Intel's Mesa Shader Compiler Backend Scheduling Now ~30% Faster
With newly-merged optimizations to Mesa 24.0-devel, the Intel shader compiler back-end is seeing its scheduling code execute around 30% faster. This big speed-up comes due to overhauling how they store pass information and reusing that for multiple pre-RA scheduling modes.
This Intel compiler code benefits the company's open-source OpenGL and Vulkan drivers within Mesa. The 30% boost to the scheduling portion of the back-end compiler was measured while profiling with Cyberpunk 2077, ROTTR, and Total War: Warhammer 3.
While this back-end scheduling code is much faster, for shader execution time on Intel graphics it ultimately ends up amounting to just in the ~2% range. In any event all of these relentless performance optimizations to Mesa being pursued by Intel engineers for their Iris Gallium3D and ANV Vulkan drivers are adding up.
Those wanting to learn more about this latest optimization effort can see this merge request for all the details.
This Intel compiler code benefits the company's open-source OpenGL and Vulkan drivers within Mesa. The 30% boost to the scheduling portion of the back-end compiler was measured while profiling with Cyberpunk 2077, ROTTR, and Total War: Warhammer 3.
While this back-end scheduling code is much faster, for shader execution time on Intel graphics it ultimately ends up amounting to just in the ~2% range. In any event all of these relentless performance optimizations to Mesa being pursued by Intel engineers for their Iris Gallium3D and ANV Vulkan drivers are adding up.
Those wanting to learn more about this latest optimization effort can see this merge request for all the details.
2 Comments