Hi all, I've been trying to build octave324 on i386-darwin10 now, and I think I understand the cause of more ff2c-related failures. On both darwin8 and darwin10, the Accelerate framework needs -ff2c for the correct calling convention to system-atlas. Any fink-built atlas (gfortran) will not need it. So far, qrupdate and octave are both atlas-varianted. arpack however is not, but I think it needs to be, even if it doesn't link against atlas/lapack. Since octave/qrupdate need to be ff2c-consistent, so does arpack, but the current packaging doesn't give the option to select ff2c or without.
What I propose is to variant-ify arpack so that non-fink-atlas is uses -ff2c and is compatible with system-atlas variants, and -atlas omits -ff2c to work properly with -atlas variants. I think this will take care of funny errors we see with arpack functionality, namely in eigenvalue-related parts of the packages and their tests. Any other fortran packages that eventually link to an atlas-varianted package may need the same treatment. Thoughts? Fang > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > On 9/13/11 11:58 AM, Dominique Dhumieres wrote: >>> Did anyone happen to notice during octave324-atlas's build: >> >> If was too sleepy (2am local time), I saw that octave324-atlas >> 3.2.4-21 was built without atlas. 3.2.4-22 seems to work as >> expected (build in progress). The only glitches I have spotted so >> far are some >> >> configure:12470: flag-sort -r gcc -o conftest -g -O3 -MD >> -I/sw/include -I/sw/include -L/sw/lib conftest.c -lqrupdate >> /sw/lib/liblapack.dylib /sw/lib/libf77blas.dylib >> /sw/lib/gcc4.6/lib/libgfortran.dylib -lhdf5 -lz -lm >> -lGraphicsMagick -lmetis >&5 >> /usr/libexec/gcc/powerpc-apple-darwin8/4.0.1/ld: warning multiple >> definitions of symbol _ilaenv_ /sw/lib/libf77blas.dylib(single >> module) definition of _ilaenv_ /sw/lib/liblapack.dylib(single >> module) definition of _ilaenv_ >> /usr/libexec/gcc/powerpc-apple-darwin8/4.0.1/ld: warning suggest >> use of -bind_at_load, as lazy binding may result in errors or >> different symbols being used symbol _ilaenv_ used from dynamic >> library /sw/lib/libf77blas.dylib(single module) not from earlier >> dynamic library /sw/lib/liblapack.dylib(single module) >> >> probably harmless. >> >> The problem is more serious on x86_64-apple-darwin10, on which I >> get >> >> configure:12126: checking whether CDOTU is called correctly from >> Fortran configure:12146: /sw64/bin/gfortran-fsf-4.6 -o conftest -O3 >> -Wl,-dead_strip_dylibs -L/sw64/lib conftest.f >> -Wl,-framework,Accelerate,-dylib_file,/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib:/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib >> -lhdf5 -lz -lm -lGraphicsMagick -lmetis >&5 configure:12146: $? = >> 0 configure:12146: ./conftest ./configure: line 2499: 66156 >> Segmentation fault ./conftest$ac_exeext configure:12146: $? = >> 139 configure: program exited with status 139 configure: failed >> program was: | program main | | complex >> cdotu,a(1),b(1),w | external cdotu | a(1) = >> cmplx(1e0,1e0) | b(1) = cmplx(1e0,2e0) | w = >> cdotu(1,a,1,b,1) | if (w .ne. a(1)*b(1)) stop 1 | | >> end configure:12155: result: no >> >> and the same with ZDOTU with the result >> >> configure:12398: WARNING: A BLAS library was detected but found >> incompatible with your Fortran 77 compiler. The reference BLAS >> implementation will be used. To improve performance, consider using >> a different Fortran compiler or a switch like -ff2c to make your >> Fortran compiler use a calling convention compatible with the way >> your BLAS library was compiled, or use a different BLAS library. >> >> Manual tests show that -ff2c is required for (C|Z)DOTU with >> -Wl,-framework,Accelerate. I have checked on some codes I have that >> this flag does not hurt *GEMM. >> > > I wound up with more errors when I tested it with -ff2c than without > it on some platform. I think it was 10.6/x86_64, but I don't recall > that for a fact. I'll try it again for my available platforms. > >> Question: why the ugly >> -Wl,-framework,Accelerate,-dylib_file,/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib:/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib >> >> > while -Wl,-framework,Accelerate seems enough? > > It was like that when I took the package over :-) . I'm certainly > willing to simplify that on the next update. > >> >> Work in progress here. >> >> Cheers, >> >> Dominique > > > - -- > Alexander Hansen, Ph.D. > Fink User Liaison > http://finkakh.wordpress.com/ > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.11 (Darwin) > Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ > > iEYEARECAAYFAk5vgbAACgkQB8UpO3rKjQ8m2ACgnQGJseOf1eerSne0462kiQhe > niwAn1oYxQl0NKk+vO4eqM7AQzMddiks > =X+fD > -----END PGP SIGNATURE----- > -- David Fang http://www.csl.cornell.edu/~fang/ http://www.achronix.com/ ------------------------------------------------------------------------------ BlackBerry® DevCon Americas, Oct. 18-20, San Francisco, CA Learn about the latest advances in developing for the BlackBerry® mobile platform with sessions, labs & more. See new tools and technologies. Register for BlackBerry® DevCon today! http://p.sf.net/sfu/rim-devcon-copy1 _______________________________________________ Fink-users mailing list Fink-users@lists.sourceforge.net List archive: http://news.gmane.org/gmane.os.macosx.fink.user Subscription management: https://lists.sourceforge.net/lists/listinfo/fink-users
