[Webkit-unassigned] [Bug 162095] Speed up Heap::isMarkedConcurrently

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=162095

--- Comment #1 from JF Bastien <jfbastien at apple.com> ---
So far I've:
 - Compiled the code for A64 with barrier, and with dependency ordering;
 - Changed Heap.cpp:measurePhaseTiming to return true;
 - Ran jsc a few times on-device with a self-contained splay.js (from Octane);
 - grep "Heap::markRoots convergence took" | tail -n2

This gives the last GC:Eden and GC:Full entries for each run, which I then averaged to get the following numbers:
  GC:Eden is 93% average runtime with consume ordering than it is with the barrier.
  GC:Full is 76% average runtime with consume ordering than it is with the barrier.
These measurements are fairly noisy, though!

I ran Octane and Kraken, it reports:
  31.0664+-0.2420 30.9790+-0.1619 might be 1.0028x faster

I'm running them for longer, and added JSBench.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20160916/498a8038/attachment-0001.html>