[Webkit-unassigned] [Bug 55158] Use VFP for double to int truncate in ARMv7

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=55158


Xan Lopez <xan.lopez at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |xan.lopez at gmail.com




--- Comment #2 from Xan Lopez <xan.lopez at gmail.com>  2011-02-24 10:01:41 PST ---
Two notes:

Someone suggested to check whether using FPSCR actually worked better. I tried doing something like:


#if 0
        m_assembler.vcvtr_S32_F64(fpTempRegisterAsSingle(), src);
        m_assembler.vmov(dest, fpTempRegisterAsSingle());
        m_assembler.vmrs();
        return makeBranch(ARMv7Assembler::ConditionVS);
#endif

but 3 tests fail. Is VS checking for all kinds of over and underflows? Is there something wrong in the code?

Also, here's the SunSpider compare for reference:

TEST                   COMPARISON            FROM                 TO             DETAILS

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

** TOTAL **:           1.006x as fast    2037.0ms +/- 0.2%   2024.6ms +/- 0.2%     significant

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

  3d:                  ??                 278.9ms +/- 1.0%    282.1ms +/- 1.4%     not conclusive: might be *1.011x as slow*
    cube:              *1.006x as slow*    93.5ms +/- 0.2%     94.1ms +/- 0.4%     significant
    morph:             ??                  65.8ms +/- 1.1%     66.9ms +/- 2.1%     not conclusive: might be *1.017x as slow*
    raytrace:          ??                 119.6ms +/- 1.7%    121.1ms +/- 2.0%     not conclusive: might be *1.013x as slow*

  access:              -                  274.2ms +/- 0.8%    273.8ms +/- 0.6% 
    binary-trees:      -                   59.2ms +/- 1.7%     59.2ms +/- 1.6% 
    fannkuch:          -                  103.4ms +/- 1.2%    103.1ms +/- 0.8% 
    nbody:             ??                  73.8ms +/- 0.2%     73.9ms +/- 0.1%     not conclusive: might be *1.002x as slow*
    nsieve:            -                   37.8ms +/- 0.6%     37.5ms +/- 0.5% 

  bitops:              ??                 146.9ms +/- 0.1%    147.2ms +/- 0.2%     not conclusive: might be *1.002x as slow*
    3bit-bits-in-byte: -                   23.0ms +/- 0.0%     23.0ms +/- 0.0% 
    bits-in-byte:      -                   32.0ms +/- 0.1%     32.0ms +/- 0.0% 
    bitwise-and:       *1.027x as slow*    39.8ms +/- 0.4%     40.9ms +/- 0.5%     significant
    nsieve-bits:       1.016x as fast      52.1ms +/- 0.2%     51.3ms +/- 0.3%     significant

  controlflow:         -                   16.0ms +/- 0.0%     16.0ms +/- 0.0% 
    recursive:         -                   16.0ms +/- 0.0%     16.0ms +/- 0.0% 

  crypto:              1.004x as fast     152.1ms +/- 0.2%    151.6ms +/- 0.1%     significant
    aes:               ??                  86.4ms +/- 0.2%     86.5ms +/- 0.2%     not conclusive: might be *1.001x as slow*
    md5:               -                   39.0ms +/- 0.0%     39.0ms +/- 0.0% 
    sha1:              1.027x as fast      26.7ms +/- 0.5%     26.0ms +/- 0.2%     significant

  date:                1.003x as fast     251.1ms +/- 0.2%    250.3ms +/- 0.2%     significant
    format-tofte:      -                  124.5ms +/- 0.2%    124.3ms +/- 0.2% 
    format-xparb:      1.005x as fast     126.6ms +/- 0.3%    125.9ms +/- 0.3%     significant

  math:                1.083x as fast     194.5ms +/- 0.1%    179.5ms +/- 0.1%     significant
    cordic:            1.32x as fast       61.4ms +/- 0.2%     46.4ms +/- 0.3%     significant
    partial-sums:      -                   92.1ms +/- 0.1%     92.1ms +/- 0.1% 
    spectral-norm:     -                   41.0ms +/- 0.1%     41.1ms +/- 0.2% 

  regexp:              -                   88.8ms +/- 0.4%     88.7ms +/- 0.2% 
    dna:               -                   88.8ms +/- 0.4%     88.7ms +/- 0.2% 

  string:              *1.002x as slow*   634.4ms +/- 0.1%    635.5ms +/- 0.1%     significant
    base64:            *1.006x as slow*    79.0ms +/- 0.2%     79.5ms +/- 0.3%     significant
    fasta:             -                   76.1ms +/- 0.1%     76.1ms +/- 0.1% 
    tagcloud:          *1.003x as slow*   146.6ms +/- 0.2%    147.0ms +/- 0.2%     significant
    unpack-code:       ??                 240.3ms +/- 0.1%    240.6ms +/- 0.2%     not conclusive: might be *1.001x as slow*
    validate-input:    -                   92.3ms +/- 0.2%     92.3ms +/- 0.2%

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.