add best support SMP & SMP bench

2002-10-06
2002-10-07
  • jean-philippe perois

    I made some modification in your code:
      - optimise the C code ... on my 2xATHLON XP1800+ it is 20/25 % faster!
      - Rewrite the SMP part I use a complete synchonised computation:
       the man thread only start the compute(s) one whith a FlushEvent()!
       and wait all thread to finish (WaitForMultipleObjects)
       and then print the point!
         => no more Sleep(), and the balance is best!
      - take the compute in a new fonction and call it in the benchs and the readraw!
         => the benchs use SMP computer correctly!

    But I can't compile the MMX and the 3DNOW! code if you can tel me what to do for that!

     
    • Daniele Paccaloni

      Nice work, i'll test it asap.
      ---SMP---
      I recommend to test the SMP code faking the detection of the number of CPUs (just fake in 2000 CPUs==threads and see what happens).
      It will be integrated in the next release.
      ---C code---
      It will be integrated in the next release.
      ---compiling---
      If you are using VisualStudio v6, please download ServicePack 5 for the new instructions support.
      VisualStudio v7 (.NET) should have it integrated.

      Thanks,
      .Daniele

       
    • jean-philippe perois

      OK send you my code (.h and .cpp files...)
        Test it on a 2CPU (AMD Athlon XP1800+)
      and 20 thread!

      /==== my score ;-) are =================================
      /==== new code:  =======================================
      FFFF v3.xxxx BENCHMARK (Using 2 CPU, no render)
      Sys perf timer freq: 1526210000 Hz
      size:  500*500
      maxiters:      9999
      rangex:        -2.00 to 1.00
      rangey:        -1.50 to 1.50  FPU ASM benchmark:
          2.005 sec
          209.678 MegaIters/sec
        FPU C benchmark:
          4.282 sec
          98.189 MegaIters/sec
        DEV benchmark:
          1.949 sec
          215.703 MegaIters/sec
      /=======================================================
      the DEV is a test with FPU ASM... juste a liter bether for me so...
      the C FPU is the new one... (old one give me ~75?)

      /==== v3.1.0b BENCHMARK ================================
      FFFF v3.1.0b BENCHMARK (Using 1 CPU, no render)
      Sys perf timer freq: 1526210000 Hz
      size:  500*500
      maxiters:      9999
      rangex:        -2.00 to 1.00
      rangey:        -1.50 to 1.50
        SSE benchmark:
          1.929 sec
          218.456 MegaIters/sec
        3DNow! benchmark:
          2.460 sec
          171.292 MegaIters/sec
        FPU ASM benchmark:
          4.027 sec
          104.410 MegaIters/sec
        FPU C benchmark:
          10.518 sec
          39.976 MegaIters/sec
      /=======================================================

       
    • jean-philippe perois

      after compiling the SSE and the 3DNows code this is the perfo with the next release (v3.2.0???)

        Later
            Djip
      ================================================================================
      FFFF v3.xxxx BENCHMARK (Using 2 CPU, no render)
      Sys perf timer freq: 1526210000 Hz
      size:  500*500
      maxiters:      9999
      rangex:        -2.00 to 1.00
      rangey:        -1.50 to 1.50
        SSE benchmark:
          0.964 sec
          437.034 MegaIters/sec
        3DNow! benchmark:
          1.230 sec
          342.489 MegaIters/sec
        FPU ASM benchmark:
          2.028 sec
          207.319 MegaIters/sec
        FPU C benchmark:
          4.306 sec
          97.650 MegaIters/sec
        DEV benchmark:
          1.949 sec
          215.755 MegaIters/sec
        GPU VP benchmark (beta! maxiters=60) on Radeon 8500 DDR x86/MMX/3DNow!/SSE:
          Not supported.
      =================================================================================

       

Log in to post a comment.

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:

JavaScript is required for this form.





No, thanks