grygielski opened a new pull request #19578: URL: https://github.com/apache/incubator-mxnet/pull/19578
## Description ## This PR adds `BN + ReLU` -> `BatchNormReLU` fuse into MKLDNN backend. It results in speedup of Resnet V2 models since they don't only have `conv+bn+relu` patterns but also `bn+relu` without convolution before it. Approximate speedups measured on i9 machine: ``` resnet18_v2: 5.0% resnet50_v2: 11.6% resnet101_v2: 10.9% resnet152_v2: 10.4% ``` ## Checklist ## ### Essentials ### - [x] PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], [FEATURE], [DOC], etc) - [x] Changes are complete (i.e. I finished coding on this PR) - [x] All changes have test coverage - [x] Code is well-documented ### Changes ### - Added `mkldnn_bn_relu_property` ## Comments ## - This fuse can be added also to `MKLDNN_QUANTIZE` backend once we restore quantization files and test it ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
