[CUDA] Add getTargetFeatures for the NVPTX toolchain

The NVPTX toolchain uses target features to determine the PTX version to
use. However this isn't exposed externally like most other toolchain
specific target features are. Add this functionaliy in preparation for
using it in for OpenMP offloading.

Reviewed By: jdoerfert, tra

Differential Revision: https://reviews.llvm.org/D122089

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index d942ec2..5bf6b24 100644 (file)
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -369,6 +369,10 @@ static void getTargetFeatures(const Driver &D, const llvm::Triple &Triple,
   case llvm::Triple::amdgcn:
     amdgpu::getAMDGPUTargetFeatures(D, Triple, Args, Features);
     break;
+  case llvm::Triple::nvptx:
+  case llvm::Triple::nvptx64:
+    NVPTX::getNVPTXTargetFeatures(D, Triple, Args, Features);
+    break;
   case llvm::Triple::m68k:
     m68k::getM68kTargetFeatures(D, Triple, Args, Features);
     break;

diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
index d853278..b162cec 100644 (file)
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -630,6 +630,43 @@ void NVPTX::OpenMPLinker::ConstructJob(Compilation &C, const JobAction &JA,
       Exec, CmdArgs, Inputs, Output));
 }
 
+void NVPTX::getNVPTXTargetFeatures(const Driver &D, const llvm::Triple &Triple,
+                                   const llvm::opt::ArgList &Args,
+                                   std::vector<StringRef> &Features,
+                                   Optional<clang::CudaVersion> CudaVersion) {
+  if (!CudaVersion) {
+    CudaInstallationDetector CudaInstallation(D, Triple, Args);
+    CudaVersion = CudaInstallation.version();
+  }
+
+  // New CUDA versions often introduce new instructions that are only supported
+  // by new PTX version, so we need to raise PTX level to enable them in NVPTX
+  // back-end.
+  const char *PtxFeature = nullptr;
+  switch (*CudaVersion) {
+#define CASE_CUDA_VERSION(CUDA_VER, PTX_VER)                                   \
+  case CudaVersion::CUDA_##CUDA_VER:                                           \
+    PtxFeature = "+ptx" #PTX_VER;                                              \
+    break;
+    CASE_CUDA_VERSION(115, 75);
+    CASE_CUDA_VERSION(114, 74);
+    CASE_CUDA_VERSION(113, 73);
+    CASE_CUDA_VERSION(112, 72);
+    CASE_CUDA_VERSION(111, 71);
+    CASE_CUDA_VERSION(110, 70);
+    CASE_CUDA_VERSION(102, 65);
+    CASE_CUDA_VERSION(101, 64);
+    CASE_CUDA_VERSION(100, 63);
+    CASE_CUDA_VERSION(92, 61);
+    CASE_CUDA_VERSION(91, 61);
+    CASE_CUDA_VERSION(90, 60);
+#undef CASE_CUDA_VERSION
+  default:
+    PtxFeature = "+ptx42";
+  }
+  Features.push_back(PtxFeature);
+}
+
 /// CUDA toolchain.  Our assembler is ptxas, and our "linker" is fatbinary,
 /// which isn't properly a linker but nonetheless performs the step of stitching
 /// together object files from the assembler into a single blob.
@@ -701,32 +738,11 @@ void CudaToolChain::addClangTargetOptions(
 
   clang::CudaVersion CudaInstallationVersion = CudaInstallation.version();
 
-  // New CUDA versions often introduce new instructions that are only supported
-  // by new PTX version, so we need to raise PTX level to enable them in NVPTX
-  // back-end.
-  const char *PtxFeature = nullptr;
-  switch (CudaInstallationVersion) {
-#define CASE_CUDA_VERSION(CUDA_VER, PTX_VER)                                   \
-  case CudaVersion::CUDA_##CUDA_VER:                                           \
-    PtxFeature = "+ptx" #PTX_VER;                                              \
-    break;
-    CASE_CUDA_VERSION(115, 75);
-    CASE_CUDA_VERSION(114, 74);
-    CASE_CUDA_VERSION(113, 73);
-    CASE_CUDA_VERSION(112, 72);
-    CASE_CUDA_VERSION(111, 71);
-    CASE_CUDA_VERSION(110, 70);
-    CASE_CUDA_VERSION(102, 65);
-    CASE_CUDA_VERSION(101, 64);
-    CASE_CUDA_VERSION(100, 63);
-    CASE_CUDA_VERSION(92, 61);
-    CASE_CUDA_VERSION(91, 61);
-    CASE_CUDA_VERSION(90, 60);
-#undef CASE_CUDA_VERSION
-  default:
-    PtxFeature = "+ptx42";
-  }
-  CC1Args.append({"-target-feature", PtxFeature});
+  std::vector<StringRef> Features;
+  NVPTX::getNVPTXTargetFeatures(getDriver(), getTriple(), DriverArgs, Features,
+                                CudaInstallationVersion);
+  for (StringRef PtxFeature : Features)
+    CC1Args.append({"-target-feature", DriverArgs.MakeArgString(PtxFeature)});
   if (DriverArgs.hasFlag(options::OPT_fcuda_short_ptr,
                          options::OPT_fno_cuda_short_ptr, false))
     CC1Args.append({"-mllvm", "--nvptx-short-ptr"});

diff --git a/clang/lib/Driver/ToolChains/Cuda.h b/clang/lib/Driver/ToolChains/Cuda.h
index a7e6e84..f4f924f 100644 (file)
--- a/clang/lib/Driver/ToolChains/Cuda.h
+++ b/clang/lib/Driver/ToolChains/Cuda.h
@@ -124,6 +124,11 @@ class LLVM_LIBRARY_VISIBILITY OpenMPLinker : public Tool {
                      const char *LinkingOutput) const override;
 };
 
+void getNVPTXTargetFeatures(const Driver &D, const llvm::Triple &Triple,
+                            const llvm::opt::ArgList &Args,
+                            std::vector<StringRef> &Features,
+                            Optional<clang::CudaVersion> CudaVersion = None);
+
 } // end namespace NVPTX
 } // end namespace tools