Hello,
If I understand correctly, XNNPACK has become the default computation engine for Pytorch mobile after the 1.6 release. I’ve built Pytorch mobile from source according to the following tutorial: https://pytorch.org/mobile/android/#building-pytorch-android-from-source with -DUSE_XNNPACK=ON, but XNNPACK engine is still not being used (verified by putting some print statements in aten/src/ATen/native/xnnpack/Convolution.cpp). Any help on how to resolve this issue would be appreciated.