1942919 - Measure performance impact of removing asm.js

Status: RESOLVED FIXED

(Core :: JavaScript: WebAssembly, task, P3)

Description

[Ryan Hunt [:rhunt]]

We would like to deprecate asm.js and remove it from our code base. This would reduce the amount of code we need to support. We do see usage trending downward overtime. Right now we're at about 0.15% of page loads using asm.js [1].

The nice part about removing asm.js support is that it's still valid JS and so the code still works, but will be slower. How much slower will determine when we can remove support or not.

I think it'd be useful to find usage of asm.js on the web and measure how much slower it is when we disable asm.js. The javascript.options.asmjs can help here.

The chromestatus link also has a sample of top sites that use asm.js that we can look through.

[1] https://chromestatus.com/metrics/feature/timeline/popularity/473

Comment 1

[Ryan Hunt [:rhunt]]

I've done some ad-hoc measurements of asm.js games from chromestatus top links and couldn't observe a difference with/without asm.js.

I also found the MASSIVE asm.js benchmark and was able to run it with/without asm.js. This is probably the absolute worst case for asm.js deprecation. Results are here. Score reductions were anywhere from 25% to 90%.

Comment 2

[Ryan Hunt [:rhunt]]

We've gathered enough data for now.