IdxList[i])) | |||
| 1802 | break; | |||
| 1803 | if (i == e) | |||
| 1804 | return Insert(Folder.CreateGetElementPtr(Ty, PC, IdxList), Name); | |||
| 1805 | } | |||
| 1806 | return Insert(GetElementPtrInst::Create(Ty, Ptr, IdxList), Name); | |||
| 1807 | } | |||
| 1808 | ||||
| 1809 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateInBoundsGEP(Value *Ptr, ArrayRef<Value *> IdxList, const Twine &Name = "") | |||
| 1810 | Value *CreateInBoundsGEP(Value *Ptr, ArrayRef<Value *> IdxList,[[deprecated("Use the version with explicit element type instead" )]] Value *CreateInBoundsGEP(Value *Ptr, ArrayRef<Value *> IdxList, const Twine &Name = "") | |||
| 1811 | const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateInBoundsGEP(Value *Ptr, ArrayRef<Value *> IdxList, const Twine &Name = "") | |||
| 1812 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateInBoundsGEP(Value *Ptr, ArrayRef<Value *> IdxList, const Twine &Name = "") { | |||
| 1813 | return CreateInBoundsGEP( | |||
| 1814 | Ptr->getType()->getScalarType()->getPointerElementType(), Ptr, IdxList, | |||
| 1815 | Name); | |||
| 1816 | } | |||
| 1817 | ||||
| 1818 | Value *CreateInBoundsGEP(Type *Ty, Value *Ptr, ArrayRef<Value *> IdxList, | |||
| 1819 | const Twine &Name = "") { | |||
| 1820 | if (auto *PC = dyn_cast<Constant>(Ptr)) { | |||
| 1821 | // Every index must be constant. | |||
| 1822 | size_t i, e; | |||
| 1823 | for (i = 0, e = IdxList.size(); i != e; ++i) | |||
| 1824 | if (!isa<Constant>(IdxList[i])) | |||
| 1825 | break; | |||
| 1826 | if (i == e) | |||
| 1827 | return Insert(Folder.CreateInBoundsGetElementPtr(Ty, PC, IdxList), | |||
| 1828 | Name); | |||
| 1829 | } | |||
| 1830 | return Insert(GetElementPtrInst::CreateInBounds(Ty, Ptr, IdxList), Name); | |||
| 1831 | } | |||
| 1832 | ||||
| 1833 | Value *CreateGEP(Type *Ty, Value *Ptr, Value *Idx, const Twine &Name = "") { | |||
| 1834 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1835 | if (auto *IC = dyn_cast<Constant>(Idx)) | |||
| 1836 | return Insert(Folder.CreateGetElementPtr(Ty, PC, IC), Name); | |||
| 1837 | return Insert(GetElementPtrInst::Create(Ty, Ptr, Idx), Name); | |||
| 1838 | } | |||
| 1839 | ||||
| 1840 | Value *CreateInBoundsGEP(Type *Ty, Value *Ptr, Value *Idx, | |||
| 1841 | const Twine &Name = "") { | |||
| 1842 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1843 | if (auto *IC = dyn_cast<Constant>(Idx)) | |||
| 1844 | return Insert(Folder.CreateInBoundsGetElementPtr(Ty, PC, IC), Name); | |||
| 1845 | return Insert(GetElementPtrInst::CreateInBounds(Ty, Ptr, Idx), Name); | |||
| 1846 | } | |||
| 1847 | ||||
| 1848 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_32(Value *Ptr, unsigned Idx0, const Twine &Name = "") | |||
| 1849 | Value *CreateConstGEP1_32(Value *Ptr, unsigned Idx0,[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_32(Value *Ptr, unsigned Idx0, const Twine &Name = "") | |||
| 1850 | const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_32(Value *Ptr, unsigned Idx0, const Twine &Name = "") | |||
| 1851 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_32(Value *Ptr, unsigned Idx0, const Twine &Name = "") { | |||
| 1852 | return CreateConstGEP1_32( | |||
| 1853 | Ptr->getType()->getScalarType()->getPointerElementType(), Ptr, Idx0, | |||
| 1854 | Name); | |||
| 1855 | } | |||
| 1856 | ||||
| 1857 | Value *CreateConstGEP1_32(Type *Ty, Value *Ptr, unsigned Idx0, | |||
| 1858 | const Twine &Name = "") { | |||
| 1859 | Value *Idx = ConstantInt::get(Type::getInt32Ty(Context), Idx0); | |||
| 1860 | ||||
| 1861 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1862 | return Insert(Folder.CreateGetElementPtr(Ty, PC, Idx), Name); | |||
| 1863 | ||||
| 1864 | return Insert(GetElementPtrInst::Create(Ty, Ptr, Idx), Name); | |||
| 1865 | } | |||
| 1866 | ||||
| 1867 | Value *CreateConstInBoundsGEP1_32(Type *Ty, Value *Ptr, unsigned Idx0, | |||
| 1868 | const Twine &Name = "") { | |||
| 1869 | Value *Idx = ConstantInt::get(Type::getInt32Ty(Context), Idx0); | |||
| 1870 | ||||
| 1871 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1872 | return Insert(Folder.CreateInBoundsGetElementPtr(Ty, PC, Idx), Name); | |||
| 1873 | ||||
| 1874 | return Insert(GetElementPtrInst::CreateInBounds(Ty, Ptr, Idx), Name); | |||
| 1875 | } | |||
| 1876 | ||||
| 1877 | Value *CreateConstGEP2_32(Type *Ty, Value *Ptr, unsigned Idx0, unsigned Idx1, | |||
| 1878 | const Twine &Name = "") { | |||
| 1879 | Value *Idxs[] = { | |||
| 1880 | ConstantInt::get(Type::getInt32Ty(Context), Idx0), | |||
| 1881 | ConstantInt::get(Type::getInt32Ty(Context), Idx1) | |||
| 1882 | }; | |||
| 1883 | ||||
| 1884 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1885 | return Insert(Folder.CreateGetElementPtr(Ty, PC, Idxs), Name); | |||
| 1886 | ||||
| 1887 | return Insert(GetElementPtrInst::Create(Ty, Ptr, Idxs), Name); | |||
| 1888 | } | |||
| 1889 | ||||
| 1890 | Value *CreateConstInBoundsGEP2_32(Type *Ty, Value *Ptr, unsigned Idx0, | |||
| 1891 | unsigned Idx1, const Twine &Name = "") { | |||
| 1892 | Value *Idxs[] = { | |||
| 1893 | ConstantInt::get(Type::getInt32Ty(Context), Idx0), | |||
| 1894 | ConstantInt::get(Type::getInt32Ty(Context), Idx1) | |||
| 1895 | }; | |||
| 1896 | ||||
| 1897 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1898 | return Insert(Folder.CreateInBoundsGetElementPtr(Ty, PC, Idxs), Name); | |||
| 1899 | ||||
| 1900 | return Insert(GetElementPtrInst::CreateInBounds(Ty, Ptr, Idxs), Name); | |||
| 1901 | } | |||
| 1902 | ||||
| 1903 | Value *CreateConstGEP1_64(Type *Ty, Value *Ptr, uint64_t Idx0, | |||
| 1904 | const Twine &Name = "") { | |||
| 1905 | Value *Idx = ConstantInt::get(Type::getInt64Ty(Context), Idx0); | |||
| 1906 | ||||
| 1907 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1908 | return Insert(Folder.CreateGetElementPtr(Ty, PC, Idx), Name); | |||
| 1909 | ||||
| 1910 | return Insert(GetElementPtrInst::Create(Ty, Ptr, Idx), Name); | |||
| 1911 | } | |||
| 1912 | ||||
| 1913 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_64(Value *Ptr, uint64_t Idx0, const Twine &Name = "") | |||
