develooper Front page | perl.perl5.porters | Postings from December 2017

Re: [perl #74864] test suite failure on powerpc with -Duselongdouble

Thread Next
From:
Zefram
Date:
December 11, 2017 01:10
Subject:
Re: [perl #74864] test suite failure on powerpc with -Duselongdouble
Message ID:
20171211011018.GJ25404@fysh.org
Niko Tyni wrote:
>I assume the 128-bit long doubles on this platform are actually a sum of
>two 64-bit doubles, which "does not conform to the IEEE floating-point
>standard" [1] and "has serious arithmetic anomalies" [2].
>
>If this something that can not be supported, perhaps Configure should
>detect such a platform and disable the longdouble setting or error out?

A double-double format *can* be well-behaved, if its two halves are
constrained to have a fixed exponent difference and all the edge cases
are handled properly.  I concur that we should detect the badly-behaved
ones and reject their use.

Can someone offer me an account on one of these PowerPC Linux systems
with the suspect long double?  Having never experienced a double-double
system before, I'd like to experiment to see exactly where the numeric
semantics break down, and thus how to efficiently detect the problem.

-zefram

Thread Next


nntp.perl.org: Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at ask@perl.org | Group listing | About