Try Out New Apple G5 Compiler

cookiescookies
Posted:
in Current Mac Hardware edited January 2014
Here's an interesting real G5 test.



I just got the new Apple gcc compiler with the huge list of G5-specific optimizations and compiled two of the most widely-used JPEG compression and decompression programs.



http://www.ijg.org/



The first programs are compiled with standard "-O2" optimizations, and the second programs are compiled for G5 only.



Download the items from here:



http://www.myersdaily.org/joseph/not...eg-time.tar.gz



Open the file named "time.command" to time the compression and decompression times. This will give exact values for the amount of real/user/system time use.



(More JPEG files can be placed into the directory named "images" in order to expand the test as much as you want.)

Comments

  • Reply 1 of 12
    scottscott Posts: 7,431member
    And did you try this on a G5 and what did you get?
  • Reply 2 of 12
    michaelbmichaelb Posts: 242member
    That's probably why he posted it here. I won't be able to test my app's optimizations either until the Dual 2.0 ships.



    Someone try them, please?
  • Reply 3 of 12
    gabidgabid Posts: 477member
    From a stock G5 1.8 GHz with a bunch of apps running and processor speed set to Automatic:



    Quote:

    images/flood-water-wedding

    decode



    real 0m0.065s

    user 0m0.030s

    sys 0m0.030s

    encode



    real 0m0.114s

    user 0m0.090s

    sys 0m0.000s

    images/foo

    decode



    real 0m0.119s

    user 0m0.050s

    sys 0m0.060s

    encode



    real 0m0.358s

    user 0m0.230s

    sys 0m0.090s

    images/kitty-3

    decode



    real 0m0.034s

    user 0m0.010s

    sys 0m0.030s

    encode



    real 0m0.062s

    user 0m0.040s

    sys 0m0.000s

    images/flood-water-wedding

    decode



    real 0m0.058s

    user 0m0.020s

    sys 0m0.020s

    encode



    real 0m0.085s

    user 0m0.040s

    sys 0m0.030s

    images/foo

    decode



    real 0m0.090s

    user 0m0.030s

    sys 0m0.050s

    encode



    real 0m0.236s

    user 0m0.180s

    sys 0m0.050s

    images/kitty-3

    decode



    real 0m0.032s

    user 0m0.010s

    sys 0m0.020s

    encode



    real 0m0.063s

    user 0m0.030s

    sys 0m0.010s

    logout

    [Process completed]





    I'll try again later with other apps off and with the Highest setting. How do these numbers look though?
  • Reply 4 of 12
    machemmachem Posts: 319member
    Quote:

    Originally posted by Gabid

    From a stock G5 1.8 GHz with a bunch of apps running and processor speed set to Automatic:









    I'll try again later with other apps off and with the Highest setting. How do these numbers look though?



    ..are you serious? They look columnar. And totally obfuscated.



    Other than that they look good.
  • Reply 5 of 12
    gabidgabid Posts: 477member
    Quote:

    Originally posted by machem

    ..are you serious? They look columnar. And totally obfuscated.



    Other than that they look good.







    Does anyone out there have comparative scores for other machine?
  • Reply 6 of 12
    cookiescookies Posts: 61member
    Quote:

    Originally posted by Gabid

    From a stock G5 1.8 GHz with a bunch of apps running and processor speed set to Automatic:









    I'll try again later with other apps off and with the Highest setting. How do these numbers look though?



    Essentially, your results mean that a fast, well-optimized program is improved in general more than 10% by G5-specific optimizations. The best result that I see is this one:



    images/foo (non-G5)



    user 0m0.230s (non-G5)

    user 0m0.180s (G5)



    That represents the real operating time of the program within the G5 chip.



    The first list represents results from -O2 optimization:



    images/flood-water-wedding



    decode

    user 0m0.030s

    encode

    user 0m0.090s



    images/foo



    decode

    user 0m0.050s

    encode

    user 0m0.230s



    images/kitty-3



    decode

    user 0m0.010s

    encode

    user 0m0.040s





    The second is from the G5.



    images/flood-water-wedding

    decode

    user 0m0.020s

    encode

    user 0m0.040s



    images/foo

    decode

    user 0m0.030s

    encode

    user 0m0.180s





    images/kitty-3

    decode

    user 0m0.010s

    encode

    user 0m0.030s
  • Reply 7 of 12
    krassykrassy Posts: 595member
    Quote:

    Originally posted by cookies

    ages/flood-water-wedding



    non G5 encode user 0m0.090s

    now G5 encode user 0m0.040s



    more than doubling the performance in this case







  • Reply 8 of 12
    mark_wilkinsmark_wilkins Posts: 154member
    How about dl'ing the public beta of IBM's XLC compiler and trying that?



    -- Mark
  • Reply 9 of 12
    carson o'geniccarson o'genic Posts: 1,279member
    Quote:

    Originally posted by machem

    ..are you serious? They look columnar. And totally obfuscated.



    Other than that they look good.



    That was very funny!
  • Reply 10 of 12
    michaelbmichaelb Posts: 242member
    Quote:

    Originally posted by mark_wilkins

    How about dl'ing the public beta of IBM's XLC compiler and trying that?



    Excellent idea.



    How easy is it to install over gcc 3.3?
  • Reply 11 of 12
    mark_wilkinsmark_wilkins Posts: 154member
    It's built to coexist with GCC 3.3 and Apple's developer tools. The README describes it all.



    -- Mark
  • Reply 12 of 12
    cookiescookies Posts: 61member
    Quote:

    Originally posted by cookies

    Here's an interesting real G5 test.





    I just made the script better so that anyone can see which time is G5-optimized and which time is normal.



    If I can download the XLC compiler I might try, but I don't need it for myself.



    This funny guy posted his note about it here:



    http://babble.risckyworkings.com/arc...m_g5_compiler/
Sign In or Register to comment.