Strange but true 8.04 updates today May 26[long]

[Manuel McLure]

On Tuesday 27 May 2008 18:37:44 Martin Laberge wrote:
> On Tuesday 27 May 2008 20:03:37 Manuel McLure wrote:
> > You don't need a 64-bit processor to do fast 64-bit floating point
> > processing. A 64 bit processor will do 64-bit *integer* processing faster
> > than a 32-bit processor, but the floating point processor in a 32-bit
> > processor handles 64-bit floating point numbers just as well as the one
> > on a 64-bit processor. This has been true from the original 80287 (which
> > handled *80* bit floating point numbers.) For floating point math, the
> > only advantage a 64-bit processor might have is wider memory bandwidth,
> > but if you're running the 64-bit processor in 32-bit mode you have that
> > anyway. The floating point registers in 32-bit mode are the same width as
> > the ones in 64-bit mode.
>
> do you know if the 80 bits floats of the 287 are the same length
> today, or are they larger??

Modern processors actually use less than 80 bits now - they all provide IEEE 
floating point which only defines up to 64 bits for floats. The x87 used 80 
bits for internal operations and then rounded to 64 in contravention of the 
standard. This is still the case if you use the x87 instructions on a modern 
processor, but many compilers for Intel/AMD now use SSE2 instructions (if 
available) for floating point. The SSE2 instructions use strict IEEE 
semantics (i.e. 64-bits throughout.)
-- 
Manuel A. McLure KE6TAW <manuel at mclure.org> <http://www.mclure.org>
...for in Ulthar, according to an ancient and significant law,
no man may kill a cat.                       -- H.P. Lovecraft