Bug 110995

Summary:	performance regression in Heroes of the Storm with Mesa 19.1.1 & Polaris
Product:	Mesa	Reporter:	tempel.julian
Component:	Drivers/Vulkan/radeon	Assignee:	mesa-dev
Status:	RESOLVED MOVED	QA Contact:	mesa-dev
Severity:	normal
Priority:	medium
Version:	19.1
Hardware:	x86-64 (AMD64)
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:

Description tempel.julian 2019-06-25 14:37:41 UTC

I lose at least 10 fps with 19.1.1 vs. 19.0 on RX 580.

19.1 (at least 168 fps):
https://abload.de/img/screenshot_20190625_16akrb.png

19.1.1 (not more than 158fps):
https://abload.de/img/screenshot_20190625_1kij8n.png


Performance regressions first occured in mesa-git around a week ago which is why I switched to stable 19.0. mesa-git was even worse though with ~150fps. Probably related to work regarding DCC?


linux 5.1
dxvk-git
both LLVM 8 & 9

Comment 1 Samuel Pitoiset 2019-06-25 14:45:57 UTC

Can you try to revert 410759091173fa61436ba46baeb20a79d5767849 ?

Btw, 158 vs 168 is really a small diff.

Comment 2 tempel.julian 2019-06-25 16:17:00 UTC

It's an additional loss on top of DXVK vs. native D3D11 performance, so it sums up to something worse. :)

I could bisect it to this commit:
https://github.com/mesa3d/mesa/commit/0a313cc285c2939de9cac07f045b0b699bc208ca
With it reverted, performance in mesa-master is normal.

Comment 3 tempel.julian 2019-07-03 12:04:42 UTC

Is there a chance to make CTS dEQP-VK.pipeline.depth_range_unrestricted.* pass without causing the performance regression?

Comment 4 tempel.julian 2019-09-18 09:24:13 UTC

This is unfortunately unchanged with latest master:
~148fps where it was ~162fps.
amdvlk scores  ~155fps, and with pageflipping + amdgpu.dc=0 ~160fps

I can't workaround this anymore by reverting 0a313cc285c2939de9cac07f045b0b699bc208ca, as radv stopped building successfully without it.

Comment 5 Samuel Pitoiset 2019-09-18 09:58:31 UTC

Should be fixed by
https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2025

Comment 6 tempel.julian 2019-09-18 10:30:57 UTC

Confirmed, performance is back to >160fps. Thanks a lot, your efforts are really outstanding!

Comment 7 Samuel Pitoiset 2019-09-18 11:23:21 UTC

Can you please wait before closing tickets? Sending a change doesn't necessarily mean that it will be pushed/accepted. Thanks!

Comment 8 tempel.julian 2019-09-18 11:34:28 UTC

Of course, sorry.

Comment 9 Samuel Pitoiset 2019-09-18 13:51:43 UTC

Can you please pull the MR and try again? Does the performance regression is still fixed? I would say yes, just want a confirmation.

Comment 10 tempel.julian 2019-09-18 14:18:04 UTC

With the updated MR, performance unfortunately is worse again.

Comment 11 Samuel Pitoiset 2019-09-18 14:39:29 UTC

Okay, that's unfortunate but I prefer correctness vs performance. I will wait for the MR to be reviewed and let you know what happens.

Comment 12 GitLab Migration User 2019-09-18 20:07:09 UTC

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/863.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.