On Sun, Sep 03, 2017 at 08:36:01PM +0200, Rafael Laboissière wrote:
> * Adrian Bunk <b...@debian.org> [2017-09-03 19:01]:
>
> > On Sun, Sep 03, 2017 at 05:52:26PM +0200, Rafael Laboissière wrote:
> >
> > > It seems that many of the unit test failures happen when the
> > > function henon is involved. This function computes the Henon map
> > > with a default transient initial period of 10000 samples. The
> > > computation in the henon function is iterative and my guess is that
> > > the results are sensitive to the architecture-specific floating
> > > point representation. Would -ffloat-store help here?
> >
> > -ffloat-store sometimes matters on i386, but on most architectures it is
> > a complete nop.
>
> Thanks for your quick reply. Are there other sources of differences in the
> representation of float numbers across the architectures, that might explain
> the different results in the henon function?
Sorry, I don't have any immediate ideas on that
(and no time to debug it properly).
> Rafael
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
