clang  9.0.0
CGCUDANV.cpp
Go to the documentation of this file.
1 //===----- CGCUDANV.cpp - Interface to NVIDIA CUDA Runtime ----------------===//
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 provides a class for CUDA code generation targeting the NVIDIA CUDA
10 // runtime library.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "CGCUDARuntime.h"
15 #include "CodeGenFunction.h"
16 #include "CodeGenModule.h"
17 #include "clang/AST/Decl.h"
18 #include "clang/Basic/Cuda.h"
21 #include "llvm/IR/BasicBlock.h"
22 #include "llvm/IR/Constants.h"
23 #include "llvm/IR/DerivedTypes.h"
24 #include "llvm/Support/Format.h"
25 
26 using namespace clang;
27 using namespace CodeGen;
28 
29 namespace {
30 constexpr unsigned CudaFatMagic = 0x466243b1;
31 constexpr unsigned HIPFatMagic = 0x48495046; // "HIPF"
32 
33 class CGNVCUDARuntime : public CGCUDARuntime {
34 
35 private:
36  llvm::IntegerType *IntTy, *SizeTy;
37  llvm::Type *VoidTy;
38  llvm::PointerType *CharPtrTy, *VoidPtrTy, *VoidPtrPtrTy;
39 
40  /// Convenience reference to LLVM Context
41  llvm::LLVMContext &Context;
42  /// Convenience reference to the current module
43  llvm::Module &TheModule;
44  /// Keeps track of kernel launch stubs emitted in this module
45  struct KernelInfo {
46  llvm::Function *Kernel;
47  const Decl *D;
48  };
49  llvm::SmallVector<KernelInfo, 16> EmittedKernels;
50  struct VarInfo {
51  llvm::GlobalVariable *Var;
52  const VarDecl *D;
53  unsigned Flag;
54  };
56  /// Keeps track of variable containing handle of GPU binary. Populated by
57  /// ModuleCtorFunction() and used to create corresponding cleanup calls in
58  /// ModuleDtorFunction()
59  llvm::GlobalVariable *GpuBinaryHandle = nullptr;
60  /// Whether we generate relocatable device code.
61  bool RelocatableDeviceCode;
62  /// Mangle context for device.
63  std::unique_ptr<MangleContext> DeviceMC;
64 
65  llvm::FunctionCallee getSetupArgumentFn() const;
66  llvm::FunctionCallee getLaunchFn() const;
67 
68  llvm::FunctionType *getRegisterGlobalsFnTy() const;
69  llvm::FunctionType *getCallbackFnTy() const;
70  llvm::FunctionType *getRegisterLinkedBinaryFnTy() const;
71  std::string addPrefixToName(StringRef FuncName) const;
72  std::string addUnderscoredPrefixToName(StringRef FuncName) const;
73 
74  /// Creates a function to register all kernel stubs generated in this module.
75  llvm::Function *makeRegisterGlobalsFn();
76 
77  /// Helper function that generates a constant string and returns a pointer to
78  /// the start of the string. The result of this function can be used anywhere
79  /// where the C code specifies const char*.
80  llvm::Constant *makeConstantString(const std::string &Str,
81  const std::string &Name = "",
82  const std::string &SectionName = "",
83  unsigned Alignment = 0) {
84  llvm::Constant *Zeros[] = {llvm::ConstantInt::get(SizeTy, 0),
85  llvm::ConstantInt::get(SizeTy, 0)};
86  auto ConstStr = CGM.GetAddrOfConstantCString(Str, Name.c_str());
87  llvm::GlobalVariable *GV =
88  cast<llvm::GlobalVariable>(ConstStr.getPointer());
89  if (!SectionName.empty()) {
90  GV->setSection(SectionName);
91  // Mark the address as used which make sure that this section isn't
92  // merged and we will really have it in the object file.
93  GV->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::None);
94  }
95  if (Alignment)
96  GV->setAlignment(Alignment);
97 
98  return llvm::ConstantExpr::getGetElementPtr(ConstStr.getElementType(),
99  ConstStr.getPointer(), Zeros);
100  }
101 
102  /// Helper function that generates an empty dummy function returning void.
103  llvm::Function *makeDummyFunction(llvm::FunctionType *FnTy) {
104  assert(FnTy->getReturnType()->isVoidTy() &&
105  "Can only generate dummy functions returning void!");
106  llvm::Function *DummyFunc = llvm::Function::Create(
107  FnTy, llvm::GlobalValue::InternalLinkage, "dummy", &TheModule);
108 
109  llvm::BasicBlock *DummyBlock =
110  llvm::BasicBlock::Create(Context, "", DummyFunc);
111  CGBuilderTy FuncBuilder(CGM, Context);
112  FuncBuilder.SetInsertPoint(DummyBlock);
113  FuncBuilder.CreateRetVoid();
114 
115  return DummyFunc;
116  }
117 
118  void emitDeviceStubBodyLegacy(CodeGenFunction &CGF, FunctionArgList &Args);
119  void emitDeviceStubBodyNew(CodeGenFunction &CGF, FunctionArgList &Args);
120  std::string getDeviceSideName(const Decl *ND);
121 
122 public:
123  CGNVCUDARuntime(CodeGenModule &CGM);
124 
125  void emitDeviceStub(CodeGenFunction &CGF, FunctionArgList &Args) override;
126  void registerDeviceVar(const VarDecl *VD, llvm::GlobalVariable &Var,
127  unsigned Flags) override {
128  DeviceVars.push_back({&Var, VD, Flags});
129  }
130 
131  /// Creates module constructor function
132  llvm::Function *makeModuleCtorFunction() override;
133  /// Creates module destructor function
134  llvm::Function *makeModuleDtorFunction() override;
135  /// Construct and return the stub name of a kernel.
