This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Regenerate x86 and x86_64 ulps
- From: Andi Kleen <andi at firstfloor dot org>
- To: "Joseph S. Myers" <joseph at codesourcery dot com>
- Cc: Jeroen <_jeroen_ at yahoo dot com>, "libc-alpha at sourceware dot org" <libc-alpha at sourceware dot org>, Andi Kleen <andi at firstfloor dot org>
- Date: Wed, 3 Jul 2013 16:37:42 +0200
- Subject: Re: Regenerate x86 and x86_64 ulps
- References: <Pine dot LNX dot 4 dot 64 dot 1307022001020 dot 28677 at digraph dot polyomino dot org dot uk> <1372836611 dot 94370 dot YahooMailNeo at web124504 dot mail dot ne1 dot yahoo dot com> <Pine dot LNX dot 4 dot 64 dot 1307031352210 dot 18167 at digraph dot polyomino dot org dot uk>
On Wed, Jul 03, 2013 at 01:54:25PM +0000, Joseph S. Myers wrote:
> On Wed, 3 Jul 2013, Jeroen wrote:
>
> > Are compiler to allow extended instruction sets (SSE*,AVX) instructions
> > allowed or even explicitly wanted in testing?
>
> I'm not clear on the question. Compilers configured with a range of
> different -march= settings and similar optimizations should all work for
> building glibc so testing with them is useful, and testing on a range of
> processors is also useful because of the multiarch variants of functions.
> -mfpmath=sse on 32-bit x86 is riskier, however, and I don't know if glibc
> built with it would work correctly.
My FC14 setup with gcc 4.5 is totally unable to pass even a fraction
of the math tests on 32bit. 64bit works. With gcc 4.7 on FC16
things also work on 32bit and 64bit.
-Andi