[clang] [llvm] [CUDA] Add support for CUDA-12.6 and sm_100 (PR #97402)

Sergey Kozub via llvm-commits llvm-commits at lists.llvm.org
Tue Jul 2 02:49:36 PDT 2024


https://github.com/sergey-kozub created https://github.com/llvm/llvm-project/pull/97402

Adds support for sm_100 (Blackwell), similar to https://github.com/llvm/llvm-project/pull/74895

One important aspect is that `sm_100` is not compatible with `sm_90a`, only with `sm_90` - note the defines in "BuiltinsNVPTX.def"

>From 02e1acff6ffa1ddc3a26b0edc7e89923ac38978f Mon Sep 17 00:00:00 2001
From: Sergey Kozub <skozub at nvidia.com>
Date: Tue, 2 Jul 2024 02:44:56 -0700
Subject: [PATCH] [CUDA] Add support for CUDA-12.6 and sm_100

---
 clang/docs/ReleaseNotes.rst                 |  3 ++-
 clang/include/clang/Basic/BuiltinsNVPTX.def | 10 ++++++++--
 clang/include/clang/Basic/Cuda.h            |  4 +++-
 clang/lib/Basic/Cuda.cpp                    |  4 ++++
 clang/lib/Basic/Targets/NVPTX.cpp           |  2 ++
 clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp    |  1 +
 clang/lib/Driver/ToolChains/Cuda.cpp        |  3 +++
 clang/test/Misc/target-invalid-cpu-note.c   |  2 +-
 llvm/lib/Target/NVPTX/NVPTX.td              |  5 +++--
 9 files changed, 27 insertions(+), 7 deletions(-)

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index c720e47dbe35b..3c10ee51550d9 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -1057,7 +1057,8 @@ CUDA/HIP Language Changes
 
 CUDA Support
 ^^^^^^^^^^^^
-- Clang now supports CUDA SDK up to 12.5
+- Clang now supports CUDA SDK up to 12.6
+- Added support for sm_100
 
