<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
<title></title>
</head>
<body bgcolor="#ffffff" text="#000000">
Hello again Michael,<br>
<br>
Gabriele Sclauzero wrote:
<blockquote cite="mid:4A32494D.6080304@sissa.it" type="cite">
<pre wrap="">Sorry Michael,
I can't understand why you're continuing having problems in compiling. It should be not
a "mission-impossible" kind of task, you simply need to have working compilers and mpi
wrappers (which you seem to have now) and maybe tweak a bit with the configure options.
</pre>
</blockquote>
<br>
A possible problem may come from a mixture with the blas and lapack
dylibs provided by Apple. If configure tells you that it finds suitable
linear algebra with -latlas and -lblas and a make pw stops at the final
link complaining about missing functions then I am afraid it is the
case. May I suggest to try the following tweak. Configure (I use
./configure --prefix=/Users/Shared/Applications/espresso-4.0.5
MPIF90=mpif90 F90=gfortran F77=gfortran CC=gcc - with correct PATH and
DYLD_LIBRARY_PATH) then edit the make.sys and modify the following 2
lines to the built-in libs BLAS_LIBS = ../flib/blas.a and LAPACK_LIBS =
../flib/lapack.a. Make pw and tell us if it goes up to the end.<br>
<blockquote cite="mid:4A32494D.6080304@sissa.it" type="cite">
<pre wrap="">
I compiled openmpi 1.3 on top of gfortran and Apple's gcc on MacOSX 10.5.7 (although I
really _strongly_ suggest you to use Intel compilers if you want to do serious
calculations on your Intel desktop... you may discover that you're wasting half of the
time just to save an hundred dollars or so), and I haven't met any problem.
</pre>
</blockquote>
<br>
I don't want to start a flamewar about the virtues of free and
commercial compilers. G95 produces dead slow binaries. Recent 4.3 or
4.4 gfortran are quite competitive compared to Intel ifort when nicely
tuned ; the difference is only a few percent although it is always
possible to optimize one and misconfigure the other in order to show a
preference. Intel MKL libraries are nicely optimized and will work
faster than plain blas and lapack.<br>
<br>
Kind regards,<br>
<br>
Alain<br>
<br>
<br>
<blockquote cite="mid:4A32494D.6080304@sissa.it" type="cite">
<pre wrap="">
Michael Hayden wrote:
</pre>
<blockquote type="cite">
<pre wrap="">Alain,
Again, thanks for your efforts and time.
I
have successfully installed and tested OpenMPI. I also configured Quant Espresso and received the "Parallel environment detected successfully, configured for compilation of parallel executables"
response at the end of the config process.
I then compiled it,but the end of a very long output read:
ld: in /usr/local/lib, can't map file, errno=22
collect2: ld returned 1 exit status
make[1]: *** [pw.x] Error 1
make: *** [pw] Error 2
</pre>
</blockquote>
<pre wrap=""><!---->
Most probably you have a problem in your environment variables (check if PATH,
LD_LIBRARY_PATH and DYLD_LIBRARY_PATH are what you expect them to be, i.e. with the path
to the MPI version you want to use listed _first_) or you miscompiled openmpi.
Honestly I don't see any need to feed so many parameters as suggested by Alain to the
configure of openmpi (unless you know exactly what they mean, maybe Alain does, I do
not...). You simply need to specify CC, CXX, F77, FC (and --prefix option, since it is not
wise to overwrite Apple's compiler files). If something goes wrong in the configure you
may need to add some others.
Cheers
GS
</pre>
<blockquote type="cite">
<pre wrap="">This doesn't look good and I can't find any list of these errors in the
user guide. I tried example01 and received (of course) this output:
ERROR: /usr/local/espresso-4.0.5/bin/pw.x not existent or not executable
Aborting
Any ideas?
cheers,
Mike
On Jun 10, 2009, at 7:05 PM, Alain Jacques wrote:
</pre>
<blockquote type="cite">
<pre wrap="">Hello Mike,
Humm, sorry it didn't work as expected. My development platform runs
OSX 10.5.7. I have compiled many working openmpi versions with the
following recipe:
export LD_LIBRARY_PATH=/opt/gcc-4.3.2/lib:$LD_LIBRARY_PATH && export
PATH=/opt/gcc-4.3.2/bin:$PATH
./configure --prefix=/opt/openmpi-1.2.8_gcc-4.3.2_m32 --enable-static
--enable-shared --disable-dependency-tracking --with-devel-headers
--with-wrapper-ldflags="-Wl,-search_paths_first"
CC=/opt/gcc-4.3.2/bin/gcc LDFLAGS="-L/opt/gcc-4.3.2/lib"
FC=/opt/gcc-4.3.2/bin/gfortran F77=/opt/gcc-4.3.2/bin/gfortran
CXX=/opt/gcc-4.3.2/bin/g++ CPP=/opt/gcc-4.3.2/bin/cpp CFLAGS="-m32 -O3
-march=native" CXXFLAGS="-m32 -O3 -march=native" FCFLAGS="-m32 -O3
-march=native" FFLAGS="-m32 -O3 -march=native"
As you understand, my homemade gcc/gfortran lives in
/opt/gcc-4.3.2/bin and I install the corresponding openmpi in
/opt/openmpi-1.2.8_gcc-4.3.2_m32. Same story with openmpi up to
version 1.3.2 and gcc up to 4.4.0. The -m32 flag is there to specify
32bit binaries (-m64 will build 64bit openmpi for the corresponding
binaries, adapt the LDFLAGS too). I am quite convinced that the error
message you receive comes from a mixture with Apple provided openmpi
libs without gfortran support.
And when it is done, in order to compile or run mpi software, I add
the path to the new openmpi binaries and libs in front with a:
export PATH=/opt/openmpi-1.2.8_gcc-4.3.2_m32/bin:$PATH && export
DYLD_LIBRARY_PATH=/opt/openmpi-1.2.8_gcc-4.3.2_m32/lib:$DYLD_LIBRARY_PATH
i.e. I end with a
PATH=/opt/openmpi-1.2.8_gcc-4.3.2_m32/bin:/opt/gcc-4.3.2/bin:xxx and
DYLD_LIBRARY_PATH=/opt/openmpi-1.2.8_gcc-4.3.2_m32/lib:/opt/gcc-4.3.2:xxx
where xxx is the original content of the variables.
Kind regards,
Alain
</pre>
</blockquote>
<pre wrap="">
------------------------------------------------------------------------
_______________________________________________
Pw_forum mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Pw_forum@pwscf.org">Pw_forum@pwscf.org</a>
<a class="moz-txt-link-freetext" href="http://www.democritos.it/mailman/listinfo/pw_forum">http://www.democritos.it/mailman/listinfo/pw_forum</a>
</pre>
</blockquote>
<pre wrap=""><!---->
</pre>
</blockquote>
<br>
</body>
</html>