Closed Bug 1451203 Opened 3 years ago Closed 3 years ago
https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=e450e7797b7f8d5d8d286ed72e6fbc1676728db2&framework=1&filter=tsvgx%20opt&selectedTimeRange=172800 This is basically svg invalidation. Blob image invalidation will help here but we'll probably need to improve things a bit more.
A current profile on mac with my current patches: https://perfht.ml/2GY8XYk
One reason we'll be worse than current Firefox is because we invalidate a rect, while Firefox invalidates a region. This test case gets hurt because of that.
3 years ago
Priority: -- → P1
Still very bad: https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=654f5f3b2eb3cf3de3fa89418b2ca46b4590e989&framework=1&showOnlyImportant=1&selectedTimeRange=172800 70% regression on linux, 160% regression on windows.
Comparison against gecko with direct2d disabled (try push still pending): https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=5c24f72331677ccf375a7d85760189ff8b8cef25&newProject=try&newRevision=654f5f3b2eb3cf3de3fa89418b2ca46b4590e989&framework=1
Still a large regression. This regression is bigger than I'd expect which suggests there's something unusual going on. We should try to get profiles/reproduce locally.
Here's a profile https://perfht.ml/2Ju1RrT It looks like we're blocked on vsync whilst asking to composite at a fantastic rate.
Actually we don't seem blocked on vsync in the profile above, but are in this profile: https://perfht.ml/2HuRNOV I don't know why this would be different.
this got way worse, tsvgx is up to 220% on windows, 130% on other platforms
Let's use bug 1416652 for tracking the tsvgx regression.
Status: NEW → RESOLVED
Closed: 3 years ago
Resolution: --- → DUPLICATE
Duplicate of bug: 1416652
You need to log in before you can comment on or make changes to this bug.