| 1 | //===-- SPIRVPrepareFunctions.cpp - modify function signatures --*- 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 pass modifies function signatures containing aggregate arguments |
| 10 | // and/or return value before IRTranslator. Information about the original |
| 11 | // signatures is stored in metadata. It is used during call lowering to |
| 12 | // restore correct SPIR-V types of function arguments and return values. |
| 13 | // This pass also substitutes some llvm intrinsic calls with calls to newly |
| 14 | // generated functions (as the Khronos LLVM/SPIR-V Translator does). |
| 15 | // |
| 16 | // NOTE: this pass is a module-level one due to the necessity to modify |
| 17 | // GVs/functions. |
| 18 | // |
| 19 | //===----------------------------------------------------------------------===// |
| 20 | |
| 21 | #include "SPIRV.h" |
| 22 | #include "SPIRVSubtarget.h" |
| 23 | #include "SPIRVTargetMachine.h" |
| 24 | #include "SPIRVUtils.h" |
| 25 | #include "llvm/ADT/StringExtras.h" |
| 26 | #include "llvm/Analysis/ValueTracking.h" |
| 27 | #include "llvm/CodeGen/IntrinsicLowering.h" |
| 28 | #include "llvm/IR/IRBuilder.h" |
| 29 | #include "llvm/IR/IntrinsicInst.h" |
| 30 | #include "llvm/IR/Intrinsics.h" |
| 31 | #include "llvm/IR/IntrinsicsSPIRV.h" |
| 32 | #include "llvm/Transforms/Utils/Cloning.h" |
| 33 | #include "llvm/Transforms/Utils/LowerMemIntrinsics.h" |
| 34 | #include <regex> |
| 35 | |
| 36 | using namespace llvm; |
| 37 | |
| 38 | namespace { |
| 39 | |
| 40 | class SPIRVPrepareFunctions : public ModulePass { |
| 41 | const SPIRVTargetMachine &TM; |
| 42 | bool substituteIntrinsicCalls(Function *F); |
| 43 | Function *removeAggregateTypesFromSignature(Function *F); |
| 44 | |
| 45 | public: |
| 46 | static char ID; |
| 47 | SPIRVPrepareFunctions(const SPIRVTargetMachine &TM) |
| 48 | : ModulePass(ID), TM(TM) {} |
| 49 | |
| 50 | bool runOnModule(Module &M) override; |
| 51 | |
| 52 | StringRef getPassName() const override { return "SPIRV prepare functions" ; } |
| 53 | |
| 54 | void getAnalysisUsage(AnalysisUsage &AU) const override { |
| 55 | ModulePass::getAnalysisUsage(AU); |
| 56 | } |
| 57 | }; |
| 58 | |
| 59 | } // namespace |
| 60 | |
| 61 | char SPIRVPrepareFunctions::ID = 0; |
| 62 | |
| 63 | INITIALIZE_PASS(SPIRVPrepareFunctions, "prepare-functions" , |
| 64 | "SPIRV prepare functions" , false, false) |
| 65 | |
| 66 | static std::string lowerLLVMIntrinsicName(IntrinsicInst *II) { |
| 67 | Function *IntrinsicFunc = II->getCalledFunction(); |
| 68 | assert(IntrinsicFunc && "Missing function" ); |
| 69 | std::string FuncName = IntrinsicFunc->getName().str(); |
| 70 | llvm::replace(Range&: FuncName, OldValue: '.', NewValue: '_'); |
| 71 | FuncName = "spirv." + FuncName; |
| 72 | return FuncName; |
| 73 | } |
| 74 | |
| 75 | static Function *getOrCreateFunction(Module *M, Type *RetTy, |
| 76 | ArrayRef<Type *> ArgTypes, |
| 77 | StringRef Name) { |
