Re: Turning on stack protection by default

[Tobias Nygren <tnn%NetBSD.org@localhost>]

On Sun, 25 Oct 2009 12:15:59 +0000
Matthias Scheler <tron%zhadum.org.uk@localhost> wrote:

> So unless somebody has the time to repeat Andrew Doran's NetBSD 5.0
> benchmarking and a way to figure out how to benchmark Firefox
> I'm afraid nothing is going to be done about this. I'm sorry.

Below are two runs of SunSpider on firefox-3.5.3, NetBSD-5.99.21-amd64
before and after installation of a USE_SSP=yes userland. As you can see, 
the performance difference is negligible and within the margin of error.
During build of pkgsrc/devel/glib2 I see a performance penalty of 1-2% in
real time spent, averaged over a few rounds.

TEST                   COMPARISON            FROM                 TO            
 DETAILS

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

** TOTAL **:           -                 2017.2ms +/- 0.2%   2018.6ms +/- 0.4% 

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

  3d:                  -                  240.6ms +/- 0.3%    239.8ms +/- 0.4% 
    cube:              1.01x as fast       90.8ms +/- 0.6%     90.0ms +/- 0.0%  
   significant
    morph:             -                   69.8ms +/- 0.8%     69.2ms +/- 0.8% 
    raytrace:          *1.01x as slow*     80.0ms +/- 0.0%     80.6ms +/- 1.8%  
   significant

  access:              *1.01x as slow*    339.0ms +/- 0.4%    341.8ms +/- 1.5%  
   significant
    binary-trees:      -                   32.6ms +/- 2.1%     32.2ms +/- 1.7% 
    fannkuch:          *1.01x as slow*    138.0ms +/- 0.0%    139.0ms +/- 1.3%  
   significant
    nbody:             *1.02x as slow*    128.8ms +/- 0.8%    131.2ms +/- 3.0%  
   significant
    nsieve:            -                   39.6ms +/- 1.7%     39.4ms +/- 1.7% 

  bitops:              ??                 259.0ms +/- 1.1%    261.2ms +/- 3.0%  
   not conclusive: might be *1.01x as slow*
    3bit-bits-in-byte: -                   39.6ms +/- 1.7%     39.0ms +/- 2.3% 
    bits-in-byte:      -                   57.8ms +/- 1.0%     57.6ms +/- 1.2% 
    bitwise-and:       ??                  94.6ms +/- 2.7%     97.8ms +/- 7.4%  
   not conclusive: might be *1.03x as slow*
    nsieve-bits:       1.00x as fast       67.0ms +/- 0.0%     66.8ms +/- 0.8%  
   significant

  controlflow:         -                   29.4ms +/- 2.3%     29.2ms +/- 1.9% 
    recursive:         -                   29.4ms +/- 2.3%     29.2ms +/- 1.9% 

  crypto:              -                  124.6ms +/- 0.5%    124.0ms +/- 0.7% 
    aes:               -                   50.2ms +/- 1.1%     50.0ms +/- 1.8% 
    md5:               -                   36.2ms +/- 1.5%     36.0ms +/- 0.0% 
    sha1:              -                   38.2ms +/- 1.5%     38.0ms +/- 0.0% 

  date:                *1.01x as slow*    167.8ms +/- 0.6%    169.8ms +/- 1.0%  
   significant
    format-tofte:      -                   71.4ms +/- 1.0%     71.2ms +/- 0.8% 
    format-xparb:      *1.02x as slow*     96.4ms +/- 1.2%     98.6ms +/- 1.4%  
   significant

  math:                1.01x as fast      238.4ms +/- 0.6%    236.6ms +/- 0.6%  
   significant
    cordic:            *1.00x as slow*     95.0ms +/- 0.0%     95.2ms +/- 0.6%  
   significant
    partial-sums:      -                  100.8ms +/- 1.8%     99.0ms +/- 1.3% 
    spectral-norm:     -                   42.6ms +/- 1.6%     42.4ms +/- 1.6% 

  regexp:              -                  191.4ms +/- 2.2%    188.8ms +/- 0.9% 
    dna:               -                  191.4ms +/- 2.2%    188.8ms +/- 0.9% 

  string:              -                  427.0ms +/- 0.5%    427.4ms +/- 0.3% 
    base64:            *1.02x as slow*     38.4ms +/- 1.8%     39.2ms +/- 1.4%  
   significant
    fasta:             -                  105.2ms +/- 1.0%    104.4ms +/- 0.7% 
    tagcloud:          ??                  85.6ms +/- 1.9%     86.8ms +/- 0.6%  
   not conclusive: might be *1.01x as slow*
    unpack-code:       -                  135.4ms +/- 2.1%    134.4ms +/- 1.1% 
    validate-input:    ??                  62.4ms +/- 1.1%     62.6ms +/- 1.1%  
   not conclusive: might be *1.00x as slow*