[gmx-users] ic vs gcc : strange results

[Erik Lindahl]

Hi,

If you get such a big difference it is almost certainly because you are 
using the 64-bit mode, without SSE assembly loops. We have 64-bit 
loops, and they will appear soon, but in the meantime you will get 
better performance by adding flags to compile in 32-bit mode.

(Or perhaps that's what the intel compiler is doing...)

Anyway, check your logfile and see if the simulation is using SSE loops.

Cheers,

Erik


On Feb 19, 2004, at 1:30 PM, Michel Cuendet wrote:

>
> It does indeed improve the speed of about 10 - 15% on my system. As I 
> said, I am running on an Opteron 240. I used  CFLAGS  =  "-O3 -ip 
> -unroll".
>
> Cheers.
>
> Lieven Buts wrote:
>
>> On Thursday 19 February 2004 13:02, Michel Cuendet wrote:
>>
>>>  However I recompiled only mdrun with icc, whereas grompp is gcc. 
>>> Could
>>> this create problems ?
>>>
>>
>> Please note that compiling mdrun with icc is unlikely to provide any 
>> benefit, since Gromacs uses assembler inner loops on Intel, which 
>> should beat even the code generated by icc.
>>
>>
>
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the www 
> interface or send it to gmx-users-request at gromacs.org.