User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b4pre) Gecko/20100805 Minefield/4.0b4pre Build Identifier: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b4pre) Gecko/20100805 Minefield/4.0b4pre The SVG dice http://ie.microsoft.com/testdrive/Performance/SVGDice/Default.xhtml looks to be as smooth as IE9 until the dice intersect with the gradient/graphics or each other. The 'Vegas' theme has the worst performance, while 'low fidelity' maintains a fast framerate for the longest. IE9 maintains high framerate throughout Reproducible: Always Steps to Reproduce: 1. Load SVG Dice and 'roll' 2. Change 'themes' and roll again to observe framerate differences. Actual Results: Observe smooth animation of dice when they first appear. When the dice cover gradient background or another dice framerate is drastically reduced. Expected Results: Framerate should not drop dramatically over gradients and other dice D2D on.

Blocking on the investigation; if this turns out to be unimportant, feel free to unblock.

There are various SVG bugs with similar examples. The underlying issue is that when we move an object we invalidate where it once was. If two overlapping objects move we get two invalidates, 3=3 etc. The number of redraws grows rapidly with multiple overlapping objects. I think once we move to display lists for SVG (no bug for that yet) they can coalesce the invalidates.

Robert, or JWatt, could you confirm here for me then that this is -not- related to graphics performance?

I can't confirm that. I profiled the testcase on Mac, in a debug build even, and we're spending almost all our time in nested calls to nsSVGUtils::PaintFrameWithEffects. A lot of computing mask alphas and suchlike. We're also spending TONS of time in io_alltraps, which indicates memory churn probably due to temporary surfaces. Maybe we can speed up graphics stuff, maybe we should be trying to cache the alpha surface for masks. However, I do not recommend blocking FF4 on this.

The patches in bug 586954 might help here.