136  std::string getDeviceStubName(llvm::StringRef Name) const override;
137 };
138 
139 }
140 
141 std::string CGNVCUDARuntime::addPrefixToName(StringRef FuncName) const {
142  if (CGM.getLangOpts().HIP)
143  return ((Twine("hip") + Twine(FuncName)).str());
144  return ((Twine("cuda") + Twine(FuncName)).str());
145 }
146 std::string
147 CGNVCUDARuntime::addUnderscoredPrefixToName(StringRef FuncName) const {
148  if (CGM.getLangOpts().HIP)
149  return ((Twine("__hip") + Twine(FuncName)).str());
150  return ((Twine("__cuda") + Twine(FuncName)).str());
151 }
152 
153 CGNVCUDARuntime::CGNVCUDARuntime(CodeGenModule &CGM)
154  : CGCUDARuntime(CGM), Context(CGM.getLLVMContext()),
155  TheModule(CGM.getModule()),
156  RelocatableDeviceCode(CGM.getLangOpts().GPURelocatableDeviceCode),
157  DeviceMC(CGM.getContext().createMangleContext(
158  CGM.getContext().getAuxTargetInfo())) {
159  CodeGen::CodeGenTypes &Types = CGM.getTypes();
160  ASTContext &Ctx = CGM.getContext();
161 
162  IntTy = CGM.IntTy;
163  SizeTy = CGM.SizeTy;
164  VoidTy = CGM.VoidTy;
165 
166  CharPtrTy = llvm::PointerType::getUnqual(Types.ConvertType(Ctx.CharTy));
167  VoidPtrTy = cast<llvm::PointerType>(Types.ConvertType(Ctx.VoidPtrTy));
168  VoidPtrPtrTy = VoidPtrTy->getPointerTo();
169 }
170 
171 llvm::FunctionCallee CGNVCUDARuntime::getSetupArgumentFn() const {
172  // cudaError_t cudaSetupArgument(void *, size_t, size_t)
173  llvm::Type *Params[] = {VoidPtrTy, SizeTy, SizeTy};
174  return CGM.CreateRuntimeFunction(
175  llvm::FunctionType::get(IntTy, Params, false),
176  addPrefixToName("SetupArgument"));
177 }
178 
179 llvm::FunctionCallee CGNVCUDARuntime::getLaunchFn() const {
180  if (CGM.getLangOpts().HIP) {
181  // hipError_t hipLaunchByPtr(char *);
182  return CGM.CreateRuntimeFunction(
183  llvm::FunctionType::get(IntTy, CharPtrTy, false), "hipLaunchByPtr");
184  } else {
185  // cudaError_t cudaLaunch(char *);
186  return CGM.CreateRuntimeFunction(
187  llvm::FunctionType::get(IntTy, CharPtrTy, false), "cudaLaunch");
188  }
189 }
190 
191 llvm::FunctionType *CGNVCUDARuntime::getRegisterGlobalsFnTy() const {
192  return llvm::FunctionType::get(VoidTy, VoidPtrPtrTy, false);
193 }
194 
195 llvm::FunctionType *CGNVCUDARuntime::getCallbackFnTy() const {
196  return llvm::FunctionType::get(VoidTy, VoidPtrTy, false);
197 }
198 
199 llvm::FunctionType *CGNVCUDARuntime::getRegisterLinkedBinaryFnTy() const {
200  auto CallbackFnTy = getCallbackFnTy();
201  auto RegisterGlobalsFnTy = getRegisterGlobalsFnTy();
202  llvm::Type *Params[] = {RegisterGlobalsFnTy->getPointerTo(), VoidPtrTy,
203  VoidPtrTy, CallbackFnTy->getPointerTo()};
204  return llvm::FunctionType::get(VoidTy, Params, false);
205 }
206 
207 std::string CGNVCUDARuntime::getDeviceSideName(const Decl *D) {
208  auto *ND = cast<const NamedDecl>(D);
209  std::string DeviceSideName;
210  if (DeviceMC->shouldMangleDeclName(ND)) {
211  SmallString<256> Buffer;
212  llvm::raw_svector_ostream Out(Buffer);
213  DeviceMC->mangleName(ND, Out);
214  DeviceSideName = Out.str();
215  } else
216  DeviceSideName = ND->getIdentifier()->getName();
217  return DeviceSideName;
218 }
219 
220 void CGNVCUDARuntime::emitDeviceStub(CodeGenFunction &CGF,
221  FunctionArgList &Args) {
222  // Ensure either we have different ABIs between host and device compilations,
223  // says host compilation following MSVC ABI but device compilation follows
224  // Itanium C++ ABI or, if they follow the same ABI, kernel names after
225  // mangling should be the same after name stubbing. The later checking is
226  // very important as the device kernel name being mangled in host-compilation
227  // is used to resolve the device binaries to be executed. Inconsistent naming
228  // result in undefined behavior. Even though we cannot check that naming
229  // directly between host- and device-compilations, the host- and
230  // device-mangling in host compilation could help catching certain ones.
231  assert((CGF.CGM.getContext().getAuxTargetInfo() &&
233  CGF.CGM.getContext().getTargetInfo().getCXXABI())) ||
234  getDeviceStubName(getDeviceSideName(CGF.CurFuncDecl)) ==
235  CGF.CurFn->getName());
236 
237  EmittedKernels.push_back({CGF.CurFn, CGF.CurFuncDecl});
240  emitDeviceStubBodyNew(CGF, Args);
241  else
242  emitDeviceStubBodyLegacy(CGF, Args);
243 }
244 
245 // CUDA 9.0+ uses new way to launch kernels. Parameters are packed in a local
246 // array and kernels are launched using cudaLaunchKernel().
247 void CGNVCUDARuntime::emitDeviceStubBodyNew(CodeGenFunction &CGF,
248  FunctionArgList &Args) {
249  // Build the shadow stack entry at the very start of the function.
250 
251  // Calculate amount of space we will need for all arguments. If we have no
252  // args, allocate a single pointer so we still have a valid pointer to the
253  // argument array that we can pass to runtime, even if it will be unused.
254  Address KernelArgs = CGF.CreateTempAlloca(
255  VoidPtrTy, CharUnits::fromQuantity(16), "kernel_args",
256  llvm::ConstantInt::get(SizeTy, std::max<size_t>(1, Args.size())));
257  // Store pointers to the arguments in a locally allocated launch_args.
258  for (unsigned i = 0; i < Args.size(); ++i) {
259  llvm::Value* VarPtr = CGF.GetAddrOfLocalVar(Args[i]).getPointer();
260  llvm::Value *VoidVarPtr = CGF.Builder.CreatePointerCast(VarPtr, VoidPtrTy);
262  VoidVarPtr, CGF.Builder.CreateConstGEP1_32(KernelArgs.getPointer(), i));
263  }
264 
265  llvm::BasicBlock *EndBlock = CGF.createBasicBlock("setup.end");
266 
267  // Lookup cudaLaunchKernel function.
