TaoLv commented on a change in pull request #13749: Add NHWC layout support to Pooling (cpu, gpu cuda, gpu cuDNN) URL: https://github.com/apache/incubator-mxnet/pull/13749#discussion_r253267148
########## File path: src/operator/nn/mkldnn/mkldnn_pooling-inl.h ########## @@ -104,7 +104,8 @@ class MKLDNNPoolingBwd { inline bool SupportMKLDNNPooling(const PoolingParam ¶m) { return param.kernel.ndim() == 2 && (param.pool_type == pool_enum::kMaxPooling || - param.pool_type == pool_enum::kAvgPooling); + param.pool_type == pool_enum::kAvgPooling) && + (!param.layout.has_value() || param.layout.value() == mshadow::kNCHW); Review comment: Thank you for changing this. Just to clarify, MKL-DNN pooling primitive does support NHWC input format. But when we integrated MKL-DNN backend into MXNet, we assumed that user input data should always be NCHW for 4D tesnsor. We need re-evaluated the workflow and integration to see if we need support NHWC here. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services