Re: [lmi] Unit test for C99 round()

[Greg Chicares]

On 2008-06-03 18:33Z, Vadim Zeitlin wrote:
> On Tue, 03 Jun 2008 16:34:26 +0000 Greg Chicares <address@hidden> wrote:
> 
> GC> Would you please run the new 'round_test' on GNU/Linux,
> 
>  Thanks for adding it to Makefile.am BTW!
> 
> GC>  - without modification; and
> GC>  - with the patch below [1];
> GC> and describe its results here?
> 
>  The result is the same in both cases:
> 
> % ./test_round
[...]
> .... 1128 tests succeeded
> no errors detected
> 
> So glibc (version 2.3.6) seems to be ok too.

Great, thanks.

>  I didn't test under x86 (this was under amd64) but can also rebuild LMI in
> chroot if you think it's going to be useful.

Thanks, but that's not necessary. I just wanted to be sure
this unit test and the code it was written to test hadn't
co-evolved in such a way that no other implementation could
hope to pass all the tests. And I think the algorithm we
chose is unusual enough that no one else would be using it.
(FRNDINT with both RC bits = 0 gives the right answer in all
cases except for a sparse point set, so detect and adjust
only those points--transforming the hardware's unbiased
rounding into the degenerate thing C99 7.12.9.6 requires.)