Peter Jeremy wrote: > On 2009-Dec-29 13:51:43 +0000, "Dr. David Kirkby" <david.kir...@onetel.net> > wrote: >> Then when I thought about it more, I do not know if its worth the hassle. Why >> not simply ask the user to export SAGE64 to "yes" on Open Solaris, and forget >> about a 32-bit build? > > It looks like OpenSolaris/SPARC defaults to 32-bit mode (looking at the > installed executables - I don't have a development system installed on > my OSol box yet).
Yes it does. And I can understand why, since for 99% of programs, there is no advantage to 64-bit, but some disadvantages (larger pointers, let fit in cache etc). But for Sage, I would have thought the only advantage of 32-bit would be the fact that gcc is less buggy. And I admit, that is a pretty significant advantage. > Note that the current assumption that Sage is building in 32-bit mode > unless 'SAGE64' is set is not universally true: Tru64, FreeBSD/amd64 > and FreeBSD/SPARC64 (at least) default to building 64-bit executables > and have minimal, if any, support for building 32-bit executables. I'm not making that assumption. >> http://trac.sagemath.org/sage_trac/ticket/7505 >> >> could get a positive review and be integrated into Sage asap. Two people have >> looked at it, and both agree it is OK. > > I like the idea but the actual implementation seems somewhat convoluted. > Why not make '${CC} -E ${TESTFILE}' directly output the wanted identifier, > rather than running $CC multiple times and grepping for strings in the > output. This approach also seems likely to cause maintenance issues as > changes need to be implemented in three places (including the usage). I > have attached a suggested alternative (testcxx could be similarly adapted). I have not tested it, but I agree it is cleaner. >> 1) Determine what compiler is in use by the use of the scripts in #7505 >> 2) Set CFLAGS to have the appropriate option for a 64-bit build in sage-env >> if >> SAGE64 is set to "yes". That will be -m64 with Sun Studio or GCC, but will be >> different for other compilers. >> 3) Remove all the SAGE64 junk from every spkg-install, because: >> * It's unnecessary duplication >> * Currently has a mix of those that only work on OS X and those that work on >> Solaris, and those that only work on OS X. >> * It's incorrect to assume the flag to build 64-bit code will always be >> -m64, as some compilers use a different flag. > > The code also needs to handle the situation where a 64-bit Sage will > be built by default. I'm not convinced there needs to be anything to handle the case of 64-bit being built by default, unless we wish to be able to force 32-bit builds on platforms where 64-bit is the default. If 64-bit is the default, one would not bother setting SAGE64, -m64 (or equivalent) flag would not get added, and the code should work. Have I overlooked something? > Ideally, the 32 vs 64 bit code should be > centralised and the relevant cc/c++/fortran/linker flags passed into > each spkg-install. I'm just in the process of trying to do that in sage-env if [ "x$SAGE64" = "xyes" ] ; then if [ "x`$SAGE_LOCAL/bin/testcc.sh`" = "xGCC" ] ; then # Flags for a 64-bit build. CFLAG64="-m64" CXXFLAG64="-m64" FCFLAG64="-m64" elif [ "x`$SAGE_LOCAL/bin/testcc.sh`" = "xSun_Studio" ] ; then CFLAG64="-m64" CXXFLAG64="-m64" FCFLAG64="-m64" elif [ "x`$SAGE_LOCAL/bin/testcc.sh`" = "xIBM_on_AIX" ] ; then # http://www-01.ibm.com/support/docview.wss?uid=swg27012860&aid=1 echo "The appropriate 64-bit option for IBM's C++ and Fortran" echo "compilers are assumed to be the same as the" echo "C compiler. This has not been checked" echo "Adding -q64 to generate 64-bit code" CFLAG64="-q64" CXXFLAG64="-q64" FCFLAG64="-q64" etc then later on: CFLAGS="$CFLAGS $CFLAG64" CXXFLAGS="$CXXFLAGS $CXXFLAG64" FCFLAGS="$FCFLAGS $FCFLAG64" That way, we have as a variable (CFLAG64) in isolation, as well as in CFLAGS. I think ECL needs needed -m64 at one point during the link process, so not as a CFLAG. We can also add in other things for a specific compiler, unrelated to 64-bit. One we need is the right flags for building shared libraries. I'd propose we call the flags a simpilar name to their GNU counterparts, so it makes substitution of hard-coded GCC-specific flags to variable names. if [ "x`$SAGE_LOCAL/bin/testcc.sh`" = "xGCC" ] ; then # These are the names of the environment variables, and their GCC # name. They are not appended to CFLAGS, CXXFLAGS or FCFLAGS as they # are not always needed. FPIC_FLAG="-fPIC " SHARED_FLAG="-shared" SONAME_FLAG="-soname" WHOLE_ARCHIVE_FLAG="--whole-archive" NO_WHOLE_ARCHIVE_FLAG="--no-whole-archive" # Enable warnings on gcc. CFLAGS="$CFLAGS -Wall" CXXFLAGS="$CXXFLAGS -Wall" FCFLAGS="$FCFLAGS -Wall" elif [ "x`$SAGE_LOCAL/bin/testcc.sh`" = "xSun_Studio" ] ; then # -xcode=pic32 is valid on 64-bit too, despite the name. # It is equivalent to the obsolete -KPIC on SPARC and -kPIC on x86 # These flags for Solaris should work in most circumstances. # Identical flags have been used to build a 32-bit version of Sage # on SPARC. FPIC_FLAG="-xcode=pic32" SHARED_FLAG="-G" SONAME_FLAG="-h" WHOLE_ARCHIVE_FLAG="-z allextract" NO_WHOLE_ARCHIVE_FLAG="-z defaultextract" Any thoughts on that? dave -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