| 78 | FunctionType *FT = FunctionType::get(Result: RetTy, Params: ArgTypes, isVarArg: false); |
| 79 | Function *F = M->getFunction(Name); |
| 80 | if (F && F->getFunctionType() == FT) |
| 81 | return F; |
| 82 | Function *NewF = Function::Create(Ty: FT, Linkage: GlobalValue::ExternalLinkage, N: Name, M); |
| 83 | if (F) |
| 84 | NewF->setDSOLocal(F->isDSOLocal()); |
| 85 | NewF->setCallingConv(CallingConv::SPIR_FUNC); |
| 86 | return NewF; |
| 87 | } |
| 88 | |
| 89 | static bool lowerIntrinsicToFunction(IntrinsicInst *Intrinsic) { |
| 90 | // For @llvm.memset.* intrinsic cases with constant value and length arguments |
| 91 | // are emulated via "storing" a constant array to the destination. For other |
| 92 | // cases we wrap the intrinsic in @spirv.llvm_memset_* function and expand the |
| 93 | // intrinsic to a loop via expandMemSetAsLoop(). |
| 94 | if (auto *MSI = dyn_cast<MemSetInst>(Val: Intrinsic)) |
| 95 | if (isa<Constant>(Val: MSI->getValue()) && isa<ConstantInt>(Val: MSI->getLength())) |
| 96 | return false; // It is handled later using OpCopyMemorySized. |
| 97 | |
| 98 | Module *M = Intrinsic->getModule(); |
| 99 | std::string FuncName = lowerLLVMIntrinsicName(II: Intrinsic); |
| 100 | if (Intrinsic->isVolatile()) |
| 101 | FuncName += ".volatile" ; |
| 102 | // Redirect @llvm.intrinsic.* call to @spirv.llvm_intrinsic_* |
| 103 | Function *F = M->getFunction(Name: FuncName); |
| 104 | if (F) { |
| 105 | Intrinsic->setCalledFunction(F); |
| 106 | return true; |
| 107 | } |
| 108 | // TODO copy arguments attributes: nocapture writeonly. |
| 109 | FunctionCallee FC = |
| 110 | M->getOrInsertFunction(Name: FuncName, T: Intrinsic->getFunctionType()); |
| 111 | auto IntrinsicID = Intrinsic->getIntrinsicID(); |
| 112 | Intrinsic->setCalledFunction(FC); |
| 113 | |
| 114 | F = dyn_cast<Function>(Val: FC.getCallee()); |
| 115 | assert(F && "Callee must be a function" ); |
| 116 | |
| 117 | switch (IntrinsicID) { |
| 118 | case Intrinsic::memset: { |
| 119 | auto *MSI = static_cast<MemSetInst *>(Intrinsic); |
| 120 | Argument *Dest = F->getArg(i: 0); |
| 121 | Argument *Val = F->getArg(i: 1); |
| 122 | Argument *Len = F->getArg(i: 2); |
| 123 | Argument *IsVolatile = F->getArg(i: 3); |
| 124 | Dest->setName("dest" ); |
| 125 | Val->setName("val" ); |
| 126 | Len->setName("len" ); |
| 127 | IsVolatile->setName("isvolatile" ); |
| 128 | BasicBlock *EntryBB = BasicBlock::Create(Context&: M->getContext(), Name: "entry" , Parent: F); |
| 129 | IRBuilder<> IRB(EntryBB); |
| 130 | auto *MemSet = IRB.CreateMemSet(Ptr: Dest, Val, Size: Len, Align: MSI->getDestAlign(), |
| 131 | isVolatile: MSI->isVolatile()); |
| 132 | IRB.CreateRetVoid(); |
| 133 | expandMemSetAsLoop(MemSet: cast<MemSetInst>(Val: MemSet)); |
| 134 | MemSet->eraseFromParent(); |
| 135 | break; |
| 136 | } |
| 137 | case Intrinsic::bswap: { |
| 138 | BasicBlock *EntryBB = BasicBlock::Create(Context&: M->getContext(), Name: "entry" , Parent: F); |
