571595 - TM: make sure dense_grow always gets inlined

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[David Mandelin [:dmandelin]]

Attachment: Patch

It looks like I get a 1% boost in SS:


TEST                   COMPARISON            FROM                 TO
 DETAILS

=============================================================================

** TOTAL **:           1.010x as fast    544.5ms +/- 0.9%   539.1ms +/- 0.5%
 significant

=============================================================================

  3d:                  -                  75.0ms +/- 2.3%    73.9ms +/- 0.8%
    cube:              -                  26.1ms +/- 1.6%    26.0ms +/- 1.3%
    morph:             -                  16.1ms +/- 6.8%    15.1ms +/- 1.5%
    raytrace:          -                  32.8ms +/- 1.4%    32.8ms +/- 1.4%

  access:              1.048x as fast     71.4ms +/- 1.8%    68.1ms +/- 1.4%
 significant
    binary-trees:      1.041x as fast     12.7ms +/- 2.7%    12.2ms +/- 2.5%
 significant
    fannkuch:          1.056x as fast     33.9ms +/- 1.8%    32.1ms +/- 1.6%
 significant
    nbody:             -                  16.3ms +/- 2.1%    15.9ms +/- 2.6%
    nsieve:            1.076x as fast      8.5ms +/- 4.4%     7.9ms +/- 2.9%
 significant

  bitops:              -                  25.5ms +/- 3.6%    25.1ms +/- 2.1%
    3bit-bits-in-byte: ??                  0.4ms +/- 92.3%     0.5ms +/- 75.3%
   not conclusive: might be *1.25x as slow*
    bits-in-byte:      -                   8.1ms +/- 6.5%     8.0ms +/- 4.2%
    bitwise-and:       -                   2.3ms +/- 15.0%     2.2ms +/- 13.7%
    nsieve-bits:       -                  14.7ms +/- 3.3%    14.4ms +/- 2.6%

  controlflow:         -                   5.6ms +/- 6.6%     5.5ms +/- 6.8%
    recursive:         -                   5.6ms +/- 6.6%     5.5ms +/- 6.8%

  crypto:              -                  31.8ms +/- 2.3%    31.1ms +/- 2.8%
    aes:               -                  18.4ms +/- 4.2%    17.9ms +/- 4.0%
    md5:               -                   8.5ms +/- 4.4%     8.4ms +/- 4.4%
    sha1:              -                   4.9ms +/- 4.6%     4.8ms +/- 6.3%

  date:                -                  66.4ms +/- 2.1%    66.0ms +/- 1.4%
    format-tofte:      -                  39.2ms +/- 1.7%    39.1ms +/- 1.3%
    format-xparb:      -                  27.2ms +/- 3.0%    26.9ms +/- 2.0%

  math:                -                  32.1ms +/- 1.9%    31.7ms +/- 1.5%
    cordic:            -                  18.6ms +/- 3.2%    18.1ms +/- 1.2%
    partial-sums:      -                  10.2ms +/- 3.0%     9.9ms +/- 2.3%
    spectral-norm:     ??                  3.3ms +/- 10.5%     3.7ms +/- 9.3%
  not conclusive: might be *1.121x as slow*

  regexp:              1.033x as fast     28.4ms +/- 1.8%    27.5ms +/- 1.4%
 significant
    dna:               1.033x as fast     28.4ms +/- 1.8%    27.5ms +/- 1.4%
 significant

  string:              ??                208.3ms +/- 0.9%   210.2ms +/- 0.6%
 not conclusive: might be *1.009x as slow*
    base64:            ??                 12.0ms +/- 2.8%    12.1ms +/- 1.9%
 not conclusive: might be *1.008x as slow*
    fasta:             ??                 19.6ms +/- 2.5%    19.9ms +/- 2.0%
 not conclusive: might be *1.015x as slow*
    tagcloud:          -                  70.7ms +/- 1.0%    70.4ms +/- 1.0%
    unpack-code:       ??                 57.4ms +/- 1.5%    57.9ms +/- 1.5%
 not conclusive: might be *1.009x as slow*
    validate-input:    ??                 48.6ms +/- 4.1%    49.9ms +/- 2.1%
 not conclusive: might be *1.027x as slow*

Comment 1

[Nicholas Nethercote [inactive]]

It's a good change, but I've already got it in my WIP patch for bug 566767, and dense_grow() has changed and been renamed so it will cause conflicts if you land this patch.

Do you mind holding off and letting it be part of bug 566767?

Comment 2

[Nicholas Nethercote [inactive]]

Comment on attachment 450747 [details] [diff] [review]
Patch

I've changed my mind -- there's enough uncertainty around bug 566767 that it's worth getting this improvement in now.  I'll deal with the conflicts.

Comment 3

[David Mandelin [:dmandelin]]

http://hg.mozilla.org/tracemonkey/rev/47136f7328b5

Comment 4

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/47136f7328b5