Bug 77922 – Apple/Safari: Enable WebGL multisampling on ATI cards for particular builds

• Status: RESOLVED
• Resolution: FIXED
• Component: WebGL

[Dean Jackson]

Currently GraphicsContext3DOpenGLCommon validates the incoming antialias attribute and only enables it for NVIDIA cards. Chromium allows multisampling on OS X + ATI when on 10.7.2 and above.

For more info:

http://code.google.com/codesearch#OAMlx_jo-ck/src/chrome/browser/resources/software_rendering_list/software_rendering_list.json&exact_package=chromium&ct=rc&cd=1&q=software_rendering_list.json
http://src.chromium.org/viewvc/chrome/trunk/deps/gpu/software_rendering_list/software_rendering_list.json?view=log

[Radar WebKit Bug Importer]

<rdar://problem/10817469>

[Kenneth Russell]

zmo's responsible for Chromium's GPU blacklist and could help advise on changes in this area.

Though this is far beyond the scope of this issue, it would be fantastic if we could figure out how to share the GPU blacklist information not only between WebKit ports but also between browsers.

[James Robinson]

Firefox exposes their blacklisting info on this wiki: https://wiki.mozilla.org/Blocklisting/Blocked_Graphics_Drivers. It looks like they've taken a slightly different approach to MSAA, quoting from the version currently there:

"For MSAA, we block all ATI cards except for AMD Radeon HD 6490M (device id 0x6760) and ATI Radeon HD 4670 (device id 0x9488). See Chromium issue 83153 [1]."

[Kenneth Russell]

zmo recently verified that even older ATI hardware seems to support multisampled frame buffer objects well with 10.7.2, where the same cards hung while creating a multisampled renderbuffer on earlier versions of Mac OS. That informed Chrome's decision to stop blacklisting ATI cards on 10.7.

[Dean Jackson]

Created attachment 125754 [details]
Patch

[Dean Jackson]

(In reply to comment #2)

> Though this is far beyond the scope of this issue, it would be fantastic if we could figure out how to share the GPU blacklist information not only between WebKit ports but also between browsers.

Definitely! I expect we'll (Apple'll) will always lag on this while WebGL is disabled by default. It would be so much nicer for us if there was a shared system.

[Early Warning System Bot]

Comment on attachment 125754 [details]
Patch

Attachment 125754 [details] did not pass qt-ews (qt):
Output: http://queues.webkit.org/results/11436119

[Dean Jackson]

Comment on attachment 125754 [details]
Patch

breaks QT

[Dean Jackson]

Created attachment 125756 [details]
Patch

[Dean Jackson]

http://trac.webkit.org/changeset/106956

[Alexey Proskuryakov]

Comment on attachment 125756 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=125756&action=review

> Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:63
> +    static SInt32 version;

I recommend always adding ASSERT(isMainThread()) in functions that use static variables.

> Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:69
> +    return version > 0x1072;

This doesn't match comments or description in ChangeLog (should be ">="?)

[Dean Jackson]

(In reply to comment #11)
> (From update of attachment 125756 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=125756&action=review
> 
> > Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:63
> > +    static SInt32 version;
> 
> I recommend always adding ASSERT(isMainThread()) in functions that use static variables.

I'll fix this.

> > Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:69
> > +    return version > 0x1072;
> 
> This doesn't match comments or description in ChangeLog (should be ">="?)

Indeed! Sorry.

[Dean Jackson]

Followup. http://trac.webkit.org/changeset/106964