[Mlir-commits] [mlir] [MLIR][GPU] Add support for non-portable cluster size attribute (PR #95545)

[Guray Ozen]

================
@@ -335,11 +335,13 @@ extern "C" MLIR_CUDA_WRAPPERS_EXPORT void mgpuSetDefaultDevice(int32_t device) {
 
 #if (CUDA_VERSION >= 12000)
 
-extern "C" MLIR_CUDA_WRAPPERS_EXPORT void mgpuLaunchClusterKernel(
-    CUfunction function, intptr_t clusterX, intptr_t clusterY,
-    intptr_t clusterZ, intptr_t gridX, intptr_t gridY, intptr_t gridZ,
-    intptr_t blockX, intptr_t blockY, intptr_t blockZ, int32_t smem,
-    CUstream stream, void **params, void **extra, size_t /*paramsCount*/) {
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void
+mgpuLaunchClusterKernel(CUfunction function, intptr_t clusterX,
+                        intptr_t clusterY, intptr_t clusterZ, intptr_t gridX,
+                        intptr_t gridY, intptr_t gridZ, intptr_t blockX,
+                        intptr_t blockY, intptr_t blockZ, int32_t smem,
+                        bool nonPortableClusterSize, CUstream stream,
----------------
grypp wrote:

I understand that it's low cost passing a bool here. But if we want to support other arguments, we cannot pass all of them as bool. We need alternative solution. 
This isn't blocking this PR. 

https://github.com/llvm/llvm-project/pull/95545