Problems installing gmp

Herve Pages hpages at fhcrc.org
Mon Jul 24 21:10:17 CEST 2006


Torbjorn Granlund wrote:
> Herve Pages <hpages at fhcrc.org> writes:
>
> Yeah I know... and I'd love to know how to get ld to link the
> ultrasparc3-sun-solaris2.9 build by the way!
>
> It works for me. :-}
I'm glad for you.
>
> > A 32-bit build runs at abuot 1/3
> > of the speed of a 64-bit build.
> >
>
> Not on UltraSPARC, according to these charts:
> http://www.swox.com/gmp/32vs64.html
> It looks more like the 32-bit build runs at about 60% to 70%
> of the speed of a 64-bit build, which is better than having no GMP
> at all (and consequently no Fortran front-end for GCC 4.1.1).
>
> Granted, ultrasparc is special, as explained at that page. But you'd
> find that the used fallback build is more like 20 times slower than a
> 64-build, at least on GMPbench, and that can be avoided.
>
> Here are some GMPbench 0.1 scores on a 900 MHz UltraSPARC3 system:
>
> --host=sparc-sun-solaris2.9 54
> --host=sparcv8-sun-solaris2.9 257
> --host=sparcv9-sun-solaris2.9 ABI=32 462
> --host=sparcv9-sun-solaris2.9 977
>
> The fallback build was sparc-X-Y IIRC, but the proper fallback build
> should be the ABI=32 one (letting the --host parameter be determined
> automatically is usually best).
>
> Suggested read: http://www.swox.com/gmp/manual/ABI-and-ISA.html
Interesting reading. It tells me how to compile applications that I want to
link with a 64-bit build of GMP on my Solaris system (Solaris 2.9 + gcc 
4.1.1)
i.e. with:
  gcc -m64 -mptr64 -Wa,-xarch=v9 -mcpu=v9

However, when I run 'configure' I get:

    checking build system type... ultrasparc3-sun-solaris2.9
    checking host system type... ultrasparc3-sun-solaris2.9
    ...
    checking compiler gcc -O2 -m64 -mptr64  -mcpu=ultrasparc3... yes
    ...

so I'm wondering why the flags shown by 'configure' don't match those given
in your "Suggested read" page.

Then when I run 'make check', I can see that the tests are indeed compiled
with the flags shown above by 'configure' but I get an ld warning
for each compiled test, e.g for the "t-count_zeros" test:

    gcc -DHAVE_CONFIG_H -I. -I/usr/local/src/gmp-4.2.1/tests -I..
    -I/usr/local/src/gmp-4.2.1    -O2 -m64 -mptr64 -mcpu=ultrasparc3 -c
    /usr/local/src/gmp-4.2.1/tests/t-count_zeros.c
    /usr/local/bin/bash ../libtool --mode=link gcc  -O2 -m64 -mptr64
    -mcpu=ultrasparc3   -o t-count_zeros  t-count_zeros.o libtests.la
    ../libgmp.la
    gcc -O2 -m64 -mptr64 -mcpu=ultrasparc3 -o .libs/t-count_zeros
    t-count_zeros.o  ./.libs/libtests.a
    /usr/local/src/gmp-build/.libs/libgmp.so ../.libs/libgmp.so
    -R/usr/local/lib
    ld: warning: file ../.libs/libgmp.so: linked to
    /usr/local/src/gmp-build/.libs/libgmp.so: attempted multiple
    inclusion of file
    creating t-count_zeros

and then an ld fatal error at run time for each test too:

    ld.so.1: t-count_zeros: fatal: /usr/local/lib/libgcc_s.so.1: wrong
    ELF class: ELFCLASS32
    /usr/local/bin/bash: line 1: 26990 Killed                  ${dir}$tst
    FAIL: t-count_zeros

I have the feeling that after reading your "Suggested read" the
solution to my problem should be just obvious to me... but I'm
affraid this is not the case :-/
Is there anything else you can suggest me?
Thank you,

H.


-- 
------------------------
Hervé Pagès
E-mail: hpages at fhcrc.org
Phone: (206) 667-5791
Fax: (206) 667-1319
------------------------


More information about the gmp-bugs mailing list