[CUDA] Add basic support for CUDA-10.1

[tra]

Make the function object local.

[jlebar]

clang/lib/CodeGen/CGCUDANV.cpp

624

Was confused by this sentence for a while. Maybe change to "Check whether cuda version requires ...."?

[tra]

Made a comment more readable.

clang/include/clang/Basic/Cuda.h

enum class CudaVersion {
   UNKNOWN,
   CUDA_70,
   CUDA_75,
   CUDA_80,
   CUDA_90,
   CUDA_91,
   CUDA_92,
   CUDA_100,
-  LATEST = CUDA_100,
+  CUDA_101,
+  LATEST = CUDA_101,
 };
 const char *CudaVersionToString(CudaVersion V);
 // Input is "Major.Minor"
 CudaVersion CudaStringToVersion(llvm::StringRef S);
 
 enum class CudaArch {
   UNKNOWN,
   SM_20,

 
 /// Get the latest CudaVersion that supports the given CudaArch.
 CudaVersion MaxVersionForCudaArch(CudaArch A);
 
 //  Various SDK-dependent features that affect CUDA compilation
 enum class CudaFeature {
   // CUDA-9.2+ uses a new API for launching kernels.
   CUDA_USES_NEW_LAUNCH,
+  // CUDA-10.1+ needs explicit end of GPU binary registration.
+  CUDA_USES_FATBIN_REGISTER_END,
 };
 
 bool CudaFeatureEnabled(llvm::VersionTuple, CudaFeature);
 bool CudaFeatureEnabled(CudaVersion, CudaFeature);
 
 } // namespace clang
 
 #endif

clang/lib/Basic/Cuda.cpp

const char *CudaVersionToString(CudaVersion V) {
   case CudaVersion::CUDA_90:
     return "9.0";
   case CudaVersion::CUDA_91:
     return "9.1";
   case CudaVersion::CUDA_92:
     return "9.2";
   case CudaVersion::CUDA_100:
     return "10.0";
+  case CudaVersion::CUDA_101:
+    return "10.1";
   }
   llvm_unreachable("invalid enum");
 }
 
 CudaVersion CudaStringToVersion(llvm::StringRef S) {
   return llvm::StringSwitch<CudaVersion>(S)
       .Case("7.0", CudaVersion::CUDA_70)
       .Case("7.5", CudaVersion::CUDA_75)
       .Case("8.0", CudaVersion::CUDA_80)
       .Case("9.0", CudaVersion::CUDA_90)
       .Case("9.1", CudaVersion::CUDA_91)
       .Case("9.2", CudaVersion::CUDA_92)
-      .Case("10.0", CudaVersion::CUDA_100);
+      .Case("10.0", CudaVersion::CUDA_100)
+      .Case("10.1", CudaVersion::CUDA_101);
 }
 
 const char *CudaArchToString(CudaArch A) {
   switch (A) {
   case CudaArch::LAST:
     break;
   case CudaArch::UNKNOWN:
     return "unknown";
static CudaVersion ToCudaVersion(llvm::VersionTuple Version) {
   case 90:
     return CudaVersion::CUDA_90;
   case 91:
     return CudaVersion::CUDA_91;
   case 92:
     return CudaVersion::CUDA_92;
   case 100:
     return CudaVersion::CUDA_100;
+  case 101:
+    return CudaVersion::CUDA_101;
   default:
     return CudaVersion::UNKNOWN;
   }
 }
 
 bool CudaFeatureEnabled(llvm::VersionTuple  Version, CudaFeature Feature) {
   return CudaFeatureEnabled(ToCudaVersion(Version), Feature);
 }
 
 bool CudaFeatureEnabled(CudaVersion Version, CudaFeature Feature) {
   switch (Feature) {
   case CudaFeature::CUDA_USES_NEW_LAUNCH:
     return Version >= CudaVersion::CUDA_92;
+  case CudaFeature::CUDA_USES_FATBIN_REGISTER_END:
+    return Version >= CudaVersion::CUDA_101;
   }
   llvm_unreachable("Unknown CUDA feature.");
 }
 } // namespace clang

clang/lib/CodeGen/CGCUDANV.cpp

llvm::Function *CGNVCUDARuntime::makeModuleCtorFunction() {
   // implementation if we don't need to register anything.
   if (RelocatableDeviceCode && !RegisterGlobalsFunc)
     RegisterGlobalsFunc = makeDummyFunction(getRegisterGlobalsFnTy());
 
