in reply to Re: Convert string to array - performance challenge
in thread Convert string to array - performance challenge

Very interesting! With the EXACT same code as Ikegami, (with the (a)* mod), I get: 
           Rate    regex    split unpack_C unpack_a
regex    3.00/s       --      -3%      -9%     -19%
split    3.09/s       3%       --      -6%     -17%
unpack_C 3.29/s      10%       7%       --     -11%
unpack_a 3.71/s      24%      20%      13%       --

[download]
I have seen before that the Ikegami machine is faster than mine on certain benchmarks. But, I am unable to understand how split() can be the fastest. I am running a Prescott processor, an old 3 Ghz hyper-threaded critter. With Active State Perl 5.10.1.

Replies are listed 'Best First'.
Re^3: Convert string to array - performance challenge
by Anonymous Monk on Apr 08, 2010 at 07:08 UTC
    I think it depends a lot on the system, c-runtime 
    ActivePerl v5.8.9 built for MSWin32-x86-multi-thread, Binary build 825
+ [288577]
           Rate unpack_C unpack_a    split    regex
unpack_C 3.48/s       --      -5%      -7%      -9%
unpack_a 3.66/s       5%       --      -3%      -4%
split    3.76/s       8%       3%       --      -1%
regex    3.82/s      10%       4%       1%       --

           Rate unpack_C    regex unpack_a    split
unpack_C 3.56/s       --      -5%      -5%      -6%
regex    3.74/s       5%       --      -0%      -1%
unpack_a 3.74/s       5%       0%       --      -1%
split    3.79/s       7%       2%       1%       --


Rate    split unpack_C    regex unpack_a
split    3.45/s       --      -4%      -9%     -25%
unpack_C 3.58/s       4%       --      -6%     -22%
regex    3.79/s      10%       6%       --     -18%
unpack_a 4.60/s      34%      29%      21%       --

    [download]
    The newer versions (built using MinGW) are more consistent between runs 
    C:\perl\5.10.1\bin\MSWin32-x86-multi-thread\perl.exe

           Rate unpack_C    regex    split unpack_a
unpack_C 3.66/s       --      -2%     -15%     -22%
regex    3.74/s       2%       --     -14%     -21%
split    4.33/s      18%      16%       --      -8%
unpack_a 4.70/s      28%      26%       9%       --


C:\perl\5.11.1\bin\MSWin32-x86-multi-thread\perl.exe


           Rate    regex unpack_C    split unpack_a
regex    2.74/s       --     -28%     -38%     -39%
unpack_C 3.82/s      39%       --     -13%     -14%
split    4.40/s      60%      15%       --      -2%
unpack_a 4.47/s      63%      17%       2%       --

    [download]
[reply]
[d/l]
[select]

      I see some variablility, but nothing like those levels using AS perls.

      AS1007 64-bit: 

      c:\test>junk
           Rate    regex unpack_C unpack_a    split
regex    7.95/s       --      -1%     -10%     -14%
unpack_C 8.01/s       1%       --     -10%     -14%
unpack_a 8.88/s      12%      11%       --      -4%
split    9.30/s      17%      16%       5%       --

c:\test>junk
           Rate    regex unpack_C unpack_a    split
regex    7.95/s       --      -0%      -8%     -12%
unpack_C 7.95/s       0%       --      -8%     -12%
unpack_a 8.68/s       9%       9%       --      -4%
split    9.09/s      14%      14%       5%       --

c:\test>junk
           Rate unpack_C    regex unpack_a    split
unpack_C 8.02/s       --      -0%      -9%     -11%
regex    8.02/s       0%       --      -9%     -11%
unpack_a 8.81/s      10%      10%       --      -2%
split    9.02/s      13%      13%       2%       --

c:\test>junk
           Rate unpack_C    regex unpack_a    split
unpack_C 7.95/s       --      -0%      -9%     -13%
regex    7.95/s       0%       --      -9%     -13%
unpack_a 8.74/s      10%      10%       --      -5%
split    9.16/s      15%      15%       5%       --

      [download]

      AS826 32-bit: 

      c:\test>\perl32\bin\perl junk.pl
           Rate unpack_C    split    regex unpack_a
unpack_C 8.38/s       --      -5%      -9%     -15%
split    8.81/s       5%       --      -5%     -11%
regex    9.23/s      10%       5%       --      -6%
unpack_a 9.86/s      18%      12%       7%       --

c:\test>\perl32\bin\perl junk.pl
           Rate unpack_C    split    regex unpack_a
unpack_C 8.45/s       --      -5%      -8%     -15%
split    8.87/s       5%       --      -3%     -11%
regex    9.16/s       8%       3%       --      -8%
unpack_a 9.94/s      18%      12%       9%       --

c:\test>\perl32\bin\perl junk.pl
           Rate unpack_C    split    regex unpack_a
unpack_C 8.38/s       --      -5%      -8%     -14%
split    8.81/s       5%       --      -4%     -10%
regex    9.16/s       9%       4%       --      -6%
unpack_a 9.78/s      17%      11%       7%       --

c:\test>\perl32\bin\perl junk.pl
           Rate unpack_C    split    regex unpack_a
unpack_C 8.38/s       --      -5%      -8%     -14%
split    8.81/s       5%       --      -4%     -10%
regex    9.16/s       9%       4%       --      -6%
unpack_a 9.78/s      17%      11%       7%       --

      [download]
      Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
      "Science is about questioning the status quo. Questioning authority".
      In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
[d/l]
[select]

        split // was optimised to not use the regex engine on 9 Aug 2007. The 5.8 build you are using probably predates that.

        I got the same variability between regex and unpack_C with 5.10. They are basically equivalent in terms of performance.

[reply]
[d/l]

In Section Seekers of Perl Wisdom