[Webkit-unassigned] [Bug 227596] REGRESSION (r279466): [Big Sur] webgl/1.0.3/conformance & webgl/2.0.0/conformance are failing

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=227596

Kenneth Russell <kbr at google.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Depends on|                            |227226

--- Comment #2 from Kenneth Russell <kbr at google.com> ---
Let's mark these as expected failures in TestExpectations and follow up. If just the arctangent function has exceptionally poor precision when fastMathEnabled is used, we could plausibly increase the tolerance on these specific WebGL conformance tests.

I think that we shouldn't revert the changes from Bug 227226 . Without fastMathEnabled, it seems likely that a significant amount of WebGL content will be slower with ANGLE's Metal backend than the OpenGL backend.

Would it be feasible or appropriate for Apple engineers to also file a bug against Metal to understand why precision of this one trigonometric function is so poor?


Referenced Bugs:

https://bugs.webkit.org/show_bug.cgi?id=227226
[Bug 227226] BabylonJS Under water demo is slower than it should be on Intel
-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20210701/6c61b586/attachment-0001.htm>