| 139 | IRBuilder<> IRB(EntryBB); |
| 140 | auto *BSwap = IRB.CreateIntrinsic(ID: Intrinsic::bswap, Types: Intrinsic->getType(), |
| 141 | Args: F->getArg(i: 0)); |
| 142 | IRB.CreateRet(V: BSwap); |
| 143 | IntrinsicLowering IL(M->getDataLayout()); |
| 144 | IL.LowerIntrinsicCall(CI: BSwap); |
| 145 | break; |
| 146 | } |
| 147 | default: |
| 148 | break; |
| 149 | } |
| 150 | return true; |
| 151 | } |
| 152 | |
| 153 | static std::string getAnnotation(Value *AnnoVal, Value *OptAnnoVal) { |
| 154 | if (auto *Ref = dyn_cast_or_null<GetElementPtrInst>(Val: AnnoVal)) |
| 155 | AnnoVal = Ref->getOperand(i_nocapture: 0); |
| 156 | if (auto *Ref = dyn_cast_or_null<BitCastInst>(Val: OptAnnoVal)) |
| 157 | OptAnnoVal = Ref->getOperand(i_nocapture: 0); |
| 158 | |
| 159 | std::string Anno; |
| 160 | if (auto *C = dyn_cast_or_null<Constant>(Val: AnnoVal)) { |
| 161 | StringRef Str; |
| 162 | if (getConstantStringInfo(V: C, Str)) |
| 163 | Anno = Str; |
| 164 | } |
| 165 | // handle optional annotation parameter in a way that Khronos Translator do |
| 166 | // (collect integers wrapped in a struct) |
| 167 | if (auto *C = dyn_cast_or_null<Constant>(Val: OptAnnoVal); |
| 168 | C && C->getNumOperands()) { |
| 169 | Value *MaybeStruct = C->getOperand(i: 0); |
| 170 | if (auto *Struct = dyn_cast<ConstantStruct>(Val: MaybeStruct)) { |
| 171 | for (unsigned I = 0, E = Struct->getNumOperands(); I != E; ++I) { |
| 172 | if (auto *CInt = dyn_cast<ConstantInt>(Val: Struct->getOperand(i_nocapture: I))) |
| 173 | Anno += (I == 0 ? ": " : ", " ) + |
| 174 | std::to_string(val: CInt->getType()->getIntegerBitWidth() == 1 |
| 175 | ? CInt->getZExtValue() |
| 176 | : CInt->getSExtValue()); |
| 177 | } |
| 178 | } else if (auto *Struct = dyn_cast<ConstantAggregateZero>(Val: MaybeStruct)) { |
| 179 | // { i32 i32 ... } zeroinitializer |
| 180 | for (unsigned I = 0, E = Struct->getType()->getStructNumElements(); |
| 181 | I != E; ++I) |
| 182 | Anno += I == 0 ? ": 0" : ", 0" ; |
| 183 | } |
| 184 | } |
| 185 | return Anno; |
| 186 | } |
| 187 | |
| 188 | static SmallVector<Metadata *> parseAnnotation(Value *I, |
| 189 | const std::string &Anno, |
| 190 | LLVMContext &Ctx, |
| 191 | Type *Int32Ty) { |
| 192 | // Try to parse the annotation string according to the following rules: |
| 193 | // annotation := ({kind} | {kind:value,value,...})+ |
| 194 | // kind := number |
| 195 | // value := number | string |
| 196 | static const std::regex R( |
| 197 | "\\{(\\d+)(?:[:,](\\d+|\"[^\"]*\")(?:,(\\d+|\"[^\"]*\"))*)?\\}" ); |
| 198 | SmallVector<Metadata *> MDs; |
| 199 | int Pos = 0; |
| 200 | for (std::sregex_iterator |
| 201 | It = std::sregex_iterator(Anno.begin(), Anno.end(), R), |
| 202 | ItEnd = std::sregex_iterator(); |
| 203 | It != ItEnd; ++It) { |
| 204 | if (It->position() != Pos) |
| 205 | return SmallVector<Metadata *>{}; |
| 206 | Pos = It->position() + It->length(); |
| 207 | std::smatch Match = *It; |
| 208 | SmallVector<Metadata *> MDsItem; |
