[gmx-users] CFLAGS

[Erik Lindahl]

Hi,

The configure script does a pretty good job of detecting things  
automatically. The appropriate assembly loops are enabled by default  
- use the switches if you want to disable them for some reason.

The same thing goes for the compiler options - just set them manually  
if the automatic detection doesn't work.

ia64 is a bit special in that the nonbonded loops are very optimized  
relative to the neighborsearching (compared to ia32 or x86_64). Since  
the neighborsearching is compiled code you probably want to use the  
intel compiler and "-O3" (should be set automatically if you define  
CC=icc) for better performance.

Cheers,

Erik



On Nov 16, 2005, at 6:18 PM, Joanne Hanna wrote:

> Joanne Hanna
> Department of Chemistry
> University of Warwick
> Coventry
> CV4 7AL
>
> J.F.Hanna at warwick.ac.uk
>>>> Florian.Haberl at chemie.uni-erlangen.de 11/16/05 17:05 PM >>>
> hi,
>
> On Wednesday 16 November 2005 17:47, Joanne Hanna wrote:
>> Hi
>>
>> So if my machine is Intel Itanium2 should i really use version  
>> 3.3? and if
>> i use 3.2.1 should i use the -O3 cflags option? Sorry I am quite  
>> new to
>> compiling!
>
> Normally you don^t use optimazition flags for compiling gromacs,  
> only if you
> want to test further improvement of the speed, as Erik Lindal  
> mentioned,
> there is not much speed improvement (i guess below 2% not more).
>
> I would use gromacs 3.3 cause this is the version with buildin  
> assembly loops,
> this means much better performance as version 3.2.1 on Itanium2  
> cpus (factor
> 2 or so overall).
>
> Okay so this should pick up the cflags automatically, as v3.2.1  
> does not and thats why i had to define them.
> Also does this apply to my machine?
> Add "--enable-sse" if you computers are i686 or x86-64.
>
> Sorry for very obvious questions.
> Jo
>
>>
>> Thanks
>> Jo
>>
>> Joanne Hanna
>> Department of Chemistry
>> University of Warwick
>> Coventry
>> CV4 7AL
>>
>> J.F.Hanna at warwick.ac.uk
>>
>>>>> lindahl at sbc.su.se 11/16/05 15:05 PM >>>
>>
>> Hi,
>>
>> The intel compilers should support both -O3 (our default when icc is
>> detected) and -O2. The difference isn't very large, though, since our
>> assembly loops bypass the compiler on both ia32, X86-64, and ia64 -
>> but then you need to use gromacs 3.3 or later!!! (3.2 only includes
>> ia32 assembly loops).
>>
>> If you're only having problems when compiling with MPI support the
>> culprit is likely the mpi wrapper script for the compiler.
>>
>> Cheers,
>>
>> Erik
>>
>> On Nov 16, 2005, at 2:53 PM, Joanne Hanna wrote:
>>> Hi
>>>
>>> I have recently compiled an mpi version of gromacs 3.2.1 with the
>>> icc compiler. I ran into troubles with this error, at the ./
>>> configure stage:
>>> * Sorry, these optimization settings don't seem to work for       *
>>> * your C compiler. Use make CFLAGS=..., or edit the top Makefile. *
>>> I decided to use the cflags option -O2.
>>> Can anyone tell me whether this is a good option to use. I
>>> understand that -O3 enables more aggressive optimisations such as
>>> loop and  memory  access transformation. and is recommended for
>>> applications that have loops with heavy use of float-ing point
>>> calculations and process large data sets (as from the manual). But
>>> i don't know which i should use.
>>> I have been told that optimising too aggressively can produce the
>>> wrong numbers.
>>>
>>> Anu advice would be much appreciated.
>>>
>>> Thanks
>>> Jo
>>>
>>>
>>> Joanne Hanna
>>> Department of Chemistry
>>> University of Warwick
>>> Coventry
>>> CV4 7AL
>>>
>>> J.F.Hanna at warwick.ac.uk
>>> _______________________________________________
>>> 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.
>>
>> -----------------------------------------------------------
>> Erik Lindahl  <lindahl at sbc.su.se>
>> Assistant Professor, Stockholm Bioinformatics Center
>> Stockholm University, SE 106 91 Stockholm
>> Phone: +46 8 5537 8564     Fax: +46 8 5537 8214
>>
>
> Greetings,
> Florian
>
> -- 
> ---------------------------------------------------------------------- 
> -----------------------------
>  Florian Haberl                             Universitaet Erlangen/ 
> Nuernberg
>  Computer-Chemie-Centrum     Naegelsbachstr. 25, D-91052 Erlangen
>
>  Mailto: florian.haberl AT chemie.uni-erlangen.de
>
> ---------------------------------------------------------------------- 
> -----------------------------
>
> _______________________________________________
> 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.
>
> _______________________________________________
> 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.
>

-----------------------------------------------------------
Erik Lindahl  <lindahl at sbc.su.se>
Assistant Professor, Stockholm Bioinformatics Center
Stockholm University, SE 106 91 Stockholm
Phone: +46 8 5537 8564     Fax: +46 8 5537 8214