Many new math failures

Andreas Jaeger aj@suse.de
Sun Jun 30 04:35:00 GMT 2002


Ulrich Drepper <drepper@redhat.com> writes:

> On Thu, 2002-06-27 at 05:29, Stephen L Moshier wrote:
>> 
>> > Test "Real part of: cacosh (0.75 + 1.25 i) ==
>> >1.13239363160530819522266333696834467 +
>> >1.11752014915610270578240049553777969 i":
>> >@@ -310,8 +308,8 @@
>> > float: 3
>> > idouble: 1
>> > ifloat: 3
>> >-ildouble: 37
>> -ldouble: 37
>> >+ildouble: 370
>> >+ldouble: 370
>> 
>> This is due to arithmetic cancellation error inside the function routine.
>> It might be better to use the identity cacosh(x) = I * cacos(x) here.
>
> No, the excerpt from the patch is misleading.  See the @@ line?  The 370
> ulps error is for cpow.

Upps.  I'm appending the complete ULPs file from gcc 2.95 so that
everybody can look at all the data.

Andreas

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: gcc-2.95.ulps
URL: <http://sourceware.org/pipermail/libc-alpha/attachments/20020630/7f716c57/attachment.ksh>
-------------- next part --------------

-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj


More information about the Libc-alpha mailing list