Zink's Mike Blumenkrantz Tackling Mesa WSI Inefficiencies
Earlier this month I ran some fresh benchmarks of Zink OpenGL-on-Vulkan against RadeonSI. While Zink in general is already quite speedy and in good shape for most workloads, those tests uncovered some troubled spots and Zink lead developer Mike Blumenkrantz of Valve has been diving into some of those issues with fixes. Another merge request is pending to deal with inefficiencies in the Mesa Vulkan windowing system integration (WSI) code.
One of the benchmarks shown where Zink performed quite poorly against RadeonSI was in the very basic/elementary glmark2 test case:
Given the large difference, Mike began looking at it and attributed it to the Mesa WSI code. For those interested, Mike has published another lengthy blog post with his usual creative writing that covers the problem. With various changes, he's cut roughly 40% off Zink's frametime. It's still behind the RadeonSI performance in this basic OpenGL test, but at least better than before.
The merge request opened for it is WSI: various unethical perf improvements that are being evaluated now for upstream Mesa.
One of the benchmarks shown where Zink performed quite poorly against RadeonSI was in the very basic/elementary glmark2 test case:
Given the large difference, Mike began looking at it and attributed it to the Mesa WSI code. For those interested, Mike has published another lengthy blog post with his usual creative writing that covers the problem. With various changes, he's cut roughly 40% off Zink's frametime. It's still behind the RadeonSI performance in this basic OpenGL test, but at least better than before.
The merge request opened for it is WSI: various unethical perf improvements that are being evaluated now for upstream Mesa.
7 Comments