| 1914 | Value *CreateConstGEP1_64(Value *Ptr, uint64_t Idx0,[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_64(Value *Ptr, uint64_t Idx0, const Twine &Name = "") | |||
| 1915 | const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_64(Value *Ptr, uint64_t Idx0, const Twine &Name = "") | |||
| 1916 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP1_64(Value *Ptr, uint64_t Idx0, const Twine &Name = "") { | |||
| 1917 | return CreateConstGEP1_64( | |||
| 1918 | Ptr->getType()->getScalarType()->getPointerElementType(), Ptr, Idx0, | |||
| 1919 | Name); | |||
| 1920 | } | |||
| 1921 | ||||
| 1922 | Value *CreateConstInBoundsGEP1_64(Type *Ty, Value *Ptr, uint64_t Idx0, | |||
| 1923 | const Twine &Name = "") { | |||
| 1924 | Value *Idx = ConstantInt::get(Type::getInt64Ty(Context), Idx0); | |||
| 1925 | ||||
| 1926 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1927 | return Insert(Folder.CreateInBoundsGetElementPtr(Ty, PC, Idx), Name); | |||
| 1928 | ||||
| 1929 | return Insert(GetElementPtrInst::CreateInBounds(Ty, Ptr, Idx), Name); | |||
| 1930 | } | |||
| 1931 | ||||
| 1932 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP1_64(Value *Ptr, uint64_t Idx0 , const Twine &Name = "") | |||
| 1933 | Value *CreateConstInBoundsGEP1_64(Value *Ptr, uint64_t Idx0,[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP1_64(Value *Ptr, uint64_t Idx0 , const Twine &Name = "") | |||
| 1934 | const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP1_64(Value *Ptr, uint64_t Idx0 , const Twine &Name = "") | |||
| 1935 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP1_64(Value *Ptr, uint64_t Idx0 , const Twine &Name = "") { | |||
| 1936 | return CreateConstInBoundsGEP1_64( | |||
| 1937 | Ptr->getType()->getScalarType()->getPointerElementType(), Ptr, Idx0, | |||
| 1938 | Name); | |||
| 1939 | } | |||
| 1940 | ||||
| 1941 | Value *CreateConstGEP2_64(Type *Ty, Value *Ptr, uint64_t Idx0, uint64_t Idx1, | |||
| 1942 | const Twine &Name = "") { | |||
| 1943 | Value *Idxs[] = { | |||
| 1944 | ConstantInt::get(Type::getInt64Ty(Context), Idx0), | |||
| 1945 | ConstantInt::get(Type::getInt64Ty(Context), Idx1) | |||
| 1946 | }; | |||
| 1947 | ||||
| 1948 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1949 | return Insert(Folder.CreateGetElementPtr(Ty, PC, Idxs), Name); | |||
| 1950 | ||||
| 1951 | return Insert(GetElementPtrInst::Create(Ty, Ptr, Idxs), Name); | |||
| 1952 | } | |||
| 1953 | ||||
| 1954 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP2_64(Value *Ptr, uint64_t Idx0, uint64_t Idx1, const Twine &Name = "") | |||
| 1955 | Value *CreateConstGEP2_64(Value *Ptr, uint64_t Idx0, uint64_t Idx1,[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP2_64(Value *Ptr, uint64_t Idx0, uint64_t Idx1, const Twine &Name = "") | |||
| 1956 | const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP2_64(Value *Ptr, uint64_t Idx0, uint64_t Idx1, const Twine &Name = "") | |||
| 1957 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstGEP2_64(Value *Ptr, uint64_t Idx0, uint64_t Idx1, const Twine &Name = "") { | |||
| 1958 | return CreateConstGEP2_64( | |||
| 1959 | Ptr->getType()->getScalarType()->getPointerElementType(), Ptr, Idx0, | |||
| 1960 | Idx1, Name); | |||
| 1961 | } | |||
| 1962 | ||||
| 1963 | Value *CreateConstInBoundsGEP2_64(Type *Ty, Value *Ptr, uint64_t Idx0, | |||
| 1964 | uint64_t Idx1, const Twine &Name = "") { | |||
| 1965 | Value *Idxs[] = { | |||
| 1966 | ConstantInt::get(Type::getInt64Ty(Context), Idx0), | |||
| 1967 | ConstantInt::get(Type::getInt64Ty(Context), Idx1) | |||
| 1968 | }; | |||
| 1969 | ||||
| 1970 | if (auto *PC = dyn_cast<Constant>(Ptr)) | |||
| 1971 | return Insert(Folder.CreateInBoundsGetElementPtr(Ty, PC, Idxs), Name); | |||
| 1972 | ||||
| 1973 | return Insert(GetElementPtrInst::CreateInBounds(Ty, Ptr, Idxs), Name); | |||
| 1974 | } | |||
| 1975 | ||||
| 1976 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP2_64(Value *Ptr, uint64_t Idx0 , uint64_t Idx1, const Twine &Name = "") | |||
| 1977 | Value *CreateConstInBoundsGEP2_64(Value *Ptr, uint64_t Idx0,[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP2_64(Value *Ptr, uint64_t Idx0 , uint64_t Idx1, const Twine &Name = "") | |||
| 1978 | uint64_t Idx1, const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP2_64(Value *Ptr, uint64_t Idx0 , uint64_t Idx1, const Twine &Name = "") | |||
| 1979 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateConstInBoundsGEP2_64(Value *Ptr, uint64_t Idx0 , uint64_t Idx1, const Twine &Name = "") { | |||
| 1980 | return CreateConstInBoundsGEP2_64( | |||
| 1981 | Ptr->getType()->getScalarType()->getPointerElementType(), Ptr, Idx0, | |||
| 1982 | Idx1, Name); | |||
| 1983 | } | |||
| 1984 | ||||
| 1985 | Value *CreateStructGEP(Type *Ty, Value *Ptr, unsigned Idx, | |||
| 1986 | const Twine &Name = "") { | |||
| 1987 | return CreateConstInBoundsGEP2_32(Ty, Ptr, 0, Idx, Name); | |||
| 1988 | } | |||
| 1989 | ||||
| 1990 | LLVM_ATTRIBUTE_DEPRECATED([[deprecated("Use the version with explicit element type instead" )]] Value *CreateStructGEP(Value *Ptr, unsigned Idx, const Twine &Name = "") | |||
| 1991 | Value *CreateStructGEP(Value *Ptr, unsigned Idx, const Twine &Name = ""),[[deprecated("Use the version with explicit element type instead" )]] Value *CreateStructGEP(Value *Ptr, unsigned Idx, const Twine &Name = "") | |||
| 1992 | "Use the version with explicit element type instead")[[deprecated("Use the version with explicit element type instead" )]] Value *CreateStructGEP(Value *Ptr, unsigned Idx, const Twine &Name = "") { | |||
| 1993 | return CreateConstInBoundsGEP2_32( | |||
| 1994 | Ptr->getType()->getScalarType()->getPointerElementType(), Ptr, 0, Idx, | |||
| 1995 | Name); | |||
| 1996 | } | |||
| 1997 | ||||
| 1998 | /// Same as CreateGlobalString, but return a pointer with "i8*" type | |||
| 1999 | /// instead of a pointer to array of i8. | |||
| 2000 | /// | |||
| 2001 | /// If no module is given via \p M, it is take from the insertion point basic | |||
