This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: numerical instability and estimate-probability
- From: mike stump <mrs at windriver dot com>
- To: rth at redhat dot com
- Cc: gcc-pdo at atrey dot karlin dot mff dot cuni dot cz, gcc at gcc dot gnu dot org, jh at suse dot cz
- Date: Thu, 15 Nov 2001 18:02:03 -0800 (PST)
- Subject: Re: numerical instability and estimate-probability
> Date: Thu, 15 Nov 2001 17:39:03 -0800
> From: Richard Henderson <rth@redhat.com>
> On Thu, Nov 15, 2001 at 11:31:52AM -0800, mike stump wrote:
> > It is advantageous to have gcc emit the same code, on different hosts
> > for the same source code. This currently works.
> No, it doesn't. We have code that is affected by the undefined
> order of evaluation of function arguments. E.g.
Ah, currently works to some degree. Sorry...
> foo (gen_reg_rtx (...), gen_reg_rtx (...))
> Note that this is not a correctness issue -- we don't really care
> what pseudo number is allocated to which register. However, the
> pseudo number _does_ affect e.g. the order in which register
> allocation considers pseudos for hard registers, which in turn
> affects which hard registers are assigned.
Would be nice to get rid of these things so that the compiler more
often generates the same code.
Thanks for the pointer to the problem.