Bug 178254 – Performance: Skip texture upload if source image and destination texture haven't changed

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Jer Noble]

Performance: Skip texture upload if source image and destination texture haven't changed

[Jer Noble]

<rdar://problem/34968181>

[Jer Noble]

Created attachment 323653 [details]
Patch for landing

[Jer Noble]

Created attachment 323676 [details]
Patch

[Jer Noble]

Created attachment 323720 [details]
Patch

[Build Bot]

Attachment 323720 [details] did not pass style-queue:


ERROR: Source/WTF/wtf/UnsafePointer.h:40:  Should be indented on a separate line, with the colon or comma first on that line.  [whitespace/indent] [4]
ERROR: Source/WTF/wtf/UnsafePointer.h:41:  Should be indented on a separate line, with the colon or comma first on that line.  [whitespace/indent] [4]
Total errors found: 2 in 8 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Simon Fraser (smfr)]

Comment on attachment 323720 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=323720&action=review

> Source/WTF/wtf/UnsafePointer.h:34
> +template<typename T>

Couldn't this just be a void* wrapper, and not even templatized? Or do you care about pointer type?

[Dean Jackson]

Comment on attachment 323720 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=323720&action=review

I'm not sure I like the "seed" terminology, but this is ok for now. I'd prefer it if all this state could be tracked directly inside VideoTextureCopierCV using a separate context, but we can do that later.

>> Source/WTF/wtf/UnsafePointer.h:34
>> +template<typename T>
> 
> Couldn't this just be a void* wrapper, and not even templatized? Or do you care about pointer type?

Yeah, I'm not sure this is needed in this bug. Maybe it could be a separate patch and just use IOSurfaceRef for now.

> Source/WebCore/platform/graphics/cv/VideoTextureCopierCV.h:113
> +    UnsafePointer<IOSurfaceRef> m_lastSurface;

I think we can just use a raw IOSurfaceRef here.

> Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:264
> +    ::glBindTexture(GL_TEXTURE_2D, m_state.boundTarget(GL_TEXTURE0) == GL_TEXTURE_2D ? m_state.boundTexture(GL_TEXTURE0) : 0);

Since boundTexture() will return 0 if the value doesn't exist, I don't think you need the conditional.

> Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:1937
> +    m_state.textureSeedCount.add(o);

Has it been seeded at this point? I don't think so.

[Jer Noble]

(In reply to Dean Jackson from comment #7)
> Comment on attachment 323720 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=323720&action=review
> 
> I'm not sure I like the "seed" terminology, but this is ok for now. I'd
> prefer it if all this state could be tracked directly inside
> VideoTextureCopierCV using a separate context, but we can do that later.

I'd prefer that too, but that's going to be a Buch 

> >> Source/WTF/wtf/UnsafePointer.h:34
> >> +template<typename T>
> > 
> > Couldn't this just be a void* wrapper, and not even templatized? Or do you care about pointer type?

I think you'd care about pointer type.

> Yeah, I'm not sure this is needed in this bug. Maybe it could be a separate
> patch and just use IOSurfaceRef for now.
> 
> > Source/WebCore/platform/graphics/cv/VideoTextureCopierCV.h:113
> > +    UnsafePointer<IOSurfaceRef> m_lastSurface;
> 
> I think we can just use a raw IOSurfaceRef here.

I'm concerned that someone might try to actually call something on it in the future and cause a use-after-free security bug. So I'd prefer to leave this in.

> > Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:264
> > +    ::glBindTexture(GL_TEXTURE_2D, m_state.boundTarget(GL_TEXTURE0) == GL_TEXTURE_2D ? m_state.boundTexture(GL_TEXTURE0) : 0);
> 
> Since boundTexture() will return 0 if the value doesn't exist, I don't think
> you need the conditional.

It's not just checking whether the value exists, but also that the target is TEXTURE2D, so I think you do need the conditional.

> > Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:1937
> > +    m_state.textureSeedCount.add(o);
> 
> Has it been seeded at this point? I don't think so.

I'd like to keep separate the concepts of "this texture exists and has a seed of 0" from "this texture does not exist and has a seed of 0".  Making the former start at "1" means those two states are distinguishable.

[Jer Noble]

(In reply to Jer Noble from comment #8)
> (In reply to Dean Jackson from comment #7)
> > Comment on attachment 323720 [details]
> > Patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=323720&action=review
> > 
> > I'm not sure I like the "seed" terminology, but this is ok for now. I'd
> > prefer it if all this state could be tracked directly inside
> > VideoTextureCopierCV using a separate context, but we can do that later.
> 
> I'd prefer that too, but that's going to be a Buch 

...bunch more work before it's technically possible. This is, I think, a good short term performance fix.

[Jer Noble]

Created attachment 323754 [details]
Patch for landing

[Build Bot]

Attachment 323754 [details] did not pass style-queue:


ERROR: Source/WTF/wtf/UnsafePointer.h:40:  Should be indented on a separate line, with the colon or comma first on that line.  [whitespace/indent] [4]
ERROR: Source/WTF/wtf/UnsafePointer.h:41:  Should be indented on a separate line, with the colon or comma first on that line.  [whitespace/indent] [4]
Total errors found: 2 in 8 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[WebKit Commit Bot]

Comment on attachment 323754 [details]
Patch for landing

Clearing flags on attachment: 323754

Committed r223315: <https://trac.webkit.org/changeset/223315>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.