torch_nn_functional_conv2d_problem

Different output on ARM and x86_64 architectures for torch.nn.functional.conv2d when applying the same input and parameters. To run the script: python3 functional_conv2d_example.py Already computed results available in results/ folder. Differences can be observed when performing a 2-way comparison between output_tensor_arm.txt and output_tensor_x86_64.txt. The ARM result was obtained on an NVIDIA Jetson Nano, Ubuntu 18.04.5 LTS, pytorch 1.4.0, nnpack backend. The x86_64 result was computed on an Intel i5-8250U, Ubuntu 18.04.4 LTS, pytorch 1.4.0, mkl-dnn backend.