Re: Function mpz_gcd() works differently on HPUX and Linux - ia64

[Kevin Ryde]

Prashanth PS <address@hidden> writes:
>
>     I would like to discuss and resolve two issues I stumble upon when
> working with HPUX/ia64.

In future, it's much easier to deal with separate messages for
separate matters.

> i)    I've been trying to install ans work with GMP on HPUX (ansi C
> Compiler) and Linux

It's surprising that you can install using only the kernel.  Perhaps
you mean the GNU/Linux system.

> Though
> installation and tests (including "make check") work fine with
> gcc/Linux, t-constants fails during the first phase ("make check"),
> with an error (5 out of 4 tests passed),
>
>         PP_INVERTED == 21cfe6cfc938b36b, but pp_inverted_calc == 0

I think we had at least two problems in 4.1.2 with hpux ia64, firstly
that we only know lp64 mode in the asm code, but didn't force
CFLAGS=+DD64 in the build, and secondly that the data in that
invert_limb.asm was for little endian, whereas hpux runs in big endian
mode.  You might like to try forcing the CFLAGS if you haven't
already, and change the data to

.LC0:   data4 0x403f8000, 0x00000000, 0x00000000, 0x00000000    C 2^64
        data4 0x407f8000, 0x00000000, 0x00000000, 0x00000000    C 2^128

This will all be addressed properly in the next release.  Until then
unfortunately we can't really claim to work on hpux ia64, or only with
the generic C code (ie. cpu "none", which will be very slow).

> The compilation works just fine on Linux with Gcc. However with
> aCC/HPUX, the conversion of assembly file from invert_limb.asm to
> tmp-invert_limb.s and then to invert_limb.o does not work (bcoz of
> .section .rodata ).

Thanks, yep, that should be just ".rodata".

> first, mp_size= 1, mp_d=1
> sec, mp_size=1, mp_d=0

Hmm.  Something has gone wrong before this point, a value 0 should be
represented with size==0.

> Req: Please donot share the test source with/on public distributions lists.

Umm, too late for that, bug-gmp is variously archived.

-- 
Note all followups to address@hidden only, please.