268  // cudaError_t cudaLaunchKernel(const void *func, dim3 gridDim, dim3 blockDim,
269  // void **args, size_t sharedMem,
270  // cudaStream_t stream);
273  IdentifierInfo &cudaLaunchKernelII =
274  CGM.getContext().Idents.get("cudaLaunchKernel");
275  FunctionDecl *cudaLaunchKernelFD = nullptr;
276  for (const auto &Result : DC->lookup(&cudaLaunchKernelII)) {
277  if (FunctionDecl *FD = dyn_cast<FunctionDecl>(Result))
278  cudaLaunchKernelFD = FD;
279  }
280 
281  if (cudaLaunchKernelFD == nullptr) {
282  CGM.Error(CGF.CurFuncDecl->getLocation(),
283  "Can't find declaration for cudaLaunchKernel()");
284  return;
285  }
286  // Create temporary dim3 grid_dim, block_dim.
287  ParmVarDecl *GridDimParam = cudaLaunchKernelFD->getParamDecl(1);
288  QualType Dim3Ty = GridDimParam->getType();
289  Address GridDim =
290  CGF.CreateMemTemp(Dim3Ty, CharUnits::fromQuantity(8), "grid_dim");
291  Address BlockDim =
292  CGF.CreateMemTemp(Dim3Ty, CharUnits::fromQuantity(8), "block_dim");
293  Address ShmemSize =
294  CGF.CreateTempAlloca(SizeTy, CGM.getSizeAlign(), "shmem_size");
295  Address Stream =
296  CGF.CreateTempAlloca(VoidPtrTy, CGM.getPointerAlign(), "stream");
297  llvm::FunctionCallee cudaPopConfigFn = CGM.CreateRuntimeFunction(
298  llvm::FunctionType::get(IntTy,
299  {/*gridDim=*/GridDim.getType(),
300  /*blockDim=*/BlockDim.getType(),
301  /*ShmemSize=*/ShmemSize.getType(),
302  /*Stream=*/Stream.getType()},
303  /*isVarArg=*/false),
304  "__cudaPopCallConfiguration");
305 
306  CGF.EmitRuntimeCallOrInvoke(cudaPopConfigFn,
307  {GridDim.getPointer(), BlockDim.getPointer(),
308  ShmemSize.getPointer(), Stream.getPointer()});
309 
310  // Emit the call to cudaLaunch
311  llvm::Value *Kernel = CGF.Builder.CreatePointerCast(CGF.CurFn, VoidPtrTy);
312  CallArgList LaunchKernelArgs;
313  LaunchKernelArgs.add(RValue::get(Kernel),
314  cudaLaunchKernelFD->getParamDecl(0)->getType());
315  LaunchKernelArgs.add(RValue::getAggregate(GridDim), Dim3Ty);
316  LaunchKernelArgs.add(RValue::getAggregate(BlockDim), Dim3Ty);
317  LaunchKernelArgs.add(RValue::get(KernelArgs.getPointer()),
318  cudaLaunchKernelFD->getParamDecl(3)->getType());
319  LaunchKernelArgs.add(RValue::get(CGF.Builder.CreateLoad(ShmemSize)),
320  cudaLaunchKernelFD->getParamDecl(4)->getType());
321  LaunchKernelArgs.add(RValue::get(CGF.Builder.CreateLoad(Stream)),
322  cudaLaunchKernelFD->getParamDecl(5)->getType());
323 
324  QualType QT = cudaLaunchKernelFD->getType();
325  QualType CQT = QT.getCanonicalType();
326  llvm::Type *Ty = CGM.getTypes().ConvertType(CQT);
327  llvm::FunctionType *FTy = dyn_cast<llvm::FunctionType>(Ty);
328 
329  const CGFunctionInfo &FI =
330  CGM.getTypes().arrangeFunctionDeclaration(cudaLaunchKernelFD);
331  llvm::FunctionCallee cudaLaunchKernelFn =
332  CGM.CreateRuntimeFunction(FTy, "cudaLaunchKernel");
333  CGF.EmitCall(FI, CGCallee::forDirect(cudaLaunchKernelFn), ReturnValueSlot(),
334  LaunchKernelArgs);
335  CGF.EmitBranch(EndBlock);
336 
337  CGF.EmitBlock(EndBlock);
338 }
339 
340 void CGNVCUDARuntime::emitDeviceStubBodyLegacy(CodeGenFunction &CGF,
341  FunctionArgList &Args) {
342  // Emit a call to cudaSetupArgument for each arg in Args.
343  llvm::FunctionCallee cudaSetupArgFn = getSetupArgumentFn();
344  llvm::BasicBlock *EndBlock = CGF.createBasicBlock("setup.end");
346  for (const VarDecl *A : Args) {
347  CharUnits TyWidth, TyAlign;
348  std::tie(TyWidth, TyAlign) =
349  CGM.getContext().getTypeInfoInChars(A->getType());
350  Offset = Offset.alignTo(TyAlign);
351  llvm::Value *Args[] = {
352  CGF.Builder.CreatePointerCast(CGF.GetAddrOfLocalVar(A).getPointer(),
353  VoidPtrTy),
354  llvm::ConstantInt::get(SizeTy, TyWidth.getQuantity()),
355  llvm::ConstantInt::get(SizeTy, Offset.getQuantity()),
356  };
357  llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(cudaSetupArgFn, Args);
358  llvm::Constant *Zero = llvm::ConstantInt::get(IntTy, 0);
359  llvm::Value *CBZero = CGF.Builder.CreateICmpEQ(CB, Zero);
360  llvm::BasicBlock *NextBlock = CGF.createBasicBlock("setup.next");
361  CGF.Builder.CreateCondBr(CBZero, NextBlock, EndBlock);
362  CGF.EmitBlock(NextBlock);
363  Offset += TyWidth;
364  }
365 
366  // Emit the call to cudaLaunch
367  llvm::FunctionCallee cudaLaunchFn = getLaunchFn();
368  llvm::Value *Arg = CGF.Builder.CreatePointerCast(CGF.CurFn, CharPtrTy);
369  CGF.EmitRuntimeCallOrInvoke(cudaLaunchFn, Arg);
370  CGF.EmitBranch(EndBlock);
371 
372  CGF.EmitBlock(EndBlock);
373 }
374 
375 /// Creates a function that sets up state on the host side for CUDA objects that
376 /// have a presence on both the host and device sides. Specifically, registers
377 /// the host side of kernel functions and device global variables with the CUDA
378 /// runtime.
379 /// \code
380 /// void __cuda_register_globals(void** GpuBinaryHandle) {
381 /// __cudaRegisterFunction(GpuBinaryHandle,Kernel0,...);
382 /// ...
