gmp 4.3.0 asm errors on ppc64

Jørn Amundsen Jorn.Amundsen at ntnu.no
Mon Apr 20 00:16:29 CEST 2009


Except for some of the xlC (C++) test to compile very slowly, it pass
the test suite (C and C++) without any failures indeed.

I can send any output you might need. Also, I am also in position to 
provide you with an account on our test system if this is of interest
to you - without any fine prints like ``we own your code''. :-)

--joern


On Sun, 19 Apr 2009, Torbjorn Granlund wrote:

> Jørn Amundsen <Jorn.Amundsen at ntnu.no> writes:
>
>  On Sun, 19 Apr 2009, Torbjorn Granlund wrote:
>  >
>  > I don't understand why you introduce EXTERN_TEXT.  If .globl is the
>  > right declaration for any externally defined symbol, why not use the
>  > existing mechanism EXTERN?  Am I missing something?
>  >
>  > --
>  > Torbjörn
>  >
>
>  This was the very first thing I tried. In divrem_1.asm, the EXTERN
>  macro expands into
>
>          .globl  __gmpn_invert_limb
>
>  , while the CALL macro expands to
>
>         bl      .__gmpn_invert_limb
>
> So the reply to my question "am I missing something" is "a dot".  :-)
>
>  I have built almost all of the previous gmp 4.x releases with xlc, and
>  it has always worked flawlessly. This is the first time I experience
>  problems. We also have the latest gcc 4.3.3 compiler
>  installed. However, my experience is xlc is performing better.
>
> I forgot to test using xlc for this release, and no automated testing is
> done using that compiler.
>
> When I tested now, I could get things to work with changes similar to
> the ones suggested by you, except that I get a check failure under Aches
> 5.1 for mpz/t-mul.c.  Did you run "make check" successfully?
>
> Users of xlc might want to update from the GMP hg repository,
> http://gmplib.org/repo-usage.html.
>
> --
> Torbjörn
>


More information about the gmp-bugs mailing list