| 2002 | /// block. | |||
| 2003 | Constant *CreateGlobalStringPtr(StringRef Str, const Twine &Name = "", | |||
| 2004 | unsigned AddressSpace = 0, | |||
| 2005 | Module *M = nullptr) { | |||
| 2006 | GlobalVariable *GV = CreateGlobalString(Str, Name, AddressSpace, M); | |||
| 2007 | Constant *Zero = ConstantInt::get(Type::getInt32Ty(Context), 0); | |||
| 2008 | Constant *Indices[] = {Zero, Zero}; | |||
| 2009 | return ConstantExpr::getInBoundsGetElementPtr(GV->getValueType(), GV, | |||
| 2010 | Indices); | |||
| 2011 | } | |||
| 2012 | ||||
| 2013 | //===--------------------------------------------------------------------===// | |||
| 2014 | // Instruction creation methods: Cast/Conversion Operators | |||
| 2015 | //===--------------------------------------------------------------------===// | |||
| 2016 | ||||
| 2017 | Value *CreateTrunc(Value *V, Type *DestTy, const Twine &Name = "") { | |||
| 2018 | return CreateCast(Instruction::Trunc, V, DestTy, Name); | |||
| 2019 | } | |||
| 2020 | ||||
| 2021 | Value *CreateZExt(Value *V, Type *DestTy, const Twine &Name = "") { | |||
| 2022 | return CreateCast(Instruction::ZExt, V, DestTy, Name); | |||
| 2023 | } | |||
| 2024 | ||||
| 2025 | Value *CreateSExt(Value *V, Type *DestTy, const Twine &Name = "") { | |||
| 2026 | return CreateCast(Instruction::SExt, V, DestTy, Name); | |||
| 2027 | } | |||
| 2028 | ||||
| 2029 | /// Create a ZExt or Trunc from the integer value V to DestTy. Return | |||
| 2030 | /// the value untouched if the type of V is already DestTy. | |||
| 2031 | Value *CreateZExtOrTrunc(Value *V, Type *DestTy, | |||
| 2032 | const Twine &Name = "") { | |||
| 2033 | assert(V->getType()->isIntOrIntVectorTy() &&((void)0) | |||
| 2034 | DestTy->isIntOrIntVectorTy() &&((void)0) | |||
| 2035 | "Can only zero extend/truncate integers!")((void)0); | |||
| 2036 | Type *VTy = V->getType(); | |||
| 2037 | if (VTy->getScalarSizeInBits() < DestTy->getScalarSizeInBits()) | |||
| 2038 | return CreateZExt(V, DestTy, Name); | |||
| 2039 | if (VTy->getScalarSizeInBits() > DestTy->getScalarSizeInBits()) | |||
| 2040 | return CreateTrunc(V, DestTy, Name); | |||
| 2041 | return V; | |||
| 2042 | } | |||
| 2043 | ||||
| 2044 | /// Create a SExt or Trunc from the integer value V to DestTy. Return | |||
| 2045 | /// the value untouched if the type of V is already DestTy. | |||
| 2046 | Value *CreateSExtOrTrunc(Value *V, Type *DestTy, | |||
| 2047 | const Twine &Name = "") { | |||
| 2048 | assert(V->getType()->isIntOrIntVectorTy() &&((void)0) | |||
| 2049 | DestTy->isIntOrIntVectorTy() &&((void)0) | |||
| 2050 | "Can only sign extend/truncate integers!")((void)0); | |||
| 2051 | Type *VTy = V->getType(); | |||
| 2052 | if (VTy->getScalarSizeInBits() < DestTy->getScalarSizeInBits()) | |||
| 2053 | return CreateSExt(V, DestTy, Name); | |||
| 2054 | if (VTy->getScalarSizeInBits() > DestTy->getScalarSizeInBits()) | |||
| 2055 | return CreateTrunc(V, DestTy, Name); | |||
| 2056 | return V; | |||
| 2057 | } | |||
| 2058 | ||||
| 2059 | Value *CreateFPToUI(Value *V, Type *DestTy, const Twine &Name = "") { | |||
| 2060 | if (IsFPConstrained) | |||
| 2061 | return CreateConstrainedFPCast(Intrinsic::experimental_constrained_fptoui, | |||
| 2062 | V, DestTy, nullptr, Name); | |||
| 2063 | return CreateCast(Instruction::FPToUI, V, DestTy, Name); | |||
| 2064 | } | |||
| 2065 | ||||
| 2066 | Value *CreateFPToSI(Value *V, Type *DestTy, const Twine &Name = "") { | |||
| 2067 | if (IsFPConstrained) | |||
| 2068 | return CreateConstrainedFPCast(Intrinsic::experimental_constrained_fptosi, | |||
| 2069 | V, DestTy, nullptr, Name); | |||
| 2070 | return CreateCast(Instruction::FPToSI, V, DestTy, Name); | |||
| 2071 | } | |||
| 2072 | ||||
| 2073 | Value *CreateUIToFP(Value *V, Type *DestTy, const Twine &Name = ""){ | |||
| 2074 | if (IsFPConstrained) | |||
| 2075 | return CreateConstrainedFPCast(Intrinsic::experimental_constrained_uitofp, | |||
| 2076 | V, DestTy, nullptr, Name); | |||
| 2077 | return CreateCast(Instruction::UIToFP, V, DestTy, Name); | |||
| 2078 | } | |||
| 2079 | ||||
| 2080 | Value *CreateSIToFP(Value *V, Type *DestTy, const Twine &Name = ""){ | |||
| 2081 | if (IsFPConstrained) | |||
| 2082 | return CreateConstrainedFPCast(Intrinsic::experimental_constrained_sitofp, | |||
| 2083 | V, DestTy, nullptr, Name); | |||
| 2084 | return CreateCast(Instruction::SIToFP, V, DestTy, Name); | |||
| 2085 | } | |||
| 2086 | ||||
| 2087 | Value *CreateFPTrunc(Value *V, Type *DestTy, | |||
| 2088 | const Twine &Name = "") { | |||
| 2089 | if (IsFPConstrained) | |||
| 2090 | return CreateConstrainedFPCast( | |||
| 2091 | Intrinsic::experimental_constrained_fptrunc, V, DestTy, nullptr, | |||
| 2092 | Name); | |||
| 2093 | return CreateCast(Instruction::FPTrunc, V, DestTy, Name); | |||
| 2094 | } | |||
| 2095 | ||||
| 2096 | Value *CreateFPExt(Value *V, Type *DestTy, const Twine &Name = "") { | |||
| 2097 | if (IsFPConstrained) | |||
| 2098 | return CreateConstrainedFPCast(Intrinsic::experimental_constrained_fpext, | |||
| 2099 | V, DestTy, nullptr, Name); | |||
| 2100 | return CreateCast(Instruction::FPExt, V, DestTy, Name); | |||
| 2101 | } | |||
| 2102 | ||||
| 2103 | Value *CreatePtrToInt(Value *V, Type *DestTy, | |||
| 2104 | const Twine &Name = "") { | |||
| 2105 | return CreateCast(Instruction::PtrToInt, V, DestTy, Name); | |||
| 2106 | } | |||
| 2107 | ||||
| 2108 | Value *CreateIntToPtr(Value *V, Type *DestTy, | |||
| 2109 | const Twine &Name = "") { | |||
| 2110 | return CreateCast(Instruction::IntToPtr, V, DestTy, Name); | |||
| 2111 | } | |||
| 2112 | ||||
| 2113 | Value *CreateBitCast(Value *V, Type *DestTy, | |||
| 2114 | const Twine &Name = "") { | |||
| 2115 | return CreateCast(Instruction::BitCast, V, DestTy, Name); | |||
| 2116 | } | |||
| 2117 | ||||
| 2118 | Value *CreateAddrSpaceCast(Value *V, Type *DestTy, | |||
| 2119 | const Twine &Name = "") { | |||
| 2120 | return CreateCast(Instruction::AddrSpaceCast, V, DestTy, Name); | |||
| 2121 | } | |||
| 2122 | ||||
| 2123 | Value *CreateZExtOrBitCast(Value *V, Type *DestTy, | |||
