[Webkit-unassigned] [Bug 274043] REGRESSION (277125 at main): 10% performance regression in Web Assembly

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=274043

--- Comment #12 from Christian Speckner <cnspeckn at googlemail.com> ---
(In reply to Justin Michaud from comment #11)
> Pull request: https://github.com/WebKit/WebKit/pull/28990

Just built this PR, and the slowdown is gone �� Speed is now on par again with current Safari, both with the call_indirect heavy version of the emulator and with the current codebase.

In case you want or need to do further benchmarking, I have extended https://cloudpilot-emu.github.io/uarm-indirect/ to include a slider to increase the MIPS cap. Note that you'll have to load the SD / ROM image files again. I have also uploaded a SD card image with two more benchmarks to https://cspeckner.de/benchmark.img .

The first new benchmark is yCPUBench. If you start it and click "Benchmark armlet" you'll get a number that can be compared (provided you set the MIPS to 300 in order to avoid the cap).

The other benchmark is TCPMP which is a video player. Start it (takes a few seconds), click the title to open the "File" menu, select "open files" from the menu, then select "slot 2" and open "ED.avi". If you then select "Benchmark" from the file menu it'll start decoding as fast as possible, and if you click the video after a few seconds you'll get a quantitative benchmark that can be compared (again, set MIPS to 300 to avoid hitting the cap).

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20240523/8ba54ce6/attachment.htm>