609698 - performance regression for android vm shell v8.5/optimized/splay of about 5%

Status: RESOLVED WONTFIX

(Tamarin Graveyard :: Virtual Machine, defect, P3)

Description

[Dan Schaffer]

the 30 day performance chart shows about a 5% regression only on android on v8.5.  Only the optimized/splay.as test seemed to regress.  Tracked down the injection to between 5338 to 5342.  I manually built 5341 and reproduced the regression and 5339 did not build properly. 

on a nexxus one with release shell
builds before show ~750-760ms
5338: 765ms
5342: 685ms
builds after show: ~670-680ms

to reproduce I was doing:
$ export AVM=tr/platform/android/android_shell.sh
$ adb push tr/platform/android/android_runner.sh /data/app (simple shell script to return exit code)
$ adb push avmshell /data/app
$ adb chmod 777 /data/app/avmshel
$ $AVM (verifies the build # if you are using a build from asteam/builds
$ $AVM tr/test/performance/v8.5/optimized/splay.abc
(outputs result in operations/sec)

Comment 1

[Dan Schaffer]

I suspect this revision as the injection point but I do not see use of ByteArray in the test.  But the change seems to affect android more than mac or win.  Is there some collateral damage from the change possibly?

suspecting:
changeset:   5341:a06e4632e581
parent:      5339:fd4c454fb615
user:        Steven Johnson <[e-mail]>
date:        Thu Oct 07 13:48:21 2010 -0700
summary:     Bug 564248 - Unify ByteArray between Tamarin and Flash Player (Pass #1) (r=wsharp)

Comment 2

[Edwin Smith]

5338 looks capable of tweaking things subtly too. CC'ing stejohns.  Have you ruled out 5338 as a possible injection point?  You said 5338-5342, but its not clear whether you mean inclusive or exclusive.

Comment 3

[Dan Schaffer]

Tracked down to 5342/5341.   

build ops/sec
5337  755
5338  766
5339  773
5340  762
5341  n/a compile errors
5342  687 (merge)  ** bingo
5343  662

changeset:   5342:6cf6894728c8
parent:      5341:a06e4632e581
parent:      5340:bda2725fd58a
user:        Steven Johnson <[e-mail]>
date:        Thu Oct 07 13:48:42 2010 -0700
summary:     merge

changeset:   5341:a06e4632e581
parent:      5339:fd4c454fb615
user:        Steven Johnson <[e-mail]>
date:        Thu Oct 07 13:48:21 2010 -0700
summary:     Bug 564248 - Unify ByteArray between Tamarin and Flash Player (Pass #1) (r=wsharp)

Comment 4

[Steven Johnson]

That's truly weird -- I can't imaging why 5342 would inject this. I'll do a diff and examine, but...

Comment 5

[Tommy Reilly]

When last worked on that test I found it be to be wildly dependent on inlining.  From the #'s isn't %10 regression more accurate?

Comment 6

[Dan Schaffer]

the overall v8.5 change was -5%, the splay change is ~-15%.

Comment 7

[Sylvestre Ledru [:Sylvestre]]

Tamarin is a dead project now. Mass WONTFIX.

Comment 8

[Sylvestre Ledru [:Sylvestre]]

Tamarin isn't maintained anymore. WONTFIX remaining bugs.