| 209 | for (std::size_t i = 1; i < Match.size(); ++i) { |
| 210 | std::ssub_match SMatch = Match[i]; |
| 211 | std::string Item = SMatch.str(); |
| 212 | if (Item.length() == 0) |
| 213 | break; |
| 214 | if (Item[0] == '"') { |
| 215 | Item = Item.substr(pos: 1, n: Item.length() - 2); |
| 216 | // Acceptable format of the string snippet is: |
| 217 | static const std::regex RStr("^(\\d+)(?:,(\\d+))*$" ); |
| 218 | if (std::smatch MatchStr; std::regex_match(s: Item, m&: MatchStr, re: RStr)) { |
| 219 | for (std::size_t SubIdx = 1; SubIdx < MatchStr.size(); ++SubIdx) |
| 220 | if (std::string SubStr = MatchStr[SubIdx].str(); SubStr.length()) |
| 221 | MDsItem.push_back(Elt: ConstantAsMetadata::get( |
| 222 | C: ConstantInt::get(Ty: Int32Ty, V: std::stoi(str: SubStr)))); |
| 223 | } else { |
| 224 | MDsItem.push_back(Elt: MDString::get(Context&: Ctx, Str: Item)); |
| 225 | } |
| 226 | } else if (int32_t Num; llvm::to_integer(S: StringRef(Item), Num, Base: 10)) { |
| 227 | MDsItem.push_back( |
| 228 | Elt: ConstantAsMetadata::get(C: ConstantInt::get(Ty: Int32Ty, V: Num))); |
| 229 | } else { |
| 230 | MDsItem.push_back(Elt: MDString::get(Context&: Ctx, Str: Item)); |
| 231 | } |
| 232 | } |
| 233 | if (MDsItem.size() == 0) |
| 234 | return SmallVector<Metadata *>{}; |
| 235 | MDs.push_back(Elt: MDNode::get(Context&: Ctx, MDs: MDsItem)); |
| 236 | } |
| 237 | return Pos == static_cast<int>(Anno.length()) ? MDs |
| 238 | : SmallVector<Metadata *>{}; |
| 239 | } |
| 240 | |
| 241 | static void lowerPtrAnnotation(IntrinsicInst *II) { |
| 242 | LLVMContext &Ctx = II->getContext(); |
| 243 | Type *Int32Ty = Type::getInt32Ty(C&: Ctx); |
| 244 | |
| 245 | // Retrieve an annotation string from arguments. |
| 246 | Value *PtrArg = nullptr; |
| 247 | if (auto *BI = dyn_cast<BitCastInst>(Val: II->getArgOperand(i: 0))) |
| 248 | PtrArg = BI->getOperand(i_nocapture: 0); |
| 249 | else |
| 250 | PtrArg = II->getOperand(i_nocapture: 0); |
| 251 | std::string Anno = |
| 252 | getAnnotation(AnnoVal: II->getArgOperand(i: 1), |
| 253 | OptAnnoVal: 4 < II->arg_size() ? II->getArgOperand(i: 4) : nullptr); |
| 254 | |
| 255 | // Parse the annotation. |
| 256 | SmallVector<Metadata *> MDs = parseAnnotation(I: II, Anno, Ctx, Int32Ty); |
| 257 | |
| 258 | // If the annotation string is not parsed successfully we don't know the |
| 259 | // format used and output it as a general UserSemantic decoration. |
| 260 | // Otherwise MDs is a Metadata tuple (a decoration list) in the format |
| 261 | // expected by `spirv.Decorations`. |
| 262 | if (MDs.size() == 0) { |
| 263 | auto UserSemantic = ConstantAsMetadata::get(C: ConstantInt::get( |
| 264 | Ty: Int32Ty, V: static_cast<uint32_t>(SPIRV::Decoration::UserSemantic))); |
| 265 | MDs.push_back(Elt: MDNode::get(Context&: Ctx, MDs: {UserSemantic, MDString::get(Context&: Ctx, Str: Anno)})); |
| 266 | } |
| 267 | |
| 268 | // Build the internal intrinsic function. |
| 269 | IRBuilder<> IRB(II->getParent()); |
