20 Dec
2002
20 Dec
'02
8:04 a.m.
On Fri, Dec 20, 2002 at 09:20:26 +0000, Mike Rawes wrote:
The only sound argument I have found for using integer
fixed point over
float can be summed up in two words:
Denormal Number
Although deft coding can alleviate it somewhat, I can only hope that
with the Itanium, Opteron or whatever, this is no longer a problem!
You can tell the SSE unit(s) to remove them automatically and SSE is the
only (or main?) way of doing arithmetic on the Itanium IIRC. I haven't
read up on the Opteron.
But even in 387 denormal numbers are easy to stamp on if you find them.
- Steve