I just landed https://bugzilla.mozilla.org/show_bug.cgi?id=572200 on mozilla-inbound, which makes a slight change to a performance vs. memory tradeoff in the style system.
In particular, in cases where we cache style data in the rule tree, this causes us to cache the data on any rule nodes that are referenced directly from a style context, rather than only caching a style struct on the highest rule node to which it's relevant. This increases instances of the nsResetStyleData (and, occasionally, nsInheritedStyleData) struct, which is an array of pointers to the different style structs that we only bother allocating at all if we have any structs. On the tp5_paint workload, the change seems to be about a 1%-2% performance improvement and a 0%-1% memory increase, depending on platform (see data in the bug). However, the real motivation for making the change is that it can give a more substantial performance improvement in the processing of dynamic style changes; for example, I measured a 14% performance improvement on a testcase reduced from panning the Gaia homescreen. Boris and I agreed that we should make this tradeoff, so I've landed the patches, but I thought it was worth sharing the motivation here, and warning others of the expected Talos notifications on dev-tree-management. -David -- 𝄞 L. David Baron http://dbaron.org/ 𝄂 𝄢 Mozilla http://www.mozilla.org/ 𝄂 _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
