Match cuBLAS sync behavior (#169)

LLNL · Nov 7, 2023 · d883ac3 · d883ac3
1 parent bc4ea53
commit d883ac3
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/src/hydrogen/device/rocBLAS_API.cpp b/src/hydrogen/device/rocBLAS_API.cpp
@@ -75,6 +75,10 @@ class ResultMgr
                 H_CHECK_HIP(hipFreeAsync(device_, stream));
 #endif // HYDROGEN_HAVE_CUB
 
+                // Sync stream to match cuBLAS behavior (cuBLAS docs here:
+                // https://docs.nvidia.com/cuda/cublas/#scalar-parameters)
+                H_CHECK_HIP(hipStreamSynchronize(stream));
+
                 // Reset pointer mode
                 H_CHECK_ROCBLAS(rocblas_set_pointer_mode(handle_, rocblas_pointer_mode_host));
             }