790663 - IonMonkey: v8 Regex is getting slower over time

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Nicolas B. Pierron [:nbp]]

(Jean Claveau from Bug 790628 comment #6)
> I often run the http://v8.googlecode.com/svn/data/benchmarks/v7/run.html
> benchmark to have an overview of Nightly optimizations.
> 
> The results I had with the build of today morning were extremly different of
> usually.
> 
> Those last 4 months my score have grown from 4900 to ~5750 with almost the
> same results : really good NovierStrokes, stable regexp at 1000pts, stable
> crypto at 12000.
> 
> When I benched today the first global score was really huge (almost 6800pts)
> but not stable : the more I relaunch the bench, the more the global and the
> regexp scores go down (each time lower and after a dozen of consecutivs
> benchs, the regexp score is at 36pts!).
> 
> The crypto bench lost almost 2500 pts and seems stable at 9500 (12000
> yesterday).
> 
> The good point is that EarlyBoyer's score increased significantly (If I have
> good memories).
> 
> 
> My reflex was to profile the bench and I've got a weird result when I
> analyzed it : Ffx crashed
> 
> Here is the report : bp-3d3e9ddd-bb9f-49f0-a178-d34f52120912 and I repeated
> it once to check if I could reproduce it
> bp-1f19349b-6ef8-4234-a974-4a3602120912
> 
> This line pushed to post here :
> 7 	libxul.so 	EnterIon 	js/src/ion/Ion.cpp:1325
> 
> 
> I didn't remember enough scores to tell you more but I prefer to communicate.
> 
> Hope it will help.

Thanks for reporting, I moved it to a dedicated bug since it does not seems related to the previous one.

Comment 1

[Jean Claveau]

This issue doesn't occur anymore on my pc and the regexp score has massively increased with the last Ion optimizations. :)

Comment 2

[Jan de Mooij [:jandem]]

(In reply to Jean Claveau from comment #1)
> This issue doesn't occur anymore on my pc and the regexp score has massively
> increased with the last Ion optimizations. :)

Yeah a lot has happened since September, so I'm closing this as WFM. Thanks!