1
0
mirror of https://github.com/cc65/cc65.git synced 2024-11-02 18:06:48 +00:00
cc65/libsrc/runtime/lshr.s
uz 53dd513176 This commit was generated by cvs2svn to compensate for changes in r2,
which included commits to RCS files with non-trunk default branches.


git-svn-id: svn://svn.cc65.org/cc65/trunk@3 b7a2c559-68d2-44c3-8de9-860c34a00d81
2000-05-28 13:40:48 +00:00

186 lines
3.0 KiB
ArmAsm

;
; Ullrich von Bassewitz, 20.09.1998
;
; CC65 runtime: right shift support for longs
;
.export tosasreax, tosshreax
.import addysp1
.importzp sp, sreg, ptr1, ptr2
; --------------------------------------------------------------------
; signed shift
.proc tosasreax
jsr getlhs ; Get the lhs from the stack
jsr checkovf ; Check for overflow
bcs L6 ; Jump if shift count too large
cpy #0 ; Shift count zero?
beq L5
; We must shift. Shift by multiples of eight if possible
tya
L1: cmp #8
bcc L3
sbc #8
ldx ptr1+1
stx ptr1
ldx ptr2
stx ptr1+1
ldy #0
ldx ptr2+1
stx ptr2
bpl L2
dey ; Get sign
L2: sty ptr2+1
jmp L1
; Shift count is now less than eight. Do a real shift.
L3: tay ; Shift count to Y
lda ptr2+1 ; Get one byte into A for speed
cpy #0
beq L4a ; Jump if done
L4: cmp #$80 ; Get sign bit into C
ror a
ror ptr2
ror ptr1+1
ror ptr1
dey
bne L4
; Put the result in place
L4a: sta sreg+1
lda ptr2
sta sreg
ldx ptr1+1
lda ptr1
L5: rts
; Jump here if shift count overflow
L6: ldx #0
lda ptr2+1 ; Check sign
bpl L7
dex
L7: stx sreg+1
stx sreg
txa
rts
.endproc
; --------------------------------------------------------------------
; unsigned shift
.proc tosshreax
jsr getlhs ; Get the lhs from the stack
jsr checkovf ; Check for overflow
bcs L6 ; Jump if shift count too large
cpy #0 ; Shift count zero?
beq L5
; We must shift. Shift by multiples of eight if possible
tya
L1: cmp #8
bcc L3
sbc #8
ldx ptr1+1
stx ptr1
ldx ptr2
stx ptr1+1
ldx ptr2+1
stx ptr2
ldx #0
stx ptr2+1
beq L1
; Shift count is now less than eight. Do a real shift.
L3: tay ; Shift count to Y
lda ptr2+1 ; Get one byte into A for speed
cpy #0
beq L4a ; Jump if done
L4: lsr a
ror ptr2
ror ptr1+1
ror ptr1
dey
bne L4
; Put the result in place
L4a: sta sreg+1
lda ptr2
sta sreg
ldx ptr1+1
lda ptr1
L5: rts
; Jump here if shift count overflow
L6: lda #0
sta sreg+1
sta sreg
tax
rts
.endproc
; --------------------------------------------------------------------
; Helpers
.proc getlhs ; Get the lhs from stack into ptr1/ptr2
pha
ldy #0
lda (sp),y
sta ptr1
iny
lda (sp),y
sta ptr1+1
iny
lda (sp),y
sta ptr2
iny
lda (sp),y
sta ptr2+1
pla
jmp addysp1
.endproc
.proc checkovf ; Check for shift overflow
tay ; Low byte of shift count into y
txa ; Get byte 2
ora sreg
ora sreg+1 ; Check high 24 bit
bne TooLarge ; Shift count too large
cpy #32
bcc Ok
TooLarge:
sec
Ok: rts
.endproc