| 2124 | const Twine &Name = "") { | |||
| 2125 | if (V->getType() == DestTy) | |||
| 2126 | return V; | |||
| 2127 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 2128 | return Insert(Folder.CreateZExtOrBitCast(VC, DestTy), Name); | |||
| 2129 | return Insert(CastInst::CreateZExtOrBitCast(V, DestTy), Name); | |||
| 2130 | } | |||
| 2131 | ||||
| 2132 | Value *CreateSExtOrBitCast(Value *V, Type *DestTy, | |||
| 2133 | const Twine &Name = "") { | |||
| 2134 | if (V->getType() == DestTy) | |||
| 2135 | return V; | |||
| 2136 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 2137 | return Insert(Folder.CreateSExtOrBitCast(VC, DestTy), Name); | |||
| 2138 | return Insert(CastInst::CreateSExtOrBitCast(V, DestTy), Name); | |||
| 2139 | } | |||
| 2140 | ||||
| 2141 | Value *CreateTruncOrBitCast(Value *V, Type *DestTy, | |||
| 2142 | const Twine &Name = "") { | |||
| 2143 | if (V->getType() == DestTy) | |||
| 2144 | return V; | |||
| 2145 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 2146 | return Insert(Folder.CreateTruncOrBitCast(VC, DestTy), Name); | |||
| 2147 | return Insert(CastInst::CreateTruncOrBitCast(V, DestTy), Name); | |||
| 2148 | } | |||
| 2149 | ||||
| 2150 | Value *CreateCast(Instruction::CastOps Op, Value *V, Type *DestTy, | |||
| 2151 | const Twine &Name = "") { | |||
| 2152 | if (V->getType() == DestTy) | |||
| 2153 | return V; | |||
| 2154 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 2155 | return Insert(Folder.CreateCast(Op, VC, DestTy), Name); | |||
| 2156 | return Insert(CastInst::Create(Op, V, DestTy), Name); | |||
| 2157 | } | |||
| 2158 | ||||
| 2159 | Value *CreatePointerCast(Value *V, Type *DestTy, | |||
| 2160 | const Twine &Name = "") { | |||
| 2161 | if (V->getType() == DestTy) | |||
| 2162 | return V; | |||
| 2163 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 2164 | return Insert(Folder.CreatePointerCast(VC, DestTy), Name); | |||
| 2165 | return Insert(CastInst::CreatePointerCast(V, DestTy), Name); | |||
| 2166 | } | |||
| 2167 | ||||
| 2168 | Value *CreatePointerBitCastOrAddrSpaceCast(Value *V, Type *DestTy, | |||
| 2169 | const Twine &Name = "") { | |||
| 2170 | if (V->getType() == DestTy) | |||
| 2171 | return V; | |||
| 2172 | ||||
| 2173 | if (auto *VC = dyn_cast<Constant>(V)) { | |||
| 2174 | return Insert(Folder.CreatePointerBitCastOrAddrSpaceCast(VC, DestTy), | |||
| 2175 | Name); | |||
| 2176 | } | |||
| 2177 | ||||
| 2178 | return Insert(CastInst::CreatePointerBitCastOrAddrSpaceCast(V, DestTy), | |||
| 2179 | Name); | |||
| 2180 | } | |||
| 2181 | ||||
| 2182 | Value *CreateIntCast(Value *V, Type *DestTy, bool isSigned, | |||
| 2183 | const Twine &Name = "") { | |||
| 2184 | if (V->getType() == DestTy) | |||
| 2185 | return V; | |||
| 2186 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 2187 | return Insert(Folder.CreateIntCast(VC, DestTy, isSigned), Name); | |||
| 2188 | return Insert(CastInst::CreateIntegerCast(V, DestTy, isSigned), Name); | |||
| 2189 | } | |||
| 2190 | ||||
| 2191 | Value *CreateBitOrPointerCast(Value *V, Type *DestTy, | |||
| 2192 | const Twine &Name = "") { | |||
| 2193 | if (V->getType() == DestTy) | |||
| 2194 | return V; | |||
| 2195 | if (V->getType()->isPtrOrPtrVectorTy() && DestTy->isIntOrIntVectorTy()) | |||
| 2196 | return CreatePtrToInt(V, DestTy, Name); | |||
| 2197 | if (V->getType()->isIntOrIntVectorTy() && DestTy->isPtrOrPtrVectorTy()) | |||
| 2198 | return CreateIntToPtr(V, DestTy, Name); | |||
| 2199 | ||||
| 2200 | return CreateBitCast(V, DestTy, Name); | |||
| 2201 | } | |||
| 2202 | ||||
| 2203 | Value *CreateFPCast(Value *V, Type *DestTy, const Twine &Name = "") { | |||
| 2204 | if (V->getType() == DestTy) | |||
| 2205 | return V; | |||
| 2206 | if (auto *VC = dyn_cast<Constant>(V)) | |||
| 2207 | return Insert(Folder.CreateFPCast(VC, DestTy), Name); | |||
| 2208 | return Insert(CastInst::CreateFPCast(V, DestTy), Name); | |||
| 2209 | } | |||
| 2210 | ||||
| 2211 | CallInst *CreateConstrainedFPCast( | |||
| 2212 | Intrinsic::ID ID, Value *V, Type *DestTy, | |||
| 2213 | Instruction *FMFSource = nullptr, const Twine &Name = "", | |||
| 2214 | MDNode *FPMathTag = nullptr, | |||
| 2215 | Optional<RoundingMode> Rounding = None, | |||
| 2216 | Optional<fp::ExceptionBehavior> Except = None); | |||
| 2217 | ||||
| 2218 | // Provided to resolve 'CreateIntCast(Ptr, Ptr, "...")', giving a | |||
| 2219 | // compile time error, instead of converting the string to bool for the | |||
| 2220 | // isSigned parameter. | |||
| 2221 | Value *CreateIntCast(Value *, Type *, const char *) = delete; | |||
| 2222 | ||||
| 2223 | //===--------------------------------------------------------------------===// | |||
| 2224 | // Instruction creation methods: Compare Instructions | |||
| 2225 | //===--------------------------------------------------------------------===// | |||
| 2226 | ||||
| 2227 | Value *CreateICmpEQ(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2228 | return CreateICmp(ICmpInst::ICMP_EQ, LHS, RHS, Name); | |||
| 2229 | } | |||
| 2230 | ||||
| 2231 | Value *CreateICmpNE(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2232 | return CreateICmp(ICmpInst::ICMP_NE, LHS, RHS, Name); | |||
| 2233 | } | |||
| 2234 | ||||
| 2235 | Value *CreateICmpUGT(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2236 | return CreateICmp(ICmpInst::ICMP_UGT, LHS, RHS, Name); | |||
| 2237 | } | |||
| 2238 | ||||
| 2239 | Value *CreateICmpUGE(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2240 | return CreateICmp(ICmpInst::ICMP_UGE, LHS, RHS, Name); | |||
| 2241 | } | |||
| 2242 | ||||
| 2243 | Value *CreateICmpULT(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2244 | return CreateICmp(ICmpInst::ICMP_ULT, LHS, RHS, Name); | |||
| 2245 | } | |||
| 2246 | ||||
| 2247 | Value *CreateICmpULE(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2248 | return CreateICmp(ICmpInst::ICMP_ULE, LHS, RHS, Name); | |||
| 2249 | } | |||
| 2250 | ||||
| 2251 | Value *CreateICmpSGT(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2252 | return CreateICmp(ICmpInst::ICMP_SGT, LHS, RHS, Name); | |||
