GigaFLOPS

--=-agUS2KbiVO1spnbSPaBO
Content-Type: text/plain
Content-Transfer-Encoding: quoted-printable

On Fri, 2003-08-08 at 16:23, Bryce C wrote:
> Anyone else got something SIMPLE? Or does someone want to tell me how to
> do this HPL with LAM on my nice little box here beside me?

It really depends on how acurate you want to be. I believe that you can
download the SPECfp tests, and those will give you reasonable numbers -
but you can get those from your processor vendor. Do you want
benchmarks that are just processor based, or do you want some that work
with the memory subsystem? Benchmarks are a tricky game.

If you just want a number try this:

int main (int argc, char * argv[]) {
    float mynum =3D 1.0;
    int i;

    for (i =3D 0; i < 1000000; i++) {
        mynum +=3D 0.1;
    }

    printf("%f\n", mynum);
    return 0;
}

Put that in a file 'myfile.c' and do this:

gcc -O0 -o myfile myfile.c
time ./myfile

You'll then know how long it takes to do roughly a million floating
point adds, which could be considered 'floating point operations'. This
doesn't take into account anything like cache latency, which effects
real world performance pretty significantly (and is one of the key
factors between workstations and PCs :)

        --Ted

--=-agUS2KbiVO1spnbSPaBO
Content-Type: application/pgp-signature; name=signature.asc
Content-Description: This is a digitally signed message part

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)

iD8DBQA/NEj2LE335pRPGp0RAkgYAKCk7Rh39TUf2c10wU8utICxxRgkvgCguz+u
CVWle9TdXEJfJm1AcXmR/mQ=
=Uk52
-----END PGP SIGNATURE-----

--=-agUS2KbiVO1spnbSPaBO--