With the upcoming integration of Firefox Translations into Firefox by default, we would like to have the intgemm library [1] and its correspondent Webassembly intrinsics [2] to ride the trains to release and enabled by default so that translation users could benefit from its performance improvements.

Based on the conditions listed here https://bugzilla.mozilla.org/show_bug.cgi?id=1722102#c16, I am ni'ing the key people in order to have discussion started. Please, feel free to add more (or the correct set of) peers as needed.

@yury what do you think we might need beyond the list of conditions listed in the link above?

[1] https://bugzilla.mozilla.org/show_bug.cgi?id=1722102
[2] https://bugzilla.mozilla.org/show_bug.cgi?id=1746631

Is this available directly in content somehow? If so, how is it exposed?

Is this available directly in content somehow? If so, how is it exposed?

Integration tests are available at bug 1746631.

Notice that interface is only exposed for WebAssembly global with system principal. This is ensured by /dom/tests/mochitest/integer-gemm/test_unavailable_for_webpage.html

As long as this feature is not exposed to content at all, it has low fuzzing priority and it shouldn't be a blocker to have fuzzing for it.

Thanks @decoder. So if we don't need fuzzing, then only security review is enough, @yury? Or there are more architectural changes required?

then only security review is enough, @yury?

intgemm is a third-party library somebody on Firefox and/or SM needs to look at it with this question in mind

Or there are more architectural changes required?

There is always a room for improvements. As far as I know the goal was to compile intgemm to wasm as well to provide fallback equivalent for the intrinsic functions -- I'm not sure it was done.

We are going with another route (bug 1815790).