Created attachment 48791 [details] read-front.png mesa: 6bde225b8b5791588837295b3b89ac132095a6f7 (master) Run piglit read-front on softpipe or llvmpipe. The test no longer passes. $ ./bin/read-front Probe at (0,0) Expected: 1.000000 0.000000 0.000000 Observed: 0.000000 0.000000 0.000000 Probe at (50,0) Expected: 0.000000 1.000000 0.000000 Observed: 0.000000 0.000000 0.000000
git bisect good 1a7e17e44a1129bbd6a0f454fe95b3ce8240cd45 git bisect bad a221807dc5d69598afd1d0e0a4e715fb82a9f30d In between the above commits the builds are affected by bug 38896.
Thanks for bisecting, Vinson. The only reasonable explanation for this problem is that with the glx state tracker, an X roundtrip is required each and every time the state tracker (mesa or vega) checks whether to validate its drawables. I think that was the old behaviour. That will become pretty costly.
These other piglit tests are also failing and have the same bisect result as comment #1. fp-fragment-position fp-incomplete-tex fp-kil fp-lit-mask trinity-fp1
Created attachment 48852 [details] [review] Patch to fix the problem This was actually a bug that was previously masked by the fact that the glx state tracker always re-invalidated the drawable on each validate() call. Now when that doesn't happen anymore, there was nothing to mark the drawable invalid when the mesa state tracker added a new color renderbuffer. Patch is out for review.
This issue should now be resolved in mesa master.
mesa: fc98444bd58960e6cab28423365923bc7e7af3e1 (master) Verified fixed.
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.