Re: gfortran

[Dominique Rousset <dominique.rousset@free.fr>]

Le Lundi 24 Octobre 2005 23:00, Lennart Sorensen a écrit :
> Do the results seem to match as well numerically?  I imagine you are
> probably doing floating point, which is certainly expected to be faster
> in 64bit mode since it uses sse rather than x87 instructions.
Further investigations indicate an extra time step on AMD64 which result in 
one extra floating point value per record. 
The number of time steps is computed using floating point arithmetics, so very 
small differences may result in a difference of 1.
The results may be correct.
I cannot check now.
> Could you post your code?  Might be easier for someone to help if they
> knew what the code was.
>
> Len Sorensen
It will be on the net tomorrow, I hope.
I was doing some bechmarking in order to write a small note on y work web 
pages.

I tried three different hardware (IBM/Power 3, Sun/UltraSparc2 and Pentium 4) 
config and two compilers on intel 32 bits with the same results.
The only difference is from AMD64.

I don't know now which one is the most accurate !

D.
-- 
Dominique Rousset
dominique.rousset@free.fr