[CUDA] add support for targeting sm_103/sm_121 with CUDA-12.9 (#151587)

Author: Artem-B

clang/include/clang/Basic/BuiltinsNVPTX.td

@@ -21,13 +21,17 @@ class SM<string version, list<SMFeatures> newer_list> : SMFeatures {
                         !strconcat(f, "|", newer.Features));
 }
 
+let Features = "sm_121a" in def SM_121a : SMFeatures;
 let Features = "sm_120a" in def SM_120a : SMFeatures;
+let Features = "sm_103a" in def SM_103a : SMFeatures;
 let Features = "sm_101a" in def SM_101a : SMFeatures;
 let Features = "sm_100a" in def SM_100a : SMFeatures;
 let Features = "sm_90a" in def SM_90a : SMFeatures;
 
-def SM_120 : SM<"120", [SM_120a]>;
-def SM_101 : SM<"101", [SM_101a, SM_120]>;
+def SM_121 : SM<"121", [SM_121a]>;
+def SM_120 : SM<"120", [SM_120a, SM_121]>;
+def SM_103 : SM<"103", [SM_103a, SM_120]>;
+def SM_101 : SM<"101", [SM_101a, SM_103]>;
 def SM_100 : SM<"100", [SM_100a, SM_101]>;
 def SM_90 : SM<"90", [SM_90a, SM_100]>;
 def SM_89 : SM<"89", [SM_90]>;
@@ -50,8 +54,9 @@ class PTX<string version, PTXFeatures newer> : PTXFeatures {
   let Features = !strconcat("ptx", version, "|", newer.Features);
 }
 
-let Features = "ptx87" in def PTX87 : PTXFeatures;
+let Features = "ptx88" in def PTX88 : PTXFeatures;
 
+def PTX87 : PTX<"87", PTX88>;
 def PTX86 : PTX<"86", PTX87>;
 def PTX85 : PTX<"85", PTX86>;
 def PTX84 : PTX<"84", PTX85>;

clang/include/clang/Basic/Cuda.h

@@ -47,9 +47,10 @@ enum class CudaVersion {
   CUDA_125,
   CUDA_126,
   CUDA_128,
-  FULLY_SUPPORTED = CUDA_123,
+  CUDA_129,
+  FULLY_SUPPORTED = CUDA_128,
   PARTIALLY_SUPPORTED =
-      CUDA_128, // Partially supported. Proceed with a warning.
+      CUDA_129, // Partially supported. Proceed with a warning.
   NEW = 10000,  // Too new. Issue a warning, but allow using it.
 };
 const char *CudaVersionToString(CudaVersion V);

clang/include/clang/Basic/OffloadArch.h

@@ -45,8 +45,12 @@ enum class OffloadArch {
   SM_100a,
   SM_101,
   SM_101a,
+  SM_103,
+  SM_103a,
   SM_120,
   SM_120a,
+  SM_121,
+  SM_121a,
   GFX600,
   GFX601,
   GFX602,

clang/lib/Basic/Cuda.cpp

@@ -44,6 +44,7 @@ static const CudaVersionMapEntry CudaNameVersionMap[] = {
     CUDA_ENTRY(12, 5),
     CUDA_ENTRY(12, 6),
     CUDA_ENTRY(12, 8),
+    CUDA_ENTRY(12, 9),
     {"", CudaVersion::NEW, llvm::VersionTuple(std::numeric_limits<int>::max())},
     {"unknown", CudaVersion::UNKNOWN, {}} // End of list tombstone.
 };
@@ -119,6 +120,11 @@ CudaVersion MinVersionForOffloadArch(OffloadArch A) {
   case OffloadArch::SM_120:
   case OffloadArch::SM_120a:
     return CudaVersion::CUDA_128;
+  case OffloadArch::SM_103:
+  case OffloadArch::SM_103a:
+  case OffloadArch::SM_121:
+  case OffloadArch::SM_121a:
+    return CudaVersion::CUDA_129;
   default:
     llvm_unreachable("invalid enum");
   }

