Closed
Bug 1351102
Opened 8 years ago
Closed 8 years ago
Up to 8 times slower JS performance compared to Chrome in attached code
Categories
(Core :: JavaScript Engine: JIT, defect)
Tracking
()
RESOLVED
FIXED
mozilla55
| Tracking | Status | |
|---|---|---|
| firefox55 | --- | fixed |
People
(Reporter: Boris.Letocha, Assigned: jandem)
References
Details
Attachments
(2 files)
|
(deleted),
text/plain
|
Details | |
|
(deleted),
patch
|
h4writer
:
review+
|
Details | Diff | Splinter Review |
User Agent: Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/57.0.2987.110 Safari/537.36
Steps to reproduce:
Run attached JS code in simple page:
<html>
<body>
<script src="code.js"></script>
</body>
</html>
Actual results:
On my Intel notebook Windows Firefox 50.1 displays this in console:
Bobx observable get+set 1M: timer started
Bobx observable get+set 1M: 213.97ms
1010000
Bobx observable get+set 1M: timer started
Bobx observable get+set 1M: 203.21ms
1010000
Expected results:
I would like to see much better performace.
Same code in latest Chrome 57 it writes to console (run it with closed dev tools):
Bobx observable get+set 1M: 39.5ms
1010000
Bobx observable get+set 1M: 25.8ms
1010000
Sorry for still relatively long code to reproduce.
|
||
Updated•8 years ago
|
Component: Untriaged → JavaScript Engine
Product: Firefox → Core
|
Assignee | |
Updated•8 years ago
|
Flags: needinfo?(jdemooij)
|
|
Assignee | |
Comment 1•8 years ago
|
||
Thanks for filing this bug. (Note that Firefox 50.1 is no longer supported.)
I get the following numbers on OS X:
Firefox 52.0.1: 120 ms
Safari: 54 ms
Nightly (55): 45 ms
Chrome: 20 ms
So our Nightly builds are almost 3x faster than Firefox 52, but still slower than Chrome.
|
|
Assignee | |
Comment 2•8 years ago
|
||
This is really weird. In our JS shell and sometimes in my Nightly build I get numbers like 3-8 ms. I wonder if it's some inlining or GC quirk.
Boris, would you mind trying a Nightly build on Windows and measuring with the console closed?
|
|
Assignee | |
Updated•8 years ago
|
Status: UNCONFIRMED → NEW
Component: JavaScript Engine → JavaScript Engine: JIT
Ever confirmed: true
|
|
Assignee | |
Comment 3•8 years ago
|
||
Hm there seems to be a problem with console.time/timeEnd. If I replace that with:
var t = new Date;
...
console.log(new Date - t);
I consistently get 3-8 ms on Firefox Nightly and Chrome is still at 23 ms with that, so we're way faster.
I'll do some more digging.
|
|
Assignee | |
Comment 4•8 years ago
|
||
Tracked it down to the same-compartment check in GetFinishedBuilder. Replacing that with a same-runtime check gives us the 4-5x faster than Chrome numbers all the time.
(What was happening is that console.time() (also reproducible with console.log) entered system JS, then we got the interrupt and the finished off-thread compilation got ignored.)
|
||
Comment 5•8 years ago
|
||
Comment on attachment 8851973 [details] [diff] [review]
Patch
Review of attachment 8851973 [details] [diff] [review]:
-----------------------------------------------------------------
Good find!
Attachment #8851973 -
Flags: review?(hv1989) → review+
|
Reporter | |
Comment 6•8 years ago
|
||
I can confirm that in current nightly (55.0a1 2017-03-28) (64bit) I am seeing randomly 65ms or 7ms. So if this patch will make it consitently faster way, it will be tremendous.
Now maybe I should report it to V8 guys :-)
Pushed by jandemooij@gmail.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/70fcc672ffd5
Fix GetFinishedBuilder to check the runtime instead of the compartment. r=h4writer
|
||
Comment 8•8 years ago
|
||
| bugherder | ||
Status: NEW → RESOLVED
Closed: 8 years ago
status-firefox55:
--- → fixed
Resolution: --- → FIXED
Target Milestone: --- → mozilla55
You need to log in
before you can comment on or make changes to this bug.
Description
•