[PATCH] D16097: [CUDA] Add explicit mapping from sm_XX to compute_YY.

[Justin Lebar via cfe-commits]

jlebar updated this revision to Diff 44585.
jlebar added a comment.

Update to latest set of architectures, according to ptxas --help from cuda 7.5.


http://reviews.llvm.org/D16097

Files:
  include/clang/Driver/Action.h
  lib/Driver/Action.cpp
  test/Driver/cuda-bad-arch.cu

Index: test/Driver/cuda-bad-arch.cu
===================================================================
--- test/Driver/cuda-bad-arch.cu
+++ test/Driver/cuda-bad-arch.cu
@@ -5,28 +5,16 @@
 
 // RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=compute_20 -c %s 2>&1 \
 // RUN: | FileCheck -check-prefix BAD %s
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=foo_20 -c %s 2>&1 \
-// RUN: | FileCheck -check-prefix BAD %s
 // RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm20 -c %s 2>&1 \
 // RUN: | FileCheck -check-prefix BAD %s
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_abc -c %s 2>&1 \
-// RUN: | FileCheck -check-prefix BAD %s
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_20a -c %s 2>&1 \
-// RUN: | FileCheck -check-prefix BAD %s
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_a20 -c %s 2>&1 \
-// RUN: | FileCheck -check-prefix BAD %s
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=ssm_20 -c %s 2>&1 \
-// RUN: | FileCheck -check-prefix BAD %s
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_ -c %s 2>&1 \
+// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_19 -c %s 2>&1 \
 // RUN: | FileCheck -check-prefix BAD %s
 
 // BAD: error: Unsupported CUDA gpu architecture
 
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_2 -c %s 2>&1 \
-// RUN: | FileCheck -check-prefix OK %s
 // RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_20 -c %s 2>&1 \
 // RUN: | FileCheck -check-prefix OK %s
-// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_999 -c %s 2>&1 \
+// RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_52 -c %s 2>&1 \
 // RUN: | FileCheck -check-prefix OK %s
 // RUN: %clang -### -target x86_64-linux-gnu -c %s 2>&1 \
 // RUN: | FileCheck -check-prefix OK %s
Index: lib/Driver/Action.cpp
===================================================================
--- lib/Driver/Action.cpp
+++ lib/Driver/Action.cpp
@@ -50,6 +50,24 @@
 BindArchAction::BindArchAction(Action *Input, const char *_ArchName)
     : Action(BindArchClass, Input), ArchName(_ArchName) {}
 
+// Converts CUDA GPU architecture, e.g. "sm_21", to its corresponding virtual
+// compute arch, e.g. "compute_20".  Returns null if the input arch is null or
+// doesn't match an existing arch.
+static const char* GpuArchToComputeName(const char *ArchName) {
+  if (!ArchName)
+    return nullptr;
+  llvm::StringRef A(ArchName);
+  if (A == "sm_20" || A == "sm_21") return "compute_20";
+  if (A == "sm_30") return "compute_30";
+  if (A == "sm_32") return "compute_32";
+  if (A == "sm_35") return "compute_35";
+  if (A == "sm_37") return "compute_37";
+  if (A == "sm_50") return "compute_50";
+  if (A == "sm_52") return "compute_52";
+  if (A == "sm_53") return "compute_53";
+  return nullptr;
+}
+
 void CudaDeviceAction::anchor() {}
 
 CudaDeviceAction::CudaDeviceAction(Action *Input, const char *ArchName,
@@ -59,9 +77,12 @@
   assert(IsValidGpuArchName(GpuArchName));
 }
 
+const char *CudaDeviceAction::getComputeArchName() const {
+  return GpuArchToComputeName(GpuArchName);
+}
+
 bool CudaDeviceAction::IsValidGpuArchName(llvm::StringRef ArchName) {
-  static llvm::Regex RE("^sm_[0-9]+$");
-  return RE.match(ArchName);
+  return GpuArchToComputeName(ArchName.data()) != nullptr;
 }
 
 void CudaHostAction::anchor() {}
Index: include/clang/Driver/Action.h
===================================================================
--- include/clang/Driver/Action.h
+++ include/clang/Driver/Action.h
@@ -146,6 +146,10 @@
   CudaDeviceAction(Action *Input, const char *ArchName, bool AtTopLevel);
 
   const char *getGpuArchName() const { return GpuArchName; }
+
+  /// Gets the compute_XX that corresponds to getGpuArchName().
+  const char *getComputeArchName() const;
+
   bool isAtTopLevel() const { return AtTopLevel; }
 
   static bool IsValidGpuArchName(llvm::StringRef ArchName);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D16097.44585.patch
Type: text/x-patch
Size: 3960 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20160112/4ea3a28a/attachment.bin>