master
[master] Avoid dots, full path to a file. (#20751)
[master] Unified oneDNN implementation calls for quantized operators (#20987)
Restore quantized RNN to master (#20952)
[master] Remove dnnl_ops-inl.h file (#20997)
Add quantized batch norm operator fused with ReLU (#21137)
[Master] Decouple OneDNN data structures in MXNet C++ API (#20624)
Requantize scale fix (#21100)
Refactor SupportDNNL functions (#21032)
Reduce overhead in sg_onednn_fully_connected for floats (#21092)
[submodule] Remove soon to be obsolete dnnl nomenclature from mxnet (#20606)
[master] Implemented oneDNN Backward Adaptive Pooling kernel (#20825)
Get rid of warnings (#21099)