Re: Last IRIX "make check" problem.

Re: Last IRIX "make check" problem.

[Brian Gough]

Lowell Johnson writes:
 > FAIL: gsl_isinf(inf) (0 observed vs 1 expected)
 > ...
 > 
 > It appears that the following lines in sys/infnan.c
 > 
 >   int status = (x != x);
 >   double y = x - x;
 > 
 > are not serving their intended purposes?

Thanks for the bug report. Are you using the irix compiler?
If so, is this the problem? (from the cc manpage),

 IEEE_comparisons[=(OFF|ON)] Force all comparisons to yield
   results conforming to the IEEE 754 standard for NaN and Inf operands.
   If this option is OFF (the default case), certain optimizations will
   produce non- IEEE results for comparisons involving NaN and Inf
   values, e.g. 'x==x' will be treated as TRUE without executing a test.

Re: Last IRIX "make check" problem.

[Lowell Johnson]

On Tue, 04 Sep 2001, Brian Gough wrote:
> Lowell Johnson writes:
>  > FAIL: gsl_isinf(inf) (0 observed vs 1 expected)
>  > ...
>  > 
>  > It appears that the following lines in sys/infnan.c
>  > 
>  >   int status = (x != x);
>  >   double y = x - x;
>  > 
>  > are not serving their intended purposes?
> 
> Thanks for the bug report. Are you using the irix compiler?
> If so, is this the problem? (from the cc manpage),
> 
>  IEEE_comparisons[=(OFF|ON)] Force all comparisons to yield
>    results conforming to the IEEE 754 standard for NaN and Inf operands.
>    If this option is OFF (the default case), certain optimizations will
>    produce non- IEEE results for comparisons involving NaN and Inf
>    values, e.g. 'x==x' will be treated as TRUE without executing a test.

Yes, we are using the IRIX MIPSpro C compiler, and yes, the
IEEE_comparisons=ON did the trick.  Thanks.

Unfortunately, it appears that our cc man page may be a bit out of date. 
When we first started looking at compiler options for our problem, we
didn't come across anything.  Then when you suggested the IEEE_comparisons
option, I did a search in the man page and ... it wasn't there!  Then we
noticed the man page header says "Last changed: 4-15-99".  We know that
our compiler version is newer than that, so we figured the IEEE_comparisons
option is available but didn't know how to invoke it.   After a couple of
trial and errors, we got it to work with -OPT:IEEE_comparisons=ON.

What is the last changed date on your IRIX cc manpage?

Thanks again for the fix.


    Lowell

-- 

Re: Last IRIX "make check" problem.

[Brian Gough]

Lowell Johnson writes:
 >  I did a search in the man page and ... it wasn't there!  Then we
 > noticed the man page header says "Last changed: 4-15-99".  We know that
 > our compiler version is newer than that, so we figured the IEEE_comparisons
 > option is available but didn't know how to invoke it.   After a couple of
 > trial and errors, we got it to work with -OPT:IEEE_comparisons=ON.

Ok, I'll add this to the MACHINES file.

 > What is the last changed date on your IRIX cc manpage?

I found it from a web search, it said 'last changed 4-7-99', but the
manpage was actually for opt(5).

Reading more closely it looks like IEEE_comparisons is deprecated
and -OPT:IEEE_NaN_inf=ON is now the preferred option.

Brian

Last IRIX "make check" problem.

[Lowell Johnson]

Only one test issue remains in our IRIX GSL 0.9.2 build:  The sys directory
testing failed with the following:

...
FAIL: gsl_isinf(inf) (0 observed vs 1 expected)
FAIL: gsl_isinf(-inf) (0 observed vs -1 expected)
...
FAIL: gsl_isnan(nan) (0 observed vs 1 expected)
...
FAIL: gsl_finite(inf) (1 observed vs 0 expected)
FAIL: gsl_finite(nan) (1 observed vs 0 expected)
...

It appears that the following lines in sys/infnan.c

  int status = (x != x);
  double y = x - x;

are not serving their intended purposes?


    Lowell Johnson
--