Re: Clang as default compiler November 4th

On Tue, Sep 11, 2012 at 08:12:30AM -0700, Steve Kargl wrote:
> On Tue, Sep 11, 2012 at 04:27:55PM +0200, Tijl Coosemans wrote:
> > On 11-09-2012 16:10, Dimitry Andric wrote:
> > > On 2012-09-11 15:24, Steve Kargl wrote:
> > >> What is important is whether software built with clang functions
> > >> correctly.  See for example,
> > >>
> > >> http://math-atlas.sourceforge.net/errata.html#WhatComp
> > > 
> > > Yes, maths support, specifically precision, is admittedly still one of
> > > clang's (really llvm's) weaker points.  It is currently not really a
> > > high priority item for upstream.
> > > 
> > > This is obviously something that a certain part of our userbase will
> > > care a lot about, while most of the time they won't care so much about
> > > licensing or politics.  So those people are probably better off using
> > > gcc for the time being.
> > 
> > Does it affect the accuracy of libm functions?
> > 
> 
> I'm not sure if anyone has done any extensive testing.
> I've started to run some of my test codes to compare
> certain functions in a clang-compiled libm, gcc-compiled
> libm, and reference solutions generated from math/mpfr.
> For a locally patched j0f, I found that clang gave
> much worse accuracy.  If I revert the local patch,
> clang and gcc are to give the same results.  Unfortnately,
> an unpatched j0f gives 500000 ULP errors.

Steve,

Can you please provide a small self contained test case that shows
that clang is doing worse on accuracy than gcc?

So that we can analyze it and decide if it's a bug in the code or
in the compiler. So far we know absolutely nothing.

Thank you, Roman