4D/25 sp complex fft timings:
# points	time (msec)	butterfly (usec)
   64		  0.81		 2.109
  128		  1.72		 1.920
  256		  3.46		 1.689
  512		  7.41		 1.608
 1024		 16.3		 1.592
 2048		 34.7		 1.540   ***
 4096		 78.0		 1.587
 8192		239.0		 2.244   \___ Cache
16384		522.0		 2.276   /    Limit?
note that 8MB/sec was apparently the non-cache memory limit for a 4d/25

4D/35 sp complex fft timings:
# points	time (msec)	butterfly (usec)
   64		  0.41		 1.068
  128		  0.84		 0.938
  256		  1.80		 0.879
  512		  3.82		 0.829
 1024		  8.16		 0.797
 2048		 17.46		 0.775
 4096		 37.20		 0.757
 8192		 79.36		 0.745   ***
16384		219.80		 0.958	 >--- Cache Limit ?

R3000 sp Indigo complex fft timings:
# points	time (msec)	butterfly (usec)
   64		  0.44		 1.146
  128		  0.92		 1.027
  256		  1.98		 0.967
  512		  4.26		 0.924
 1024		  9.12		 0.891
 2048		 19.48		 0.865
 4096		 41.56		 0.846   ***
 8192		117.12		 1.100	 \___ Cache
16384		271.60		 1.184   /    Limit?

R4000 sp Indigo complex fft timings:
# points	time (msec)	butterfly (usec)
   64		  0.17		 0.443
  128		  0.35		 0.391
  256		  0.76		 0.371
  512		  1.60		 0.347
 1024		  3.40		 0.332  ***
 2048		  8.71		 0.387  >--- out of primary cache
 4096		 19.60		 0.399
 8192		 43.34		 0.407
16384		 98.36		 0.429

Combined SP MFLOP Ratings from FFT timings:
note: assumes n*(5*log n + 2) floating point calculations required.
      also, there is a one-time sin/cos table calculation not accounted for.
			Indigo		Indigo
# points	4D/25	R3000	4D/35	R4000
   64		2.53	4.65	5.00	12.05
  128		2.75	5.15	5.64	13.53
  256		3.11	5.43	5.97	14.15
  512		3.25	5.65	6.30	15.04
 1024		3.27	5.84	6.53	15.66
 2048		3.36	5.99	6.69	13.40
 4096		3.26	6.11	6.83	12.96
 8192		2.30	4.69	6.92	12.66
16384		2.26	4.34	5.37	11.99

