[CUDA] Add explicit mapping from sm_XX to compute_YY.

[jlebar]

Summary

This is used by D16082 when it invokes fatbinary.

[jlebar]

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

[echristo]

Seems reasonable. Why remove all of the tests though?

[jlebar]

In D16097#325093, @echristo wrote:

Seems reasonable. Why remove all of the tests though?

Previously we were checking that the arch was of the form /^sm_\d+$/ -- now we're checking that it's one of some list of acceptable things. I figured that the edge-casey tests aren't as useful in that latter case, but I can add them back if you like, not a big deal.

[echristo]

As long as the list of BAD things is somewhat conclusive I don't mind either way. Just making sure that this was the reason :)

-eric

[jlebar]

Use llvm::StringSwitch instead of a sequence of if statements.

[echristo]

Still good. much nicer now.

include/clang/Driver/Action.h

class CudaDeviceAction : public Action {
   /// True when action results are not consumed by the host action (e.g when
   /// -fsyntax-only or --cuda-device-only options are used).
   bool AtTopLevel;
 
 public:
   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);
 
   static bool classof(const Action *A) {
     return A->getKind() == CudaDeviceClass;
   }
 };

lib/Driver/Action.cpp

InputAction::InputAction(const Arg &_Input, types::ID _Type)
   : Action(InputClass, _Type), Input(_Input) {
 }
 
 void BindArchAction::anchor() {}
 
 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" || A == "sm_32") return "compute_30";
+  if (A == "sm_35") return "compute_35";
+  if (A == "sm_50") return "compute_50";
+  if (A == "sm_52") return "compute_52";
+  return nullptr;
+}
+
 void CudaDeviceAction::anchor() {}
 
 CudaDeviceAction::CudaDeviceAction(Action *Input, const char *ArchName,
                                    bool AtTopLevel)
     : Action(CudaDeviceClass, Input), GpuArchName(ArchName),
       AtTopLevel(AtTopLevel) {
   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() {}
 
 CudaHostAction::CudaHostAction(Action *Input, const ActionList &DeviceActions)
     : Action(CudaHostClass, Input), DeviceActions(DeviceActions) {}
 
 void JobAction::anchor() {}

test/Driver/cuda-bad-arch.cu

 // Checks errors generated by passing a bad value for --cuda-gpu-arch.
 // REQUIRES: clang-driver
 // REQUIRES: x86-registered-target
 // REQUIRES: nvptx-registered-target
 
 // 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
 
 // OK-NOT: error: Unsupported CUDA gpu architecture