(j3.2006) background reading

[Bill Long]

On 1/24/12 3:49 PM, Van Snyder wrote:
>
> On Fri, 2012-01-13 at 17:34 -0800, Dan Nagle wrote:
>> Hi,
>>
>> While our discussions at 197 will most likely be based
>> on 11-256r1, the following might have some interest:
>>
>> 08-126, 08-131r1, N1835, N1856 and the tutorials
>> about coscalars and the Rice U CAF group presentation
>> in the Tutorials folder.
>
> 11-256r1 proposes adding CO_MAX.
>
> As with MAXVAL and MAXLOC, for linear algebra (and many other cases as
> well) the usually-interesting calculation is the value or location of
> the maximum absolute value.
>
> Should one expect CO_MAX(abs(...)) to be as efficient as a CO_MAXABS?
>

If you do this, you will need to specify a separate RESULT variable, 
since without that the result overwrites the input, and the input here 
is an expression.  The implementation could initially copy the input in 
to the result, doing the ABS on the way, and then perform the reduction 
in place on the result variable.   So I do not see why a significant 
performance hit would be necessary.  This would seem like a generally 
applicable optimization whenever the input argument is an expression.

Cheers,
Bill


>
> _______________________________________________
> J3 mailing list
> J3 at j3-fortran.org
> http://j3-fortran.org/mailman/listinfo/j3

-- 
Bill Long                                           longb at cray.com
Fortran Technical Support    &                 voice: 651-605-9024
Bioinformatics Software Development            fax:   651-605-9142
Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101