673364 - SunSpider regression on ARM Cortex-A8.

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Jacob Bramley [:jbramley]]

I've noticed a SunSpider regression from early June. It affects only the Cortex-A8 boards I've tested, and only on SunSpider. (At least, not on V8. I didn't check Kraken.) The A8 boards have much less memory than my A9 board, so this could be memory-related somehow.

The regression occurred somewhere between 7e6f3b179644 and cc612a118e7c. Plenty of stuff went on between those revisions, though the majority of the changesets got pulled in with a merge from mozilla-central.

Comment 1

[Mike Shaver (:shaver emeritus)]

How big is the regression, and on which tests?  Might help narrow it down.

Comment 2

[Jacob Bramley [:jbramley]]

Good questions. The overall regression was about 8.5%, but only on
SunSpider and only for my A8 boards.

Between 7e6f3b179644 and cc612a118e7c:


TEST                   COMPARISON      
=======================================
** TOTAL **:           *1.083x as slow*
=======================================

  3d:                  *1.104x as slow*
    cube:              *1.089x as slow*
    morph:             *1.067x as slow*
    raytrace:          *1.133x as slow*

  access:              *1.076x as slow*
    binary-trees:      *1.029x as slow*
    fannkuch:          *1.014x as slow*
    nbody:             *1.29x as slow* 
    nsieve:            *1.010x as slow*

  bitops:              *1.012x as slow*
    3bit-bits-in-byte: *1.040x as slow*
    bits-in-byte:      ??              
    bitwise-and:       *1.011x as slow*
    nsieve-bits:       ??              

  controlflow:         *1.032x as slow*
    recursive:         *1.032x as slow*

  crypto:              *1.076x as slow*
    aes:               *1.071x as slow*
    md5:               *1.088x as slow*
    sha1:              *1.073x as slow*

  date:                *1.073x as slow*
    format-tofte:      *1.055x as slow*
    format-xparb:      *1.093x as slow*

  math:                *1.23x as slow* 
    cordic:            1.020x as fast  
    partial-sums:      *1.47x as slow* 
    spectral-norm:     *1.148x as slow*

  regexp:              1.148x as fast  
    dna:               1.148x as fast  

  string:              *1.054x as slow*
    base64:            *1.085x as slow*
    fasta:             *1.075x as slow*
    tagcloud:          *1.049x as slow*
    unpack-code:       *1.041x as slow*
    validate-input:    *1.054x as slow*


(I skipped out the absolute times to keep the bug log clean.)

Note that a couple of tests got faster! I'm not here next week, but I'll
investigate when I get back if no-one else gets there first.

Comment 3

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee didn't login in Bugzilla in the last 7 months, so the assignee is being reset.

Comment 4

[Mayank Bansal]

Sunspider is low priority these days.
Closing as incomplete.