DSPRelated.com
Forums

LOG2NRM.ASM

Started by "ralph.harry" February 26, 2008
Hi to all,

I was looking at a piece of code available on the net to compute log2(x)
for any number x between 2^-23 and 1.

This code is a macro included in the file called LOG2NRM.ASM:

log2nrm macro
;
; This program calculates the base 2 logarithm of an unnormalized
; 24 bit fraction "x" in register A and returns a scaled fraction
; "y" in register A.
;
; y = log2(x)/32.0 where 2**(-23) =< x < 1.0
; -23/32 =< y < 0.0
;
; Note - "x" must be a non-zero, positive fraction.
;
; Three steps are required.
;
; 1. Normalize "x" so that 0.5 =< A < 1.0.
; 2. Calculate the log2(A).
; 3. Divide the result by 32.0
;
;
; Step 1 - Normalize A to get value between .5 and 1.0
;
clb a,b
normf b1,a
move #pcoef,r1 ;point to polynomial coefficients for log2
move a,x0 ;put normalized number in x0
;
; Step 2 - Calculate LOG2 by polynomial approximation. 8 Bit
accuracy.
;
; LOG2(x) = 4.0* (-.3372223 x*x + .9981958 x - .6626105)
; a2 a1 a0
; where 0.5 <= x < 1.0
;
; r1 initially points to the coefficients in y memory in the
; order: a1,a2,a0
;
nop
nop
mpyr x0,x0,a y:(r1)+,y0 ;x**2, get a1
nop
mpy x0,y0,a a,x1 y:(r1)+,y0 ;a1*x, mv x**2, get a2
mac x1,y0,a y:(r1)+,y0 ;a2* x**2, get a0
add y0,a ;add in a0

;
; Step 3 - Divide result by 32.
;
asl #3,a,a
move b1,b0
dec b
nop
move b0,a2 ;new sign = characteristic

asr #6,a,a
rnd a ;round result
endm
I don't understand precisely what's happening in Step 3. Especially this
part commented as 'new sign = characteristic'...

Would someone be so kind to explain me?

Thanks - Ralph