[CUDA] allow sm_50,52,53 GPUs

[tra]

Summary

LLVM accepts them since r233575.

lib/Basic/Targets.cpp

class NVPTXTargetInfo : public TargetInfo {
   // The GPU profiles supported by the NVPTX backend
   enum GPUKind {
     GK_NONE,
     GK_SM20,
     GK_SM21,
     GK_SM30,
     GK_SM35,
     GK_SM37,
+    GK_SM50,
+    GK_SM52,
+    GK_SM53,
   } GPU;
 
 public:
   NVPTXTargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts)
       : TargetInfo(Triple) {
     BigEndian = false;
     TLSSupported = false;
     LongWidth = LongAlign = 64;
if (Opts.CUDAIsDevice) {
         CUDAArchCode = "300";
         break;
       case GK_SM35:
         CUDAArchCode = "350";
         break;
       case GK_SM37:
         CUDAArchCode = "370";
         break;
+      case GK_SM50:
+        CUDAArchCode = "500";
+        break;
+      case GK_SM52:
+        CUDAArchCode = "520";
+        break;
+      case GK_SM53:
+        CUDAArchCode = "530";
+        break;
       default:
         llvm_unreachable("Unhandled target CPU");
       }
       Builder.defineMacro("__CUDA_ARCH__", CUDAArchCode);
     }
   }
   ArrayRef<Builtin::Info> getTargetBuiltins() const override {
     return llvm::makeArrayRef(BuiltinInfo,
public:
   }
   bool setCPU(const std::string &Name) override {
     GPU = llvm::StringSwitch<GPUKind>(Name)
               .Case("sm_20", GK_SM20)
               .Case("sm_21", GK_SM21)
               .Case("sm_30", GK_SM30)
               .Case("sm_35", GK_SM35)
               .Case("sm_37", GK_SM37)
+              .Case("sm_50", GK_SM50)
+              .Case("sm_52", GK_SM52)
+              .Case("sm_53", GK_SM53)
               .Default(GK_NONE);
 
     return GPU != GK_NONE;
   }
 };
 
 const Builtin::Info NVPTXTargetInfo::BuiltinInfo[] = {
 #define BUILTIN(ID, TYPE, ATTRS)                                               \

lib/Driver/ToolChains.cpp

for (llvm::sys::fs::directory_iterator LI(CudaLibDevicePath, EC), LE;
         CudaLibDeviceMap["sm_20"] = FilePath;
         CudaLibDeviceMap["sm_21"] = FilePath;
       } else if (GpuArch == "compute_30") {
         CudaLibDeviceMap["sm_30"] = FilePath;
         CudaLibDeviceMap["sm_32"] = FilePath;
       } else if (GpuArch == "compute_35") {
         CudaLibDeviceMap["sm_35"] = FilePath;
         CudaLibDeviceMap["sm_37"] = FilePath;
+      } else if (GpuArch == "compute_50") {
+        CudaLibDeviceMap["sm_50"] = FilePath;
+        CudaLibDeviceMap["sm_52"] = FilePath;
+        CudaLibDeviceMap["sm_53"] = FilePath;
       }
     }
 
     IsValid = true;
     break;
   }
 }
 

test/CodeGen/nvptx-cpus.c

 // RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_20 -O3 -S -o %t %s -emit-llvm
 // RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_21 -O3 -S -o %t %s -emit-llvm
 // RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_30 -O3 -S -o %t %s -emit-llvm
 // RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_35 -O3 -S -o %t %s -emit-llvm
 // RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_37 -O3 -S -o %t %s -emit-llvm
+// RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_50 -O3 -S -o %t %s -emit-llvm
+// RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_52 -O3 -S -o %t %s -emit-llvm
+// RUN: %clang_cc1 -triple nvptx-unknown-unknown -target-cpu sm_53 -O3 -S -o %t %s -emit-llvm
 
 // Make sure clang accepts all supported architectures.
 
 void foo(float* a,
          float* b) {
   a[0] = b[0];
 }