Use context sync as well

src/targets/gpu/hip.cpp

@@ -10,7 +10,11 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
-void gpu_sync() { hipDeviceSynchronize(); }
+void gpu_sync() 
+{ 
+    hipDeviceSynchronize();
+    hipCtxSynchronize();
+}
 
 using hip_ptr = MIGRAPHX_MANAGE_PTR(void, hipFree);
 

src/targets/gpu/include/migraphx/gpu/context.hpp

@@ -34,8 +34,7 @@ struct hip_device
         static hip_stream_ptr create_stream()
         {
             hipStream_t result = nullptr;
-            auto status        = hipStreamCreate(&result);
-            // auto status        = hipStreamCreateWithFlags(&result, hipStreamNonBlocking);
+            auto status        = hipStreamCreateWithFlags(&result, hipStreamNonBlocking);
 
             if(status != hipSuccess)
                 MIGRAPHX_THROW("Failed to allocate stream");