Open Bug 1431220 Opened 2 years ago Updated 2 years ago

SpiderMonkey runs `source-map` library benchmarks slower and with more variation than V8 and JSC do

Categories

(Core :: JavaScript Engine, defect, P2)

defect

Tracking

()

Tracking Status
firefox60 --- affected

People

(Reporter: fitzgen, Unassigned)

Details

Attachments

(6 files)

Sometimes much slower, often with 5-10x the variance that V8/JSC have.

See the Firefox + JavaScript entries in the graphs I'm attaching.

STR:

* Clone this repo and check out the "control" branch: https://github.com/fitzgen/source-map/tree/control

* From the root of the repo, run `npm install && npm run build && python -m SimpleHTTPServer`

* Navigate to "localhost:8000/bench/bench.html"

* Click the button for the benchmark you want to run
Note the last data point in JS + Firefox vs JS + Safari

bbouvier mentions that this is spending a lot of time in String.prototype.split with a regexp argument, which doesn't get JITed, and we'll fix that in the library, but the perf difference vs safari is scary.
Note the variance vs others here.
Jan, would you take a look?
Flags: needinfo?(jdemooij)
Priority: -- → P2
mraleph and I looked at this a while ago, bug 1437471 came out of that. This code has been ported to Rust/Wasm since so I assume it's not blocking anything.
Flags: needinfo?(jdemooij)
FWIW, jsc seems to do a little bit better than SM on the "set first breakpoint" benchmark still: http://fitzgeraldnick.com/media/set.first.breakpoint.with.sub.sorting.scalajs.svg

Don't know if it is worth leaving this open for that or not.
You need to log in before you can comment on or make changes to this bug.