[CUDA] Add support for sm101 and sm120 target architectures (llvm#127187)

Add support for sm101 and sm120 target architectures. It requires CUDA
12.8.

---------

Co-authored-by: Sebastian Jodlowski <[email protected]>

Author: 2 people

clang/include/clang/Basic/Cuda.h

@@ -44,6 +44,8 @@ enum class CudaVersion {
   CUDA_124,
   CUDA_125,
   CUDA_126,
+  CUDA_127,
+  CUDA_128,
   FULLY_SUPPORTED = CUDA_123,
   PARTIALLY_SUPPORTED =
       CUDA_126, // Partially supported. Proceed with a warning.
@@ -80,6 +82,11 @@ enum class OffloadArch {
   SM_90,
   SM_90a,
   SM_100,
+  SM_100a,
+  SM_101,
+  SM_101a,
+  SM_120,
+  SM_120a,
   GFX600,
   GFX601,
   GFX602,

clang/lib/Basic/Cuda.cpp

@@ -44,6 +44,8 @@ static const CudaVersionMapEntry CudaNameVersionMap[] = {
     CUDA_ENTRY(12, 4),
     CUDA_ENTRY(12, 5),
     CUDA_ENTRY(12, 6),
+    CUDA_ENTRY(12, 7),
+    CUDA_ENTRY(12, 8),
     {"", CudaVersion::NEW, llvm::VersionTuple(std::numeric_limits<int>::max())},
     {"unknown", CudaVersion::UNKNOWN, {}} // End of list tombstone.
 };
@@ -98,6 +100,11 @@ static const OffloadArchToStringMap arch_names[] = {
     SM(90),                          // Hopper
     SM(90a),                         // Hopper
     SM(100),                         // Blackwell
+    SM(100a),                        // Blackwell
+    SM(101),                         // Blackwell
+    SM(101a),                        // Blackwell
+    SM(120),                         // Blackwell
+    SM(120a),                        // Blackwell
     GFX(600),  // gfx600
     GFX(601),  // gfx601
     GFX(602),  // gfx602
@@ -226,8 +233,12 @@ CudaVersion MinVersionForOffloadArch(OffloadArch A) {
   case OffloadArch::SM_90a:
     return CudaVersion::CUDA_120;
   case OffloadArch::SM_100:
-    return CudaVersion::NEW; // TODO: use specific CUDA version once it's
-                             // public.
+  case OffloadArch::SM_100a:
+  case OffloadArch::SM_101:
+  case OffloadArch::SM_101a:
+  case OffloadArch::SM_120:
+  case OffloadArch::SM_120a:
+    return CudaVersion::CUDA_128;
   default:
     llvm_unreachable("invalid enum");
   }

clang/lib/Basic/Targets/NVPTX.cpp

@@ -180,7 +180,7 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
 
   if (Opts.CUDAIsDevice || Opts.OpenMPIsTargetDevice || !HostTarget) {
     // Set __CUDA_ARCH__ for the GPU specified.
-    std::string CUDAArchCode = [this] {
+    llvm::StringRef CUDAArchCode = [this] {
       switch (GPU) {
       case OffloadArch::GFX600:
       case OffloadArch::GFX601:
@@ -285,12 +285,27 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
         return "900";
       case OffloadArch::SM_100:
         return "1000";
+      case OffloadArch::SM_101:
+      case OffloadArch::SM_101a:
+         return "1010";
+      case OffloadArch::SM_120:
+      case OffloadArch::SM_120a:
+         return "1200";
       }
       llvm_unreachable("unhandled OffloadArch");
     }();
     Builder.defineMacro("__CUDA_ARCH__", CUDAArchCode);
-    if (GPU == OffloadArch::SM_90a)
-      Builder.defineMacro("__CUDA_ARCH_FEAT_SM90_ALL", "1");
+    switch(GPU) {
+      case OffloadArch::SM_90a:
+      case OffloadArch::SM_100a:
+      case OffloadArch::SM_101a:
+      case OffloadArch::SM_120a:
+        Builder.defineMacro("__CUDA_ARCH_FEAT_SM" + CUDAArchCode.drop_back() + "_ALL", "1");
+        break;
+      default:
+        // Do nothing if this is not an enhanced architecture.
+        break;
+    }
   }
 }
 

clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp

@@ -2277,6 +2277,11 @@ void CGOpenMPRuntimeGPU::processRequiresDirective(const OMPRequiresDecl *D) {
       case OffloadArch::SM_90:
       case OffloadArch::SM_90a:
       case OffloadArch::SM_100:
+      case OffloadArch::SM_100a:
+      case OffloadArch::SM_101:
+      case OffloadArch::SM_101a:
+      case OffloadArch::SM_120:
+      case OffloadArch::SM_120a:
       case OffloadArch::GFX600:
       case OffloadArch::GFX601:
       case OffloadArch::GFX602:

clang/test/Misc/target-invalid-cpu-note/nvptx.c

@@ -27,6 +27,11 @@
 // CHECK-SAME: {{^}}, sm_90
 // CHECK-SAME: {{^}}, sm_90a
 // CHECK-SAME: {{^}}, sm_100
+// CHECK-SAME: {{^}}, sm_100a
+// CHECK-SAME: {{^}}, sm_101
+// CHECK-SAME: {{^}}, sm_101a
+// CHECK-SAME: {{^}}, sm_120
+// CHECK-SAME: {{^}}, sm_120a
 // CHECK-SAME: {{^}}, gfx600
 // CHECK-SAME: {{^}}, gfx601
 // CHECK-SAME: {{^}}, gfx602