I agree 9!:19[0 is a very strong condition, original default
comparison tolerance is 2^_44. We found that depending on
platforms and CFLAGS, either

g421e can pass with zero tolerance or 

pass with default tolerance but with random failures
unless using a larger tolerance.

The zero tolerance was intended to identify which type of
success it belonged.  You may ignore the failure.

I suspect -O1 in ppc64 may also fail g421e with default
tolerance if the test is repeated many times.

Are there any compilation options in ppc64 so that -O1 can work?

Пт, 30 сен 2016, Xiao-Yong Jin написал(а):
> commit 50cf6e1 changed g421e to do 9!:19[0
> 
> On ppc64, it caused the test to fail with -O1 to gcc 4.4.7.
> 
>    + test2 *"0 'biIdxqz'
> |assertion failure: RUN1
> |   xx    (u//.@(v/)-:u/"_/.@(v/"_))yy
>    xx
> 24.2186 475.758 _105.497 336.422 _413.12 72.2156 _160.262 33.4909 _85.4288 
> _19.932 _350.381 _20.6579 110.159 232.242 56.6778 _436.011 370.545
>    yy
> 87.8893 343.818 _488.396 234.015 386.031 _205.93 416.216 11.2203 213.67 
> 292.035 46.3149 _153.041 _262.872 99.3587 _207.526 _80.8204 _112.822 _165.617 
> _351.292 _86.3578 _209.097 32.4487 139.001
>    xx(+//.@(*/) - +/"_/.@(*/"_))yy
> 0 0 0 _2.91038e_11 0 0 _2.91038e_11 2.91038e_11 0 0 0 6.59384e_12 0 0 
> 2.91038e_11 0 7.27596e_11 0 _5.82077e_11 0 0 0 0 0 2.91038e_11 _4.36557e_11 0 
> 0 2.91038e_11 0 0 1.45519e_11 0 _7.27596e_12 0 0 0 0 0
>    xx(([:+//.*/) - [:+/"_/.*/"_)yy
> 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
> 
> Only compiling with -O0 passes the test.  It doesn't seems to be wrong, so I 
> guess gcc optimized the special code for @ with a different execution order 
> and introduced some floating point precision loss.
> 
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm

-- 
regards,
====================================================
GPG key 1024D/4434BAB3 2008-08-24
gpg --keyserver subkeys.pgp.net --recv-keys 4434BAB3
gpg --keyserver subkeys.pgp.net --armor --export 4434BAB3
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to