chi x*x+y*y-x*y 16bit in 8051

chi x*x + y*y - x*y (16 bit).asm

;inputs
mov dptr,#0
movx a,@dptr
mov r0,a
inc dptr
movx a,@dptr
mov r1,a
inc dptr
movx a,@dptr
mov r2,a
inc dptr
movx a,@dptr
mov r3,a

;x*x
mov 30h,r0
mov 31h,r1
mov 32h,r0
mov 33h,r1
call mul

;move x*x
mov 20h,#0
mov 21h,40h
mov 22h,41h
mov 23h,42h
mov 24h,43h

;y*y
mov 30h,r2
mov 31h,r3
mov 32h,r2
mov 33h,r3
call mul

;add to prev
mov a,24h
add a,43h
mov 24h,a
mov a,23h
addc a,42h
mov 23h,a
mov a,22h
addc a,41h
mov 22h,a
mov a,21h
addc a,40h
mov 21h,a
mov a,#0
addc a,20h
mov 20h,a



;x*y
mov 30h,r0
mov 31h,r1
mov 32h,r2
mov 33h,r3
call mul

;sub
clr c
mov a,24h
subb a,43h
mov 24h,a
mov a,23h
subb a,42h
mov 23h,a
mov a,22h
subb a,41h
mov 22h,a
mov a,21h
subb a,40h
mov 21h,a
mov a,#0
subb a,20h
mov 20h,a


endp:
sjmp endp

mul:
;clear
mov 40h,#0
mov 41h,#0
mov 42h,#0
mov 43h,#0

;last two bytes
mov a,31h
mov b,33h
mul ab
mov 43h,a
mov 42h,b

;first two bytes
mov a,30h
mov b,32h
mul ab
mov 41h,a
mov 40h,b

;x2*y1
mov a,31h
mov b,32h
mul ab
add a,42h
mov 42h,a
mov a,b
addc a,41h
mov 41h,a


;x1*y2
mov a,30h
mov b,33h
mul ab
add a,42h
mov 42h,a
mov a,b
addc a,41h
mov 41h,a
mov a,#0
addc a,40h
mov 40h,a


ret