More Zink Improvements Arrive For Mesa 22.3, Less Stuttering For RADV
Development on Zink as the OpenGL implementation atop the Vulkan API within Mesa remains very vibrant with a lot of optimizations and other improvements ongoing.
Hitting Mesa 22.3-devel this morning is implementing dynamic patch control points for the RADV driver. The merge request by Valve open-source developer Samuel Pitoiset explains that this should be useful for less stuttering with Zink.
Another recently merged improvement to Mesa 22.3-devel for Zink's benefit is implementing async pixel buffer object (PBO) shader compilation. Zink lead developer Mike Blumenkrantz (working under contract for Valve) took care of this merge request for async PBO shader compilation to help with both compile and execution speed.
Stemming from the Zink work. Mike Blumenkrantz has also been looking at Vulkan overhead. To analyze the Vulkan API call overhead, he developed vkoverhead as a benchmark for analyzing different Vulkan API calls. Yes, I'll be experimenting with this Vulkan benchmark in some future articles.
Meanwhile still pending for Zink in Mesa are some small optimizations, using SPIR-V 1.6, handling split acquire/present, and other fixes and optimizations.
Considering the already great numbers from last month's Zink benchmarking on Radeon, it will be interesting to see how much better Zink/Mesa is by the time of the 22.3 release in late Q4.
Hitting Mesa 22.3-devel this morning is implementing dynamic patch control points for the RADV driver. The merge request by Valve open-source developer Samuel Pitoiset explains that this should be useful for less stuttering with Zink.
Another recently merged improvement to Mesa 22.3-devel for Zink's benefit is implementing async pixel buffer object (PBO) shader compilation. Zink lead developer Mike Blumenkrantz (working under contract for Valve) took care of this merge request for async PBO shader compilation to help with both compile and execution speed.
Stemming from the Zink work. Mike Blumenkrantz has also been looking at Vulkan overhead. To analyze the Vulkan API call overhead, he developed vkoverhead as a benchmark for analyzing different Vulkan API calls. Yes, I'll be experimenting with this Vulkan benchmark in some future articles.
Meanwhile still pending for Zink in Mesa are some small optimizations, using SPIR-V 1.6, handling split acquire/present, and other fixes and optimizations.
Considering the already great numbers from last month's Zink benchmarking on Radeon, it will be interesting to see how much better Zink/Mesa is by the time of the 22.3 release in late Q4.
1 Comment