383 /// __cudaRegisterFunction(GpuBinaryHandle,KernelM,...);
384 /// __cudaRegisterVar(GpuBinaryHandle, GlobalVar0, ...);
385 /// ...
386 /// __cudaRegisterVar(GpuBinaryHandle, GlobalVarN, ...);
387 /// }
388 /// \endcode
389 llvm::Function *CGNVCUDARuntime::makeRegisterGlobalsFn() {
390  // No need to register anything
391  if (EmittedKernels.empty() && DeviceVars.empty())
392  return nullptr;
393 
394  llvm::Function *RegisterKernelsFunc = llvm::Function::Create(
395  getRegisterGlobalsFnTy(), llvm::GlobalValue::InternalLinkage,
396  addUnderscoredPrefixToName("_register_globals"), &TheModule);
397  llvm::BasicBlock *EntryBB =
398  llvm::BasicBlock::Create(Context, "entry", RegisterKernelsFunc);
399  CGBuilderTy Builder(CGM, Context);
400  Builder.SetInsertPoint(EntryBB);
401 
402  // void __cudaRegisterFunction(void **, const char *, char *, const char *,
403  // int, uint3*, uint3*, dim3*, dim3*, int*)
404  llvm::Type *RegisterFuncParams[] = {
405  VoidPtrPtrTy, CharPtrTy, CharPtrTy, CharPtrTy, IntTy,
406  VoidPtrTy, VoidPtrTy, VoidPtrTy, VoidPtrTy, IntTy->getPointerTo()};
407  llvm::FunctionCallee RegisterFunc = CGM.CreateRuntimeFunction(
408  llvm::FunctionType::get(IntTy, RegisterFuncParams, false),
409  addUnderscoredPrefixToName("RegisterFunction"));
410 
411  // Extract GpuBinaryHandle passed as the first argument passed to
412  // __cuda_register_globals() and generate __cudaRegisterFunction() call for
413  // each emitted kernel.
414  llvm::Argument &GpuBinaryHandlePtr = *RegisterKernelsFunc->arg_begin();
415  for (auto &&I : EmittedKernels) {
416  llvm::Constant *KernelName = makeConstantString(getDeviceSideName(I.D));
417  llvm::Constant *NullPtr = llvm::ConstantPointerNull::get(VoidPtrTy);
418  llvm::Value *Args[] = {
419  &GpuBinaryHandlePtr,
420  Builder.CreateBitCast(I.Kernel, VoidPtrTy),
421  KernelName,
422  KernelName,
423  llvm::ConstantInt::get(IntTy, -1),
424  NullPtr,
425  NullPtr,
426  NullPtr,
427  NullPtr,
428  llvm::ConstantPointerNull::get(IntTy->getPointerTo())};
429  Builder.CreateCall(RegisterFunc, Args);
430  }
431 
432  // void __cudaRegisterVar(void **, char *, char *, const char *,
433  // int, int, int, int)
434  llvm::Type *RegisterVarParams[] = {VoidPtrPtrTy, CharPtrTy, CharPtrTy,
435  CharPtrTy, IntTy, IntTy,
436  IntTy, IntTy};
437  llvm::FunctionCallee RegisterVar = CGM.CreateRuntimeFunction(
438  llvm::FunctionType::get(IntTy, RegisterVarParams, false),
439  addUnderscoredPrefixToName("RegisterVar"));
440  for (auto &&Info : DeviceVars) {
441  llvm::GlobalVariable *Var = Info.Var;
442  unsigned Flags = Info.Flag;
443  llvm::Constant *VarName = makeConstantString(getDeviceSideName(Info.D));
444  uint64_t VarSize =
445  CGM.getDataLayout().getTypeAllocSize(Var->getValueType());
446  llvm::Value *Args[] = {
447  &GpuBinaryHandlePtr,
448  Builder.CreateBitCast(Var, VoidPtrTy),
449  VarName,
450  VarName,
451  llvm::ConstantInt::get(IntTy, (Flags & ExternDeviceVar) ? 1 : 0),
452  llvm::ConstantInt::get(IntTy, VarSize),
453  llvm::ConstantInt::get(IntTy, (Flags & ConstantDeviceVar) ? 1 : 0),
454  llvm::ConstantInt::get(IntTy, 0)};
455  Builder.CreateCall(RegisterVar, Args);
456  }
457 
458  Builder.CreateRetVoid();
459  return RegisterKernelsFunc;
460 }
461 
462 /// Creates a global constructor function for the module:
463 ///
464 /// For CUDA:
465 /// \code
466 /// void __cuda_module_ctor(void*) {
467 /// Handle = __cudaRegisterFatBinary(GpuBinaryBlob);
468 /// __cuda_register_globals(Handle);
469 /// }
470 /// \endcode
471 ///
472 /// For HIP:
473 /// \code
474 /// void __hip_module_ctor(void*) {
475 /// if (__hip_gpubin_handle == 0) {
476 /// __hip_gpubin_handle = __hipRegisterFatBinary(GpuBinaryBlob);
477 /// __hip_register_globals(__hip_gpubin_handle);
478 /// }
479 /// }
480 /// \endcode
481 llvm::Function *CGNVCUDARuntime::makeModuleCtorFunction() {
482  bool IsHIP = CGM.getLangOpts().HIP;
483  bool IsCUDA = CGM.getLangOpts().CUDA;
484  // No need to generate ctors/dtors if there is no GPU binary.
485  StringRef CudaGpuBinaryFileName = CGM.getCodeGenOpts().CudaGpuBinaryFileName;
486  if (CudaGpuBinaryFileName.empty() && !IsHIP)
487  return nullptr;
488  if ((IsHIP || (IsCUDA && !RelocatableDeviceCode)) && EmittedKernels.empty() &&
489  DeviceVars.empty())
490  return nullptr;
491 
492  // void __{cuda|hip}_register_globals(void* handle);
493  llvm::Function *RegisterGlobalsFunc = makeRegisterGlobalsFn();
494  // We always need a function to pass in as callback. Create a dummy
495  // implementation if we don't need to register anything.
