ReC98/libs/BorlandC/f_lxmul.asm

63 lines
2.0 KiB
NASM

; *Not* the original file, but an edit to turn it into an includable slice.
; Changes include:
; * removal of RULES.ASI to eliminate redundancy
; * removal of the 'CODE' segment declaration (for obvious reasons)
; * the @@ prefix on any local labels
;[]-----------------------------------------------------------------[]
;| F_LXMUL.ASM -- long multiply routine |
;[]-----------------------------------------------------------------[]
;
; C/C++ Run Time Library - Version 5.0
;
; Copyright (c) 1987, 1992 by Borland International
; All Rights Reserved.
;
; calls to this routine are generated by the compiler to perform
; long multiplications.
; There is no check for overflow. Consequently, the same routine
; is used for both signed and unsigned long multiplies.
;
; in:
; (dx:ax) - 32bit arg1
; (cx:bx) - 32bit arg2
; out:
; (dx:ax) - 32bit product
;
; reg use: bx,cx destroyed, all others preserved or contain result.
;
; hi(result) := lo(hi(arg1) * lo(arg2)) +
; lo(hi(arg2) * lo(arg1)) +
; hi(lo(arg1) * lo(arg2))
; lo(result) := lo(lo(arg1) * lo(arg2))
;
;
public LXMUL@
public F_LXMUL@
LXMUL@ PROC FAR
F_LXMUL@:
push si
xchg si,ax ; save lo1
xchg ax,dx
test ax,ax ; skip mul if hi1==0
jz @@nohi1
mul bx ; hi1 * lo2
@@nohi1: ; if we jumped here, ax==0 so the following swap works
jcxz @@nohi2 ; skip mul if hi2==0
xchg cx, ax ; result <-> hi2
mul si ; lo1 * hi2
add ax, cx ; ax = hi1*lo2 + hi2*lo1
@@nohi2:
xchg ax,si
mul bx ; lo1 * lo2
add dx,si ; hi order result += partials
pop si
ret
LXMUL@ ENDP