(In reply to lpy750 from comment #13) > And yes, using Mozregression, I had to manually set gfx.webrender.dcomp-win.enabled to false each time it downloaded a different build. That's too late. The pref must be set before Firefox starts otherwise its change won't have an effect. You can set a custom preference on the second page in Mozregression: https://youtu.be/IwrWot3jVFI?t=137 At the moment it seems like you have found a performance regression of WEBRENDER_COMPOSITOR (which uses the OS compositor instead of WebRender's internal draw compositor), but not the performance regression of Firefox Stable. It would be good to be sure if the regression range you have found is for gfx.webrender.dcomp-win.enabled=false or true. Both configurations might have a performance regression with the same or different cause. Firefox 90 + 91 have another bug: With gfx.webrender.dcomp-win.enabled=false (restart required after changing the pref) you should normally not see any artifacts (bug 1638709 should only occur with true), but it is currently affected by bug 1712969 comment 5.
Bug 1720263 Comment 14 Edit History
Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.
(In reply to lpy750 from comment #13) > And yes, using Mozregression, I had to manually set gfx.webrender.dcomp-win.enabled to false each time it downloaded a different build. That's too late. The pref must be set before Firefox starts otherwise its change won't have an effect. You can set a custom preference on the second page in Mozregression: https://youtu.be/IwrWot3jVFI?t=137 At the moment it seems like you have found a performance regression of WEBRENDER_COMPOSITOR (which uses the OS compositor instead of WebRender's internal draw compositor), but not the performance regression of Firefox Stable. It would be good to be sure if the regression range you have found is for gfx.webrender.dcomp-win.enabled=false or true. Both configurations might have a performance regression with the same or different cause. Firefox 90 and later have another bug: With gfx.webrender.dcomp-win.enabled=false (restart required after changing the pref) you should normally not see any artifacts (bug 1638709 should only occur with true), but it is currently affected by bug 1712969 comment 5.
(In reply to lpy750 from comment #13) > And yes, using Mozregression, I had to manually set gfx.webrender.dcomp-win.enabled to false each time it downloaded a different build. That's too late. The pref must be set before Firefox starts otherwise its change won't have an effect. You can set a custom preference on the second page in Mozregression: https://youtu.be/IwrWot3jVFI?t=137 At the moment it seems like you have found a performance regression of WEBRENDER_COMPOSITOR (which uses the OS compositor instead of WebRender's internal draw compositor), but not the performance regression of Firefox Stable. It would be good to be sure if the regression range you have found is for gfx.webrender.dcomp-win.enabled=false or true. Both configurations might have a performance regression with the same or different cause. Since Firefox 90 there is another bug: With gfx.webrender.dcomp-win.enabled=false (restart required after changing the pref) you should normally not see any artifacts (bug 1638709 should only occur with true), but it is currently affected by bug 1712969 comment 5.