https://bugs.freedesktop.org/show_bug.cgi?id=79842
Priority: medium Bug ID: 79842 Assignee: dri-devel@lists.freedesktop.org Summary: Performance drop since mesa 10.1.4 Severity: normal Classification: Unclassified OS: All Reporter: kontakt@ib-guder.de Hardware: Other Status: NEW Version: 10.1 Component: Drivers/Gallium/radeonsi Product: Mesa
Created attachment 100755 --> https://bugs.freedesktop.org/attachment.cgi?id=100755&action=edit The results from fumark benchmark
Hello.
I've noticed a performancedrop since mesa 10.1.4 (and also with 10.2.1 now). I ran fumark_benchmark and unigine_valley. I append the results for fumark. I don't know if it is kernel or mesa problem.
I use radeonsi.
Best regards! Tom
https://bugs.freedesktop.org/show_bug.cgi?id=79842
--- Comment #1 from Tom Guder kontakt@ib-guder.de --- Created attachment 100756 --> https://bugs.freedesktop.org/attachment.cgi?id=100756&action=edit Unigine Valley good
https://bugs.freedesktop.org/show_bug.cgi?id=79842
--- Comment #2 from Tom Guder kontakt@ib-guder.de --- Created attachment 100757 --> https://bugs.freedesktop.org/attachment.cgi?id=100757&action=edit Unigine Valley worse
https://bugs.freedesktop.org/show_bug.cgi?id=79842
--- Comment #3 from Alex Deucher agd5f@yahoo.com --- Can you bisect?
https://bugs.freedesktop.org/show_bug.cgi?id=79842
--- Comment #4 from Grigori Goronzy greg@chown.ath.cx --- Probably related to this:
http://cgit.freedesktop.org/mesa/mesa/commit/?h=10.1&id=1ba2298131924daf...
Anisotropic filtering was broken in radeonsi, and that fixed it. Anisotropic filtering has some overhead, which explains the slightly reduced performance.
This patch was applied to 10.1 and 10.2 branches as well as master.
https://bugs.freedesktop.org/show_bug.cgi?id=79842
Marek Olšák maraeo@gmail.com changed:
What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |RESOLVED Resolution|--- |WONTFIX
--- Comment #5 from Marek Olšák maraeo@gmail.com --- (In reply to Grigori Goronzy from comment #4)
Probably related to this:
http://cgit.freedesktop.org/mesa/mesa/commit/?h=10. 1&id=1ba2298131924daf34b4504ba748a782c5189f48
Anisotropic filtering was broken in radeonsi, and that fixed it. Anisotropic filtering has some overhead, which explains the slightly reduced performance.
This patch was applied to 10.1 and 10.2 branches as well as master.
Makes sense. Anisotropic filtering can indeed decrease performance a lot. Closing.
dri-devel@lists.freedesktop.org