Mesa Lands Work Around Async glFlush - Should Help Workstation Performance

While vendors are increasingly just focused on Vulkan (and Direct3D under Windows), there still is plenty of OpenGL software out there especially in the workstation space where software vendors are slow to adapt. Well known RadeonSI OpenGL developer Marek Olšák of AMD landed another performance optimization this week that should benefit the likes of SPECviewperf.

This year we have seen Marek and AMD's other open-source driver developers working on workstation optimizations with optimizing their Gallium3D driver against SPECviewperf and making other improvements in this year.

While some CAD software vendors have expressed interest in Vulkan and ultimately more workstation software vendors will employ that next-gen API, for now OpenGL remains quite important. It's also this area where AMD's proprietary OpenGL driver component used across Windows and Linux has traditionally performed better than the open-source Mesa OpenGL driver. But in recent months there has been a clear and solid push towards improving Mesa for such use-cases.

The newest merge request to point out is Marek working on asynchronous glFlush around the Mesa state tracker and OpenGL threading code. This change to common Mesa code allows for executing glFlush asynchronously when no image has been imported/exported.

Marek does note that this improves viewperf performance but didn't go into any details over the expected boost in quantifiable terms. In any case, the latest RadeonSI optimizations will be out next quarter in Mesa 21.2.