| 2253 | } | |||
| 2254 | ||||
| 2255 | Value *CreateICmpSGE(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2256 | return CreateICmp(ICmpInst::ICMP_SGE, LHS, RHS, Name); | |||
| 2257 | } | |||
| 2258 | ||||
| 2259 | Value *CreateICmpSLT(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2260 | return CreateICmp(ICmpInst::ICMP_SLT, LHS, RHS, Name); | |||
| 2261 | } | |||
| 2262 | ||||
| 2263 | Value *CreateICmpSLE(Value *LHS, Value *RHS, const Twine &Name = "") { | |||
| 2264 | return CreateICmp(ICmpInst::ICMP_SLE, LHS, RHS, Name); | |||
| 2265 | } | |||
| 2266 | ||||
| 2267 | Value *CreateFCmpOEQ(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2268 | MDNode *FPMathTag = nullptr) { | |||
| 2269 | return CreateFCmp(FCmpInst::FCMP_OEQ, LHS, RHS, Name, FPMathTag); | |||
| 2270 | } | |||
| 2271 | ||||
| 2272 | Value *CreateFCmpOGT(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2273 | MDNode *FPMathTag = nullptr) { | |||
| 2274 | return CreateFCmp(FCmpInst::FCMP_OGT, LHS, RHS, Name, FPMathTag); | |||
| 2275 | } | |||
| 2276 | ||||
| 2277 | Value *CreateFCmpOGE(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2278 | MDNode *FPMathTag = nullptr) { | |||
| 2279 | return CreateFCmp(FCmpInst::FCMP_OGE, LHS, RHS, Name, FPMathTag); | |||
| 2280 | } | |||
| 2281 | ||||
| 2282 | Value *CreateFCmpOLT(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2283 | MDNode *FPMathTag = nullptr) { | |||
| 2284 | return CreateFCmp(FCmpInst::FCMP_OLT, LHS, RHS, Name, FPMathTag); | |||
| 2285 | } | |||
| 2286 | ||||
| 2287 | Value *CreateFCmpOLE(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2288 | MDNode *FPMathTag = nullptr) { | |||
| 2289 | return CreateFCmp(FCmpInst::FCMP_OLE, LHS, RHS, Name, FPMathTag); | |||
| 2290 | } | |||
| 2291 | ||||
| 2292 | Value *CreateFCmpONE(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2293 | MDNode *FPMathTag = nullptr) { | |||
| 2294 | return CreateFCmp(FCmpInst::FCMP_ONE, LHS, RHS, Name, FPMathTag); | |||
| 2295 | } | |||
| 2296 | ||||
| 2297 | Value *CreateFCmpORD(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2298 | MDNode *FPMathTag = nullptr) { | |||
| 2299 | return CreateFCmp(FCmpInst::FCMP_ORD, LHS, RHS, Name, FPMathTag); | |||
| 2300 | } | |||
| 2301 | ||||
| 2302 | Value *CreateFCmpUNO(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2303 | MDNode *FPMathTag = nullptr) { | |||
| 2304 | return CreateFCmp(FCmpInst::FCMP_UNO, LHS, RHS, Name, FPMathTag); | |||
| 2305 | } | |||
| 2306 | ||||
| 2307 | Value *CreateFCmpUEQ(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2308 | MDNode *FPMathTag = nullptr) { | |||
| 2309 | return CreateFCmp(FCmpInst::FCMP_UEQ, LHS, RHS, Name, FPMathTag); | |||
| 2310 | } | |||
| 2311 | ||||
| 2312 | Value *CreateFCmpUGT(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2313 | MDNode *FPMathTag = nullptr) { | |||
| 2314 | return CreateFCmp(FCmpInst::FCMP_UGT, LHS, RHS, Name, FPMathTag); | |||
| 2315 | } | |||
| 2316 | ||||
| 2317 | Value *CreateFCmpUGE(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2318 | MDNode *FPMathTag = nullptr) { | |||
| 2319 | return CreateFCmp(FCmpInst::FCMP_UGE, LHS, RHS, Name, FPMathTag); | |||
| 2320 | } | |||
| 2321 | ||||
| 2322 | Value *CreateFCmpULT(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2323 | MDNode *FPMathTag = nullptr) { | |||
| 2324 | return CreateFCmp(FCmpInst::FCMP_ULT, LHS, RHS, Name, FPMathTag); | |||
| 2325 | } | |||
| 2326 | ||||
| 2327 | Value *CreateFCmpULE(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2328 | MDNode *FPMathTag = nullptr) { | |||
| 2329 | return CreateFCmp(FCmpInst::FCMP_ULE, LHS, RHS, Name, FPMathTag); | |||
| 2330 | } | |||
| 2331 | ||||
| 2332 | Value *CreateFCmpUNE(Value *LHS, Value *RHS, const Twine &Name = "", | |||
| 2333 | MDNode *FPMathTag = nullptr) { | |||
| 2334 | return CreateFCmp(FCmpInst::FCMP_UNE, LHS, RHS, Name, FPMathTag); | |||
| 2335 | } | |||
| 2336 | ||||
| 2337 | Value *CreateICmp(CmpInst::Predicate P, Value *LHS, Value *RHS, | |||
| 2338 | const Twine &Name = "") { | |||
| 2339 | if (auto *LC = dyn_cast<Constant>(LHS)) | |||
| 2340 | if (auto *RC = dyn_cast<Constant>(RHS)) | |||
| 2341 | return Insert(Folder.CreateICmp(P, LC, RC), Name); | |||
| 2342 | return Insert(new ICmpInst(P, LHS, RHS), Name); | |||
| 2343 | } | |||
| 2344 | ||||
| 2345 | // Create a quiet floating-point comparison (i.e. one that raises an FP | |||
| 2346 | // exception only in the case where an input is a signaling NaN). | |||
| 2347 | // Note that this differs from CreateFCmpS only if IsFPConstrained is true. | |||
| 2348 | Value *CreateFCmp(CmpInst::Predicate P, Value *LHS, Value *RHS, | |||
| 2349 | const Twine &Name = "", MDNode *FPMathTag = nullptr) { | |||
| 2350 | return CreateFCmpHelper(P, LHS, RHS, Name, FPMathTag, false); | |||
| 2351 | } | |||
| 2352 | ||||
| 2353 | Value *CreateCmp(CmpInst::Predicate Pred, Value *LHS, Value *RHS, | |||
| 2354 | const Twine &Name = "", MDNode *FPMathTag = nullptr) { | |||
| 2355 | return CmpInst::isFPPredicate(Pred) | |||
| 2356 | ? CreateFCmp(Pred, LHS, RHS, Name, FPMathTag) | |||
| 2357 | : CreateICmp(Pred, LHS, RHS, Name); | |||
| 2358 | } | |||
| 2359 | ||||
| 2360 | // Create a signaling floating-point comparison (i.e. one that raises an FP | |||
| 2361 | // exception whenever an input is any NaN, signaling or quiet). | |||
| 2362 | // Note that this differs from CreateFCmp only if IsFPConstrained is true. | |||
| 2363 | Value *CreateFCmpS(CmpInst::Predicate P, Value *LHS, Value *RHS, | |||
| 2364 | const Twine &Name = "", MDNode *FPMathTag = nullptr) { | |||
| 2365 | return CreateFCmpHelper(P, LHS, RHS, Name, FPMathTag, true); | |||
| 2366 | } | |||
| 2367 | ||||
| 2368 | private: | |||
| 2369 | // Helper routine to create either a signaling or a quiet FP comparison. | |||
| 2370 | Value *CreateFCmpHelper(CmpInst::Predicate P, Value *LHS, Value *RHS, | |||
| 2371 | const Twine &Name, MDNode *FPMathTag, | |||
| 2372 | bool IsSignaling); | |||
| 2373 | ||||