496  if (RelocatableDeviceCode && !RegisterGlobalsFunc)
497  RegisterGlobalsFunc = makeDummyFunction(getRegisterGlobalsFnTy());
498 
499  // void ** __{cuda|hip}RegisterFatBinary(void *);
500  llvm::FunctionCallee RegisterFatbinFunc = CGM.CreateRuntimeFunction(
501  llvm::FunctionType::get(VoidPtrPtrTy, VoidPtrTy, false),
502  addUnderscoredPrefixToName("RegisterFatBinary"));
503  // struct { int magic, int version, void * gpu_binary, void * dont_care };
504  llvm::StructType *FatbinWrapperTy =
505  llvm::StructType::get(IntTy, IntTy, VoidPtrTy, VoidPtrTy);
506 
507  // Register GPU binary with the CUDA runtime, store returned handle in a
508  // global variable and save a reference in GpuBinaryHandle to be cleaned up
509  // in destructor on exit. Then associate all known kernels with the GPU binary
510  // handle so CUDA runtime can figure out what to call on the GPU side.
511  std::unique_ptr<llvm::MemoryBuffer> CudaGpuBinary = nullptr;
512  if (!CudaGpuBinaryFileName.empty()) {
513  llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> CudaGpuBinaryOrErr =
514  llvm::MemoryBuffer::getFileOrSTDIN(CudaGpuBinaryFileName);
515  if (std::error_code EC = CudaGpuBinaryOrErr.getError()) {
516  CGM.getDiags().Report(diag::err_cannot_open_file)
517  << CudaGpuBinaryFileName << EC.message();
518  return nullptr;
519  }
520  CudaGpuBinary = std::move(CudaGpuBinaryOrErr.get());
521  }
522 
523  llvm::Function *ModuleCtorFunc = llvm::Function::Create(
524  llvm::FunctionType::get(VoidTy, VoidPtrTy, false),
526  addUnderscoredPrefixToName("_module_ctor"), &TheModule);
527  llvm::BasicBlock *CtorEntryBB =
528  llvm::BasicBlock::Create(Context, "entry", ModuleCtorFunc);
529  CGBuilderTy CtorBuilder(CGM, Context);
530 
531  CtorBuilder.SetInsertPoint(CtorEntryBB);
532 
533  const char *FatbinConstantName;
534  const char *FatbinSectionName;
535  const char *ModuleIDSectionName;
536  StringRef ModuleIDPrefix;
537  llvm::Constant *FatBinStr;
538  unsigned FatMagic;
539  if (IsHIP) {
540  FatbinConstantName = ".hip_fatbin";
541  FatbinSectionName = ".hipFatBinSegment";
542 
543  ModuleIDSectionName = "__hip_module_id";
544  ModuleIDPrefix = "__hip_";
545 
546  if (CudaGpuBinary) {
547  // If fatbin is available from early finalization, create a string
548  // literal containing the fat binary loaded from the given file.
549  FatBinStr = makeConstantString(CudaGpuBinary->getBuffer(), "",
550  FatbinConstantName, 8);
551  } else {
552  // If fatbin is not available, create an external symbol
553  // __hip_fatbin in section .hip_fatbin. The external symbol is supposed
554  // to contain the fat binary but will be populated somewhere else,
555  // e.g. by lld through link script.
556  FatBinStr = new llvm::GlobalVariable(
557  CGM.getModule(), CGM.Int8Ty,
558  /*isConstant=*/true, llvm::GlobalValue::ExternalLinkage, nullptr,
559  "__hip_fatbin", nullptr,
560  llvm::GlobalVariable::NotThreadLocal);
561  cast<llvm::GlobalVariable>(FatBinStr)->setSection(FatbinConstantName);
562  }
563 
564  FatMagic = HIPFatMagic;
565  } else {
566  if (RelocatableDeviceCode)
567  FatbinConstantName = CGM.getTriple().isMacOSX()
568  ? "__NV_CUDA,__nv_relfatbin"
569  : "__nv_relfatbin";
570  else
571  FatbinConstantName =
572  CGM.getTriple().isMacOSX() ? "__NV_CUDA,__nv_fatbin" : ".nv_fatbin";
573  // NVIDIA's cuobjdump looks for fatbins in this section.
574  FatbinSectionName =
575  CGM.getTriple().isMacOSX() ? "__NV_CUDA,__fatbin" : ".nvFatBinSegment";
576 
577  ModuleIDSectionName = CGM.getTriple().isMacOSX()
578  ? "__NV_CUDA,__nv_module_id"
579  : "__nv_module_id";
580  ModuleIDPrefix = "__nv_";
581 
582  // For CUDA, create a string literal containing the fat binary loaded from
583  // the given file.
584  FatBinStr = makeConstantString(CudaGpuBinary->getBuffer(), "",
585  FatbinConstantName, 8);
586  FatMagic = CudaFatMagic;
587  }
588 
589  // Create initialized wrapper structure that points to the loaded GPU binary
591  auto Values = Builder.beginStruct(FatbinWrapperTy);
592  // Fatbin wrapper magic.
593  Values.addInt(IntTy, FatMagic);
594  // Fatbin version.
595  Values.addInt(IntTy, 1);
596  // Data.
597  Values.add(FatBinStr);
598  // Unused in fatbin v1.
599  Values.add(llvm::ConstantPointerNull::get(VoidPtrTy));
600  llvm::GlobalVariable *FatbinWrapper = Values.finishAndCreateGlobal(
601  addUnderscoredPrefixToName("_fatbin_wrapper"), CGM.getPointerAlign(),
602  /*constant*/ true);
603  FatbinWrapper->setSection(FatbinSectionName);
604 
605  // There is only one HIP fat binary per linked module, however there are
606  // multiple constructor functions. Make sure the fat binary is registered
607  // only once. The constructor functions are executed by the dynamic loader
608  // before the program gains control. The dynamic loader cannot execute the
609  // constructor functions concurrently since doing that would not guarantee
610  // thread safety of the loaded program. Therefore we can assume sequential
611  // execution of constructor functions here.
612  if (IsHIP) {
613  auto Linkage = CudaGpuBinary ? llvm::GlobalValue::InternalLinkage :
614  llvm::GlobalValue::LinkOnceAnyLinkage;
615  llvm::BasicBlock *IfBlock =
616  llvm::BasicBlock::Create(Context, "if", ModuleCtorFunc);
617  llvm::BasicBlock *ExitBlock =
618  llvm::BasicBlock::Create(Context, "exit", ModuleCtorFunc);
619  // The name, size, and initialization pattern of this variable is part
620  // of HIP ABI.
