I have written a patch to radeonhd-1.2.5 that speeds up EXA acceleration between 5% to 10%. It won't be incorporated upstream because it makes the code to hard to maintain. It would be just fine for an ebuild though. People with fast video cards, slow cpus, or both will benefit the most. Its great for laptops that have their cpus throttled back to save power. Reproducible: Always Steps to Reproduce: 1. benchmark with the command gtkperf -c 1000 2. save results 3. <optional> benchmark with command x11perf -v1.3 -all > unpatched.log 4. <optional> wait for two and a half hours for x11perf to complete 5. apply patch, compile, and install 6. benchmark with the command gtkperf -c 1000 7. compare with results obtained in step 1 8. <optional> benchmark with command x11perf -v1.3 -all > patched.log 9. <optional> wait for two and a half hours for x11perf to complete 10. <optional> compare x11perf logs with the command x11perfcomp -r unpatched.log patched.log > results.txt 11. <optional> view log with the command less results.txt 12. enjoy the faster driver.
Created attachment 193019 [details] speed up patch
Thanks for the patch, but if upstream haven't accepted this because it's too difficult to maintain, how can the gentoo developers be expected to maintain it? Each time there's a version bump, users will lose your speed up, and the devs will start getting bugs filed against the performance drop until they adapt the patch. Given that it's a 168k patch, I doubt any dev will want the task... I strongly urge that you try to address the concerns upstream have over accepting your patch and get it integrated into the main source. I'll assign this to the appropriate developers, just in case they're interested, but don't be surprised if this bug gets closed with UPSTREAM or similar.
(In reply to comment #2) > Thanks for the patch, but if upstream haven't accepted this because it's too > difficult to maintain, how can the gentoo developers be expected to maintain > it? Each time there's a version bump, users will lose your speed up, and the > devs will start getting bugs filed against the performance drop until they > adapt the patch. Given that it's a 168k patch, I doubt any dev will want the > task... I don't expect that the gentoo developers maintain it. I also can see why you don't want to include it if users will gripe when the next version bump comes. I am currently maintaining my patch and keeping it up to date with the mainline git. > > I strongly urge that you try to address the concerns upstream have over > accepting your patch and get it integrated into the main source. > > I'll assign this to the appropriate developers, just in case they're > interested, but don't be surprised if this bug gets closed with UPSTREAM or > similar. > Unfortunately to get the speed I have had to write the code to in such a way that it will probably never get accepted up stream. I have done massive amounts of manual inlining and coded things in such away to work around compiler short comings. Sadly changing these things would drop the gains to 1 or 2% at best. My main reason for posting this patch here was done because others requested that I do it. http://forums.gentoo.org/viewtopic-t-767363.html I can fully understand why people won't want to include it in an official ebuild. Perhaps someone could use it to make an unofficial experimental ebuild though.
x11 team does not need to be cced here, we definetly wont apply something like this to driver. Maintainer can decide otherwise but then it is up to him to support the stuff :] Cheers
Or an overlay might be a good place for this.
