section "vars", WRAM0, align[1]
src: ds 2
dst: ds 2
cnt: ds 2

section "code", ROM0
; After https://en.wikipedia.org/wiki/Intel_8008#Example_code
memcpy_8008:
  ld l, low(cnt)  ; ld bc, [cnt]
  ld h, high(cnt) ; (Pino) Bug fixed
  ld c, [hl]
  inc l
  ld b, [hl]
  .loop:
    ld a, c       ; Return if BC is 0
    or b
    ret z
    ld a, c       ; dec bc
    sub 1
    ld c, a
    ld a, b
    sbc 0
    ld b, a
    ld l, low(src)   ; ld de, [src]
    ld h, high(src)  ; (Pino) Bug fixed
    ld e, [hl]
    inc l
    ld d, [hl]
    ld a, c          ; ld hl, de+bc
    add e
    ld l, a
    ld a, b
    adc d
    ld h, a
    ld a, [hl]       ; Read the byte
    ld l, low(dst)   ; ld hl, [dst]
    ld h, high(dst)  ; (Pino) Bug fixed
    ld e, [hl]
    inc l
    ld d, [hl]
    ld l, e
    ld h, d
    ld d, a          ; save the read byte
    ld a, c          ; add hl, bc
    add l
    ld l, a
    ld a, b
    adc h
    ld h, a
    ld [hl], d       ; write the byte
    jp .loop

memcpy_8080:
  ld hl, [cnt]
  ld b, h
  ld c, l
  ld hl, [dst]
  ex de, hl
  ld hl, [src]
  .loop:
    ld a, c
    or b
    ret z
    ld a, [hl]
    inc hl
    ld [de], a
    inc de
    dec bc
    jp .loop