Patch to enable build on s390x
marc.glisse at inria.fr
Mon Sep 19 09:56:11 CEST 2011
On Sat, 17 Sep 2011, Mike Frysinger wrote:
> On Saturday, September 17, 2011 11:07:42 Marc Glisse wrote:
>> On Fri, 16 Sep 2011, Andreas Krebbel wrote:
>>>> Do these systems support a 32-bit ABI? If they do, the compiler
>>>> should have an option controlling ABI. Please try it, using something
>>>> trivial like hello.c.
>>> Yes. The toolchain for System z supports 2 differents ABIs:
>>> -m31 for 32 bit GPRs, 31 bit addressing - s390 <-- machine id (uname -r)
>>> -m64 for 64 bit GPRs, 64 bit addressing - s390x
>> Does something like this work (ie doesn't change anything on s390, works
>> on s390x and with ABI=32 works like s390 on s390x)?
>> --- a/configure.in Sun Aug 21 20:08:38 2011 +0200
>> +++ b/configure.in Sat Sep 17 16:45:31 2011 +0200
>> @@ -1119,6 +1119,13 @@
>> gcc_cflags="$gcc_cflags $fomit_frame_pointer"
>> + case $host_cpu in
>> + s390x)
>> + abilist="64 32"
>> + gcc_64_cflags="$gcc_cflags"
>> + gcc_32_cflags="$gcc_cflags -m31"
>> + ;;
>> + esac
> this doesnt help with the problem the original patch fixes:
> /home/vapier/gmp-5.0.2/build/.libs/libgmp.so: undefined reference to
I thought the problem with sdiv_qrnnd was caused by
extra_functions="udiv_w_sdiv". When there is an abilist, the 64 bit ABI
should only look at extra_functions_64 whereas the 32 bit one still reads
extra_functions. So what I wrote should in effect remove
extra_functions="udiv_w_sdiv" (and the same for path) for s390x.
I am not sure what I missed...
More information about the gmp-bugs