 AIX Support
 ^^^^^^^^^^^
diff --git a/clang/include/clang/Basic/BuiltinsNVPTX.def b/clang/include/clang/Basic/BuiltinsNVPTX.def
index 504314d8d96e9..3f383bc89ee70 100644
--- a/clang/include/clang/Basic/BuiltinsNVPTX.def
+++ b/clang/include/clang/Basic/BuiltinsNVPTX.def
@@ -27,8 +27,10 @@
 #pragma push_macro("SM_89")
 #pragma push_macro("SM_90")
 #pragma push_macro("SM_90a")
+#pragma push_macro("SM_100")
+#define SM_100 "sm_100"
 #define SM_90a "sm_90a"
-#define SM_90 "sm_90|" SM_90a
+#define SM_90 "sm_90|" SM_90a "|" SM_100
 #define SM_89 "sm_89|" SM_90
 #define SM_87 "sm_87|" SM_89
 #define SM_86 "sm_86|" SM_87
@@ -63,7 +65,9 @@
 #pragma push_macro("PTX83")
 #pragma push_macro("PTX84")
 #pragma push_macro("PTX85")
-#define PTX85 "ptx85"
+#pragma push_macro("PTX86")
+#define PTX86 "ptx86"
+#define PTX85 "ptx85|" PTX86
 #define PTX84 "ptx84|" PTX85
 #define PTX83 "ptx83|" PTX84
 #define PTX82 "ptx82|" PTX83
@@ -1075,6 +1079,7 @@ TARGET_BUILTIN(__nvvm_getctarank_shared_cluster, "iv*3", "", AND(SM_90,PTX78))
 #pragma pop_macro("SM_89")
 #pragma pop_macro("SM_90")
 #pragma pop_macro("SM_90a")
+#pragma pop_macro("SM_100")
 #pragma pop_macro("PTX42")
 #pragma pop_macro("PTX60")
 #pragma pop_macro("PTX61")
@@ -1097,3 +1102,4 @@ TARGET_BUILTIN(__nvvm_getctarank_shared_cluster, "iv*3", "", AND(SM_90,PTX78))
 #pragma pop_macro("PTX83")
 #pragma pop_macro("PTX84")
 #pragma pop_macro("PTX85")
+#pragma pop_macro("PTX86")
diff --git a/clang/include/clang/Basic/Cuda.h b/clang/include/clang/Basic/Cuda.h
index 83699f8897f66..a18e62620dd5d 100644
--- a/clang/include/clang/Basic/Cuda.h
+++ b/clang/include/clang/Basic/Cuda.h
@@ -43,9 +43,10 @@ enum class CudaVersion {
   CUDA_123,
   CUDA_124,
   CUDA_125,
+  CUDA_126,
   FULLY_SUPPORTED = CUDA_123,
   PARTIALLY_SUPPORTED =
-      CUDA_125, // Partially supported. Proceed with a warning.
+      CUDA_126, // Partially supported. Proceed with a warning.
   NEW = 10000,  // Too new. Issue a warning, but allow using it.
 };
 const char *CudaVersionToString(CudaVersion V);
@@ -78,6 +79,7 @@ enum class OffloadArch {
   SM_89,
   SM_90,
   SM_90a,
+  SM_100,
   GFX600,
   GFX601,
   GFX602,
diff --git a/clang/lib/Basic/Cuda.cpp b/clang/lib/Basic/Cuda.cpp
index faf3878f064d2..72d9bd89c36e7 100644
--- a/clang/lib/Basic/Cuda.cpp
+++ b/clang/lib/Basic/Cuda.cpp
@@ -43,6 +43,7 @@ static const CudaVersionMapEntry CudaNameVersionMap[] = {
     CUDA_ENTRY(12, 3),
     CUDA_ENTRY(12, 4),
     CUDA_ENTRY(12, 5),
+    CUDA_ENTRY(12, 6),
     {"", CudaVersion::NEW, llvm::VersionTuple(std::numeric_limits<int>::max())},
     {"unknown", CudaVersion::UNKNOWN, {}} // End of list tombstone.
 };
@@ -96,6 +97,7 @@ static const OffloadArchToStringMap arch_names[] = {
     SM(89),                          // Ada Lovelace
     SM(90),                          // Hopper
     SM(90a),                         // Hopper
+    SM(100),                         // Blackwell
     GFX(600),  // gfx600
     GFX(601),  // gfx601
     GFX(602),  // gfx602
@@ -221,6 +223,8 @@ CudaVersion MinVersionForOffloadArch(OffloadArch A) {
     return CudaVersion::CUDA_118;
   case OffloadArch::SM_90a:
     return CudaVersion::CUDA_120;
+  case OffloadArch::SM_100:
+    return CudaVersion::CUDA_126;
   default:
     llvm_unreachable("invalid enum");
   }
diff --git a/clang/lib/Basic/Targets/NVPTX.cpp b/clang/lib/Basic/Targets/NVPTX.cpp
index 43b653dc52ce0..88a0dbde52d52 100644
--- a/clang/lib/Basic/Targets/NVPTX.cpp
+++ b/clang/lib/Basic/Targets/NVPTX.cpp
@@ -281,6 +281,8 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
       case OffloadArch::SM_90:
       case OffloadArch::SM_90a:
         return "900";
+      case OffloadArch::SM_100:
+        return "1000";
       }
       llvm_unreachable("unhandled OffloadArch");
     }();
diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
index f5bd4a141cc2d..198b64bd7e1d4 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -2276,6 +2276,7 @@ void CGOpenMPRuntimeGPU::processRequiresDirective(const OMPRequiresDecl *D) {
       case OffloadArch::SM_89:
       case OffloadArch::SM_90:
       case OffloadArch::SM_90a:
+      case OffloadArch::SM_100:
       case OffloadArch::GFX600:
       case OffloadArch::GFX601:
       case OffloadArch::GFX602:
diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
index 08a4633902654..81a3703dc0da7 100644
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -86,6 +86,8 @@ CudaVersion getCudaVersion(uint32_t raw_version) {
     return CudaVersion::CUDA_124;
   if (raw_version < 12060)
     return CudaVersion::CUDA_125;
+  if (raw_version < 12070)
+    return CudaVersion::CUDA_126;
   return CudaVersion::NEW;
 }
 
@@ -692,6 +694,7 @@ void NVPTX::getNVPTXTargetFeatures(const Driver &D, const llvm::Triple &Triple,
   case CudaVersion::CUDA_##CUDA_VER:                                           \
     PtxFeature = "+ptx" #PTX_VER;                                              \
     break;
+    CASE_CUDA_VERSION(126, 86);
     CASE_CUDA_VERSION(125, 85);
     CASE_CUDA_VERSION(124, 84);
     CASE_CUDA_VERSION(123, 83);
diff --git a/clang/test/Misc/target-invalid-cpu-note.c b/clang/test/Misc/target-invalid-cpu-note.c
index a5f9ffa21220a..92b28c9a16520 100644
--- a/clang/test/Misc/target-invalid-cpu-note.c
+++ b/clang/test/Misc/target-invalid-cpu-note.c
@@ -29,7 +29,7 @@
 