| 270 | IRB.SetInsertPoint(II); |
| 271 | IRB.CreateIntrinsic( |
| 272 | ID: Intrinsic::spv_assign_decoration, Types: {PtrArg->getType()}, |
| 273 | Args: {PtrArg, MetadataAsValue::get(Context&: Ctx, MD: MDNode::get(Context&: Ctx, MDs))}); |
| 274 | II->replaceAllUsesWith(V: II->getOperand(i_nocapture: 0)); |
| 275 | } |
| 276 | |
| 277 | static void lowerFunnelShifts(IntrinsicInst *FSHIntrinsic) { |
| 278 | // Get a separate function - otherwise, we'd have to rework the CFG of the |
| 279 | // current one. Then simply replace the intrinsic uses with a call to the new |
| 280 | // function. |
| 281 | // Generate LLVM IR for i* @spirv.llvm_fsh?_i* (i* %a, i* %b, i* %c) |
| 282 | Module *M = FSHIntrinsic->getModule(); |
| 283 | FunctionType *FSHFuncTy = FSHIntrinsic->getFunctionType(); |
| 284 | Type *FSHRetTy = FSHFuncTy->getReturnType(); |
| 285 | const std::string FuncName = lowerLLVMIntrinsicName(II: FSHIntrinsic); |
| 286 | Function *FSHFunc = |
| 287 | getOrCreateFunction(M, RetTy: FSHRetTy, ArgTypes: FSHFuncTy->params(), Name: FuncName); |
| 288 | |
| 289 | if (!FSHFunc->empty()) { |
| 290 | FSHIntrinsic->setCalledFunction(FSHFunc); |
| 291 | return; |
| 292 | } |
| 293 | BasicBlock *RotateBB = BasicBlock::Create(Context&: M->getContext(), Name: "rotate" , Parent: FSHFunc); |
| 294 | IRBuilder<> IRB(RotateBB); |
| 295 | Type *Ty = FSHFunc->getReturnType(); |
| 296 | // Build the actual funnel shift rotate logic. |
| 297 | // In the comments, "int" is used interchangeably with "vector of int |
| 298 | // elements". |
| 299 | FixedVectorType *VectorTy = dyn_cast<FixedVectorType>(Val: Ty); |
| 300 | Type *IntTy = VectorTy ? VectorTy->getElementType() : Ty; |
| 301 | unsigned BitWidth = IntTy->getIntegerBitWidth(); |
| 302 | ConstantInt *BitWidthConstant = IRB.getInt(AI: {BitWidth, BitWidth}); |
| 303 | Value *BitWidthForInsts = |
| 304 | VectorTy |
| 305 | ? IRB.CreateVectorSplat(NumElts: VectorTy->getNumElements(), V: BitWidthConstant) |
| 306 | : BitWidthConstant; |
| 307 | Value *RotateModVal = |
| 308 | IRB.CreateURem(/*Rotate*/ LHS: FSHFunc->getArg(i: 2), RHS: BitWidthForInsts); |
| 309 | Value *FirstShift = nullptr, *SecShift = nullptr; |
| 310 | if (FSHIntrinsic->getIntrinsicID() == Intrinsic::fshr) { |
| 311 | // Shift the less significant number right, the "rotate" number of bits |
| 312 | // will be 0-filled on the left as a result of this regular shift. |
| 313 | FirstShift = IRB.CreateLShr(LHS: FSHFunc->getArg(i: 1), RHS: RotateModVal); |
| 314 | } else { |
| 315 | // Shift the more significant number left, the "rotate" number of bits |
| 316 | // will be 0-filled on the right as a result of this regular shift. |
| 317 | FirstShift = IRB.CreateShl(LHS: FSHFunc->getArg(i: 0), RHS: RotateModVal); |
| 318 | } |
| 319 | // We want the "rotate" number of the more significant int's LSBs (MSBs) to |
| 320 | // occupy the leftmost (rightmost) "0 space" left by the previous operation. |
