On Tue, Mar 16, 2010, Volodymyr Babin wrote:
>
> opteron/linux: ifort 9.1 + gcc 4.3.4 -- 4 tests formally failed,
> but it looks like roundoff; ptraj.MPI passed on 4 cpus
> [after replacing icc by gcc in configure]
Is there some reason you did not use icc? This is a very non-standard
configuration -- but glad that it seems to work.
>
> xeon32/linux: ifort 10.1.022 + icc 10.1.022 + mkl 9.1.023 (probably; not
> sure since this is not my computer {university cluster}; the folder name
> is mkl91023)
>
> there were some difficulties with MKL:
Did configure report "MKL Version 9.x or earlier assumed"? [Update: I thought
we were automatically figuring this out, but now I see that you have to add
the "-oldmkl" flag -- did you do that?]
> once I replaced FLIBSF with
>
> FLIBSF= $(LIBDIR)/arpack.a -L/usr/local/intel/mkl91023/lib/32 -lsvml
> -lmkl_lapack -lmkl -lpthread
What was FLISBF before you edited it?
>
> I added -lmkl_ia32 to FLIBSF, and it passed further (I also had to change
> FLIBS accordingly)
It looks from the configure script like it should have added -lmkl_ia32
automatically (line 740 of configure). Can you see why this is not happening?
>
> then it stopped again with
>
> if [ "/gpfs_share/vbabin/amber11/lib/libfftw2xf_intel.a" =
> '/gpfs_share/vbabin/amber11/lib/libfftw2xf_intel.a' ]; then\
> cd /usr/local/intel/mkl91023//interfaces/fftw2xf && make libi686 \
> compiler=intel PRECISION=MKL_DOUBLE \
> INSTALL_DIR=/gpfs_share/vbabin/amber11/lib; \
> else \
> cd ../fftw-2.1.5 && make install; \
> fi
> make[2]: Entering directory `/usr/local/intel/mkl91023/interfaces/fftw2xf'
I'm guessing that Tyler didn't ever test the MKL/fftw interface on old
MKL libraries with 32-bit machines.
>
> and I decided to stop with intel+mkl; will try gnu on the same machine
> tonight
Better: try intel without mkl. If you didn't already do so, try adding
"-oldmkl" to configure. But I suspect we may still have problems with using
the MKL wrappers to fftw.
...thanks for the report....dac
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Mar 16 2010 - 15:30:04 PDT
