convolution-benchmark core dumped if output-subsampling isn't 1x1

[wangshankun]

bin/convolution-benchmark -ic 3 -is 231 231 -ks 11 11 -oc 96 --output-subsampling 4 4 Batch size: 1 Input channels: 3 Output channels: 96 Input: 231x231 with implicit padding 0 Kernel: 11x11 Subsampling: 4x4 Algorithm: auto Threads: 8 Iterations: 3 Segmentation fault (core dumped)

[Maratyszcza]

By default, convolution-benchmark is running in output mode, i.e. calls nnp_convolution_output which does not support convolution with strides. Only the inference function nnp_convolution_inference currently supports strides. You can switch to inference mode using -m inference