| 321 | // Therefore, subtract the "rotate" number from the integer bitsize... |
| 322 | Value *SubRotateVal = IRB.CreateSub(LHS: BitWidthForInsts, RHS: RotateModVal); |
| 323 | if (FSHIntrinsic->getIntrinsicID() == Intrinsic::fshr) { |
| 324 | // ...and left-shift the more significant int by this number, zero-filling |
| 325 | // the LSBs. |
| 326 | SecShift = IRB.CreateShl(LHS: FSHFunc->getArg(i: 0), RHS: SubRotateVal); |
| 327 | } else { |
| 328 | // ...and right-shift the less significant int by this number, zero-filling |
| 329 | // the MSBs. |
| 330 | SecShift = IRB.CreateLShr(LHS: FSHFunc->getArg(i: 1), RHS: SubRotateVal); |
| 331 | } |
| 332 | // A simple binary addition of the shifted ints yields the final result. |
| 333 | IRB.CreateRet(V: IRB.CreateOr(LHS: FirstShift, RHS: SecShift)); |
| 334 | |
| 335 | FSHIntrinsic->setCalledFunction(FSHFunc); |
| 336 | } |
| 337 | |
| 338 | static void lowerExpectAssume(IntrinsicInst *II) { |
| 339 | // If we cannot use the SPV_KHR_expect_assume extension, then we need to |
| 340 | // ignore the intrinsic and move on. It should be removed later on by LLVM. |
| 341 | // Otherwise we should lower the intrinsic to the corresponding SPIR-V |
| 342 | // instruction. |
| 343 | // For @llvm.assume we have OpAssumeTrueKHR. |
| 344 | // For @llvm.expect we have OpExpectKHR. |
| 345 | // |
| 346 | // We need to lower this into a builtin and then the builtin into a SPIR-V |
| 347 | // instruction. |
| 348 | if (II->getIntrinsicID() == Intrinsic::assume) { |
| 349 | Function *F = Intrinsic::getOrInsertDeclaration( |
| 350 | M: II->getModule(), id: Intrinsic::SPVIntrinsics::spv_assume); |
| 351 | II->setCalledFunction(F); |
| 352 | } else if (II->getIntrinsicID() == Intrinsic::expect) { |
| 353 | Function *F = Intrinsic::getOrInsertDeclaration( |
| 354 | M: II->getModule(), id: Intrinsic::SPVIntrinsics::spv_expect, |
| 355 | Tys: {II->getOperand(i_nocapture: 0)->getType()}); |
| 356 | II->setCalledFunction(F); |
| 357 | } else { |
| 358 | llvm_unreachable("Unknown intrinsic" ); |
| 359 | } |
| 360 | } |
| 361 | |
| 362 | static bool toSpvOverloadedIntrinsic(IntrinsicInst *II, Intrinsic::ID NewID, |
| 363 | ArrayRef<unsigned> OpNos) { |
| 364 | Function *F = nullptr; |
| 365 | if (OpNos.empty()) { |
| 366 | F = Intrinsic::getOrInsertDeclaration(M: II->getModule(), id: NewID); |
| 367 | } else { |
| 368 | SmallVector<Type *, 4> Tys; |
| 369 | for (unsigned OpNo : OpNos) |
| 370 | Tys.push_back(Elt: II->getOperand(i_nocapture: OpNo)->getType()); |
| 371 | F = Intrinsic::getOrInsertDeclaration(M: II->getModule(), id: NewID, Tys); |
| 372 | } |
| 373 | II->setCalledFunction(F); |
| 374 | return true; |
| 375 | } |
| 376 | |
| 377 | // Substitutes calls to LLVM intrinsics with either calls to SPIR-V intrinsics |
| 378 | // or calls to proper generated functions. Returns True if F was modified. |
| 379 | bool SPIRVPrepareFunctions::substituteIntrinsicCalls(Function *F) { |
