1060441 - Odinmonkey - SIMD: add support for Float32x4Array and Int32x4Array

Status: RESOLVED WONTFIX

(Core :: JavaScript Engine: JIT, defect)

Description

[Benjamin Bouvier [:bbouvier] (inactive)]

function f(glob, ffi, heap) {
  "use asm";
  var f4 = glob.SIMD.float32x4;
  var view = new glob.Float32x4Array(heap); // or glob.SIMD.Float32x4Array?
  function g() {
  var i = 0;
  return f4(view[i >> 4]);
  }
}

Comment 1

[Lars T Hansen [:lth]]

In an (well, my) ideal world you would do this on top of the patches for bug 1054841.  If I'm efficient I may be able to line those up for landing this week, and then you wouldn't have to worry about it.

Comment 2

[Benjamin Bouvier [:bbouvier] (inactive)]

As in bug 1060438, I would recommend nobody to work on this, as the spec is really not stable at this point and typeX4Arrays might even disappear in the future.

Comment 3

[Luke Wagner [:luke]]

Is this WONTFIX now?

Comment 4

[Benjamin Bouvier [:bbouvier] (inactive)]

(In reply to Luke Wagner [:luke] from comment #3)
> Is this WONTFIX now?

Good question.  The spec polyfill has Float32x4Array and Int32x4Array, but no Float64x2Array, no Int8x16Array and no Int16x8Array as well.  Dan, has it been discussed recently during the face-to-face meetings? If not, should it? :)

Comment 5

[Benjamin Bouvier [:bbouvier] (inactive)]

Opened an issue on the github repo as well: https://github.com/johnmccutchan/ecmascript_simd/issues/113

Comment 6

[Luke Wagner [:luke]]

I guess I was mostly asking about the integration of (Int|Float)32x4Array into Odin; I was thinking we were fine with .load/store.

Comment 7

[Benjamin Bouvier [:bbouvier] (inactive)]

Ha right, I read too fast and didn't see the explicit reference to Odin.  Agreed we're fine with .load/store here, which is way more powerful.  Moving the discussion to bug 1060438.