621  GpuBinaryHandle = new llvm::GlobalVariable(
622  TheModule, VoidPtrPtrTy, /*isConstant=*/false,
623  Linkage,
624  /*Initializer=*/llvm::ConstantPointerNull::get(VoidPtrPtrTy),
625  "__hip_gpubin_handle");
626  GpuBinaryHandle->setAlignment(CGM.getPointerAlign().getQuantity());
627  // Prevent the weak symbol in different shared libraries being merged.
629  GpuBinaryHandle->setVisibility(llvm::GlobalValue::HiddenVisibility);
630  Address GpuBinaryAddr(
631  GpuBinaryHandle,
632  CharUnits::fromQuantity(GpuBinaryHandle->getAlignment()));
633  {
634  auto HandleValue = CtorBuilder.CreateLoad(GpuBinaryAddr);
635  llvm::Constant *Zero =
636  llvm::Constant::getNullValue(HandleValue->getType());
637  llvm::Value *EQZero = CtorBuilder.CreateICmpEQ(HandleValue, Zero);
638  CtorBuilder.CreateCondBr(EQZero, IfBlock, ExitBlock);
639  }
640  {
641  CtorBuilder.SetInsertPoint(IfBlock);
642  // GpuBinaryHandle = __hipRegisterFatBinary(&FatbinWrapper);
643  llvm::CallInst *RegisterFatbinCall = CtorBuilder.CreateCall(
644  RegisterFatbinFunc,
645  CtorBuilder.CreateBitCast(FatbinWrapper, VoidPtrTy));
646  CtorBuilder.CreateStore(RegisterFatbinCall, GpuBinaryAddr);
647  CtorBuilder.CreateBr(ExitBlock);
648  }
649  {
650  CtorBuilder.SetInsertPoint(ExitBlock);
651  // Call __hip_register_globals(GpuBinaryHandle);
652  if (RegisterGlobalsFunc) {
653  auto HandleValue = CtorBuilder.CreateLoad(GpuBinaryAddr);
654  CtorBuilder.CreateCall(RegisterGlobalsFunc, HandleValue);
655  }
656  }
657  } else if (!RelocatableDeviceCode) {
658  // Register binary with CUDA runtime. This is substantially different in
659  // default mode vs. separate compilation!
660  // GpuBinaryHandle = __cudaRegisterFatBinary(&FatbinWrapper);
661  llvm::CallInst *RegisterFatbinCall = CtorBuilder.CreateCall(
662  RegisterFatbinFunc,
663  CtorBuilder.CreateBitCast(FatbinWrapper, VoidPtrTy));
664  GpuBinaryHandle = new llvm::GlobalVariable(
665  TheModule, VoidPtrPtrTy, false, llvm::GlobalValue::InternalLinkage,
666  llvm::ConstantPointerNull::get(VoidPtrPtrTy), "__cuda_gpubin_handle");
667  GpuBinaryHandle->setAlignment(CGM.getPointerAlign().getQuantity());
668  CtorBuilder.CreateAlignedStore(RegisterFatbinCall, GpuBinaryHandle,
669  CGM.getPointerAlign());
670 
671  // Call __cuda_register_globals(GpuBinaryHandle);
672  if (RegisterGlobalsFunc)
673  CtorBuilder.CreateCall(RegisterGlobalsFunc, RegisterFatbinCall);
674 
675  // Call __cudaRegisterFatBinaryEnd(Handle) if this CUDA version needs it.
678  // void __cudaRegisterFatBinaryEnd(void **);
679  llvm::FunctionCallee RegisterFatbinEndFunc = CGM.CreateRuntimeFunction(
680  llvm::FunctionType::get(VoidTy, VoidPtrPtrTy, false),
681  "__cudaRegisterFatBinaryEnd");
682  CtorBuilder.CreateCall(RegisterFatbinEndFunc, RegisterFatbinCall);
683  }
684  } else {
685  // Generate a unique module ID.
686  SmallString<64> ModuleID;
687  llvm::raw_svector_ostream OS(ModuleID);
688  OS << ModuleIDPrefix << llvm::format("%" PRIx64, FatbinWrapper->getGUID());
689  llvm::Constant *ModuleIDConstant =
690  makeConstantString(ModuleID.str(), "", ModuleIDSectionName, 32);
691 
692  // Create an alias for the FatbinWrapper that nvcc will look for.
694  Twine("__fatbinwrap") + ModuleID, FatbinWrapper);
695 
696  // void __cudaRegisterLinkedBinary%ModuleID%(void (*)(void *), void *,
697  // void *, void (*)(void **))
698  SmallString<128> RegisterLinkedBinaryName("__cudaRegisterLinkedBinary");
699  RegisterLinkedBinaryName += ModuleID;
700  llvm::FunctionCallee RegisterLinkedBinaryFunc = CGM.CreateRuntimeFunction(
701  getRegisterLinkedBinaryFnTy(), RegisterLinkedBinaryName);
702 
703  assert(RegisterGlobalsFunc && "Expecting at least dummy function!");
704  llvm::Value *Args[] = {RegisterGlobalsFunc,
705  CtorBuilder.CreateBitCast(FatbinWrapper, VoidPtrTy),
706  ModuleIDConstant,
707  makeDummyFunction(getCallbackFnTy())};
708  CtorBuilder.CreateCall(RegisterLinkedBinaryFunc, Args);
709  }
710 
711  // Create destructor and register it with atexit() the way NVCC does it. Doing
712  // it during regular destructor phase worked in CUDA before 9.2 but results in
713  // double-free in 9.2.
714  if (llvm::Function *CleanupFn = makeModuleDtorFunction()) {
715  // extern "C" int atexit(void (*f)(void));
716  llvm::FunctionType *AtExitTy =
717  llvm::FunctionType::get(IntTy, CleanupFn->getType(), false);
718  llvm::FunctionCallee AtExitFunc =
719  CGM.CreateRuntimeFunction(AtExitTy, "atexit", llvm::AttributeList(),
720  /*Local=*/true);
721  CtorBuilder.CreateCall(AtExitFunc, CleanupFn);
722  }
723 
724  CtorBuilder.CreateRetVoid();
725  return ModuleCtorFunc;
726 }
727 
728 /// Creates a global destructor function that unregisters the GPU code blob
729 /// registered by constructor.
730 ///
731 /// For CUDA:
732 /// \code
733 /// void __cuda_module_dtor(void*) {
734 /// __cudaUnregisterFatBinary(Handle);
735 /// }
736 /// \endcode
737 ///
738 /// For HIP:
739 /// \code
740 /// void __hip_module_dtor(void*) {
741 /// if (__hip_gpubin_handle) {
742 /// __hipUnregisterFatBinary(__hip_gpubin_handle);
743 /// __hip_gpubin_handle = 0;
744 /// }
745 /// }
746 /// \endcode
747 llvm::Function *CGNVCUDARuntime::makeModuleDtorFunction() {
748  // No need for destructor if we don't have a handle to unregister.
