[CUDA] Propagate detected version of CUDA to cc1

[tra]

Summary

..and use it to control that parts of CUDA compilation
that depend on the specific version of CUDA SDK.

This patch has a placeholder for a 'new launch API' support
which is in a separate patch (D57488).
The list will be further extended in the upcoming patch to support CUDA-10.1.

[tra]

Updated the comment about SDKVersion use.

[jlebar]

clang/include/clang/Basic/Cuda.h

108

Should this enum be documented?

clang/lib/Driver/ToolChains/Clang.cpp

3475

nit, s/as/, as/

[tra]

Addressed Justin's comments.

clang/include/clang/Basic/Cuda.h

 //===--- Cuda.h - Utilities for compiling CUDA code  ------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_CLANG_BASIC_CUDA_H
 #define LLVM_CLANG_BASIC_CUDA_H
 
 namespace llvm {
 class StringRef;
+class VersionTuple;
 } // namespace llvm
 
 namespace clang {
 
 enum class CudaVersion {
   UNKNOWN,
   CUDA_70,
   CUDA_75,
   CUDA_80,
   CUDA_90,
   CUDA_91,
   CUDA_92,
   CUDA_100,
   LATEST = CUDA_100,
 };
 const char *CudaVersionToString(CudaVersion V);
-
-// No string -> CudaVersion conversion function because there's no canonical
-// spelling of the various CUDA versions.
+// Input is "Major.Minor"
+CudaVersion CudaStringToVersion(llvm::StringRef S);
 
 enum class CudaArch {
   UNKNOWN,
   SM_20,
   SM_21,
   SM_30,
   SM_32,
   SM_35,

 CudaVirtualArch VirtualArchForCudaArch(CudaArch A);
 
 /// Get the earliest CudaVersion that supports the given CudaArch.
 CudaVersion MinVersionForCudaArch(CudaArch A);
 
 /// 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_USES_NEW_LAUNCH,
+};
+
+bool CudaFeatureEnabled(llvm::VersionTuple, CudaFeature);
+bool CudaFeatureEnabled(CudaVersion, CudaFeature);
+
 } // namespace clang
 
 #endif

clang/lib/Basic/Cuda.cpp

 #include "clang/Basic/Cuda.h"
 
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/VersionTuple.h"
 
 namespace clang {
 
 const char *CudaVersionToString(CudaVersion V) {
   switch (V) {
   case CudaVersion::UNKNOWN:
     return "unknown";
   case CudaVersion::CUDA_70:
const char *CudaVersionToString(CudaVersion V) {
   case CudaVersion::CUDA_92:
     return "9.2";
   case CudaVersion::CUDA_100:
     return "10.0";
   }
   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);
+}
+
 const char *CudaArchToString(CudaArch A) {
   switch (A) {
   case CudaArch::LAST:
     break;
   case CudaArch::UNKNOWN:
     return "unknown";
   case CudaArch::SM_20:
     return "sm_20";
CudaVersion MaxVersionForCudaArch(CudaArch A) {
   case CudaArch::GFX900:
   case CudaArch::GFX902:
     return CudaVersion::CUDA_80;
   default:
     return CudaVersion::LATEST;
   }
 }
 
+static CudaVersion ToCudaVersion(llvm::VersionTuple Version) {
+  int IVer =
+      Version.getMajor() * 10 + Version.getMinor().getValueOr(0);
+  switch(IVer) {
+  case 70:
+    return CudaVersion::CUDA_70;
+  case 75:
+    return CudaVersion::CUDA_75;
+  case 80:
+    return CudaVersion::CUDA_80;
+  case 90:
+    return CudaVersion::CUDA_90;
+  case 91:
+    return CudaVersion::CUDA_91;
+  case 92:
+    return CudaVersion::CUDA_92;
+  case 100:
+    return CudaVersion::CUDA_100;
+  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;
+  }
+  llvm_unreachable("Unknown CUDA feature.");
+}
 } // namespace clang

clang/lib/Driver/ToolChains/Clang.cpp

if (IsCuda || IsHIP) {
     // We have to pass the triple of the host if compiling for a CUDA/HIP device
     // and vice-versa.
     std::string NormalizedTriple;
     if (JA.isDeviceOffloading(Action::OFK_Cuda) ||
         JA.isDeviceOffloading(Action::OFK_HIP))
       NormalizedTriple = C.getSingleOffloadToolChain<Action::OFK_Host>()
                              ->getTriple()
                              .normalize();
-    else
+    else {
+      // Host-side compilation.
       NormalizedTriple =
           (IsCuda ? C.getSingleOffloadToolChain<Action::OFK_Cuda>()
                   : C.getSingleOffloadToolChain<Action::OFK_HIP>())
               ->getTriple()
               .normalize();
-
+      if (IsCuda) {
+        // We need to figure out which CUDA version we're compiling for as that
+        // determines how we load and launch GPU kernels.
+        auto *CTC = static_cast<const toolchains::CudaToolChain *>(
+            C.getSingleOffloadToolChain<Action::OFK_Cuda>());
+        assert(CTC && "Expected valid CUDA Toolchain.");
+        if (CTC && CTC->CudaInstallation.version() != CudaVersion::UNKNOWN)
+          CmdArgs.push_back(Args.MakeArgString(
+              Twine("-target-sdk-version=") +
+              CudaVersionToString(CTC->CudaInstallation.version())));
+      }
+    }
     CmdArgs.push_back("-aux-triple");
     CmdArgs.push_back(Args.MakeArgString(NormalizedTriple));
   }
 
