diff --git a/paddle/phi/core/memory/memcpy.cc b/paddle/phi/core/memory/memcpy.cc index 9d0becd7df7983..741f0441183bdb 100644 --- a/paddle/phi/core/memory/memcpy.cc +++ b/paddle/phi/core/memory/memcpy.cc @@ -399,6 +399,7 @@ void Copy(phi::XPUPinnedPlace dst_place, reinterpret_cast(stream)); } else { + cudaDeviceSynchronize(); phi::RecordEvent record_event( "cudaMemcpy:XPU->XPUPinned", phi::TracerEventType::UserDefined, 1); cudaMemcpy(dst, src, num, cudaMemcpyDeviceToHost); @@ -435,6 +436,7 @@ void Copy(phi::XPUPlace dst_place, cudaMemcpyHostToDevice, reinterpret_cast(stream)); } else { + cudaDeviceSynchronize(); phi::RecordEvent record_event( "cudaMemcpy:XPUPinned->XPU", phi::TracerEventType::UserDefined, 1); cudaMemcpy(dst, src, num, cudaMemcpyHostToDevice);