winch(aarch64): add v128.const support

[macovedj]

This pr adds support for v128.const to winch for aarch64.

I'd like to tackle the larger list of missing SIMD support, and figured I should start with something small.

In mov_ir the v128 immediate takes the add_constant path rather than the emit path since that's what the x64 backend does for these load_fp_const.

If this goes through I could go ahead and tackle the rest of v128, followed by the others.

[cfallin]

LGTM, thanks!

There's a warning now about the now-unused UnsupportedImm error -- CI should be green once that's removed, then I'm happy to merge.

[macovedj]

LGTM, thanks!

There's a warning now about the now-unused UnsupportedImm error -- CI should be green once that's removed, then I'm happy to merge.

Thanks! Hope to have some more up soon!

[github-actions]

Subscribe to Label Action

cc @saulecabrera

Details

This issue or pull request has been labeled: "winch"

Thus the following users have been cc'd because of the following labels:

• saulecabrera: winch

To subscribe or unsubscribe from this label, edit the .github/subscribe-to-label.json configuration file.

Learn more.

[macovedj]

Looks like there was another CI failure from expecting v128_load_lane_invalid_address to fail, which passes with v128.const support. Getting rid of the blanket simd should-fail check only required adding two additional tests to the unsupported list, so I figured the change I pushed lets me remove expected failures from the unsupported list as I go.

[macovedj]

Looks like the simd check is also used by other test suites. Removing it caused a couple of tests/all winch tests to fail in the merge queue. Added it back and moved the wast exceptions to a now_supported list instead. As far as I can tell there's no way to avoid keeping that list and manually moving items over from unsupported as they start passing, but happy to change the approach if there's a better one.

[cfallin]

test-runner fixes LGTM; thanks!