749  if (!GpuBinaryHandle)
750  return nullptr;
751 
752  // void __cudaUnregisterFatBinary(void ** handle);
753  llvm::FunctionCallee UnregisterFatbinFunc = CGM.CreateRuntimeFunction(
754  llvm::FunctionType::get(VoidTy, VoidPtrPtrTy, false),
755  addUnderscoredPrefixToName("UnregisterFatBinary"));
756 
757  llvm::Function *ModuleDtorFunc = llvm::Function::Create(
758  llvm::FunctionType::get(VoidTy, VoidPtrTy, false),
760  addUnderscoredPrefixToName("_module_dtor"), &TheModule);
761 
762  llvm::BasicBlock *DtorEntryBB =
763  llvm::BasicBlock::Create(Context, "entry", ModuleDtorFunc);
764  CGBuilderTy DtorBuilder(CGM, Context);
765  DtorBuilder.SetInsertPoint(DtorEntryBB);
766 
767  Address GpuBinaryAddr(GpuBinaryHandle, CharUnits::fromQuantity(
768  GpuBinaryHandle->getAlignment()));
769  auto HandleValue = DtorBuilder.CreateLoad(GpuBinaryAddr);
770  // There is only one HIP fat binary per linked module, however there are
771  // multiple destructor functions. Make sure the fat binary is unregistered
772  // only once.
773  if (CGM.getLangOpts().HIP) {
774  llvm::BasicBlock *IfBlock =
775  llvm::BasicBlock::Create(Context, "if", ModuleDtorFunc);
776  llvm::BasicBlock *ExitBlock =
777  llvm::BasicBlock::Create(Context, "exit", ModuleDtorFunc);
778  llvm::Constant *Zero = llvm::Constant::getNullValue(HandleValue->getType());
779  llvm::Value *NEZero = DtorBuilder.CreateICmpNE(HandleValue, Zero);
780  DtorBuilder.CreateCondBr(NEZero, IfBlock, ExitBlock);
781 
782  DtorBuilder.SetInsertPoint(IfBlock);
783  DtorBuilder.CreateCall(UnregisterFatbinFunc, HandleValue);
784  DtorBuilder.CreateStore(Zero, GpuBinaryAddr);
785  DtorBuilder.CreateBr(ExitBlock);
786 
787  DtorBuilder.SetInsertPoint(ExitBlock);
788  } else {
789  DtorBuilder.CreateCall(UnregisterFatbinFunc, HandleValue);
790  }
791  DtorBuilder.CreateRetVoid();
792  return ModuleDtorFunc;
793 }
794 
795 std::string CGNVCUDARuntime::getDeviceStubName(llvm::StringRef Name) const {
796  if (!CGM.getLangOpts().HIP)
797  return Name;
798  return (Name + ".stub").str();
799 }
800 
802  return new CGNVCUDARuntime(CGM);
803 }
const llvm::DataLayout & getDataLayout() const
ReturnValueSlot - Contains the address where the return value of a function can be stored...
Definition: CGCall.h:363
CharUnits alignTo(const CharUnits &Align) const
alignTo - Returns the next integer (mod 2**64) that is greater than or equal to this quantity and is ...
Definition: CharUnits.h:183
llvm::StoreInst * CreateDefaultAlignedStore(llvm::Value *Val, llvm::Value *Addr, bool IsVolatile=false)
Definition: CGBuilder.h:121
Represents a function declaration or definition.
Definition: Decl.h:1748
llvm::IntegerType * IntTy
int
External linkage, which indicates that the entity can be referred to from other translation units...
Definition: Linkage.h:59
CanQualType VoidPtrTy
Definition: ASTContext.h:1042
A (possibly-)qualified type.
Definition: Type.h:643
const CodeGenOptions & getCodeGenOpts() const
Address CreateMemTemp(QualType T, const Twine &Name="tmp", Address *Alloca=nullptr)
CreateMemTemp - Create a temporary memory object of the given type, with appropriate alignmen and cas...
Definition: CGExpr.cpp:139
The standard implementation of ConstantInitBuilder used in Clang.
Decl - This represents one declaration (or definition), e.g.
Definition: DeclBase.h:88
DiagnosticBuilder Report(SourceLocation Loc, unsigned DiagID)
Issue the message to the client.
Definition: Diagnostic.h:1297
const TargetInfo & getTargetInfo() const
Definition: ASTContext.h:693
llvm::IntegerType * Int8Ty
i8, i16, i32, and i64
Address GetAddrOfLocalVar(const VarDecl *VD)
GetAddrOfLocalVar - Return the address of a local variable.
Represents a variable declaration or definition.
Definition: Decl.h:812
Objects with "hidden" visibility are not seen by the dynamic linker.
Definition: Visibility.h:36
DiagnosticsEngine & getDiags() const
llvm::Value * getPointer() const
Definition: Address.h:37
Represents a parameter to a function.
Definition: Decl.h:1564
Linkage
Describes the different kinds of linkage (C++ [basic.link], C99 6.2.2) that an entity may have...
Definition: Linkage.h:23
long i
Definition: xmmintrin.h:1456
void add(RValue rvalue, QualType type)
Definition: CGCall.h:287
One of these records is kept for each identifier that is lexed.
TargetCXXABI getCXXABI() const
Get the C++ ABI currently in use.
Definition: TargetInfo.h:1054
CodeGenFunction - This class organizes the per-function state that is used while generating LLVM code...
llvm::Type * ConvertType(QualType T)
ConvertType - Convert type T into a llvm::Type.
Holds long-lived AST nodes (such as types and decls) that can be referred to throughout the semantic ...
Definition: ASTContext.h:154
static CharUnits Zero()
Zero - Construct a CharUnits quantity of zero.
Definition: CharUnits.h:52
IdentifierTable & Idents
Definition: ASTContext.h:569
CharUnits - This is an opaque type for sizes expressed in character units.
Definition: CharUnits.h:37
llvm::BasicBlock * createBasicBlock(const Twine &name="", llvm::Function *parent=nullptr, llvm::BasicBlock *before=nullptr)
createBasicBlock - Create an LLVM basic block.
llvm::AllocaInst * CreateTempAlloca(llvm::Type *Ty, const Twine &Name="tmp", llvm::Value *ArraySize=nullptr)
CreateTempAlloca - This creates an alloca and inserts it into the entry block if ArraySize is nullptr...
