• [^] # Re: Performances ?

    Posté par (site web personnel) . En réponse à la dépêche LLVM 2.2 : Un concurrent pour GCC ?. Évalué à 6.

    Ah oui, l'url semble HS maintenant. Une copie avec Google Cache :

    gcc 4.0.1(apple) (latest gcc from Xcode Tools)
    $ gcc -o bmt -O3 -DSMALL himenoBMTxps.c
    $ ./bmt
    ...
    Gosa : 1.167853e-04
    MFLOPS measured : 544.017453 cpu : 56.726971
    Score based on Pentium III 600MHz : 6.634359

    gcc 4.2
    $ gcc-4.2 -o bmt -O3 -DSMALL himenoBMTxps.c
    $ ./bmt
    ...
    Gosa : 1.753087e-05
    MFLOPS measured : 953.891485 cpu : 54.242544
    Score based on Pentium III 600MHz : 11.632823

    Emit LLVM code with llvm-gcc-4-2.0, then exec it with LLVM JIT.
    $ ~/src/llvm-gcc4-2.0-x86-darwin8/bin/llvm-gcc -emit-llvm \
    -O3 -DSMALL -c himenoBMTxps.c
    $ ~/src/llvm-2.0/Release/bin/lli himenoBMTxps.o

    Gosa : 2.229028e-05
    MFLOPS measured : 1147.841139 cpu : 42.767418
    Score based on Pentium III 600MHz : 13.998063

    The result tells me that LLVM JIT does good job(20% faster than natively geneted code by gcc-4.2),
    altough we must pay attention that "Gosa"(which means computational error in Japanese) is a bit higher than gcc-4.2’s result.
    gcc-4.0.1 is fairly bad... I don’t know why...