From: Donald G Porter <dgp@em...> - 2003-12-12 16:08:12
> * *#132 "Floating-Point Conversion Improvement"* - Apparently this
> is a real bear to implement. :^(
I want this one to happen. I want it enough, I'll probably end up
taking it on just so that I can be sure it gets done. Will be grateful
for help, of course.
FWIW, I think the spec on this proposal is finished. Only implementation
seems to be the difficulty, so I think we could go ahead and vote on
this to move it to Accepted status, as an additional sign/motivation
that implementation work is worthwhile.
| Don Porter Mathematical and Computational Sciences Division |
| donald.porter@... Information Technology Laboratory |
| http://math.nist.gov/~DPorter/ NIST |
From: Donal K. Fellows <donal.fellows@ma...> - 2003-12-16 15:55:53
Donald G Porter wrote:
>I want this one to happen. I want it enough, I'll probably end up
>taking it on just so that I can be sure it gets done. Will be grateful
>for help, of course.
I think you'll be able to ask me for help, though it'll be as a general
coder-type person and not as someone who knows the guts of
floating-point handling very well.
>FWIW, I think the spec on this proposal is finished. Only implementation
>seems to be the difficulty, so I think we could go ahead and vote on
>this to move it to Accepted status, as an additional sign/motivation
>that implementation work is worthwhile.
Modulo the references to a particular implementation of the
floating-point code within the TIP, I'd largely agree. However, I'm
less keen on mandating the use of David Gay's code. Instead, perhaps
that should just be taken as being illustrative of what is desired. I
wonder whether it is possible to do this code in a platform-independent
way though; I thought the messy nature of the DG codebase was a major
implementation sticking point...? Did I misunderstand what KBK was
saying earlier this year? Is something less than perfect but still good
enough for most purposes (and portable) likely to be a reasonable