Polyhedron performance regression

[FX Coudert]

> On x86_ia64/FC5 with:
> /svn-4.3/bin/gfortran -static -O3 -ffast-math  -funroll-loops ac.f90
> time ./a.out
>
>   revision        time
>   118474        9.77s
>   118475        12.3s
>    current        33.0s !

On x86_64-linux "Dual Core AMD Opteron(tm) Processor 265", 1.8 GHz,  
with gfortran -static -O2 -funroll-loops -ftree-vectorize ac.f90

118474  21.695
118480  21.710
118485  21.794
118491  21.771
118492  54.635

I'm ashamed that revision 118492 (http://gcc.gnu.org/viewcvs? 
view=rev&revision=118492), which introduced this performance  
regression, is the MOD/MODULO patch by Paul Thomas and myself.

While with -ffast-math:

rev. 118474		rev. 118492
17.730			55.599			-O2 -funroll-loops ac.f90 -ffast-math
17.776			53.062			-O3 -funroll-loops ac.f90 -ffast-math
25.549			59.131			-O2
17.554			52.921			-O3 -ftree-vectorize -ffast-math -funroll-loops


Still one thing I don't understand is the difference between Paul  
results and mine for the fwprop: it looks like it impacted with -O3 - 
ftree-vectorize -fno-fast-math.

FX