Igor Schein on Wed, 11 May 2005 17:19:56 +0200


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: PARI/GP build failures with gcc-4.0


On Mon, May 09, 2005 at 07:25:14PM +0200, Bill Allombert wrote:
> Hello PARI developers,
> 
> For those of you who keep an eye on the the PARI/GP buildlogs,
> ( http://pari.math.u-bordeaux.fr/buildlog.html ) you have probably seen
> that I have added 4 amd64 builds and 4 gcc-4.0 (on x86) builds.
> 
> The amd64 builds are OK, but the gcc 4.0 ones exhibit two problems:
> 1) matker is broken
> 2) divri_with_gmp is broken.
> 
> For now, I am tempted to believe this is a compiler bug.
> This only happens when compiled with -O3 -fomit-frame-pointer.

If you configure  with CFLAGS=-fno-omit-frame-pointer, the problem
goes away, but you lose performance.  If you configure with
CFLAGS=-fno-inline-functions, make dobench passes, but the following
occurs:

? thue(thueinit(x^7-401),88);
  *** thue: Warning: Non trivial conditional class group.
  *** May miss solutions of the norm equation.
  *** thue: bug in GP (Segmentation Fault), please report

So the best workaround is to use CFLAGS=-fno-gcse-after-reload with
gcc 4.0.

I also tried gcc mainline compiler:

gcc version 4.1.0 20050508 (experimental)

The situation is even worse there.

Hope that helps.

Igor