Definition: CGExpr.cpp:106
lookup_result lookup(DeclarationName Name) const
lookup - Find the declarations (if any) with the given Name in this context.
Definition: DeclBase.cpp:1602
QuantityType getQuantity() const
getQuantity - Get the raw integer representation of this quantity.
Definition: CharUnits.h:178
unsigned Offset
Definition: Format.cpp:1713
static CGCallee forDirect(llvm::Constant *functionPtr, const CGCalleeInfo &abstractInfo=CGCalleeInfo())
Definition: CGCall.h:133
llvm::PointerType * getType() const
Return the type of the pointer value.
Definition: Address.h:43
static CharUnits fromQuantity(QuantityType Quantity)
fromQuantity - Construct a CharUnits quantity from a raw integer type.
Definition: CharUnits.h:62
const TargetInfo & getTarget() const
const LangOptions & getLangOpts() const
ASTContext & getContext() const
The l-value was considered opaque, so the alignment was determined from a type.
const llvm::VersionTuple & getSDKVersion() const
Definition: TargetInfo.h:1356
Address CreateBitCast(Address Addr, llvm::Type *Ty, const llvm::Twine &Name="")
Definition: CGBuilder.h:141
QualType getCanonicalType() const
Definition: Type.h:6181
static DeclContext * castToDeclContext(const TranslationUnitDecl *D)
Definition: Decl.h:129
IdentifierInfo & get(StringRef Name)
Return the identifier token info for the specified named identifier.
RValue EmitCall(const CGFunctionInfo &CallInfo, const CGCallee &Callee, ReturnValueSlot ReturnValue, const CallArgList &Args, llvm::CallBase **callOrInvoke, SourceLocation Loc)
EmitCall - Generate a call of the given function, expecting the given result type, and using the given argument list which specifies both the LLVM arguments and the types they were derived from.
Definition: CGCall.cpp:3776
const ParmVarDecl * getParamDecl(unsigned i) const
Definition: Decl.h:2312
An aligned address.
Definition: Address.h:24
bool CudaFeatureEnabled(llvm::VersionTuple, CudaFeature)
Definition: Cuda.cpp:388
constexpr XRayInstrMask None
Definition: XRayInstr.h:37
void Error(SourceLocation loc, StringRef error)
Emit a general error that something can&#39;t be done.
std::pair< CharUnits, CharUnits > getTypeInfoInChars(const Type *T) const
FunctionArgList - Type for representing both the decl and type of parameters to a function...
Definition: CGCall.h:358
CanQualType CharTy
Definition: ASTContext.h:1016
CGFunctionInfo - Class to encapsulate the information about a function definition.
This class organizes the cross-function state that is used while generating LLVM code.
std::string CudaGpuBinaryFileName
Name of file passed with -fcuda-include-gpubinary option to forward to CUDA runtime back-end for inco...
Dataflow Directional Tag Classes.
DeclContext - This is used only as base class of specific decl types that can act as declaration cont...
Definition: DeclBase.h:1271
const CGFunctionInfo & arrangeFunctionDeclaration(const FunctionDecl *FD)
Free functions are functions that are compatible with an ordinary C function pointer type...
Definition: CGCall.cpp:433
std::unique_ptr< DiagnosticConsumer > create(StringRef OutputFile, DiagnosticOptions *Diags, bool MergeChildRecords=false)
Returns a DiagnosticConsumer that serializes diagnostics to a bitcode file.
llvm::LoadInst * CreateLoad(Address Addr, const llvm::Twine &Name="")
Definition: CGBuilder.h:69
llvm::Module & getModule() const
Indicates that the tracking object is a descendant of a referenced-counted OSObject, used in the Darwin kernel.
This class organizes the cross-module state that is used while lowering AST types to LLVM types...
Definition: CodeGenTypes.h:59
Internal linkage, which indicates that the entity can be referred to from within the translation unit...
Definition: Linkage.h:31
llvm::FunctionCallee CreateRuntimeFunction(llvm::FunctionType *Ty, StringRef Name, llvm::AttributeList ExtraAttrs=llvm::AttributeList(), bool Local=false)
Create or return a runtime function declaration with the specified type and name. ...
void EmitBlock(llvm::BasicBlock *BB, bool IsFinished=false)
EmitBlock - Emit the given block.
Definition: CGStmt.cpp:454
CGCUDARuntime * CreateNVCUDARuntime(CodeGenModule &CGM)
Creates an instance of a CUDA runtime class.
Definition: CGCUDANV.cpp:801
const Decl * CurFuncDecl
CurFuncDecl - Holds the Decl for the current outermost non-closure context.
TranslationUnitDecl * getTranslationUnitDecl() const
Definition: ASTContext.h:1007
void EmitBranch(llvm::BasicBlock *Block)
EmitBranch - Emit a branch to the specified basic block from the current insert block, taking care to avoid creation of branches from dummy blocks.
Definition: CGStmt.cpp:474
The top declaration context.
Definition: Decl.h:107
static RValue get(llvm::Value *V)
Definition: CGValue.h:85
QualType getType() const
Definition: Decl.h:647
static RValue getAggregate(Address addr, bool isVolatile=false)
Definition: CGValue.h:106
const TargetInfo * getAuxTargetInfo() const
Definition: ASTContext.h:694
llvm::CallBase * EmitRuntimeCallOrInvoke(llvm::FunctionCallee callee, ArrayRef< llvm::Value *> args, const Twine &name="")
Emits a call or invoke instruction to the given runtime function.
Definition: CGCall.cpp:3736
CallArgList - Type for representing both the value and type of arguments in a call.
Definition: CGCall.h:262
SourceLocation getLocation() const
Definition: DeclBase.h:429
static OMPLinearClause * Create(const ASTContext &C, SourceLocation StartLoc, SourceLocation LParenLoc, OpenMPLinearClauseKind Modifier, SourceLocation ModifierLoc, SourceLocation ColonLoc, SourceLocation EndLoc, ArrayRef< Expr *> VL, ArrayRef< Expr *> PL, ArrayRef< Expr *> IL, Expr *Step, Expr *CalcStep, Stmt *PreInit, Expr *PostUpdate)
Creates clause with a list of variables VL and a linear step Step.
const llvm::Triple & getTriple() const