LAPACK Archives

[Lapack] suggestion for dlamch


Dear LAPACKers,

I would like to suggest moving function dlamc3 out of dlamch.f and 
packaged as a separate source file.

Function dlamc3(a,b) just returns a+b.

dlamch.f calls dlamc3 to "fool" the compiler to ensure that relevant 
values are stored and not held in registers or affected by optimizers. 
By having dlamc3 in a separate file instead of within dlamch.f, there is 
less chance of inlining dlamc3 by optimizing compilers.


Another suggestion may be providing an alternate dlamch that uses 
Fortran 90 intrinsics to directly provide the characteristics of the 
floating point system.




-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://lists.cs.utk.edu/private/lapack/attachments/20070306/e08847e1/attachment.htm
 

<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or