added stream synch after hipMemset (#609)

[ROCm/rccl commit: 06bce9d0c9]

projects/rccl/src/include/alloc.h

@@ -93,6 +93,7 @@ static ncclResult_t ncclCudaCallocDebug(const char *filefunc, int line, T** ptr,
   CUDACHECK(hipStreamDestroy(stream));
 #else
   CUDACHECK(hipMemset(*ptr, 0, nelem*sizeof(T)));
+  CUDACHECK(hipStreamSynchronize(NULL));
 #endif
   INFO(NCCL_ALLOC, "%s:%d Cuda Alloc Size %ld pointer %p", filefunc, line, nelem*sizeof(T), *ptr);
   int dev;