Dr. David Kirkby wrote: > Before building gcc 4.4.0 on 't2' I needed to build mpfr, as it is > perquisite for gcc. > > I built the latest version of mpfr (2.4.1) using the Sun supplied gcc > 3.4.2 in /usr/sfw/bin. mpfr built and passed all 148 tests. > > > With the aid: > > 1) The aid of a patch to atlas > http://trac.sagemath.org/sage_trac/ticket/6276 > (still needs review) > > and > > > 2) An update to binutils 2.19 > > the following Sage packages all build ok. > > atlas-3.8.3.p4 > blas-20070724 > bzip2-1.0.5 > conway_polynomials-0.2 > dir-0.1 > eclib-20080310.p7 > elliptic_curves-0.1 > extcode-4.0.2 > f2c-20070816.p1 > flint-1.3.0.p1 > fortran-20071120.p5 > freetype-2.3.5.p0 > gd-2.0.35.p1 > gdmodule-0.56.p5 > givaro-3.2.13rc2 > gnutls-2.2.1.p1 > graphs-20070722 > gsl-1.10.p1 > iml-1.0.1.p11 > ipython-0.9.1 > lapack-20071123.p0 > libgcrypt-1.4.3.p0 > libgpg_error-1.6.p0 > libpng-1.2.35 > linbox-1.1.6 > matplotlib-0.98.5.3rc0-svn6910.p3 > mercurial-1.1.2 > mpir-1.2.p4 > ntl-5.4.2.p7 > numpy-1.3.0 > opencdk-0.6.6 > pari-2.3.3.p0 > prereq-0.3 > python-2.5.4.p1 > readline-5.2.p6 > sage_scripts-4.0.2 > sqlite-3.5.3.p4 > termcap-1.3.1.p0 > zlib-1.2.3.p4 > > > However, Sage includes mpfr 2.4.1 as one of its packages (the exact same > version I used to build gcc). But in Sage, mpfr fails 20 of the tests. > But Sage is not to blame!! > > Trying to build mpfr 2.4.1 outside of Sage with the gcc 4.4.0 compiler > results in 20 test failures too. An attempt to reduce optimisation did > not help. So to date, I'm unable to build mpfr 2.3.1 with gcc 4.4.0. > > This is of course one of those things that could have many reasons for > the problem - mpfr, libtool, gcc, gnu assembler/linker ... etc etc. > > Last time I looked (an hour ago), the mpfr website was down, so I can't > report this, or look to see if others had the same issue. >
For what it is worth, I just built gcc 4.4.0 on my Sun Blade 2000 (Solaris 10 update 6) but using the Sun assembler and linker. I configured it with: $ gcc -v Using built-in specs. Target: sparc-sun-solaris2.10 Configured with: ../gcc-4.4.0/configure CC=/usr/sfw/bin/gcc --prefix=/export/home/drkirkby/install-gcc-4.4.0-Sun-as-ld --with-as=/usr/ccs/bin/as --without-gnu-as --with-ld=/usr/ccs/bin/ld --without-gnu-ld --enable-languages=c,c++,fortran --with-mpfr-include=/export/home/drkirkby/install-mpfr-2.3.1/include --with-mpfr-lib=/export/home/drkirkby/install-mpfr-2.3.1/lib --with-gmp-include=/export/home/drkirkby/install-gmp-4.3.1/include --with-gmp-lib=/export/home/drkirkby/install-gmp-4.3.1/lib --with-libiconv-prefix=/usr/local Thread model: posix gcc version 4.4.0 (GCC) drkir...@kestrel:[~/mp] $ After that, I used that gcc to build mpfr 2.4.1, which had failed miserably with gcc 4.4.0 on 't2'. That resulted in all tests passed on my Blade 2000. ==================== All 148 tests passed ==================== I'm just in the process of downloading sage 4.0.2 to my Blade 2000 and will try to compile Sage 4.0.2 with the gcc 4.4.0 I built using the Sun linker and assembler, rather than the GNU ones as Micheal's toolchain used, and what I used when building gcc 4.4.0 on 't2'. There are a few things to make me suspect that gcc built with the Sun linker and assembler (/usr/ccs/bin/as and /usr/ccs/bin/ld), rather than the GBU ones from binutils, may be a better choice. These are: * Sun ship gcc, which uses their own linker and assembler. * The developers of libtool, who clearly know a thing or two about linking programs, think the GNU tools are a disaster on SPARC. * Some comments from a Sun employee. * General comments I've seen over the years. Does anyone know if Michael had a good reason for choosing the GNU utils? It's very possible there is some code in Sage which assumes the use of the GNU tools, and starts passing GNU-specific flags. In which case, he may have felt he had no choice. I don;t know, but there is no harm in trying to build with the Sun linker and assembler. Certainly for mpfr, it was successful on my personal SPARC. Dave --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---