Bug 142417 – [CG] Have Canvas use the IOSurfacePool

• Status: RESOLVED
• Resolution: FIXED
• Component: Canvas

[Chris Dumez]

Have Canvas use the IOSurfacePool to avoid repeatedly creating IOSurfaces.

Radar: <rdar://problem/20044440>

[Chris Dumez]

Created attachment 248119 [details]
Patch

[Tim Horton]

Comment on attachment 248119 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=248119&action=review

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:50
> +    if (cachedSurface->contextSize() != contextSize)

I thought you mentioned that this was slow. Did further testing invalidate that result?

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:162
> +    // Release the graphics context and update the context size. Next time, the graphics context is

no comma between time and the

[Chris Dumez]

Comment on attachment 248119 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=248119&action=review

>> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:50
>> +    if (cachedSurface->contextSize() != contextSize)
> 
> I thought you mentioned that this was slow. Did further testing invalidate that result?

I was wrong, I had a bug and it was causing the slow down.

[Chris Dumez]

Created attachment 248180 [details]
Patch

[Darin Adler]

Comment on attachment 248180 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=248180&action=review

> Source/WebCore/platform/graphics/cg/ImageBufferDataCG.h:52
>  class ImageBufferData {
>  public:
>      ImageBufferData();
> +    ~ImageBufferData();

Since all the data members are public, we should change this to a struct and remove all the "m_" prefixes.

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:46
> +    std::unique_ptr<IOSurface> cachedSurface = IOSurfacePool::sharedPool().takeSurface(size, colorSpace);

I suggest using auto here instead of std::unique_ptr<IOSurface>.

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:51
> +    if (cachedSurface->contextSize() != contextSize)
> +        cachedSurface->setContextSize(contextSize);

Why not put the == check inside setContextSize instead?

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:57
> +    if (std::unique_ptr<IOSurface> cachedSurface = surfaceFromPool(size, size, colorSpace))

I suggest auto here instead of naming the type.

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:59
>      return std::unique_ptr<IOSurface>(new IOSurface(size, colorSpace));

I suggest make_unique here.

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:64
> +    if (std::unique_ptr<IOSurface> cachedSurface = surfaceFromPool(size, contextSize, colorSpace))

I suggest auto here instead of naming the type.

> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:66
>      return std::unique_ptr<IOSurface>(new IOSurface(size, contextSize, colorSpace));

I suggest make_unique here.

[Chris Dumez]

Comment on attachment 248180 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=248180&action=review

>> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:51
>> +        cachedSurface->setContextSize(contextSize);
> 
> Why not put the == check inside setContextSize instead?

The cost of calling the setContextSize() method was showing on the performance test. Since setContextSize() is private, I could also mark it as inline and move the contextSize check to the method if you'd prefer.

[Chris Dumez]

Comment on attachment 248180 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=248180&action=review

>>> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:51
>>> +        cachedSurface->setContextSize(contextSize);
>> 
>> Why not put the == check inside setContextSize instead?
> 
> The cost of calling the setContextSize() method was showing on the performance test. Since setContextSize() is private, I could also mark it as inline and move the contextSize check to the method if you'd prefer.

Never mind, I tested again today and moving the check to setContextSize() did not affect the results on Canvas/reuse.html. I'll make this change.

[Chris Dumez]

Comment on attachment 248180 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=248180&action=review

>> Source/WebCore/platform/graphics/cocoa/IOSurface.mm:59
>>      return std::unique_ptr<IOSurface>(new IOSurface(size, colorSpace));
> 
> I suggest make_unique here.

In order to use make_unique<>(), I would need to make the constructors public. However, we really don't want people to call these constructors directly as this would bypass the IOSurfacePool.

[Chris Dumez]

Created attachment 248207 [details]
Patch

[Chris Dumez]

Created attachment 248208 [details]
Patch

[Build Bot]

Comment on attachment 248208 [details]
Patch

Attachment 248208 [details] did not pass mac-ews (mac):
Output: http://webkit-queues.appspot.com/results/5132014336344064

Number of test failures exceeded the failure limit.

[Build Bot]

Created attachment 248211 [details]
Archive of layout-test-results from ews100 for mac-mavericks

The attached test failures were seen while running run-webkit-tests on the mac-ews.
Bot: ews100  Port: mac-mavericks  Platform: Mac OS X 10.9.5

[Build Bot]

Comment on attachment 248208 [details]
Patch

Attachment 248208 [details] did not pass mac-wk2-ews (mac-wk2):
Output: http://webkit-queues.appspot.com/results/4850539359633408

Number of test failures exceeded the failure limit.

[Build Bot]

Created attachment 248214 [details]
Archive of layout-test-results from ews104 for mac-mavericks-wk2

The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews.
Bot: ews104  Port: mac-mavericks-wk2  Platform: Mac OS X 10.9.5

[Chris Dumez]

Created attachment 248215 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 248215 [details]
Patch

Clearing flags on attachment: 248215

Committed r181301: <http://trac.webkit.org/changeset/181301>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.