Problem building zn_poly on top of MPIR 2.6.0

[Jean-Pierre Flori]

Dear all,

While updating MPIR in Sage, we encountered problems with zn_poly (ok it is old and unmaintained, but it will need some work before we can get rid of it).
See
http://trac.sagemath.org/sage_trac/ticket/13137#comment:70

I see different possibilities:
* bug in GCC,
* problem with the new FFT (I assume what the nuss_mul test does is to compare the zn_poly FFT with the GMP/MPIR (just checked, it is indeed)).

Would you mind having a look?
From what we gathered the problem appears on
* MacOSX using FSF GCC,
* Cygwin on Windows XP
I did not get this problem on a usual 64 bits Ubuntu 12.04.1 with Uguntu GCC 4.6.3.

Cheers,
JP

[Bill Hart]

Hi,

thanks for reporting this. I had heard there was an issue.

I need some more information. In particular, does this only occur on
32 bit machines and if so what compiler version. Does dialing back the
optimisation level on the compiler to -O1 fix the problem?

Bill.

> --
> You received this message because you are subscribed to the Google Groups
> "mpir-devel" group.
> To view this discussion on the web visit
> https://groups.google.com/d/msg/mpir-devel/-/Ivp9uQn1WDcJ.
> To post to this group, send email to mpir-...@googlegroups.com.
> To unsubscribe from this group, send email to
> mpir-devel+...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/mpir-devel?hl=en.

[Jean-Pierre Flori]

On Wednesday, January 9, 2013 6:36:57 PM UTC+1, Bill Hart wrote:

Hi,

thanks for reporting this. I had heard there was an issue.

I need some more information. In particular, does this only occur on
32 bit machines and if so what compiler version. Does dialing back the
optimisation level on the compiler to -O1 fix the problem?

I'll give it a shot on my Windows 7 install (Windows is 64 bits but Cygwin is 32).
But I don't remember having trouble before, so...

I'll ask for more details for the Mac hardware used.

[Jean-Pierre Flori]

On Wednesday, January 9, 2013 7:34:33 PM UTC+1, Jean-Pierre Flori wrote:

On Wednesday, January 9, 2013 6:36:57 PM UTC+1, Bill Hart wrote:

Hi,

thanks for reporting this. I had heard there was an issue.

I need some more information. In particular, does this only occur on
32 bit machines and if so what compiler version. Does dialing back the
optimisation level on the compiler to -O1 fix the problem?

I'll give it a shot on my Windows 7 install (Windows is 64 bits but Cygwin is 32).
But I don't remember having trouble before, so...

Still no problems on my setup, so let's wait for other users feedback.

[Jean-Pierre Flori]

(I'm putting this back on mpir-devel.)

2013/1/9 Bill Hart <goodwi...@googlemail.com>:

>>> But I don't remember having trouble before, so...
>>
>> Still no problems on my setup, so let's wait for other users feedback.
>

> Then it is bound to be a compiler bug or configuration problem.
Yes but this seems to happen on quite different systems (Cygwin on 32
bits Windows XP and Mac OS X 10.7 (I'd say Intel 64 bits, asking for
confirmation) with quite different compiler (Cygwin 5.4.3 and FSF
5.6.3 respectively), but the same user, so we can hope for a config
problem :)
I'll try to setup a Win XP Cygwin in the next couple of weeks but
cannot promise anything.
I could have access to a Mac OS this afternoon, not sure though.
>
> The obvious thing to suspect is the "new" FFT. But we'd have probably
> noticed bugs by now in flint if that were the case. It's still
> possible though I guess.
Yeah I hope it is not that but it looks like the best suspect so far.
>
> Bill.



--
Jean-Pierre Flori

[Jean-Pierre Flori]

It looks more and more likely that the problem is really in zn_poly (or compilers not being keen with it) as it seems to happen with MPIR 2.4.x as well, not yet sure about that, will report back when I gather more info.
--
Jean-Pierre Flori

[Jean-Pierre Flori]

On Friday, January 11, 2013 10:03:53 AM UTC+1, Jean-Pierre Flori wrote:

It looks more and more likely that the problem is really in zn_poly (or compilers not being keen with it) as it seems to happen with MPIR 2.4.x as well, not yet sure about that, will report back when I gather more info.

Ok just found an old build log here:
https://groups.google.com/d/topic/sage-release/ngw_EiBaY0Q/discussion
which was definitely not with MPIR 2.6.0 (which wasn't released yet :) ).
So no problem with the new FFT!

I'd say we have to blame zn_poly, sorry for the noise here.

Best,
JP