This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Minimum floating-point requirements
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: David Edelsohn <dje dot gcc at gmail dot com>
- Cc: <libc-alpha at sourceware dot org>, Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>, Steve Munroe <sjmunroe at us dot ibm dot com>
- Date: Thu, 30 Jan 2014 00:54:27 +0000
- Subject: Re: Minimum floating-point requirements
- Authentication-results: sourceware.org; auth=none
- References: <CAGWvnym4yN=7rLrm0RRtNN++T=xwx8r3MUKJOfz4r+H=Z9zd7Q at mail dot gmail dot com>
On Wed, 29 Jan 2014, David Edelsohn wrote:
> As we discussed on the GCC Mailing List, I appreciate your desire for
> all GLIBC targets and ports to conform to relevant standards (IEEE,
> ISO C, etc.), but I think it is a very bad precedent to use standards
> conformance and community rules to override port maintainers in
> port-specific decisions. GLIBC should strive to be welcoming of all
My view as de facto maintainer of the soft-float powerpc support is that
using a version of the IBM long double support code from libgcc that makes
the basic arithmetic follow the normal accuracy goals of glibc libm
functions is the most practical way to keep testsuite failures under
control, allow function implementations for various functions such as
those in <complex.h> to be shared among all long double formats rather
than needing separate variants for different formats, and maintain the
ability to use the testsuite to identify bugs in glibc.
I don't believe there is any port-specific decision that glibc functions
on powerpc should not work in cases where they are expected to work on
other platforms, such as cbrtl (LDBL_MAX). Instead, I understand that
Adhemerval has been using my libgcc patch when regenerating libm-test-ulps
for hard-float powerpc for 2.19 and investigating remaining failures.
--
Joseph S. Myers
joseph@codesourcery.com