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

Re: I vote NO on motion 21...



On 2010-12-04 19:15:08 -0800, Dan Zuras Intervals wrote:
> 	Folks,
> 
> 	I vote NO on motion 21.
> 
> 	My reason has nothing to do with the set of
> 	comparisons proposed.  Any reasonable set is
> 	fine with me.
> 
> 	The reason I vote NO is that the motion is
> 	written in a manner that suggests (or even
> 	implies) the existence of state attached to
> 	the act of making a comparison.

This is wrong: it doesn't suggest or imply anything like that (the
word "state" is used purely in an abstract specification; there are
no data formats with some state in them). Perhaps the wording isn't
perfect, but we are not voting on the wording.

> 	We wrote comparisons in a similar manner in
> 	754-1985 & it led to implementers actually
> 	CREATING such state as 4 global variable
> 	bits usually in a PSW attached to the thread
> 	of execution in question.
> 
> 	At the time & in the era of 4 MegaHz 8-bit
> 	microprocessors that implemented their
> 	floating-point in software more often than
> 	not, this was not considered bad.

So, what's wrong at that time?

> 	But in the years that have followed it has
> 	become clear that requiring state is a bad
> 	thing in the era of multi threaded multi
> 	GigaHz 32 or 64 bit microprocessors with
> 	multiple on chip floating-point units.  It
> 	creates an interlock choke point similar to
> 	a branch that slows everything down & makes
> 	added headaches for the hardware designer.

They are not forced to use the same kind of implementation today.

-- 
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 / Arénaire project (LIP, ENS-Lyon)