| 2374 | public: | |||
| 2375 | CallInst *CreateConstrainedFPCmp( | |||
| 2376 | Intrinsic::ID ID, CmpInst::Predicate P, Value *L, Value *R, | |||
| 2377 | const Twine &Name = "", Optional<fp::ExceptionBehavior> Except = None); | |||
| 2378 | ||||
| 2379 | //===--------------------------------------------------------------------===// | |||
| 2380 | // Instruction creation methods: Other Instructions | |||
| 2381 | //===--------------------------------------------------------------------===// | |||
| 2382 | ||||
| 2383 | PHINode *CreatePHI(Type *Ty, unsigned NumReservedValues, | |||
| 2384 | const Twine &Name = "") { | |||
| 2385 | PHINode *Phi = PHINode::Create(Ty, NumReservedValues); | |||
| 2386 | if (isa<FPMathOperator>(Phi)) | |||
| 2387 | setFPAttrs(Phi, nullptr /* MDNode* */, FMF); | |||
| 2388 | return Insert(Phi, Name); | |||
| 2389 | } | |||
| 2390 | ||||
| 2391 | CallInst *CreateCall(FunctionType *FTy, Value *Callee, | |||
| 2392 | ArrayRef<Value *> Args = None, const Twine &Name = "", | |||
| 2393 | MDNode *FPMathTag = nullptr) { | |||
| 2394 | CallInst *CI = CallInst::Create(FTy, Callee, Args, DefaultOperandBundles); | |||
| 2395 | if (IsFPConstrained) | |||
| 2396 | setConstrainedFPCallAttr(CI); | |||
| 2397 | if (isa<FPMathOperator>(CI)) | |||
| 2398 | setFPAttrs(CI, FPMathTag, FMF); | |||
| 2399 | return Insert(CI, Name); | |||
| 2400 | } | |||
| 2401 | ||||
| 2402 | CallInst *CreateCall(FunctionType *FTy, Value *Callee, ArrayRef<Value *> Args, | |||
| 2403 | ArrayRef<OperandBundleDef> OpBundles, | |||
| 2404 | const Twine &Name = "", MDNode *FPMathTag = nullptr) { | |||
| 2405 | CallInst *CI = CallInst::Create(FTy, Callee, Args, OpBundles); | |||
| 2406 | if (IsFPConstrained) | |||
| 2407 | setConstrainedFPCallAttr(CI); | |||
| 2408 | if (isa<FPMathOperator>(CI)) | |||
| 2409 | setFPAttrs(CI, FPMathTag, FMF); | |||
| 2410 | return Insert(CI, Name); | |||
| 2411 | } | |||
| 2412 | ||||
| 2413 | CallInst *CreateCall(FunctionCallee Callee, ArrayRef<Value *> Args = None, | |||
| 2414 | const Twine &Name = "", MDNode *FPMathTag = nullptr) { | |||
| 2415 | return CreateCall(Callee.getFunctionType(), Callee.getCallee(), Args, Name, | |||
| 2416 | FPMathTag); | |||
| 2417 | } | |||
| 2418 | ||||
| 2419 | CallInst *CreateCall(FunctionCallee Callee, ArrayRef<Value *> Args, | |||
| 2420 | ArrayRef<OperandBundleDef> OpBundles, | |||
| 2421 | const Twine &Name = "", MDNode *FPMathTag = nullptr) { | |||
| 2422 | return CreateCall(Callee.getFunctionType(), Callee.getCallee(), Args, | |||
| 2423 | OpBundles, Name, FPMathTag); | |||
| 2424 | } | |||
| 2425 | ||||
| 2426 | CallInst *CreateConstrainedFPCall( | |||
| 2427 | Function *Callee, ArrayRef<Value *> Args, const Twine &Name = "", | |||
| 2428 | Optional<RoundingMode> Rounding = None, | |||
| 2429 | Optional<fp::ExceptionBehavior> Except = None); | |||
| 2430 | ||||
| 2431 | Value *CreateSelect(Value *C, Value *True, Value *False, | |||
| 2432 | const Twine &Name = "", Instruction *MDFrom = nullptr); | |||
| 2433 | ||||
| 2434 | VAArgInst *CreateVAArg(Value *List, Type *Ty, const Twine &Name = "") { | |||
| 2435 | return Insert(new VAArgInst(List, Ty), Name); | |||
| 2436 | } | |||
| 2437 | ||||
| 2438 | Value *CreateExtractElement(Value *Vec, Value *Idx, | |||
| 2439 | const Twine &Name = "") { | |||
| 2440 | if (auto *VC = dyn_cast<Constant>(Vec)) | |||
| 2441 | if (auto *IC = dyn_cast<Constant>(Idx)) | |||
| 2442 | return Insert(Folder.CreateExtractElement(VC, IC), Name); | |||
| 2443 | return Insert(ExtractElementInst::Create(Vec, Idx), Name); | |||
| 2444 | } | |||
| 2445 | ||||
| 2446 | Value *CreateExtractElement(Value *Vec, uint64_t Idx, | |||
| 2447 | const Twine &Name = "") { | |||
| 2448 | return CreateExtractElement(Vec, getInt64(Idx), Name); | |||
| 2449 | } | |||
| 2450 | ||||
| 2451 | Value *CreateInsertElement(Value *Vec, Value *NewElt, Value *Idx, | |||
| 2452 | const Twine &Name = "") { | |||
| 2453 | if (auto *VC = dyn_cast<Constant>(Vec)) | |||
| 2454 | if (auto *NC = dyn_cast<Constant>(NewElt)) | |||
| 2455 | if (auto *IC = dyn_cast<Constant>(Idx)) | |||
| 2456 | return Insert(Folder.CreateInsertElement(VC, NC, IC), Name); | |||
| 2457 | return Insert(InsertElementInst::Create(Vec, NewElt, Idx), Name); | |||
| 2458 | } | |||
| 2459 | ||||
| 2460 | Value *CreateInsertElement(Value *Vec, Value *NewElt, uint64_t Idx, | |||
| 2461 | const Twine &Name = "") { | |||
| 2462 | return CreateInsertElement(Vec, NewElt, getInt64(Idx), Name); | |||
| 2463 | } | |||
| 2464 | ||||
| 2465 | Value *CreateShuffleVector(Value *V1, Value *V2, Value *Mask, | |||
| 2466 | const Twine &Name = "") { | |||
| 2467 | SmallVector<int, 16> IntMask; | |||
| 2468 | ShuffleVectorInst::getShuffleMask(cast<Constant>(Mask), IntMask); | |||
| 2469 | return CreateShuffleVector(V1, V2, IntMask, Name); | |||
| 2470 | } | |||
| 2471 | ||||
| 2472 | LLVM_ATTRIBUTE_DEPRECATED(Value *CreateShuffleVector(Value *V1, Value *V2,[[deprecated("Pass indices as 'int' instead")]] Value *CreateShuffleVector (Value *V1, Value *V2, ArrayRef<uint32_t> Mask, const Twine &Name = "") | |||
| 2473 | ArrayRef<uint32_t> Mask,[[deprecated("Pass indices as 'int' instead")]] Value *CreateShuffleVector (Value *V1, Value *V2, ArrayRef<uint32_t> Mask, const Twine &Name = "") | |||
| 2474 | const Twine &Name = ""),[[deprecated("Pass indices as 'int' instead")]] Value *CreateShuffleVector (Value *V1, Value *V2, ArrayRef<uint32_t> Mask, const Twine &Name = "") | |||
| 2475 | "Pass indices as 'int' instead")[[deprecated("Pass indices as 'int' instead")]] Value *CreateShuffleVector (Value *V1, Value *V2, ArrayRef<uint32_t> Mask, const Twine &Name = "") { | |||
| 2476 | SmallVector<int, 16> IntMask; | |||
| 2477 | IntMask.assign(Mask.begin(), Mask.end()); | |||
| 2478 | return CreateShuffleVector(V1, V2, IntMask, Name); | |||
| 2479 | } | |||
| 2480 | ||||
| 2481 | /// See class ShuffleVectorInst for a description of the mask representation. | |||
| 2482 | Value *CreateShuffleVector(Value *V1, Value *V2, ArrayRef<int> Mask, | |||
