mbutts@mntgfx.mentor.com (Mike Butts) (11/08/88)
The Analog Devices ADSP-3264 is a VLSI pipelined floating point
chip which runs 31 MFLOPS on single or double precision IEEE
adds or multiplies. It also has hardware division and square
root, which take multiple cycles and block the multiplier pipe
until complete. Exact single precision divide takes 360 nanoseconds.
I agree that pipelined SW implementations of those operations
is usually more appropriate, and is the standard practice in scientific
computing. Important scientific algorithms have been developed
to avoid division in, say, matrix processing. However, the AD
chip makes divisions and square roots available relatively
cheaply, for when you can't or don't want to avoid them.
AD chips also support fixed point integer operations, including
fast pipelined integer multiplies. I have found this property
of AD chips very useful in the past, giving me fast integer
multiplies essentially for free, given that I had the FP chip(s).
The Weitek chips (at least a few years back) didn't do integers.
I don't know why the AD chips aren't more popular; they're very well
designed, and their digital CMOS process is very fast.
--
Mike Butts, Research Engineer KC7IT 503-626-1302
Mentor Graphics Corp., 8500 SW Creekside Place, Beaverton OR 97005
...!{sequent,tessi,apollo}!mntgfx!mbutts OR mbutts@pdx.MENTOR.COM
These are my opinions, & not necessarily those of Mentor Graphics.