Motorola DSP96002 - Manual (Page 628)

MOTOROLA DSP96002 USER’S MANUAL B-109
_clip2_znn
move d6.s,d0.s y:(r1),d6.s ;Zo 1 1
fsub.s d0,d6 d6.s,d0.s ; 1 1
fseedd d6,d4 ; 1 1
fmpy.s d6,d4,d6 d9.s,d2.s ; 1 1
fmpy d0,d4,d0 fsub.s d6,d2 d2.s,d3.s ; 1 1
fmpy.s d6,d2,d6 d2.s,d4.s ; 1 1
fmpy d0,d4,d0 fsub.s d6,d3 ; 1 1
fmpy.s d0,d3,d0 ; 1 1
fcmp d7,d0 ; 1 1
ftfr.s d0,d7 fflt ; 1 1
rts ; 2 2
B.1.45 Walsh-Hadamard Transforms
The Walsh-Hadamard transform (WHT) is an orthogonal transform requiring only additions and subtrac-
tions. The transform can be decomposed similar to the fast fourier transform (FFT) to yield a fast imple-
mentation of the WHT.
B.1.45.1 In-place WHT
Since the WHT requires 2 loads and 2 stores per butterfly, the maximum throughput for a WHT butterfly is
4 cycles. This implementation executes 2 butterflies in 8 cycles on the inner loop for a 4N per butterfly
execution speed. The last stage is split out and also executes 2 butterflies in 8 cycles for each pass of the
loop.
In this example, a 16 point transform is performed. The input data are in X:0-f and the output is in x:0-f in
bit reversed order.
Execution speed for a 1024 point WHT is 1.68 milliseconds at 13.5 MIPS.
page 132,60,1,1
;
; Implements the Walsh-Hadamard Transform
;
iord equ 4 ;order of transform=log2(npoints)
n equ 1<<iord ;length of transform
org x:0
data