Hi Ross (and everyone),
Yeah, I replaced "-fast" with "-ipo -O3 ..." and recompiled everything
successfully last night. As far as I can tell, all of the programs
are working correctly, and my non-rigorous benchmarking suggests that
the optimized pmemd code is quite fast (at least compared to either
pgi or gnu compilers on similar hardware). Thanks for your help!
Best,
Paul
On Sep 11, 2010, at 10:09 AM, Ross Walker wrote:
> Hi Paul
>
>> To add one more piece of data to the mix, remember I stated that
>> sander.MPI compiles OK despite the fact that pmemd.MPI does not.
>> Well
>> the one obvious difference between the two is that sander.MPI is
>> compiled with "mpif90 -o sander.MPI ...", while pmemd.MPI is compiled
>> with "mpif90 -fast -o pmemd.MPI ..." When I remove the "-fast"
>> optimization flag, pmemd.MPI compiles with no problem and passes all
>> of the test cases (test.parallel.pmemd). (So I suppose in some weird
>> way, this is related to bugfix 8.) From the Intel compiler manual
>> (p.
>> 577) I see that "-fast" sets the following options "-ipo, -O3, -no-
>> precdiv, *-static*, and -xHost". It seems to me that specifying "-
>> fast" is asking ifort to, among other things, build with static
>> libraries. Could this be putting it at odds with the dynamic mpi
>> libraries?
>
> Ah, that explains it... This did not occur to me when releasing
> bugfix.8. So
> -fast forces static linking which means you must have the .a version
> of the
> MPI library. You have two options. Either recompile your MPI and add
> support
> for static linking. Normally you would give this as an argument to the
> openMPI configure run (FFLAGS=-static or similar). This should then
> build
> you the .a version of the library.
>
> Alternatively I would edit your config.h file and change all
> occurances of
> -fast to be:
>
> -ipo -O3 -no-precdiv -xHost
>
> Then make clean, make and it should link okay this time. I will have
> a think
> about how we might improve things in the Makefile to detect if the
> static
> libraries are available or not.
>
> All the best
> Ross
>
>
> /\
> \/
> |\oss Walker
>
> ---------------------------------------------------------
> | Assistant Research Professor |
> | San Diego Supercomputer Center |
> | Adjunct Assistant Professor |
> | Dept. of Chemistry and Biochemistry |
> | University of California San Diego |
> | NVIDIA Fellow |
> | http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> ---------------------------------------------------------
>
> Note: Electronic Mail is not secure, has no guarantee of delivery,
> may not
> be read every day, and should not be used for urgent or sensitive
> issues.
>
>
>
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Sep 11 2010 - 16:30:03 PDT
