🚀 go-pugleaf

In article <sZcrr1G00hMNI3EF5i@cs.cmu.edu> Daniel.Stodolsky@cs.cmu.edu writes:
>
>
>So why not put some of big memory killer micros to
>work and have a 16 by 16 multiply lookup table? It would consume 10 megs
>of core, but that's nothing for a KILLER MICRO. Assume memory access
>with  a cache miss is around 3 cycles and one can schedule to avoid
>register interlocking (as in HP-PA), it seems possible to do 32x32 -> 64
>( results in registers) in about 20 cycles.  
>

  The idea above proposes to use 80 million bits of RAM and 20 clock cycles
  to compute a 32b integer multiply.  This in noncompetitive when compared
  to killer micros, which multiply more quickly and consume far less real
  estate.  Instead of lookup tables they implement dedicated hardware:

       R6000:     16 cycles      32x32 -> 64
       R3000:     12 cycles      32x32 -> 64
       M88000:     4 cycles      32x32 -> 32  **

  **88k computes the 32 lsb's of the 64b product (upper bits are discarded).

  If, as rumored, a new imul instruction is added to SPARC, you can bet
  its implementation in hardware will be just as fast as, or perhaps faster
  than, those above.  This is what Doctor Ross is good at.  And of course the
  T.I. folks have lots of DSP experience with blazing fast hw multipliers.
-- 
 -- Mark Johnson	
 	MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086
	(408) 991-0208    mark@mips.com  {or ...!decwrl!mips!mark}