 // RUN: not %clang_cc1 -triple nvptx--- -target-cpu not-a-cpu -fsyntax-only %s 2>&1 | FileCheck %s --check-prefix NVPTX
 // NVPTX: error: unknown target CPU 'not-a-cpu'
-// NVPTX-NEXT: note: valid target CPU values are: sm_20, sm_21, sm_30, sm_32, sm_35, sm_37, sm_50, sm_52, sm_53, sm_60, sm_61, sm_62, sm_70, sm_72, sm_75, sm_80, sm_86, sm_87, sm_89, sm_90, sm_90a, gfx600, gfx601, gfx602, gfx700, gfx701, gfx702, gfx703, gfx704, gfx705, gfx801, gfx802, gfx803, gfx805, gfx810, gfx9-generic, gfx900, gfx902, gfx904, gfx906, gfx908, gfx909, gfx90a, gfx90c, gfx940, gfx941, gfx942, gfx10-1-generic, gfx1010, gfx1011, gfx1012, gfx1013, gfx10-3-generic, gfx1030, gfx1031, gfx1032, gfx1033, gfx1034, gfx1035, gfx1036, gfx11-generic, gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx12-generic, gfx1200, gfx1201, amdgcnspirv{{$}}
+// NVPTX-NEXT: note: valid target CPU values are: sm_20, sm_21, sm_30, sm_32, sm_35, sm_37, sm_50, sm_52, sm_53, sm_60, sm_61, sm_62, sm_70, sm_72, sm_75, sm_80, sm_86, sm_87, sm_89, sm_90, sm_90a, sm_100, gfx600, gfx601, gfx602, gfx700, gfx701, gfx702, gfx703, gfx704, gfx705, gfx801, gfx802, gfx803, gfx805, gfx810, gfx9-generic, gfx900, gfx902, gfx904, gfx906, gfx908, gfx909, gfx90a, gfx90c, gfx940, gfx941, gfx942, gfx10-1-generic, gfx1010, gfx1011, gfx1012, gfx1013, gfx10-3-generic, gfx1030, gfx1031, gfx1032, gfx1033, gfx1034, gfx1035, gfx1036, gfx11-generic, gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx12-generic, gfx1200, gfx1201, amdgcnspirv{{$}}
 
 // RUN: not %clang_cc1 -triple r600--- -target-cpu not-a-cpu -fsyntax-only %s 2>&1 | FileCheck %s --check-prefix R600
 // R600: error: unknown target CPU 'not-a-cpu'
diff --git a/llvm/lib/Target/NVPTX/NVPTX.td b/llvm/lib/Target/NVPTX/NVPTX.td
index bb4549a5e6078..9af8715ef52ae 100644
--- a/llvm/lib/Target/NVPTX/NVPTX.td
+++ b/llvm/lib/Target/NVPTX/NVPTX.td
@@ -35,14 +35,14 @@ class FeaturePTX<int version>:
                     "Use PTX version " # version>;
 
 foreach sm = [20, 21, 30, 32, 35, 37, 50, 52, 53,
-              60, 61, 62, 70, 72, 75, 80, 86, 87, 89, 90] in
+              60, 61, 62, 70, 72, 75, 80, 86, 87, 89, 90, 100] in
   def SM#sm: FeatureSM<""#sm, !mul(sm, 10)>;
 
 def SM90a: FeatureSM<"90a", 901>;
 
 foreach version = [32, 40, 41, 42, 43, 50, 60, 61, 62, 63, 64, 65,
                    70, 71, 72, 73, 74, 75, 76, 77, 78,
-                   80, 81, 82, 83, 84, 85] in
+                   80, 81, 82, 83, 84, 85, 86] in
   def PTX#version: FeaturePTX<version>;
 
 //===----------------------------------------------------------------------===//
@@ -73,6 +73,7 @@ def : Proc<"sm_87", [SM87, PTX74]>;
 def : Proc<"sm_89", [SM89, PTX78]>;
 def : Proc<"sm_90", [SM90, PTX78]>;
 def : Proc<"sm_90a", [SM90a, PTX80]>;
+def : Proc<"sm_100", [SM100, PTX86]>;
 
 def NVPTXInstrInfo : InstrInfo {
 }



More information about the llvm-commits mailing list