[Webkit-unassigned] [Bug 159928] [GStreamer] Support a direct GPU-to-GPU copy of video textures to WebGL

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=159928

--- Comment #17 from Olivier Blin <olivier.blin at softathome.com> ---
(In reply to comment #15)
> Comment on attachment 289463 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=289463&action=review
> 
> > Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp:700
> > +bool MediaPlayerPrivateGStreamerBase::paintToCairoSurface(RefPtr<cairo_surface_t> outputSurface, cairo_device_t* device, GstVideoInfo* videoInfo)
> 
> I don't think you need to pass the smart pointer to this function, and in
> that case I would pass a reference instead of a copy, but I think you can
> simply pass the pointer.

Right, I have also simplified the callers, the WebGL case does not need a smart pointer at all.

> > Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp:702
> > +    // m_sampleMutex should be locked by the caller
> 
> I would move this comment before the function, saying something like: //
> This should be called with the sample mutex locked.
> 
> > Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp:710
> > -    IntSize size = IntSize(GST_VIDEO_INFO_WIDTH(&videoInfo), GST_VIDEO_INFO_HEIGHT(&videoInfo));
> > +    IntSize size = IntSize(GST_VIDEO_INFO_WIDTH(videoInfo), GST_VIDEO_INFO_HEIGHT(videoInfo));
> 
> Both callers do this, maybe you could pass this as a const reference.

Ok

> > Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp:713
> >      IntSize rotatedSize = m_videoSourceOrientation.usesWidthAsHeight() ? size.transposedSize() : size;
> 
> Ditto.

Ok

> > Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp:773
> > +    GLContext* glContext = PlatformDisplay::sharedDisplayForCompositing().sharingGLContext();
> > +    cairo_device_t* device = glContext->cairoDevice();
> > +
> > +    // Thread-awareness is a huge performance hit on non-Intel drivers.
> > +    cairo_gl_device_set_thread_aware(device, FALSE);
> > +
> > +    IntSize size = IntSize(GST_VIDEO_INFO_WIDTH(&videoInfo), GST_VIDEO_INFO_HEIGHT(&videoInfo));
> > +    IntSize rotatedSize = m_videoSourceOrientation.usesWidthAsHeight() ? size.transposedSize() : size;
> > +
> 
> Don't you need to make the context current in this case?

See Miguel's answer.

> > Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp:813
> > +    GstVideoInfo videoInfo;
> > +    WTF::GMutexLocker<GMutex> lock(m_sampleMutex);
> > +    if (!getSampleVideoInfo(m_sample.get(), videoInfo))
> > +        return nullptr;
> > +
> > +    GLContext* context = PlatformDisplay::sharedDisplayForCompositing().sharingGLContext();
> > +    context->makeContextCurrent();
> > +    cairo_device_t* device = context->cairoDevice();
> > +
> > +    // Thread-awareness is a huge performance hit on non-Intel drivers.
> > +    cairo_gl_device_set_thread_aware(device, FALSE);
> > +
> > +    IntSize size = IntSize(GST_VIDEO_INFO_WIDTH(&videoInfo), GST_VIDEO_INFO_HEIGHT(&videoInfo));
> > +    IntSize rotatedSize = m_videoSourceOrientation.usesWidthAsHeight() ? size.transposedSize() : size;
> 
> All this is mostly duplicated code, could we move it to a helper function?
> maybe a function that prepares the context and returns the GLContext as
> return value, and video info, and sizes as output parameters

I am not sure it is worth it, it is not much more readable after being extracted, but I will do so in the next patch proposal.
Also, the cairo_device needs to be kept around as well.

> > Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.h:119
> > +    bool paintToCairoSurface(RefPtr<cairo_surface_t> outputSurface, cairo_device_t*, GstVideoInfo*);
> 
> This should be private.

Yes, moved it to together with other protected methods.

Thanks for your feedback!

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20160922/e940dbf3/attachment-0001.html>