In reply to @morganalyssa "so i profiled my": are you using the new `convolve`
API (cc @ezquerra ) https://github.com/mratsim/Arraymancer/pull/617

using SIMD with convolution requires refactoring
but it's possible that just asking `omp simd` in that inner loop helps:https://
github.com/AngelEzquerra/Arraymancer/blob/4a22c333e24063709c4e798831d46935dd98e1
df/src/arraymancer/tensor/math_functions.nim#L142

Otherwise for the very adventurous, I've detailed how to do fast convolution /
cross-correlation here:
https://github.com/mratsim/laser/wiki/Convolution-optimisation-resources

And there is a high-level implementation that can be modified for SIMD here:

https://github.com/SciNim/impulse/blob/master/benchmarks/image_filters/filter2d_
separable.nim#L124-L149

And nother approach is to have convolve use matrix multiplication like this dep
learning convolution here:
https://github.com/mratsim/Arraymancer/blob/master/src/arraymancer/nn_primitives
/fallback/conv.nim#L81-L106