cuDNNTransformTensor is used to relayout tensor on memory. I notice there are two kinds of memcpy API with sync and async version. I wonder is there similar operation in TransformTensor? Thanks.
Is cuDNNTransformTensor a sync or async kernel?
Someone told me cuMemcpy and suMemCpyAsync is a runtimeAPI, while cuDNNTransformTensor is a DNN API. There is no sync or async attribute for DNN API due to it is a kernel, this question could be closed.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.