| 380 | bool Changed = false; |
| 381 | const SPIRVSubtarget &STI = TM.getSubtarget<SPIRVSubtarget>(F: *F); |
| 382 | for (BasicBlock &BB : *F) { |
| 383 | for (Instruction &I : BB) { |
| 384 | auto Call = dyn_cast<CallInst>(Val: &I); |
| 385 | if (!Call) |
| 386 | continue; |
| 387 | Function *CF = Call->getCalledFunction(); |
| 388 | if (!CF || !CF->isIntrinsic()) |
| 389 | continue; |
| 390 | auto *II = cast<IntrinsicInst>(Val: Call); |
| 391 | switch (II->getIntrinsicID()) { |
| 392 | case Intrinsic::memset: |
| 393 | case Intrinsic::bswap: |
| 394 | Changed |= lowerIntrinsicToFunction(Intrinsic: II); |
| 395 | break; |
| 396 | case Intrinsic::fshl: |
| 397 | case Intrinsic::fshr: |
| 398 | lowerFunnelShifts(FSHIntrinsic: II); |
| 399 | Changed = true; |
| 400 | break; |
| 401 | case Intrinsic::assume: |
| 402 | case Intrinsic::expect: |
| 403 | if (STI.canUseExtension(E: SPIRV::Extension::SPV_KHR_expect_assume)) |
| 404 | lowerExpectAssume(II); |
| 405 | Changed = true; |
| 406 | break; |
| 407 | case Intrinsic::lifetime_start: |
| 408 | if (!STI.isShader()) { |
| 409 | Changed |= toSpvOverloadedIntrinsic( |
| 410 | II, NewID: Intrinsic::SPVIntrinsics::spv_lifetime_start, OpNos: {1}); |
| 411 | } |
| 412 | break; |
| 413 | case Intrinsic::lifetime_end: |
| 414 | if (!STI.isShader()) { |
| 415 | Changed |= toSpvOverloadedIntrinsic( |
| 416 | II, NewID: Intrinsic::SPVIntrinsics::spv_lifetime_end, OpNos: {1}); |
| 417 | } |
| 418 | break; |
| 419 | case Intrinsic::ptr_annotation: |
| 420 | lowerPtrAnnotation(II); |
| 421 | Changed = true; |
| 422 | break; |
| 423 | } |
| 424 | } |
| 425 | } |
| 426 | return Changed; |
| 427 | } |
| 428 | |
| 429 | // Returns F if aggregate argument/return types are not present or cloned F |
| 430 | // function with the types replaced by i32 types. The change in types is |
| 431 | // noted in 'spv.cloned_funcs' metadata for later restoration. |
| 432 | Function * |
| 433 | SPIRVPrepareFunctions::removeAggregateTypesFromSignature(Function *F) { |
| 434 | bool IsRetAggr = F->getReturnType()->isAggregateType(); |
| 435 | // Allow intrinsics with aggregate return type to reach GlobalISel |
| 436 | if (F->isIntrinsic() && IsRetAggr) |
| 437 | return F; |
| 438 | |
| 439 | IRBuilder<> B(F->getContext()); |
| 440 | |
| 441 | bool HasAggrArg = llvm::any_of(Range: F->args(), P: [](Argument &Arg) { |
| 442 | return Arg.getType()->isAggregateType(); |
| 443 | }); |
| 444 | bool DoClone = IsRetAggr || HasAggrArg; |
| 445 | if (!DoClone) |
| 446 | return F; |
| 447 | SmallVector<std::pair<int, Type *>, 4> ChangedTypes; |
| 448 | Type *RetType = IsRetAggr ? B.getInt32Ty() : F->getReturnType(); |
| 449 | if (IsRetAggr) |
| 450 | ChangedTypes.push_back(Elt: std::pair<int, Type *>(-1, F->getReturnType())); |
| 451 | SmallVector<Type *, 4> ArgTypes; |
| 452 | for (const auto &Arg : F->args()) { |
| 453 | if (Arg.getType()->isAggregateType()) { |
| 454 | ArgTypes.push_back(Elt: B.getInt32Ty()); |
