- cross-posted to:
- [email protected]
- [email protected]
- cross-posted to:
- [email protected]
- [email protected]
This is the best summary I could come up with:
This week prior to the Mesa 24.0 feature freeze / code branching, a notable merge request landed that had been worked on the past few months by one of Valve’s open-source Linux graphics driver developers.
Daniel Schürmann of Valve’s Linux graphics team spent the past several months working on a new post-RA scheduler for instruction level parallelism (ILP) with ACO.
ACO as a reminder is the compiler started by Valve engineers as an alternative to the AMDGPU LLVM back-end.
ACO is used by default on RADV and has worked out very well for delivering quicker game load times and better in-game performance for this open-source Radeon Vulkan driver than using the LLVM back-end.
"This pass implements a simple forward list-scheduler which works on a small partial DAG of 16 nodes at any time.
With testing many different shaders over dozens of games, the generated instruction count was down by about 2.5%, the code size around 2% lower, etc.
The original article contains 345 words, the summary contains 158 words. Saved 54%. I’m a bot and I’m open source!