Native implementation of the wasm matrix multiply interface for Intel architecture

[abhi-agg]

Feature description

As per browsermt/bergamot-translator#205, we would like to have an optimized implementation of the generic matrix multiply interface for Intel architecture

cc @andrenatal @kpu @XapaJIaMnu

[kpu]

I remain confused as to the point of this exercise. We had a working matrix multiply interface. Every change is a tax on the working implementation. Yet the justification for coming in to do a redesign is very limited and it's not clear what advantage the new interface has. Why isn't this just comments on top of the existing one?

Index to int32_t ok fine I get that, easy. ARM I get but, as discussed, that's opaque to the user anyway and not a change to the interface. Why are we changing the order of arguments to functions?

If you want a working version of your interface, wrap the existing calls. The separation of somebody drawing up a spec and somebody implementing, which seems to be implied by this issue, just results in a spec that doesn't quite fit the implementation. And a lot more work on the implementation that I am struggling to understand or justify.

[abhi-agg]

As per #49 (comment) and #49 (comment), I have clarified that the interface in #49 is not a redesign but a well documented version of intgemm interface catering to both intel and arm architecture.

[abhi-agg]

We already have the native gemm implementation of this interface for intel with intgemm as the backend. So, closing this one. The landing is already tracked here: https://bugzilla.mozilla.org/show_bug.cgi?id=1722102