Howdy Amberites,
After banging my head against it for a few days, I have a working, tested
serial compilation of Amber 10 for OS X Intel using the Intel MKL and
compilers.
A few notes:
I reused the -p4 flag to set optimizations relevant for Intel Macs.
They're all basically Core2Duo or better, and I set the compiler to
optimize for that. If you have one of the first generation Core Duos,
you may need to change the -ax flag.
I (ab)used the -static flag for this build to make it "as static as
possible" given Apple's refusal to support static binaries on OS X.
This part may not be necessary, because I wasn't able to get the
binaries to run witout setting the DYLD_LIBRARY_PATH, which was the
goal in the first place.
So in spite of the -static-intel flags everywhere I could put them, the MKL
libraries seem to still need the DYLD_LIBRARY_PATH (or
DYLD_FALLBACK_LIBRARY_PATH) set in order to run. Without that set, I was
getting errors like:
cd dmp && ./Run.dmp
/programs/i386-mac/ambermd/10/test/dmp
MKL FATAL ERROR: dlopen(libmkl_p4p.dylib, 1): image not found
./Run.dmp: Program error
The same goes for the Intel compiler libraries and the Amber Tools:
Running test to make dna duplex:
dyld: Library not loaded: libsvml.dylib
Referenced from: /private/var/automount/nfs/programs/i386-mac/ambermd/10/test/nab/./duplex
Reason: image not found
If I add the compiler lib directory to the DYLD_LIBRARY_PATH, it works
fine. With the above flags, settings, etc, the build passes all the tests
for both Amber and Amber Tools.
HTH.
-ben
==
1. Download source, patches, untar, patch, set AMBERHOME, MKL_HOME, etc.
2. Edit configure_at and change:
--- configure_at.orig 2008-12-10 12:21:41.000000000 -0500
+++ configure_at 2008-12-10 12:35:43.000000000 -0500
.. -318,5 +318,5 ..
blas=skip
flibs_nab="\$(LIBDIR)/libsym.a \$(LIBDIR)/arpack.a \$(LIBDIR)/f2c.a"
- flibs_mkl="\$(MKL_HOME)/lib/32/libmkl_lapack.a \$(MKL_HOME)/lib/32/libmkl_ia
32.a -L\$(MKL_HOME)/lib/32 -lguide -lpthread"
+ flibs_mkl="-L\$(MKL_HOME)/lib/32 -lmkl_intel -lmkl_core -lmkl_lapack -lguide
-lpthread"
if [ "$x86_64" = "yes" ]; then
flibs_mkl="\$(MKL_HOME)/lib/em64t/libmkl_lapack.a \$(MKL_HOME)/lib/em64t
/libmkl_em64t.a -L\$(MKL_HOME)/lib/em64t -lguide -lpthread"
.. -335,6 +335,6 ..
if [ $p4 = "yes" ]; then
- ocflags=" -O3 -ip -axN"
- offlags=" -w95 -tpp7 -mp1 -ip -O3 -axN"
+ ocflags=" -O3 -ip -axT"
+ offlags=" -w95 -tpp7 -mp1 -ip -O3 -axT"
flibs_arch="$flibs_arch -lsvml"
fi
.. -352,6 +352,6 ..
if [ $static = "yes" ]; then
- flibs_nab="$flibs_nab -static"
- ldflags='-static'
+ flibs_nab="$flibs_nab -static-intel"
+ ldflags='-static-intel'
fi
3. Configure and build:
./configure_at -openmp -static -p4 icc
make -f Makefile_at
4. Test
cd ../test && make -f Makefile_at test, compare results, etc
5. Edit configure_amber and change:
--- configure_amber.orig 2008-12-04 18:00:18.000000000 -0500
+++ configure_amber 2008-12-09 11:46:11.000000000 -0500
.. -222,7 +222,7 ..
loadcc="$cc"
loadlib=
if [ $static = "yes" ]; then
- static_flag='-static'
+ static_flag='-static-intel'
else
static_flag=
fi
.. -1069,8 +1069,8 ..
;;
ifort_macosx)
- cc="gcc -m32"
- cplusplus=g++
+ cc="icc"
+ cplusplus=icpc
cflags="$cflags -O2"
loadcc="$cc $static_flag"
fppflags="$fppflags"
.. -1082,11 +1082,11 ..
fi
fflags="$fflags $verbosity"
- foptflags="$fflags -mp1 -ip -O3"
+ foptflags="$fflags -mp1 -ip -O3 -axT"
fflags="$fflags -mp1 -O0"
if [ $static = "yes" ]; then
- static_flag='-i-static'
+ static_flag='-static-intel'
else
static_flag=
fi
.. -1115,7 +1115,7 ..
loadlib="$loadlib $mkll/libmkl_lapack.a $mkll/libmkl_ia32.a -L$mkll -lguide -lpthread"
else
# IA-32 dynamic linking of double-precision-LAPACK and kernels
- loadlib="$loadlib -L$mkll -lmkl_lapack -lmkl_ia32 -lguide -lpthread"
+ loadlib="$loadlib -L$mkll -lmkl_intel -lmkl_core -lmkl_lapack -lmkl_intel_thread -lguide -lpthread"
fi
fppflags="$fppflags -DMKL"
fi
6. Configure and build
./configure_amber -p4 -static -nosanderidc ifort_macosx
make serial
7. Test, compare, etc
cd ../test
make test
--
Ben Eisenbraun
Structural Biology Grid Harvard Medical School
http://sbgrid.org http://hms.harvard.edu
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
to majordomo.scripps.edu
Received on Fri Dec 12 2008 - 01:10:49 PST
