Updates:
Blockedon: v8:3663 v8:3664 v8:3665
Comment #23 on issue 2935 by [email protected]: Poor performance in Ember
app Discourse
https://code.google.com/p/v8/issues/detail?id=2935
I looked at performance on MacBook and on Nexus7. Besides weak references
in optimized code (described by Slava in comment #18), I found two more
sources of poor performance:
- clearing of polymorphic ICs at each GC,
- code deoptimization caused by allocation site pretenuring.
I opened three issues to track each of these sources (v8:3663, v8:3664,
v8:3665).
MacOS results:
- Chrome with default flags: 203.53ms
- Chrome without weak references, IC clearing, allocation site pretenuring
[1]: 150.51ms
- Safari: 95.2ms
Nexus7 results:
- Chrome with default flags: 1845.93ms
- Chrome without weak references, IC clearing, allocation site pretenuring
[1]: 1426.85ms
[1]: with the following flags:
--js-flags="--noweak-embedded-maps-in-optimized-code
--nocleanup-code-caches-at-gc --noallocation-site-pretenuring"
Note that there is still unexplained time difference on MacOS and even more
on Nexus7. I am investigating this.
sam.saffron, stefan.penner: thanks! I am sure there is a way to make it
friendly to the optimizing compiler. I'll get in touch once I have deeper
understanding of Ember code. In the meantime I'd like to remove DOM from
the equation and get standalone test that can be run in d8 (v8 shell). Do
you think it would possible without investing huge effort?
--
You received this message because this project is configured to send all
issue notifications to this address.
You may adjust your notification preferences at:
https://code.google.com/hosting/settings
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
