RADV Zink Sees Another "Massive" Optimization On The Way
In going through my recent RADV-Zink vs. RadeonSI OpenGL benchmarks, Valve's Mike Blumenkrantz has already been landing optimizations/fixes and there is another one on the way as a result.
Blumenkrantz has been profiling ET: Legacy as the open-source Enemy Territory game that was shown in my latest round of Zink testing and an area where this generic OpenGL-on-Vulkan implementation could improve. He found the fragment shader epilog handling was quite consuming while running ET: Legacy and thus was off to working on optimizing the RADV driver code path.
There's now an open merge request to "massively" improve the prolog/eiplog look-up performance. Blumenkrantz noted in that merge request:
In a new blog post Mike commented in his usual style of writing, "How Much Faster Is It? If I had to give a number, probably at least 1000% faster. Just as a rough estimate." In any event, good news for RADV/Zink.
Blumenkrantz has been profiling ET: Legacy as the open-source Enemy Territory game that was shown in my latest round of Zink testing and an area where this generic OpenGL-on-Vulkan implementation could improve. He found the fragment shader epilog handling was quite consuming while running ET: Legacy and thus was off to working on optimizing the RADV driver code path.
There's now an open merge request to "massively" improve the prolog/eiplog look-up performance. Blumenkrantz noted in that merge request:
"these use the same mechanism, and that mechanism is wildly inefficient for CPU-bound applications"
In a new blog post Mike commented in his usual style of writing, "How Much Faster Is It? If I had to give a number, probably at least 1000% faster. Just as a rough estimate." In any event, good news for RADV/Zink.
14 Comments