1
0
mirror of https://github.com/cc65/cc65.git synced 2024-11-03 10:07:02 +00:00
cc65/libsrc/runtime/mul.s
uz acd1233bf2 Greatly improved multiplication routine. Optimized the generic 16x16 one and
added special cases for 8x16 and 8x8. The former is directly called by the
compiler as tosmula0 and tosumula0 resp.


git-svn-id: svn://svn.cc65.org/cc65/trunk@4036 b7a2c559-68d2-44c3-8de9-860c34a00d81
2009-08-17 20:48:28 +00:00

70 lines
1.8 KiB
ArmAsm

;
; Ullrich von Bassewitz, 2009-08-17
;
; CC65 runtime: multiplication for ints
;
.export tosumulax, tosmulax
.import mul8x16, mul8x16a ; in mul8.s
.import popsreg
.importzp sreg, tmp1, ptr4
;---------------------------------------------------------------------------
; 16x16 multiplication routine
tosmulax:
tosumulax:
sta ptr4
txa ; High byte zero
beq @L3 ; Do 8x16 multiplication if high byte zero
stx ptr4+1 ; Save right operand
jsr popsreg ; Get left operand
; Do ptr4:ptr4+1 * sreg:sreg+1 --> AX
lda #0
ldx sreg+1 ; Get high byte into register for speed
beq @L4 ; -> we can do 8x16 after swap
sta tmp1
ldy #16 ; Number of bits
lsr ptr4+1
ror ptr4 ; Get first bit into carry
@L0: bcc @L1
clc
adc sreg
pha
txa ; hi byte of left op
adc tmp1
sta tmp1
pla
@L1: ror tmp1
ror a
ror ptr4+1
ror ptr4
dey
bne @L0
lda ptr4 ; Load the result
ldx ptr4+1
rts ; Done
; High byte of rhs is zero, jump to the 8x16 routine instead
@L3: jmp mul8x16
; If the high byte of rhs is zero, swap the operands and use the 8x16
; routine. On entry, A and X are zero
@L4: ldy sreg ; Save right operand (8 bit)
ldx ptr4 ; Copy left 16 bit operand to right
stx sreg
ldx ptr4+1 ; Don't store, this is done later
sty ptr4 ; Copy low 8 bit of right op to left
ldy #8
jmp mul8x16a