mirror of https://github.com/nmlgc/ReC98.git
63 lines
1.2 KiB
NASM
63 lines
1.2 KiB
NASM
public SCORE_ADD
|
|
score_add proc far
|
|
|
|
@@pid = byte ptr 6
|
|
@@points = word ptr 8
|
|
@@bcd_p equ bx
|
|
@@po10_p equ si
|
|
@@score_p equ si
|
|
|
|
push bp
|
|
mov bp, sp
|
|
push si
|
|
mov @@bcd_p, offset _temp_lebcd
|
|
|
|
; Since the delta can have at most 5 digits, we only have to work on the
|
|
; range from _temp_lebcd[4] (highest) to _temp_lebcd[0] (lowest).
|
|
mov word ptr [@@bcd_p+6], 0
|
|
mov byte ptr [@@bcd_p+5], 0
|
|
|
|
mov @@po10_p, offset _FIVE_DIGIT_POWERS_OF_10
|
|
; 4 divisions, the units place doesn't need a separate one.
|
|
add @@bcd_p, 4
|
|
mov cx, [bp+@@points]
|
|
|
|
@@points_to_bcd:
|
|
mov ax, cx
|
|
xor dx, dx
|
|
div word ptr [@@po10_p]
|
|
mov cx, dx
|
|
mov [@@bcd_p], al
|
|
dec @@bcd_p
|
|
add si, word
|
|
cmp word ptr [@@po10_p], 1
|
|
ja short @@points_to_bcd
|
|
mov [@@bcd_p], cl
|
|
mov @@score_p, offset _score_lebcd
|
|
mov ch, [bp+@@pid]
|
|
or ch, ch
|
|
jz short @@p1
|
|
add @@score_p, SCORE_DIGITS
|
|
|
|
@@p1:
|
|
; The last digit doesn't need special BCD treatment
|
|
mov cx, SCORE_DIGITS - 1
|
|
xor ax, ax
|
|
|
|
@@add_next_digit_to_score:
|
|
mov al, [@@bcd_p]
|
|
add al, [@@score_p]
|
|
aaa
|
|
mov [@@score_p], al
|
|
inc @@bcd_p
|
|
inc @@score_p
|
|
add [@@score_p], ah
|
|
mov ah, 0
|
|
loop @@add_next_digit_to_score
|
|
mov al, [@@bcd_p]
|
|
add [@@score_p], al
|
|
pop si
|
|
pop bp
|
|
retf 4
|
|
score_add endp
|