Avoid dispatch overhead in call to mkldnn convolution (#52614)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/52614
This can speed up models by 5% (~.5-1% from the base, but ~5% after they've been sped up with mkldnn).
Test Plan: Imported from OSS
Reviewed By: navahgar
Differential Revision: D26696693
Pulled By: eellison
fbshipit-source-id: bfed55242524a4c2f1ae5d63e76d6803016d986d
