DeSmuME

Cocoa Port:
- Optimize VBO uploading of the HUD layer.

Render3D:
- Provide a means of specifying which specific framebuffers need to be flushed for each frame.

GPU:
- Fix bug where a swapped screen state was not being properly reset. (Regression from r5340.)

OpenGL Renderer:
- If PBOs are supported, avoid doing an extra framebuffer copy and read the PBO directly.

Cocoa Port:
- In the OpenGL blitter, use DMA texture uploads for all possible video source cases. Doing this removes a longstanding MAJOR performance bottleneck.
- Native-sized video sees up to a 15% performance improvement, while higher-resolution video can see up to a 100% performance improvement!!!!!

GPU:
- Allow clients to specify their own framebuffer storage when calling GPUSubsystem::SetCustomFramebufferSize().

GPU:
- Apply SSE2 optimizations to all of the BG layer modes. (Related to r5332.)
- Do some code refactoring and cleanup.

Render3D:
- Revert r5331, since it results in incorrect alpha testing. Add additional notes to remind myself why I wrote the code that way in the first place.

GPU:
- Optimize the loading of destination blending masks if SSSE3 is available.

Cocoa Port:
- Return to using Snow Leopard style Audio Components. Requires building with Xcode v7.2 or later, since Xcode v7.0 and v7.1 have bugs that will cause AudioUnits to crash. (Related to r5280.)