clang/lib/Basic/OffloadArch.cpp

@@ -33,8 +33,12 @@ static const OffloadArchToStringMap ArchNames[] = {
     SM(100a),                        // Blackwell
     SM(101),                         // Blackwell
     SM(101a),                        // Blackwell
+    SM(103),                         // Blackwell
+    SM(103a),                        // Blackwell
     SM(120),                         // Blackwell
     SM(120a),                        // Blackwell
+    SM(121),                         // Blackwell
+    SM(121a),                        // Blackwell
     GFX(600),  // gfx600
     GFX(601),  // gfx601
     GFX(602),  // gfx602

clang/lib/Basic/Targets/NVPTX.cpp

@@ -295,10 +295,16 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
         return "1000";
       case OffloadArch::SM_101:
       case OffloadArch::SM_101a:
-         return "1010";
+        return "1010";
+      case OffloadArch::SM_103:
+      case OffloadArch::SM_103a:
+        return "1030";
       case OffloadArch::SM_120:
       case OffloadArch::SM_120a:
-         return "1200";
+        return "1200";
+      case OffloadArch::SM_121:
+      case OffloadArch::SM_121a:
+        return "1210";
       }
       llvm_unreachable("unhandled OffloadArch");
     }();
@@ -307,7 +313,9 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
       case OffloadArch::SM_90a:
       case OffloadArch::SM_100a:
       case OffloadArch::SM_101a:
+      case OffloadArch::SM_103a:
       case OffloadArch::SM_120a:
+      case OffloadArch::SM_121a:
         Builder.defineMacro("__CUDA_ARCH_FEAT_SM" + CUDAArchCode.drop_back() + "_ALL", "1");
         break;
       default:

clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp

@@ -2278,8 +2278,12 @@ void CGOpenMPRuntimeGPU::processRequiresDirective(const OMPRequiresDecl *D) {
       case OffloadArch::SM_100a:
       case OffloadArch::SM_101:
       case OffloadArch::SM_101a:
+      case OffloadArch::SM_103:
+      case OffloadArch::SM_103a:
       case OffloadArch::SM_120:
       case OffloadArch::SM_120a:
+      case OffloadArch::SM_121:
+      case OffloadArch::SM_121a:
       case OffloadArch::GFX600:
       case OffloadArch::GFX601:
       case OffloadArch::GFX602:

clang/lib/Driver/ToolChains/Cuda.cpp

@@ -88,6 +88,8 @@ CudaVersion getCudaVersion(uint32_t raw_version) {
     return CudaVersion::CUDA_126;
   if (raw_version < 12090)
     return CudaVersion::CUDA_128;
+  if (raw_version < 13000)
+    return CudaVersion::CUDA_129;
   return CudaVersion::NEW;
 }
 
@@ -683,6 +685,7 @@ void NVPTX::getNVPTXTargetFeatures(const Driver &D, const llvm::Triple &Triple,
   case CudaVersion::CUDA_##CUDA_VER:                                           \
     PtxFeature = "+ptx" #PTX_VER;                                              \
     break;
+    CASE_CUDA_VERSION(129, 88);
     CASE_CUDA_VERSION(128, 87);
     CASE_CUDA_VERSION(126, 85);
     CASE_CUDA_VERSION(125, 85);

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

@@ -30,8 +30,12 @@
 // CHECK-SAME: {{^}}, sm_100a
 // CHECK-SAME: {{^}}, sm_101
 // CHECK-SAME: {{^}}, sm_101a
+// CHECK-SAME: {{^}}, sm_103
+// CHECK-SAME: {{^}}, sm_103a
 // CHECK-SAME: {{^}}, sm_120
 // CHECK-SAME: {{^}}, sm_120a
+// CHECK-SAME: {{^}}, sm_121
+// CHECK-SAME: {{^}}, sm_121a
 // CHECK-SAME: {{^}}, gfx600
 // CHECK-SAME: {{^}}, gfx601
 // CHECK-SAME: {{^}}, gfx602