1322724 - Regression caused by the splitting of Ionbuilder

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect, P1)

Comment 1

[Hannes Verschore [:h4writer]]

The splitting caused some regressions. Tracking in this bug:
https://treeherder.mozilla.org/perf.html#/alerts?id=4482
https://treeherder.mozilla.org/perf.html#/alerts?id=4483
https://treeherder.mozilla.org/perf.html#/alerts?id=4480


https://treeherder.mozilla.org/perf.html#/alerts?status=0&framework=5&filter=1322724&hideTo=1&page=1

Comment 2

[Hannes Verschore [:h4writer]]

Attachment: Disable optimization

Disable this optimization I added during the ionbuilder split. It seems to introduce some regressions. I think it is related to adding a block in a bad place that interrupts the flow of the true branch. I'll open follow-up bug to try to enable it again.

Comment 3

[Hannes Verschore [:h4writer]]

Attachment: Patch

Fix for this.
1) We didn't set the hit counts on the extra block added in visitTest to allow for optimizing the false branch. Which resulted in pruning not taking place. This was the bug I encountered when investigating.
2) Ditto for table switch
3) Improvement on branches where a second block was added on the false branch and make sure it gets added just before the false branch. Instead of just after the true branch.

Comment 4

[Jan de Mooij [:jandem]]

Comment on attachment 8817703 [details] [diff] [review]
Patch

Review of attachment 8817703 [details] [diff] [review]:
-----------------------------------------------------------------

Good finds.

::: js/src/jit/IonBuilder.cpp
@@ +1513,5 @@
>      if (!setCurrentAndSpecializePhis(mblock))
>          return false;
>      graph().addBlock(mblock);
>  
> +

Nit: rm second blank line.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/b877875abb3a
IonMonkey - Add the hit count information on the extra false branch blocks, r=jandem

Comment 6

[Phil Ringnalda (:philor)]

https://hg.mozilla.org/mozilla-central/rev/b877875abb3a

Comment 7

[Joel Maher ( :jmaher ) (UTC -8)]

this looks to have given is a perf boost on android perf tests:
== Change summary for alert #4567 (as of December 20 2016 21:47 UTC) ==

Improvements:

  3%  local-blank summary android-4-4-armv7-api15 opt      1287.25 -> 1253.89
  2%  remote-blank summary android-4-4-armv7-api15 opt     1307.07 -> 1277.98
  2%  local-blank summary android-4-4-armv7-api15 opt      1285.41 -> 1258.49

For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=4567

I see wins in many of our other tests, that is just the alerts we have :)