> Only if KNI allows double precision floating point (which it almost
> surely won't, since multimedia doesn't need 15-digit precision).
> Now, if KNI has very fast 32-bit to 64-bit unsigned integer multiplies it
> does open up the chance of a pretty fast number-theoretic DWT squaring,
> but I wouldn't hold my breath that it has that, either.
Anyone interested can read the PDF on Intel's entire instruction set at
ftp://download.intel.com/design/PentiumII/manuals/24319102.PDF
It's pretty big, 6MB. The normal instructions are using the same PII core,
so don't expect a speed improvement beyond the obvious boost to 500MHz (and
faster once .18 fabs come online).
As for the new instructions, you can always get slightly better performance
by optimizing your assembly to use the SIMD stuff, though you won't get a
HUGE boost from that. Otherwise, there are new instructions for multiple
integer ops that *may* help.
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