| 2483 | const Twine &Name = "") { | |||
| 2484 | if (auto *V1C = dyn_cast<Constant>(V1)) | |||
| 2485 | if (auto *V2C = dyn_cast<Constant>(V2)) | |||
| 2486 | return Insert(Folder.CreateShuffleVector(V1C, V2C, Mask), Name); | |||
| 2487 | return Insert(new ShuffleVectorInst(V1, V2, Mask), Name); | |||
| 2488 | } | |||
| 2489 | ||||
| 2490 | /// Create a unary shuffle. The second vector operand of the IR instruction | |||
| 2491 | /// is poison. | |||
| 2492 | Value *CreateShuffleVector(Value *V, ArrayRef<int> Mask, | |||
| 2493 | const Twine &Name = "") { | |||
| 2494 | return CreateShuffleVector(V, PoisonValue::get(V->getType()), Mask, Name); | |||
| 2495 | } | |||
| 2496 | ||||
| 2497 | Value *CreateExtractValue(Value *Agg, | |||
| 2498 | ArrayRef<unsigned> Idxs, | |||
| 2499 | const Twine &Name = "") { | |||
| 2500 | if (auto *AggC = dyn_cast<Constant>(Agg)) | |||
| 2501 | return Insert(Folder.CreateExtractValue(AggC, Idxs), Name); | |||
| 2502 | return Insert(ExtractValueInst::Create(Agg, Idxs), Name); | |||
| 2503 | } | |||
| 2504 | ||||
| 2505 | Value *CreateInsertValue(Value *Agg, Value *Val, | |||
| 2506 | ArrayRef<unsigned> Idxs, | |||
| 2507 | const Twine &Name = "") { | |||
| 2508 | if (auto *AggC = dyn_cast<Constant>(Agg)) | |||
| 2509 | if (auto *ValC = dyn_cast<Constant>(Val)) | |||
| 2510 | return Insert(Folder.CreateInsertValue(AggC, ValC, Idxs), Name); | |||
| 2511 | return Insert(InsertValueInst::Create(Agg, Val, Idxs), Name); | |||
| 2512 | } | |||
| 2513 | ||||
| 2514 | LandingPadInst *CreateLandingPad(Type *Ty, unsigned NumClauses, | |||
| 2515 | const Twine &Name = "") { | |||
| 2516 | return Insert(LandingPadInst::Create(Ty, NumClauses), Name); | |||
| 2517 | } | |||
| 2518 | ||||
| 2519 | Value *CreateFreeze(Value *V, const Twine &Name = "") { | |||
| 2520 | return Insert(new FreezeInst(V), Name); | |||
| 2521 | } | |||
| 2522 | ||||
| 2523 | //===--------------------------------------------------------------------===// | |||
| 2524 | // Utility creation methods | |||
| 2525 | //===--------------------------------------------------------------------===// | |||
| 2526 | ||||
| 2527 | /// Return an i1 value testing if \p Arg is null. | |||
| 2528 | Value *CreateIsNull(Value *Arg, const Twine &Name = "") { | |||
| 2529 | return CreateICmpEQ(Arg, Constant::getNullValue(Arg->getType()), | |||
| 2530 | Name); | |||
| 2531 | } | |||
| 2532 | ||||
| 2533 | /// Return an i1 value testing if \p Arg is not null. | |||
| 2534 | Value *CreateIsNotNull(Value *Arg, const Twine &Name = "") { | |||
| 2535 | return CreateICmpNE(Arg, Constant::getNullValue(Arg->getType()), | |||
| 2536 | Name); | |||
| 2537 | } | |||
| 2538 | ||||
| 2539 | /// Return the i64 difference between two pointer values, dividing out | |||
| 2540 | /// the size of the pointed-to objects. | |||
| 2541 | /// | |||
| 2542 | /// This is intended to implement C-style pointer subtraction. As such, the | |||
| 2543 | /// pointers must be appropriately aligned for their element types and | |||
| 2544 | /// pointing into the same object. | |||
| 2545 | Value *CreatePtrDiff(Value *LHS, Value *RHS, const Twine &Name = ""); | |||
| 2546 | ||||
| 2547 | /// Create a launder.invariant.group intrinsic call. If Ptr type is | |||
| 2548 | /// different from pointer to i8, it's casted to pointer to i8 in the same | |||
| 2549 | /// address space before call and casted back to Ptr type after call. | |||
| 2550 | Value *CreateLaunderInvariantGroup(Value *Ptr); | |||
| 2551 | ||||
| 2552 | /// \brief Create a strip.invariant.group intrinsic call. If Ptr type is | |||
| 2553 | /// different from pointer to i8, it's casted to pointer to i8 in the same | |||
| 2554 | /// address space before call and casted back to Ptr type after call. | |||
| 2555 | Value *CreateStripInvariantGroup(Value *Ptr); | |||
| 2556 | ||||
| 2557 | /// Return a vector value that contains the vector V reversed | |||
| 2558 | Value *CreateVectorReverse(Value *V, const Twine &Name = ""); | |||
| 2559 | ||||
| 2560 | /// Return a vector splice intrinsic if using scalable vectors, otherwise | |||
| 2561 | /// return a shufflevector. If the immediate is positive, a vector is | |||
| 2562 | /// extracted from concat(V1, V2), starting at Imm. If the immediate | |||
| 2563 | /// is negative, we extract -Imm elements from V1 and the remaining | |||
| 2564 | /// elements from V2. Imm is a signed integer in the range | |||
| 2565 | /// -VL <= Imm < VL (where VL is the runtime vector length of the | |||
| 2566 | /// source/result vector) | |||
| 2567 | Value *CreateVectorSplice(Value *V1, Value *V2, int64_t Imm, | |||
| 2568 | const Twine &Name = ""); | |||
| 2569 | ||||
| 2570 | /// Return a vector value that contains \arg V broadcasted to \p | |||
| 2571 | /// NumElts elements. | |||
| 2572 | Value *CreateVectorSplat(unsigned NumElts, Value *V, const Twine &Name = ""); | |||
| 2573 | ||||
| 2574 | /// Return a vector value that contains \arg V broadcasted to \p | |||
| 2575 | /// EC elements. | |||
| 2576 | Value *CreateVectorSplat(ElementCount EC, Value *V, const Twine &Name = ""); | |||
| 2577 | ||||
| 2578 | /// Return a value that has been extracted from a larger integer type. | |||
| 2579 | Value *CreateExtractInteger(const DataLayout &DL, Value *From, | |||
| 2580 | IntegerType *ExtractedTy, uint64_t Offset, | |||
| 2581 | const Twine &Name); | |||
| 2582 | ||||
| 2583 | Value *CreatePreserveArrayAccessIndex(Type *ElTy, Value *Base, | |||
| 2584 | unsigned Dimension, unsigned LastIndex, | |||
| 2585 | MDNode *DbgInfo); | |||
| 2586 | ||||
| 2587 | Value *CreatePreserveUnionAccessIndex(Value *Base, unsigned FieldIndex, | |||
| 2588 | MDNode *DbgInfo); | |||
| 2589 | ||||
| 2590 | Value *CreatePreserveStructAccessIndex(Type *ElTy, Value *Base, | |||
| 2591 | unsigned Index, unsigned FieldIndex, | |||
| 2592 | MDNode *DbgInfo); | |||
| 2593 | ||||
| 2594 | private: | |||
| 2595 | /// Helper function that creates an assume intrinsic call that | |||
| 2596 | /// represents an alignment assumption on the provided pointer \p PtrValue | |||
