Bug 238914 – Max Limits have been reduced

• Status: RESOLVED
• Resolution: FIXED
• Component: ANGLE

[Dean Jackson]

An internal reporter says:

It looks like WebGL capabilities have also changed between 15.3 and 15.4 (on MacOS + Metal, at any rate).
I’m seeing GL_MAX_FRAGMENT_UNIFORM_VECTORS (and VERTEX too) drop from 1024 (15.3) to 256 (15.4) on several devices:
MacBook Pro intel
MacBook Pro radeon
MacBook Pro M1 Max
Presumably another side effect of the ANGLE bump.

(Apple folks - there is some discussion on Slack about what might have gone wrong - see radar)

[Radar WebKit Bug Importer]

<rdar://problem/91384766>

[Kyle Piddington]

Created attachment 456994 [details]
Patch

[EWS Watchlist]

Note that there are important steps to take when updating ANGLE. See https://trac.webkit.org/wiki/UpdatingANGLE

[Dean Jackson]

Comment on attachment 456994 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=456994&action=review

> Source/ThirdParty/ANGLE/src/libANGLE/renderer/metal/ProgramMtl.mm:1294
> +
> +

Nit: Blank lines.

[Kyle Piddington]

Created attachment 456997 [details]
Patch

[Dean Jackson]

Comment on attachment 456997 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=456997&action=review

> Source/ThirdParty/ANGLE/src/libANGLE/renderer/metal/ProgramMtl.mm:1286
> +            ANGLE_TRY(getBufferPool(context)->allocate(context,
> +                uniformBlock.uniformData.size(), &ptrOut, &mtlBufferOut, &offsetOut));

I'm not familiar with ANGLE but I expect you don't need to worry about deallocation?

> Source/ThirdParty/ANGLE/src/libANGLE/renderer/metal/ProgramMtl.mm:1288
> +            memcpy(ptrOut, uniformBlock.uniformData.data(), uniformBlock.uniformData.size());

And we don't need to worry about reading out of bounds here?

[Kyle Piddington]

Comment on attachment 456997 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=456997&action=review

>> Source/ThirdParty/ANGLE/src/libANGLE/renderer/metal/ProgramMtl.mm:1286
>> +                uniformBlock.uniformData.size(), &ptrOut, &mtlBufferOut, &offsetOut));
> 
> I'm not familiar with ANGLE but I expect you don't need to worry about deallocation?

That's correct, deallocation will happen when the buffer pool is destroyed. Any in-flight buffers can be reused.

>> Source/ThirdParty/ANGLE/src/libANGLE/renderer/metal/ProgramMtl.mm:1288
>> +            memcpy(ptrOut, uniformBlock.uniformData.data(), uniformBlock.uniformData.size());
> 
> And we don't need to worry about reading out of bounds here?

Also correct, ANGLE_TRY will return early if we fail to allocate a buffer of proper size.

[Gphone]

the kDefaultUniformsMaxSize = 16 * 1024; is just one UBO buffer max size.
does it support that if i define many UBOs in one shader program each size is 16 * 1024?

[EWS]

Committed r293494 (250027@main): <https://commits.webkit.org/250027@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 456997 [details].

[Brent Fulgham]

*** Bug 239116 has been marked as a duplicate of this bug. ***