PARI/GP Bug report logs - #1252
Mac OS X 'Lion' default compiler miscompiles bezout.

Toggle useless messages

Message #5 received at submit@pari.math.u-bordeaux.fr (full text, mbox, reply):

Package: pari
Version: 2.5.0

The function bezout applied to integers gives the wrong output. 
For very small numbers (like (4,6) or (9,14)) it works fine. 
However, for example:
  bezout(32,432)
gives
  [1,0,32].

But 32 is not the gcd --- in fact, the input gcd(32,432) does give the correct answer: 16.
A lot of other random trials (like bezout(9,20)) also result in wrong solutions. 

This is the version-information pari supplies:
                          GP/PARI CALCULATOR Version 2.5.0 (released)
                      i386 running darwin (x86-64 kernel) 64-bit version
   compiled: Nov 22 2011, gcc-4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)
                        (readline v6.2 enabled, extended help enabled)
I downloaded this version yesterday (nov 22) from the Pari website.

Message #10 received at 1252@pari.math.u-bordeaux.fr (full text, mbox, reply):

On Wed, Nov 23, 2011 at 03:09:34PM +0000, Theo van den Bogaart wrote:
> Package: pari
> Version: 2.5.0
> 
> The function bezout applied to integers gives the wrong output. 
> For very small numbers (like (4,6) or (9,14)) it works fine. 
> However, for example:
>   bezout(32,432)
> gives
>   [1,0,32].
> 
> But 32 is not the gcd --- in fact, the input gcd(32,432) does give the correct answer: 16.
> A lot of other random trials (like bezout(9,20)) also result in wrong solutions. 
> 
> This is the version-information pari supplies:
>                           GP/PARI CALCULATOR Version 2.5.0 (released)
>                       i386 running darwin (x86-64 kernel) 64-bit version
>    compiled: Nov 22 2011, gcc-4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)
>                         (readline v6.2 enabled, extended help enabled)
> I downloaded this version yesterday (nov 22) from the Pari website.

Hello Theo,

This is a well-known bug in OS X Lion default compiler which has been reported several
time to the PARI lists. We suggest you use GCC 4.5 which is much more reliable.
If you can report the problem to Apple, please do so.

Cheers,
Bill.

Message #17 received at 1252@pari.math.u-bordeaux.fr (full text, mbox, reply):

On Wed, Nov 23, 2011 at 04:45:45PM +0100, Bill Allombert wrote:
> On Wed, Nov 23, 2011 at 03:09:34PM +0000, Theo van den Bogaart wrote:
> > Package: pari
> > Version: 2.5.0
> > 
> > The function bezout applied to integers gives the wrong output. 
> > For very small numbers (like (4,6) or (9,14)) it works fine. 
> > However, for example:
> >   bezout(32,432)
> > gives
> >   [1,0,32].
> > 
> > But 32 is not the gcd --- in fact, the input gcd(32,432) does give the correct answer: 16.
> > A lot of other random trials (like bezout(9,20)) also result in wrong solutions. 
> > 
> > This is the version-information pari supplies:
> >                           GP/PARI CALCULATOR Version 2.5.0 (released)
> >                       i386 running darwin (x86-64 kernel) 64-bit version
> >    compiled: Nov 22 2011, gcc-4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)
> >                         (readline v6.2 enabled, extended help enabled)
> > I downloaded this version yesterday (nov 22) from the Pari website.
> 
> Hello Theo,
> 
> This is a well-known bug in OS X Lion default compiler which has been reported several
> time to the PARI lists. We suggest you use GCC 4.5 which is much more reliable.
> If you can report the problem to Apple, please do so.

I finally found a fix, see

<http://pari.math.u-bordeaux.fr/archives/pari-dev-1312/threads.html>

If one changes the definition of divll in src/kernel/x86_64/asm0.h

#define divll(a,b) \
__extension__ ({ ulong __value, __arg1 = (a), __arg2 = (b); \
   __asm__ ("divq %4" \
        : "=a" /* %eax */ (__value), "=&d" /* %edx */ (hiremainder) \
                                     ^^^^^
        : "0" /* %eax */ (__arg1), "1" /* %edx */ (hiremainder), "mr" (__arg2)); \
   __value; \
})

by replacing "=&d" by "=d", then GP works fine (make test-all succeed).

Cheers,
Bill.

Message #22 received at 1252-done@pari.math.u-bordeaux.fr (full text, mbox, reply):

Hello,

A number of people have reported success with the fix applied
(GIT rev 4f94c01fdac9386cc3857a98b778e9205f0a002b), so I close this report.

Sorry it took so much time!
Bill.

Send a report that this bug log contains spam.