| 2597 | /// with offset \p OffsetValue and alignment value \p AlignValue. | |||
| 2598 | CallInst *CreateAlignmentAssumptionHelper(const DataLayout &DL, | |||
| 2599 | Value *PtrValue, Value *AlignValue, | |||
| 2600 | Value *OffsetValue); | |||
| 2601 | ||||
| 2602 | public: | |||
| 2603 | /// Create an assume intrinsic call that represents an alignment | |||
| 2604 | /// assumption on the provided pointer. | |||
| 2605 | /// | |||
| 2606 | /// An optional offset can be provided, and if it is provided, the offset | |||
| 2607 | /// must be subtracted from the provided pointer to get the pointer with the | |||
| 2608 | /// specified alignment. | |||
| 2609 | CallInst *CreateAlignmentAssumption(const DataLayout &DL, Value *PtrValue, | |||
| 2610 | unsigned Alignment, | |||
| 2611 | Value *OffsetValue = nullptr); | |||
| 2612 | ||||
| 2613 | /// Create an assume intrinsic call that represents an alignment | |||
| 2614 | /// assumption on the provided pointer. | |||
| 2615 | /// | |||
| 2616 | /// An optional offset can be provided, and if it is provided, the offset | |||
| 2617 | /// must be subtracted from the provided pointer to get the pointer with the | |||
| 2618 | /// specified alignment. | |||
| 2619 | /// | |||
| 2620 | /// This overload handles the condition where the Alignment is dependent | |||
| 2621 | /// on an existing value rather than a static value. | |||
| 2622 | CallInst *CreateAlignmentAssumption(const DataLayout &DL, Value *PtrValue, | |||
| 2623 | Value *Alignment, | |||
| 2624 | Value *OffsetValue = nullptr); | |||
| 2625 | }; | |||
| 2626 | ||||
| 2627 | /// This provides a uniform API for creating instructions and inserting | |||
| 2628 | /// them into a basic block: either at the end of a BasicBlock, or at a specific | |||
| 2629 | /// iterator location in a block. | |||
| 2630 | /// | |||
| 2631 | /// Note that the builder does not expose the full generality of LLVM | |||
| 2632 | /// instructions. For access to extra instruction properties, use the mutators | |||
| 2633 | /// (e.g. setVolatile) on the instructions after they have been | |||
| 2634 | /// created. Convenience state exists to specify fast-math flags and fp-math | |||
| 2635 | /// tags. | |||
| 2636 | /// | |||
| 2637 | /// The first template argument specifies a class to use for creating constants. | |||
| 2638 | /// This defaults to creating minimally folded constants. The second template | |||
| 2639 | /// argument allows clients to specify custom insertion hooks that are called on | |||
| 2640 | /// every newly created insertion. | |||
| 2641 | template <typename FolderTy = ConstantFolder, | |||
| 2642 | typename InserterTy = IRBuilderDefaultInserter> | |||
| 2643 | class IRBuilder : public IRBuilderBase { | |||
| 2644 | private: | |||
| 2645 | FolderTy Folder; | |||
| 2646 | InserterTy Inserter; | |||
| 2647 | ||||
| 2648 | public: | |||
| 2649 | IRBuilder(LLVMContext &C, FolderTy Folder, InserterTy Inserter = InserterTy(), | |||
| 2650 | MDNode *FPMathTag = nullptr, | |||
| 2651 | ArrayRef<OperandBundleDef> OpBundles = None) | |||
| 2652 | : IRBuilderBase(C, this->Folder, this->Inserter, FPMathTag, OpBundles), | |||
| 2653 | Folder(Folder), Inserter(Inserter) {} | |||
| 2654 | ||||
| 2655 | explicit IRBuilder(LLVMContext &C, MDNode *FPMathTag = nullptr, | |||
| 2656 | ArrayRef<OperandBundleDef> OpBundles = None) | |||
| 2657 | : IRBuilderBase(C, this->Folder, this->Inserter, FPMathTag, OpBundles) {} | |||
| 2658 | ||||
| 2659 | explicit IRBuilder(BasicBlock *TheBB, FolderTy Folder, | |||
| 2660 | MDNode *FPMathTag = nullptr, | |||
| 2661 | ArrayRef<OperandBundleDef> OpBundles = None) | |||
| 2662 | : IRBuilderBase(TheBB->getContext(), this->Folder, this->Inserter, | |||
| 2663 | FPMathTag, OpBundles), Folder(Folder) { | |||
| 2664 | SetInsertPoint(TheBB); | |||
| 2665 | } | |||
| 2666 | ||||
| 2667 | explicit IRBuilder(BasicBlock *TheBB, MDNode *FPMathTag = nullptr, | |||
| 2668 | ArrayRef<OperandBundleDef> OpBundles = None) | |||
| 2669 | : IRBuilderBase(TheBB->getContext(), this->Folder, this->Inserter, | |||
| 2670 | FPMathTag, OpBundles) { | |||
| 2671 | SetInsertPoint(TheBB); | |||
| 2672 | } | |||
| 2673 | ||||
| 2674 | explicit IRBuilder(Instruction *IP, MDNode *FPMathTag = nullptr, | |||
| 2675 | ArrayRef<OperandBundleDef> OpBundles = None) | |||
| 2676 | : IRBuilderBase(IP->getContext(), this->Folder, this->Inserter, | |||
| 2677 | FPMathTag, OpBundles) { | |||
| 2678 | SetInsertPoint(IP); | |||
| 2679 | } | |||
| 2680 | ||||
| 2681 | IRBuilder(BasicBlock *TheBB, BasicBlock::iterator IP, FolderTy Folder, | |||
| 2682 | MDNode *FPMathTag = nullptr, | |||
| 2683 | ArrayRef<OperandBundleDef> OpBundles = None) | |||
| 2684 | : IRBuilderBase(TheBB->getContext(), this->Folder, this->Inserter, | |||
| 2685 | FPMathTag, OpBundles), Folder(Folder) { | |||
| 2686 | SetInsertPoint(TheBB, IP); | |||
| 2687 | } | |||
| 2688 | ||||
| 2689 | IRBuilder(BasicBlock *TheBB, BasicBlock::iterator IP, | |||
| 2690 | MDNode *FPMathTag = nullptr, | |||
| 2691 | ArrayRef<OperandBundleDef> OpBundles = None) | |||
| 2692 | : IRBuilderBase(TheBB->getContext(), this->Folder, this->Inserter, | |||
| 2693 | FPMathTag, OpBundles) { | |||
| 2694 | SetInsertPoint(TheBB, IP); | |||
| 2695 | } | |||
| 2696 | ||||
| 2697 | /// Avoid copying the full IRBuilder. Prefer using InsertPointGuard | |||
| 2698 | /// or FastMathFlagGuard instead. | |||
| 2699 | IRBuilder(const IRBuilder &) = delete; | |||
| 2700 | ||||
| 2701 | InserterTy &getInserter() { return Inserter; } | |||
| 2702 | }; | |||
| 2703 | ||||
| 2704 | // Create wrappers for C Binding types (see CBindingWrapping.h). | |||
| 2705 | DEFINE_SIMPLE_CONVERSION_FUNCTIONS(IRBuilder<>, LLVMBuilderRef)inline IRBuilder<> *unwrap(LLVMBuilderRef P) { return reinterpret_cast <IRBuilder<>*>(P); } inline LLVMBuilderRef wrap(const IRBuilder<> *P) { return reinterpret_cast<LLVMBuilderRef >(const_cast<IRBuilder<>*>(P)); } | |||
| 2706 | ||||
| 2707 | } // end namespace llvm | |||
| 2708 | ||||
| 2709 | #endif // LLVM_IR_IRBUILDER_H |