Re: [Valgrind-users] run_innerloop error

[Tom H. <to...@co...>]

In message <200...@ac...>
          Julian Seward <js...@ac...> wrote:

> > valgrind: m_scheduler/scheduler.c:996 (vgPlain_scheduler): the
> > 'impossible' happened.
> > valgrind: VG_(scheduler), phase 3: run_innerloop detected host state
> > invariant failure
> 
> Sheesh.  That's an extremely obscure error.  I've never seen one of
> those in the wild (outside of V development land).  It indicates
> a bug in Valgrind's amd64 floating point code handling.
> 
> Are you doing something strange with floating point rounding, or
> other IEEE control word stuff (precision, exception handling) ?

We've been seeing it on amd64 machines - see my postings on the
developer list a few months ago for details.

What I see is the FPU control word changing precision from 64 bit
to 80 bit if I recall correctly. I'm pretty sure it's a kernel
bug though as I did manage to reproduce spontaneous changes in
the FPU control word in a test program without valgrind involved.

Tom

-- 
Tom Hughes (to...@co...)
http://www.compton.nu/