double rounding in x86

[Michael Meissner]

| This would seem to make the x86 for x <= 5 a second class citizen for
| scientific computing.  Anyone happen to know whether this has been
| fixed on the P6?  Maybe someone with a beta-test unit can check this (I
| signed up for one, don't know how they pick people).

Given that the x86 floating point is usually slower in real programs
than most other CPU's, it already is a second class citizen for
scientific computing.  It's only saving grace is that it is cheap and
available everywhere.  I believe that a lot of the slowness is due to
the fact that the stack architecture is hard to generate optimial code
for.  Another factor may be doing calculations in 80 bit precision
slows things down as compared to 64 bit precision.

-- 
Michael Meissner, Cygnus Support (East Coast)
Suite 105, 48 Grove Street, Somerville, MA 02144, USA
meissneracygnus.com,	617-629-3016 (office),	617-629-3010 (fax)