More performance improvements are on the way for Deus Ex and Tomb Raider on Mesa

Mesa is progressing nicely as usual and I've been keeping an eye on the mailing list for anything interesting. It seems Deus Ex: Mankind Divided and Tomb Raider have both seen some more performance tuning.

Note: All of these are now in Mesa-git.

Samuel Pitoiset from Valve has sent in a small patch that should boost performance of Deus Ex: Mankind Divided on AMD by a small amount. From the message on the public mailing list:

Quotewinsys/amdgpu: do not iterate twice when adding fence dependencies

The perf difference is very small, 3.25->2.84% in amdgpu_cs_flush() in the DXMD benchmark.

For the i965 (Intel) driver, both Tomb Raider and Deus Ex: Mankind Divided will have a regression fix from Timothy Arceri (see here):

QuotePreviously the constant array would not get copy propagated until the backend did its GLSL IR opt loop. I plan on removing that from i965 shortly which caused huge regressions in Deus-ex and Tomb Raider which have large constant arrays. Moving lowering before the opt loop in the GLSL linker fixes this and unexpectedly improves some compute shaders also.

Even before that, Kenneth Graunke sent in a patch to improve both Deus Ex: Mankind Divided and Tomb Raider even further:

QuoteHelps Deus Ex: Mankind Divided. The one shader with hurt spills/fills is from Tomb Raider at Ultra settings, but that same shader has a -39.55% reduction in instructions and -14.09% reduction in cycle counts, so it seems like a win there as well.

That one seems to affect both Intel and AMD.

On top of that, Kenneth Graunke along with Matt Turner put in another patch to cut "one shader's estimated cycles by -98.39%!", this one is for i965 (Intel):

QuoteA number of "Deus Ex: Mankind Divided" shaders use image atomics, but don't use the return value. Several of these were hitting WAW stalls for nearly 14,000 (poorly estimated) cycles a pop. Making dead code elimination null out the destination avoids this issue.

This patch cuts one shader's estimated cycles by -98.39%! Removing the message response should also help with data cluster bandwidth.

All really great to see and it looks like the Mesa release after Mesa 17 (the next stable version) is looking tasty already. Article taken from GamingOnLinux.com.