Use -fno-builtin for math tests...

Andreas Jaeger aj@suse.de
Thu Nov 13 09:29:00 GMT 2003


Jakub Jelinek <jakub@redhat.com> writes:

> On Thu, Nov 13, 2003 at 09:58:51AM +0100, Andreas Jaeger wrote:
>> 
>> As Gwenole pointed out, we do not use -fno-builtin for the math tests.
>> 
>> Now the ceil test fails:
>> testing long double (without inline functions)
>> Failure: Test: ceil (0.25) == 1.0
>> Result:
>>  is:          0.00000000000000000000e+00   0x0.00000000000000000000p-3
>>  should be:   1.00000000000000000000e+00   0x8.00000000000000000000p-3
>>  difference:  nan   nan
>>  ulp       :  nan
>>  max.ulp   :  0.0000
>> 
>> Test suite completed:
>>   2525 test cases plus 2312 tests for exception flags executed.
>>   1 errors occurred.
>> 
>> Ok to commit?
>
> Shouldn't we use both -fno-builtin and -fbuiltin (ie. compile the testcases
> twice)? People will use -fbuiltin in 99.9% of cases, so it is good if that
> is tested.  The drawback is that the math tests are CPU hogs both during
> compile time and when running them.

We compile them twice already - one with inline functions and one
without.  We can compile them another time if people think it's
worth...  I can send a patch if desired,

Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Deutschherrnstr. 15-19, 90429 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/libc-alpha/attachments/20031113/f560cc2e/attachment.sig>


More information about the Libc-alpha mailing list