Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: Motion P1788/M0047:Motion45Amendment-1 -- discussion period begins

P1788,

Normally, I am nagging people to vote.  This time, I am nagging you NOT to vote, at least until Baker officially opens Motion45Amendment-1 for voting, which should be on or after August 12.

I do not think I can count premature votes.

George Corliss
George.Corliss@xxxxxxxxxxxxx



On Aug 6, 2013, at 10:42 AM, Vincent Lefevre <vincent@xxxxxxxxxx> wrote:

> I vote NO on M0047:Motion45Amendment-1
> 
> ... for the same reasons I voted for Motion 45:
> 
> 1. CA can be efficient only in hardware, so that a requirement would
>  make sense only if hardware implementations were common or there
>  would be any plan to make them common. But there are currently no
>  hardware implementations, hardware implementations in the past have
>  been dropped, and no vendors have shown interest in implementing CA
>  in hardware. And in software, there are better alternatives to CA.
>  And CA doesn't fit well with modern computing (e.g. parallelism) as
>  it would need a lot of communication.
> 
> 2. If CA's goal is to implement other operations recommended or
>  required by P1788, it is not up to the standard to require some
>  particular implementation. A standard specifies the behavior,
>  not how features are implemented internally.
> 
> 3. There are better alternatives to EDP: exactness is not needed in
>  applications, possibly except in the worst cases, where there are
>  other ways to deal with such problems. In particular, the correctly
>  rounded dot product can be implemented without having to compute
>  the exact value.
> 
> -- 
> Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
> 100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
> Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail