628110 - Sunspider regression (350 -> 380ms)

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[JK]

User-Agent:       Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b9) Gecko/20100101 Firefox/4.0b9
Build Identifier: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b10pre) Gecko/20110122 Firefox/4.0b10pre

Working:

Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b9pre) Gecko/20110112 Firefox/4.0b9pre
http://hg.mozilla.org/tracemonkey/rev/a1d376c7a962

353.3ms +/- 0.8%

http://www2.webkit.org/perf/sunspider-0.9.1/sunspider-0.9.1/results.html?%7B%22v%22:%20%22sunspider-0.9.1%22,%20%223d-cube%22:%5B20,20,20,20,20,20,21,18,20,21%5D,%223d-morph%22:%5B13,13,13,13,12,13,13,12,13,13%5D,%223d-raytrace%22:%5B27,28,28,27,26,27,24,26,27,27%5D,%22access-binary-trees%22:%5B8,8,9,7,7,7,7,8,7,7%5D,%22access-fannkuch%22:%5B19,18,19,18,18,19,18,20,19,19%5D,%22access-nbody%22:%5B9,8,9,9,8,9,8,8,9,9%5D,%22access-nsieve%22:%5B9,10,9,8,9,9,9,9,8,9%5D,%22bitops-3bit-bits-in-byte%22:%5B1,1,1,0,1,1,0,1,1,1%5D,%22bitops-bits-in-byte%22:%5B9,11,10,9,10,9,9,10,10,10%5D,%22bitops-bitwise-and%22:%5B2,2,2,2,2,2,2,2,2,2%5D,%22bitops-nsieve-bits%22:%5B6,7,7,7,7,7,7,6,7,7%5D,%22controlflow-recursive%22:%5B3,3,3,3,4,3,3,3,3,3%5D,%22crypto-aes%22:%5B14,14,14,12,14,14,14,14,13,13%5D,%22crypto-md5%22:%5B5,6,5,6,6,5,6,5,6,5%5D,%22crypto-sha1%22:%5B4,5,4,5,5,5,5,5,5,5%5D,%22date-format-tofte%22:%5B26,22,26,23,25,26,25,25,25,26%5D,%22date-format-xparb%22:%5B29,26,28,33,28,28,28,29,27,28%5D,%22math-cordic%22:%5B11,11,11,10,11,11,13,13,10,11%5D,%22math-partial-sums%22:%5B14,14,15,13,13,14,13,14,14,14%5D,%22math-spectral-norm%22:%5B10,11,11,10,11,11,11,11,11,11%5D,%22regexp-dna%22:%5B17,17,17,17,17,18,16,17,18,17%5D,%22string-base64%22:%5B5,6,6,7,7,6,6,6,6,6%5D,%22string-fasta%22:%5B19,17,19,18,18,20,18,19,19,18%5D,%22string-tagcloud%22:%5B27,29,29,28,29,29,28,29,29,28%5D,%22string-unpack-code%22:%5B33,31,32,31,32,32,31,33,32,32%5D,%22string-validate-input%22:%5B13,13,13,12,14,13,12,13,11,12%5D%7D

Regressed:

Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b10pre) Gecko/20110113 Firefox/4.0b10pre
http://hg.mozilla.org/tracemonkey/rev/8cb2574bf7c1

380.1ms +/- 5.3%

http://www2.webkit.org/perf/sunspider-0.9.1/sunspider-0.9.1/results.html?%7B%22v%22:%20%22sunspider-0.9.1%22,%20%223d-cube%22:%5B26,25,26,26,27,26,25,26,25,28%5D,%223d-morph%22:%5B13,13,14,12,13,13,13,13,13,13%5D,%223d-raytrace%22:%5B28,28,27,29,26,26,23,27,26,28%5D,%22access-binary-trees%22:%5B7,7,7,7,6,6,6,6,7,6%5D,%22access-fannkuch%22:%5B19,19,19,19,19,20,19,19,18,19%5D,%22access-nbody%22:%5B9,9,10,10,10,9,10,9,9,10%5D,%22access-nsieve%22:%5B8,9,9,9,9,9,9,9,9,9%5D,%22bitops-3bit-bits-in-byte%22:%5B1,1,1,1,1,0,0,1,1,1%5D,%22bitops-bits-in-byte%22:%5B9,10,9,9,9,10,9,9,9,9%5D,%22bitops-bitwise-and%22:%5B3,2,2,2,2,2,2,3,2,2%5D,%22bitops-nsieve-bits%22:%5B7,7,7,7,7,7,7,7,6,7%5D,%22controlflow-recursive%22:%5B4,3,3,3,3,3,3,3,3,3%5D,%22crypto-aes%22:%5B14,14,15,13,13,14,14,12,14,12%5D,%22crypto-md5%22:%5B5,6,5,5,6,6,5,6,5,5%5D,%22crypto-sha1%22:%5B5,4,5,5,4,5,5,5,4,5%5D,%22date-format-tofte%22:%5B28,28,25,27,26,27,23,34,27,27%5D,%22date-format-xparb%22:%5B28,31,28,28,31,29,28,27,101,29%5D,%22math-cordic%22:%5B11,12,10,11,11,12,10,12,10,10%5D,%22math-partial-sums%22:%5B14,14,13,14,14,14,13,13,14,13%5D,%22math-spectral-norm%22:%5B11,12,11,11,11,10,11,11,12,10%5D,%22regexp-dna%22:%5B17,16,16,16,17,17,16,17,17,16%5D,%22string-base64%22:%5B7,6,6,6,7,6,67,6,6,5%5D,%22string-fasta%22:%5B20,20,19,19,20,20,19,19,20,19%5D,%22string-tagcloud%22:%5B29,29,29,29,100,28,29,29,30,27%5D,%22string-unpack-code%22:%5B31,31,31,31,38,31,31,30,32,31%5D,%22string-validate-input%22:%5B13,12,12,12,13,14,12,12,12,13%5D%7D

Pushlog:
http://hg.mozilla.org/tracemonkey/pushloghtml?fromchange=a1d376c7a962&tochange=8cb2574bf7c1

Reproducible: Always

Comment 1

[Boris Zbarsky [:bzbarsky]]

JK, what's the corresponding TM regression range?

I wish the awfy regress graphs were working.   :(

Comment 2

[JK]

(In reply to comment #1)
> JK, what's the corresponding TM regression range?
> 
> I wish the awfy regress graphs were working.   :(

That's the TM regression range.

Comment 3

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

I bisected and it looks like the patch for 624827 is when it started. It removed some calls to MaybeGC to reduce pause times. Unfortunately, we need these calls during the SunSpider run. They're what allow us to GC in between benchmarks, rather than during the benchmarks themselves.

I'm not sure how to resolve this. Maybe someone more familiar with Flight of the Navigator has an idea?

Comment 4

[Andreas Gal :gal]

Thanks for the bisect. Gregor and Patrick are probably best qualified to fix this.

Comment 5

[Andreas Gal :gal]

Its a major perf regression so I am afraid this is a hard blocker.

Comment 6

[Patrick Walton (:pcwalton)]

Attachment: Proposed patch.

This is due to the patch in bug 624867. Requesting approval to back out.

Comment 7

[Andreas Gal :gal]

Comment on attachment 506561 [details] [diff] [review]
Proposed patch.

Please investigate further though. Maybe there is a better fix.

Comment 8

[Patrick Walton (:pcwalton)]

http://hg.mozilla.org/mozilla-central/rev/f28ee131df7a

Moving discussion to bug 624867.