[CMake] Add CUDA Major Version Detection for Conditional Compilation (#197)

* [Feature] Add TMA Store Synchronization Support

- Introduce TMAStoreArrive and TMAStoreWait operations for CUDA TMA store synchronization
- Add new builtin operations in op/builtin.cc and op/builtin.h
- Implement TMAStoreSyncInjector to automatically inject TMA store synchronization calls
- Update CUDA codegen to support new TMA store synchronization intrinsics
- Add Python language bindings for new TMA store synchronization operations

* [CMake] Add CUDA Major Version Detection for Conditional Compilation

- Introduce CUDA_MAJOR_VERSION CMake variable to dynamically detect CUDA toolkit version
- Update runtime and transform files to use CUDA_MAJOR_VERSION for version-specific code paths
- Replace hardcoded __CUDACC_VER_MAJOR__ with dynamically set CUDA_MAJOR_VERSION
- Improve cross-version compatibility for CUDA-dependent code sections

CMakeLists.txt

@@ -156,6 +156,10 @@ if(USE_CUDA)
   endif()
 
   message(STATUS "CUDA Toolkit includes: ${CUDAToolkit_INCLUDE_DIRS}")
+  set(CUDA_MAJOR_VERSION ${CUDAToolkit_VERSION_MAJOR})
+  message(STATUS "Setting CUDA_MAJOR_VERSION=${CUDA_MAJOR_VERSION}")
+  add_compile_definitions(CUDA_MAJOR_VERSION=${CUDA_MAJOR_VERSION})
+  
   list(APPEND TILE_LANG_INCLUDES ${CUDAToolkit_INCLUDE_DIRS})
 endif(USE_CUDA)
 

src/runtime/runtime.cc

@@ -17,7 +17,7 @@ namespace tl {
 
 using namespace runtime;
 
-#if (__CUDACC_VER_MAJOR__ >= 12)
+#if (CUDA_MAJOR_VERSION >= 12)
 template <typename T> static std::string ArrayToStr(const T *ptr, size_t n) {
   std::stringstream ss;
   ss << "[";
@@ -203,7 +203,7 @@ TVM_REGISTER_GLOBAL(tvm_tensormap_create_im2col)
       }
       *ret = static_cast<int>(result);
     });
-#endif // (__CUDACC_VER_MAJOR__ >= 12)
+#endif // (CUDA_MAJOR_VERSION >= 12)
 
 } // namespace tl
 } // namespace tvm

src/runtime/runtime.h

@@ -13,12 +13,12 @@
 namespace tvm {
 namespace tl {
 
-#if (__CUDACC_VER_MAJOR__ >= 12)
+#if (CUDA_MAJOR_VERSION >= 12)
 constexpr const char *tvm_tensormap_create_tiled =
     "__tvm_tensormap_create_tiled";
 constexpr const char *tvm_tensormap_create_im2col =
     "__tvm_tensormap_create_im2col";
-#endif // (__CUDACC_VER_MAJOR__ >= 12)
+#endif // (CUDA_MAJOR_VERSION >= 12)
 } // namespace tl
 } // namespace tvm
 

src/transform/lower_hopper_intrin.cc

@@ -36,7 +36,7 @@ namespace tl {
 
 using namespace tir;
 
-#if (__CUDACC_VER_MAJOR__ >= 12)
+#if (CUDA_MAJOR_VERSION >= 12)
 class LowerHopperIntrin : public StmtExprMutator {
 public:
   static PrimFunc Substitute(PrimFunc &f) {
@@ -169,7 +169,7 @@ tvm::transform::Pass LowerHopperIntrin() {
 
 TVM_REGISTER_GLOBAL("tl.transform.LowerHopperIntrin")
     .set_body_typed(LowerHopperIntrin);
-#endif // (__CUDACC_VER_MAJOR__ >= 12)
+#endif // (CUDA_MAJOR_VERSION >= 12)
 
 } // namespace tl
 } // namespace tvm