While scrolling engadget.com with APZ and e10s enabled on a retina macbook pro, I can get 20ms+ to render at [1]. This doesn't show up at all on Windows with a GPU. The rect we're filling is quite large as well: devPxRect is (0.000000, -32254.000000) (2880.000000, 49176.000000) Draw target size: 3072, 3840 Transform is: [1.000000, 0.000000] [0.000000, 1.000000] [0.000000, 254.000000] Total time: 21.682419 The call tree through instruments looks like: Running Time Self (ms) Symbol Name 469.0ms 100.0% 0.0 nsDisplayCanvasBackgroundColor::Paint(nsDisplayListBuilder*, nsRenderingContext*) 469.0ms 100.0% 0.0 mozilla::gfx::DrawTargetTiled::FillRect(mozilla::gfx::RectTyped<mozilla::gfx::UnknownUnits> const&, mozilla::gfx::Pattern const&, mozilla::gfx::DrawOptions const&) 468.0ms 99.7% 0.0 mozilla::gfx::DrawTargetCG::FillRect(mozilla::gfx::RectTyped<mozilla::gfx::UnknownUnits> const&, mozilla::gfx::Pattern const&, mozilla::gfx::DrawOptions const&) 459.0ms 97.8% 1.0 CGContextFillRects 457.0ms 97.4% 0.0 ripc_DrawRects 445.0ms 94.8% 1.0 ripc_Render 439.0ms 93.6% 0.0 RIPLayerBltShape 438.0ms 93.3% 1.0 argb32_mark 437.0ms 93.1% 1.0 CGBlt_fillBytes 404.0ms 86.1% 404.0 _platform_memset_pattern8$VARIANT$Unknown 32.0ms 6.8% 32.0 _platform_memset_pattern16$VARIANT$Unknown Setting tile sizes to 1024x1024 makes it 4x faster, but can still take up quite a bit of time. The times it takes very long are usually when the draw target size is > 2880 by 1800, the resolution of the retina display. It might be the problem of just the first draw to a large surface is slow. Somehow, see if we can make this faster. It looks similar to the profiles I see of scrolling bugzilla and the background gradient. [1] https://dxr.mozilla.org/mozilla-central/source/layout/generic/nsCanvasFrame.cpp?from=nsDisplayCanvasBackgroundColor::Paint&case=true#339

Mason, it's been a year since this was looked at. Is performance getting any better?

(In reply to Anthony Hughes (:ashughes) [GFX][QA][Mentor] from comment #1) > Mason, it's been a year since this was looked at. Is performance getting any > better? Not sure, I haven't taken a look at it in a long time. Skia should have helped but not sure with this case. I'll close as WFM, and reopen if need be.