   // void ** __{cuda|hip}RegisterFatBinary(void *);
   llvm::FunctionCallee RegisterFatbinFunc = CGM.CreateRuntimeFunction(
       llvm::FunctionType::get(VoidPtrPtrTy, VoidPtrTy, false),
       addUnderscoredPrefixToName("RegisterFatBinary"));
+  // void __cudaRegisterFatBinaryEnd(void **);
+  llvm::FunctionCallee RegisterFatbinEndFunc = CGM.CreateRuntimeFunction(
+      llvm::FunctionType::get(VoidTy, VoidPtrPtrTy, false),
+      "__cudaRegisterFatBinaryEnd");
   // struct { int magic, int version, void * gpu_binary, void * dont_care };
   llvm::StructType *FatbinWrapperTy =
       llvm::StructType::get(IntTy, IntTy, VoidPtrTy, VoidPtrTy);
 
   // Register GPU binary with the CUDA runtime, store returned handle in a
   // global variable and save a reference in GpuBinaryHandle to be cleaned up
   // in destructor on exit. Then associate all known kernels with the GPU binary
   // handle so CUDA runtime can figure out what to call on the GPU side.
GpuBinaryHandle = new llvm::GlobalVariable(
         llvm::ConstantPointerNull::get(VoidPtrPtrTy), "__cuda_gpubin_handle");
     GpuBinaryHandle->setAlignment(CGM.getPointerAlign().getQuantity());
     CtorBuilder.CreateAlignedStore(RegisterFatbinCall, GpuBinaryHandle,
                                    CGM.getPointerAlign());
 
     // Call __cuda_register_globals(GpuBinaryHandle);
     if (RegisterGlobalsFunc)
       CtorBuilder.CreateCall(RegisterGlobalsFunc, RegisterFatbinCall);
+
+    // CUDA version requires calling __cudaRegisterFatBinaryEnd(Handle);
+    if (CudaFeatureEnabled(CGM.getTarget().getSDKVersion(),
+                           CudaFeature::CUDA_USES_FATBIN_REGISTER_END))
+      CtorBuilder.CreateCall(RegisterFatbinEndFunc, RegisterFatbinCall);
   } else {
     // Generate a unique module ID.
     SmallString<64> ModuleID;
     llvm::raw_svector_ostream OS(ModuleID);
     OS << ModuleIDPrefix << llvm::format("%" PRIx64, FatbinWrapper->getGUID());
     llvm::Constant *ModuleIDConstant =
         makeConstantString(ModuleID.str(), "", ModuleIDSectionName, 32);
 

clang/lib/Driver/ToolChains/Cuda.cpp

static CudaVersion ParseCudaVersionFile(llvm::StringRef V) {
   if (Major == 9 && Minor == 0)
     return CudaVersion::CUDA_90;
   if (Major == 9 && Minor == 1)
     return CudaVersion::CUDA_91;
   if (Major == 9 && Minor == 2)
     return CudaVersion::CUDA_92;
   if (Major == 10 && Minor == 0)
     return CudaVersion::CUDA_100;
+  if (Major == 10 && Minor == 1)
+    return CudaVersion::CUDA_101;
   return CudaVersion::UNKNOWN;
 }
 
 CudaInstallationDetector::CudaInstallationDetector(
     const Driver &D, const llvm::Triple &HostTriple,
     const llvm::opt::ArgList &Args)
     : D(D) {
   struct Candidate {

clang/lib/Headers/__clang_cuda_runtime_wrapper.h

 #pragma push_macro("__CUDA_ARCH__")
 
 // WARNING: Preprocessor hacks below are based on specific details of
 // CUDA-7.x headers and are not expected to work with any other
 // version of CUDA headers.
 #include "cuda.h"
 #if !defined(CUDA_VERSION)
 #error "cuda.h did not define CUDA_VERSION"
-#elif CUDA_VERSION < 7000 || CUDA_VERSION > 10000
+#elif CUDA_VERSION < 7000 || CUDA_VERSION > 10010
 #error "Unsupported CUDA version!"
 #endif
 
 #pragma push_macro("__CUDA_INCLUDE_COMPILER_INTERNAL_HEADERS__")
 #if CUDA_VERSION >= 10000
 #define __CUDA_INCLUDE_COMPILER_INTERNAL_HEADERS__
 #endif