Disable CUDA Graph for MUSA

Author: gaugarg-nv

ggml/src/ggml-cuda/common.cuh

@@ -678,7 +678,7 @@ struct ggml_tensor_extra_gpu {
 };
 
 
-#if (defined(GGML_CUDA_USE_GRAPHS)) || defined(GGML_HIP_GRAPHS)
+#if (defined(GGML_CUDA_USE_GRAPHS) || defined(GGML_HIP_GRAPHS))
 #define USE_CUDA_GRAPH
 #endif
 

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -2610,14 +2610,14 @@ static bool is_cuda_graph_update_required(ggml_backend_cuda_context * cuda_ctx,
 
 static void update_cuda_graph_executable(ggml_backend_cuda_context * cuda_ctx) {
 
-#if (CUDART_VERSION < 12000 || defined(__HIP_PLATFORM_AMD__))
+#if CUDART_VERSION >= 12000
+    cudaGraphExecUpdateResultInfo result_info;
+    cudaError_t stat = cudaGraphExecUpdate(cuda_ctx->cuda_graph->instance, cuda_ctx->cuda_graph->graph, &result_info);
+#else
     cudaGraphNode_t errorNode;
     cudaGraphExecUpdateResult result_info;
     cudaError_t stat = cudaGraphExecUpdate(cuda_ctx->cuda_graph->instance, cuda_ctx->cuda_graph->graph, &errorNode, &result_info);
-#else
-    cudaGraphExecUpdateResultInfo result_info;
-    cudaError_t stat = cudaGraphExecUpdate(cuda_ctx->cuda_graph->instance, cuda_ctx->cuda_graph->graph, &result_info);
-#endif // (CUDART_VERSION < 12000 || defined(__HIP_PLATFORM_AMD__))
+#endif // CUDART_VERSION >= 12000
 
     if (stat == cudaErrorGraphExecUpdateFailure) {
 #ifndef NDEBUG

ggml/src/ggml-cuda/vendors/musa.h

@@ -119,7 +119,7 @@
 #define cudaGraphExecDestroy musaGraphExecDestroy
 #define cudaGraphExec_t musaGraphExec_t
 #define cudaGraphExecUpdate musaGraphExecUpdate
-#define cudaGraphExecUpdateResultInfo musaGraphExecUpdateResult
+#define cudaGraphExecUpdateResult musaGraphExecUpdateResult
 #define cudaGraphGetNodes musaGraphGetNodes
 #define cudaGraphInstantiate musaGraphInstantiate
 #define cudaGraphKernelNodeGetParams musaGraphKernelNodeGetParams
@@ -132,6 +132,7 @@
 #define cudaGraph_t musaGraph_t
 #define cudaKernelNodeParams musaKernelNodeParams
 #define cudaStreamCaptureModeRelaxed musaStreamCaptureModeRelaxed
+#define cudaStreamBeginCapture musaStreamBeginCapture
 #define cudaStreamEndCapture musaStreamEndCapture
 
 typedef mt_bfloat16 nv_bfloat16;

ggml/src/ggml-musa/CMakeLists.txt

@@ -67,10 +67,6 @@ if (MUSAToolkit_FOUND)
     add_compile_definitions(GGML_USE_MUSA)
     add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${GGML_CUDA_PEER_MAX_BATCH_SIZE})
 
-    if (GGML_CUDA_GRAPHS)
-        add_compile_definitions(GGML_CUDA_USE_GRAPHS)
-    endif()
-
     if (GGML_CUDA_FORCE_MMQ)
         add_compile_definitions(GGML_CUDA_FORCE_MMQ)
     endif()