   if (IsOpenMPDevice) {
     // We have to pass the triple of the host if compiling for an OpenMP device.
     std::string NormalizedTriple =
         C.getSingleOffloadToolChain<Action::OFK_Host>()

clang/lib/Driver/ToolChains/Cuda.cpp

if (CudaInstallation.version() >= CudaVersion::CUDA_91) {
     // CUDA-9.0 uses new instructions that are only available in PTX6.0+
     PtxFeature = "+ptx60";
   }
   CC1Args.append({"-target-feature", PtxFeature});
   if (DriverArgs.hasFlag(options::OPT_fcuda_short_ptr,
                          options::OPT_fno_cuda_short_ptr, false))
     CC1Args.append({"-mllvm", "--nvptx-short-ptr"});
 
+  if (CudaInstallation.version() >= CudaVersion::UNKNOWN)
+    CC1Args.push_back(DriverArgs.MakeArgString(
+        Twine("-target-sdk-version=") +
+        CudaVersionToString(CudaInstallation.version())));
+
   if (DeviceOffloadingKind == Action::OFK_OpenMP) {
     SmallVector<StringRef, 8> LibraryPaths;
-
     if (const Arg *A = DriverArgs.getLastArg(options::OPT_libomptarget_nvptx_path_EQ))
       LibraryPaths.push_back(A->getValue());
 
     // Add user defined library paths from LIBRARY_PATH.
     llvm::Optional<std::string> LibPath =
         llvm::sys::Process::GetEnv("LIBRARY_PATH");
     if (LibPath) {
       SmallVector<StringRef, 8> Frags;

clang/test/Driver/cuda-detect.cu

 // RUN:     -check-prefix NOCUDAINC -check-prefix NOLIBDEVICE
 
 // Verify that C++ include paths are passed for both host and device frontends.
 // RUN: %clang -### -no-canonical-prefixes -target x86_64-linux-gnu %s \
 // RUN: --stdlib=libstdc++ --sysroot=%S/Inputs/ubuntu_14.04_multiarch_tree2 \
 // RUN: --gcc-toolchain="" 2>&1 \
 // RUN: | FileCheck %s --check-prefix CHECK-CXXINCLUDE
 
+// Verify that CUDA SDK version is propagated to the CC1 compilations.
+// RUN: %clang -### -v -target x86_64-linux-gnu --cuda-gpu-arch=sm_50 \
+// RUN:   --cuda-path=%S/Inputs/CUDA_80/usr/local/cuda %s 2>&1 \
+// RUN:   | FileCheck %s -check-prefix CUDA80
+
+// Verify that if no version file is found, we report the default of 7.0.
+// RUN: %clang -### -v -target x86_64-linux-gnu --cuda-gpu-arch=sm_50 \
+// RUN:   --cuda-path=%S/Inputs/CUDA/usr/local/cuda %s 2>&1 \
+// RUN:   | FileCheck %s -check-prefix CUDA70
+
 // CHECK: Found CUDA installation: {{.*}}/Inputs/CUDA/usr/local/cuda
 // NO-LIBDEVICE: Found CUDA installation: {{.*}}/Inputs/CUDA-nolibdevice/usr/local/cuda
 // NOCUDA-NOT: Found CUDA installation:
 
 // MISSINGLIBDEVICE: error: cannot find libdevice for sm_20.
 
 // COMMON: "-triple" "nvptx-nvidia-cuda"
 // COMMON-SAME: "-fcuda-is-device"

 // because we must search the cuda include directory first.
 // CUDAINC-SAME: "-internal-externc-isystem"
 // COMMON-SAME: "-x" "cuda"
 // CHECK-CXXINCLUDE: clang{{.*}} "-cc1" "-triple" "nvptx64-nvidia-cuda"
 // CHECK-CXXINCLUDE-SAME: {{.*}}"-internal-isystem" "{{.+}}/include/c++/4.8"
 // CHECK-CXXINCLUDE: clang{{.*}} "-cc1" "-triple" "x86_64-unknown-linux-gnu"
 // CHECK-CXXINCLUDE-SAME: {{.*}}"-internal-isystem" "{{.+}}/include/c++/4.8"
 // CHECK-CXXINCLUDE: ld{{.*}}"
+
+// CUDA80: clang{{.*}} "-cc1" "-triple" "nvptx64-nvidia-cuda"
+// CUDA80-SAME: -target-sdk-version=8.0
+// CUDA80: clang{{.*}} "-cc1" "-triple" "x86_64-unknown-linux-gnu"
+// CUDA80-SAME: -target-sdk-version=8.0
+// CUDA80: ld{{.*}}"
+
+// CUDA70: clang{{.*}} "-cc1" "-triple" "nvptx64-nvidia-cuda"
+// CUDA70-SAME: -target-sdk-version=7.0
+// CUDA70: clang{{.*}} "-cc1" "-triple" "x86_64-unknown-linux-gnu"
+// CUDA70-SAME: -target-sdk-version=7.0
+// CUDA70: ld{{.*}}"