SIMD errors on latest spidermonkey

[kripken]

asm2.test_simd_float32x4 and asm2.test_simd_int32x4 both fail on latest spidermonkey, with

asm.js type error: asm.js does not support 3-element SIMD loads or stores

Perhaps something changed in the spec? cc @juj @sunfishcode

[sunfishcode]

Yes; the load3/store3 functions were removed from asm.js.

[juj]

Why were they removed? Those functions are still part of the SIMD spec? load3, store3

[sunfishcode]

load3/store3 were removed from asm.js due to pushback from implementors because they are inconvenient to implement (they are a special case for some bounds-checking strategies because they aren't power-of-2 sized), they don't correspond directly to a single instruction on any hardware, and they can always be reintroduced in the future if there's sufficient demand.

[juj]

Is there a conversation trail or a spec document regarding this to reference to? Please include a heads-up when these kinds of changes occur, so that Emscripten will know to adapt. Is the intent to drop load3 and store3 from SIMD.js altogether?

[juj]

Authored PR #4673 to remove these.

However I feel uneasy authoring the PR, because I'm writing this based on what latest nightly SpiderMonkey validation is spitting out, I don't have any reference or changelog to read about this, or any conversation history. Are there some other SIMD.js features that have been removed? Is there a bugzilla number for these removals that I could follow?

I think it would be good to have this documentation online with the SIMD.js spec. Posted tc39/ecmascript_simd#349 about that.

[juj]

For reference, the conversation about removing them occurred in https://bugzilla.mozilla.org/show_bug.cgi?id=1298202.

[juj]

The PR that removed load3 and store3 were merged, so closing this as resolved.