[gmx-users] intel compiler + amd64: help needed.

[Jones de Andrade]

Ok, I think I got it. Compiled. ;)

About a thousand tricks. I'll publish a tutorial later. But, basically:
1 - Change all invsrt calls to gmx_inv-sqrt (57 entries, 12 files);
2 - Edit the Makefile.in in the nb_kernels directories, changing the line
"LTCCASCOMPILE = $(LIBTOOL) --mode=compile $(CCAS) $(AM_CCASFLAGS) \" by
"LTCCASCOMPILE = $(LIBTOOL) --mode=compile --tag=CXX $(CCAS) $(AM_CCASFLAGS)
\" in "Makefile.in" files;
3 - make disclean (for safety);
4 - Run the configure for "gcc", BUT with the following environment and
flags(at least):
*************************************************************************************
CPPFLAGS=-I/usr/local/lib64/fftw/gnu/include
LDFLAGS=-L/usr/local/lib64/fftw/gnu/lib
CC=/usr/bin/gcc
CXX=/usr/local/intel/bin/icpc
F77=/usr/bin/g77
CFLAGS="-O3 -ffast-math -funroll-all-loops -fpeel-loops "
CXXFLAGS="-O3 -axP -xW "
F77FLAGS="-O3 -ffast-math -funroll-all-loops -fpeel-loops "
*************************************************************************************
5 - SAVE the Makefile from nb_kernels directories; special care for the one
your compilation will use for sure (machine dependent);
6 - make distcean (for necessity)
7 - Now, run again configure for intel compilers, with the new environment
and flags (at least) below:
**************************************************************************************************
PATH="$PATH /usr/local/intel/bin/ . /usr/local/intel/lib/"
LD_LIBRARY_PATH=/usr/local/intel/lib/
CPPFLAGS=-I/usr/local/lib64/fftw/intel/include
LDFLAGS=-L/usr/local/lib64/fftw/intel/lib
CC=/usr/local/intel/bin/icc
CXX=/usr/local/intel/bin/icpc
F77=/usr/local/intel/bin/ifort
CFLAGS="-O3 -axP -xW "
CXXFLAGS="-O3 -axP -xW "
FFLAGS="-O3 -axP -xW "

configure --without-motif-libraries --without-motif-includes --without-x
**************************************************************************************************
8 - Substitute the new makefiles in nb_kernels directories for the ones you
saved in step 5;
9 - Make it.
10 - make install (finally!)
11 - make distclean (to preserve your disk space)

I think I haven't forgot anything. ;)

I can't be sure about this, but I do think that most of the changes that
were necessary to make it run on intel compiler (and, despite I don't have
an intel machine available here, I believe those changes are necessary for
their machines too) can be done with also no harm for the GCC compilation
too. Basically, I'm wasting the "CXX" environment variable with the "--tag"
for libtool, but I thik it can be done in another fashion without touching
it. Also, the substitution of "invsqrt" for "gmx_invsqrt" is harmless.
Possibly the only difficulty would be to assure that gcc was used for the
"mknb" compilation with intel.

But, as I said, it compiled. Now I'm playing a bit with the --enable-doube"
and trying to see if it understands "mkl" for blas and lapack. Will take
about an hour. But, working and running is another problem at this point. I
don't think it would really yeld problems (will know later), but it still
yelds me the following warnings:
#######################################################################################################
ghat.c(233): warning #266: function declared implicitly
    gh      = mk_rgrid(ix,iy,iz);
              ^
ghat.c(233): warning #967: conversion from "int" to "real={float} ***";
sizes do not match
    gh      = mk_rgrid(ix,iy,iz);
            ^
fftgrid.c(227): warning #188: enumerated type mixed with another type
          gmx_fft_3d_real(grid->fft_setup,dir,grid->ptr,grid->workspace);
                                          ^
gmx_fft_fftw3.c(96): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(100): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(186): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(190): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(278): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(282): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(370): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(374): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(463): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(467): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(556): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_fft_fftw3.c(560): warning #1338: arithmetic on pointer to void or
function type
      pv += 8;
         ^
gmx_dielectric.c(150): warning #266: function declared implicitly
    four1(ptr-1,nnx,-1);
    ^
gmx_nmtraj.c(134): warning #592: variable "bDMA" is used before its value is
set
    read_tps_conf(ftp2fn(efTPS,NFILE,fnm),title,&top,&xtop,NULL,box,bDMA);
########################################################################################
I'm not sure about what they mean. Possibly nothing, But I'll only be calm
if someone says it. Or should I be worried about any of them in advance
already?

Really thanks a lot to everybody for all help.It almost finished, I would
never have reached this far without all the help. Also, I'll keep my word
and publish a whole tutorial and benchmarks on the list later on! :)

Thanks again to everybody,

Jones
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060602/7536cdb9/attachment.html>