31926 - for() loops significantly slower than MSIE

Status: VERIFIED WORKSFORME

(Core :: JavaScript Engine, defect, P3)

Description

[Morten Wang]

I'll upload a test case for this, a simple for() loop that does exactly nothing, 
but does it 10 million times.  tested it in MSIE 5.01, Netscape Comm 4.72 and 
Mozilla build 200031419.  result was (all lapsed time in milliseconds):

Mozilla: 15100, 15100, 15210, 15110, 15100
IE: 8620, 8570, 8790, 8680, 8730
NC4.72: 12960, 13010, 13010, 13010, 13020

since I'm not the one to decide whether this is a significant problem or not, 
I'll leave it up to you guys.

Comment 1

[Morten Wang]

Attachment: for() loop speed test

Comment 2

[Morten Wang]

adding keyword 'testcase' as per bugathon guidelines

Comment 3

[Gwyn Judd]

hmm...interestingly I found mozilla was slightly faster than netscape 4.x. I
don't have IE though so I couldn't check that

Comment 4

[Gwyn Judd]

cc'ing me. oh by the way the reporter doesn't need to cc him/herself as it
happens automatically :)

Comment 5

[Morten Wang]

I figured out the Cc'ing thing after submitting another bug, tried to remove 
myself again now. :)

downloaded build 200042608, and as b.judd found the speed has increased 
somewhat.  I clock in at around 12500ms now, which is around half a second 
faster than NC4.72.

Comment 6

[Gwyn Judd]

hmm I just tried 2000040708 and it's slower :(. I have no real idea how to test
performance bugs though. Someone needs to try profiling this beast and seeing
what's happening (not me though I can't build).

Comment 7

[Gervase Markham [:gerv]]

14000 in Moz, 16480 in NS for me. That's a PIII 450 :-)

Confirming, adding perf keyword.

Gerv

Comment 8

[rogerl (gone)]

[SPAM] Bug triage for nsbeta3

Comment 9

[John Bandhauer]

This is a dup of the fixed bug 43902. If you try your test again I think you'll 
find the score much improved.

*** This bug has been marked as a duplicate of 43902 ***

Comment 10

[Morten Wang]

Since I cannot verify any significant speed increase as of build 2000090508, I 
reopen this bug.  testing with the attached testcase returns scores in the 
vicinity of 15000ms, give or take half a second.  I do get two alerts about a 
script slowing Mozilla down, and quickly dispatch these.

I am not sure if the problem is now associated with the alerts, or whatnot.  a 
for() loop with 1,000,000 iterations run at the same speed as my Communicator 
4.75 (~1350ms).

I'm still able to duplicate the test results with IE, so we're still 7 seconds 
behind.

Comment 11

[Phil Schwartau]

Updating QA contact -

Comment 12

[Phil Schwartau]

Attachment: for() loop speed test for standalone JS shell

Comment 13

[Phil Schwartau]

I now get these timings for ten million iterations using the two 
testcases above: 


Mozilla binary 2001030310 WinNT :  10000 ms
In optimized JS shell on WinNT  :   7900 ms


Since this represents a significant improvement over the figures
originally reported, I would like to close this as WORKSFORME.
Notice the Mozilla figure would be even lower if it weren't for 
the two warning alertboxes one has to dismiss during the test.


Morten, could you confirm this with a recent build? If you agree,
please mark this bug "Verified"; otherwise, reopen it - thanks.

Comment 14

[Morten Wang]

I'm getting consistent <10s runs in the browser testcase with build 2001052204,
therefore I mark it verified.  thanks! :)

Comment 15

[James Paige]

Anybody happen to know if there is a bug filed already for the fact that the
alert comes up more than once for the same script? I would think it ought to
remember if you say "No"