| 455 | ChangedTypes.push_back( |
| 456 | Elt: std::pair<int, Type *>(Arg.getArgNo(), Arg.getType())); |
| 457 | } else |
| 458 | ArgTypes.push_back(Elt: Arg.getType()); |
| 459 | } |
| 460 | FunctionType *NewFTy = |
| 461 | FunctionType::get(Result: RetType, Params: ArgTypes, isVarArg: F->getFunctionType()->isVarArg()); |
| 462 | Function *NewF = |
| 463 | Function::Create(Ty: NewFTy, Linkage: F->getLinkage(), N: F->getName(), M&: *F->getParent()); |
| 464 | |
| 465 | ValueToValueMapTy VMap; |
| 466 | auto NewFArgIt = NewF->arg_begin(); |
| 467 | for (auto &Arg : F->args()) { |
| 468 | StringRef ArgName = Arg.getName(); |
| 469 | NewFArgIt->setName(ArgName); |
| 470 | VMap[&Arg] = &(*NewFArgIt++); |
| 471 | } |
| 472 | SmallVector<ReturnInst *, 8> Returns; |
| 473 | |
| 474 | CloneFunctionInto(NewFunc: NewF, OldFunc: F, VMap, Changes: CloneFunctionChangeType::LocalChangesOnly, |
| 475 | Returns); |
| 476 | NewF->takeName(V: F); |
| 477 | |
| 478 | NamedMDNode *FuncMD = |
| 479 | F->getParent()->getOrInsertNamedMetadata(Name: "spv.cloned_funcs" ); |
| 480 | SmallVector<Metadata *, 2> MDArgs; |
| 481 | MDArgs.push_back(Elt: MDString::get(Context&: B.getContext(), Str: NewF->getName())); |
| 482 | for (auto &ChangedTyP : ChangedTypes) |
| 483 | MDArgs.push_back(Elt: MDNode::get( |
| 484 | Context&: B.getContext(), |
| 485 | MDs: {ConstantAsMetadata::get(C: B.getInt32(C: ChangedTyP.first)), |
| 486 | ValueAsMetadata::get(V: Constant::getNullValue(Ty: ChangedTyP.second))})); |
| 487 | MDNode *ThisFuncMD = MDNode::get(Context&: B.getContext(), MDs: MDArgs); |
| 488 | FuncMD->addOperand(M: ThisFuncMD); |
| 489 | |
| 490 | for (auto *U : make_early_inc_range(Range: F->users())) { |
| 491 | if (auto *CI = dyn_cast<CallInst>(Val: U)) |
| 492 | CI->mutateFunctionType(FTy: NewF->getFunctionType()); |
| 493 | U->replaceUsesOfWith(From: F, To: NewF); |
| 494 | } |
| 495 | |
| 496 | // register the mutation |
| 497 | if (RetType != F->getReturnType()) |
| 498 | TM.getSubtarget<SPIRVSubtarget>(F: *F).getSPIRVGlobalRegistry()->addMutated( |
| 499 | Val: NewF, Ty: F->getReturnType()); |
| 500 | return NewF; |
| 501 | } |
| 502 | |
| 503 | bool SPIRVPrepareFunctions::runOnModule(Module &M) { |
| 504 | bool Changed = false; |
| 505 | for (Function &F : M) { |
| 506 | Changed |= substituteIntrinsicCalls(F: &F); |
| 507 | Changed |= sortBlocks(F); |
| 508 | } |
| 509 | |
| 510 | std::vector<Function *> FuncsWorklist; |
| 511 | for (auto &F : M) |
| 512 | FuncsWorklist.push_back(x: &F); |
| 513 | |
| 514 | for (auto *F : FuncsWorklist) { |
| 515 | Function *NewF = removeAggregateTypesFromSignature(F); |
| 516 | |
| 517 | if (NewF != F) { |
| 518 | F->eraseFromParent(); |
| 519 | Changed = true; |
| 520 | } |
| 521 | } |
| 522 | return Changed; |
| 523 | } |
| 524 | |
| 525 | ModulePass * |
| 526 | llvm::createSPIRVPrepareFunctionsPass(const SPIRVTargetMachine &TM) { |
| 527 | return new SPIRVPrepareFunctions(TM); |
| 528 | } |
| 529 | |