MC/CUDA: reduce verbosity for cu ctx doesn't exist

openucx · Sep 24, 2024 · e111225 · e111225
1 parent 8eccd57
commit e111225
Showing 1 changed file with 6 additions and 3 deletions.
diff --git a/src/components/mc/cuda/mc_cuda.c b/src/components/mc/cuda/mc_cuda.c
@@ -356,6 +356,7 @@ ucc_status_t ucc_mc_cuda_get_resources(ucc_mc_cuda_resources_t **resources)
     CUcontext cu_ctx;
     unsigned long long int cu_ctx_id;
     ucc_status_t status;
+    CUresult cu_status;
 
     status = CUDADRV_FUNC(cuCtxGetCurrent(&cu_ctx));
     if (ucc_unlikely(status != UCC_OK)) {
@@ -366,11 +367,13 @@ ucc_status_t ucc_mc_cuda_get_resources(ucc_mc_cuda_resources_t **resources)
 #if CUDA_VERSION < 12000
     cu_ctx_id = 1;
 #else
-    status = CUDADRV_FUNC(cuCtxGetId(cu_ctx, &cu_ctx_id));
-    if (ucc_unlikely(status != UCC_OK)) {
+    cu_status = cuCtxGetId(cu_ctx, &cu_ctx_id);
+    if (ucc_unlikely(cu_status != CUDA_SUCCESS)) {
         /* worakround for pytorch, progress thread doesn't have cuda context for GPU 0*/
+        /* cuda context might be not initialized, try to proceed with default device*/
         cu_ctx_id = 0x12345;
-        mc_debug(&ucc_mc_cuda.super, "failed to get currect CUDA context ID");
+        mc_debug(&ucc_mc_cuda.super,
+                 "failed to get current CUDA context ID (%d)", cu_status);
     }
 #endif