Halide backward and forward pass

[mpmisko]

Implementation of the forward and backward passes in Halide. The schedule is created using Halide's auto scheduler.

Apart from this, the PR also vectorises the addition post processing function when AVX is present.

To use Halide for example in a forward pass, one would have to first call the singleton HalideMulFactory to obtain the operation as

const HalideMulForward &mul = HalideMulFactory::getInstance().getHalideMulForward(ip, op, groups, false);

and then call the multiplication function as

mul.execute(input_features.data<T>(), d_output_features.data<T>(), &r[0], w.data<T>(), dw.data<T>(), d_input_rows.data<T>(), nActive);

to exectute the desired matrix multiplication.

[facebook-github-bot]

Thank you for your pull request. We require contributors to sign our Contributor License Agreement, and yours has expired.

Before we can review or merge your code, we need you to email [email protected] with your details so we can update your status.