Basic CUDA-10 support.

Differential Revision: https://reviews.llvm.org/D57771

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@353232 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/Cuda.h b/include/clang/Basic/Cuda.h
index 532c23098eef4b1d43761ccca62318f258478795..d96c7e0972f99f9eb6379d8c40ecec9a37a35707 100644 (file)
--- a/include/clang/Basic/Cuda.h
+++ b/include/clang/Basic/Cuda.h
@@ -25,7 +25,8 @@ enum class CudaVersion {
   CUDA_91,
   CUDA_92,
   CUDA_100,
-  LATEST = CUDA_100,
+  CUDA_101,
+  LATEST = CUDA_101,
 };
 const char *CudaVersionToString(CudaVersion V);
 // Input is "Major.Minor"
@@ -107,6 +108,8 @@ CudaVersion MaxVersionForCudaArch(CudaArch A);
 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);

diff --git a/lib/Basic/Cuda.cpp b/lib/Basic/Cuda.cpp
index 6ba9a8437e2c92f9025dc13e01f00207290c9774..4bc8d8c5b42e5223f54c6132d4930f439ac1381d 100644 (file)
--- a/lib/Basic/Cuda.cpp
+++ b/lib/Basic/Cuda.cpp
@@ -25,6 +25,8 @@ const char *CudaVersionToString(CudaVersion V) {
     return "9.2";
   case CudaVersion::CUDA_100:
     return "10.0";
+  case CudaVersion::CUDA_101:
+    return "10.1";
   }
   llvm_unreachable("invalid enum");
 }
@@ -37,7 +39,8 @@ CudaVersion CudaStringToVersion(llvm::StringRef S) {
       .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) {
@@ -352,6 +355,8 @@ static CudaVersion ToCudaVersion(llvm::VersionTuple Version) {
     return CudaVersion::CUDA_92;
   case 100:
     return CudaVersion::CUDA_100;
+  case 101:
+    return CudaVersion::CUDA_101;
   default:
     return CudaVersion::UNKNOWN;
   }
@@ -365,6 +370,8 @@ 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.");
 }

diff --git a/lib/CodeGen/CGCUDANV.cpp b/lib/CodeGen/CGCUDANV.cpp
index 93ec94ca89bc31c1ca43f7dc5c53d5ff1d5fdaa4..68e83b939ae4f9dc6053aacf129069c44f48fb44 100644 (file)
--- a/lib/CodeGen/CGCUDANV.cpp
+++ b/lib/CodeGen/CGCUDANV.cpp
@@ -616,6 +616,16 @@ llvm::Function *CGNVCUDARuntime::makeModuleCtorFunction() {
     // Call __cuda_register_globals(GpuBinaryHandle);
     if (RegisterGlobalsFunc)
       CtorBuilder.CreateCall(RegisterGlobalsFunc, RegisterFatbinCall);
+
+    // Call __cudaRegisterFatBinaryEnd(Handle) if this CUDA version needs it.
+    if (CudaFeatureEnabled(CGM.getTarget().getSDKVersion(),
+                           CudaFeature::CUDA_USES_FATBIN_REGISTER_END)) {
+      // void __cudaRegisterFatBinaryEnd(void **);
+      llvm::FunctionCallee RegisterFatbinEndFunc = CGM.CreateRuntimeFunction(
+          llvm::FunctionType::get(VoidTy, VoidPtrPtrTy, false),
+          "__cudaRegisterFatBinaryEnd");
+      CtorBuilder.CreateCall(RegisterFatbinEndFunc, RegisterFatbinCall);
+    }
   } else {
     // Generate a unique module ID.
     SmallString<64> ModuleID;

diff --git a/lib/Driver/ToolChains/Cuda.cpp b/lib/Driver/ToolChains/Cuda.cpp
index e3fe84b10181b0856d36086f3a9db40928cac3fe..7d01f4b48d2585e494ec7812f282984ac73bc999 100644 (file)
--- a/lib/Driver/ToolChains/Cuda.cpp
+++ b/lib/Driver/ToolChains/Cuda.cpp
@@ -60,6 +60,8 @@ static CudaVersion ParseCudaVersionFile(llvm::StringRef V) {
     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;
 }
 

diff --git a/lib/Headers/__clang_cuda_runtime_wrapper.h b/lib/Headers/__clang_cuda_runtime_wrapper.h
index 4b3672463e614ac96a7e9e2ae78a044d43f077df..6b41e1741cb1a3b62c4efe89409bd9bc9d72d7d6 100644 (file)
--- a/lib/Headers/__clang_cuda_runtime_wrapper.h
+++ b/lib/Headers/__clang_cuda_runtime_wrapper.h
@@ -62,7 +62,7 @@
 #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