FR0    = $d4 ; float48
PORTB = $d301


EXTENDED_RAM = $4000 ; 16KiB bank on the XE
bankswitch = ; ???

; input in X/Y (lo/hi)
; output in FR0
; clobbers FR0
; 128 cycles
proc imul8xe
    output = FR0
    ptr = FR0 + 2

    lda #0       ; 2 cyc
    sta ptr      ; 3 cyc
    sta ptr + 1  ; 3 cyc

    ; bottom 14 bits except the LSB are the per-bank table index
    ; add $4000 for the bank pointer
    txa          ; 2 cyc
    and #$fe     ; 2 cyc
    sta ptr      ; 3 cyc
    tya          ; 2 cyc
    and #$3f     ; 2 cyc
    clc          ; 2 cyc
    adc #$40     ; 2 cyc
    sta ptr + 1  ; 3 cyc
    
    ; top 2 bits are the table bank selector
    tya          ; 2 cyc
    and #$c0     ; 2 cyc
    ; shift in extended RAM mode 2x 1 bits
    sec          ; 2 cyc
    ror          ; 2 cyc
    ror          ; 2 cyc
    ; shift in 0 bits
    asr          ; 2 cyc
    asr          ; 2 cyc
    asr          ; 2 cyc

    ; save the second param for later
    phy          ; 3 cyc

    ; disable interrupts
    lda NMIEN    ; 4 cyc
    pha          ; 3 cyc
    lda #0       ; 2 cyc
    sta NMIEN    ; 4 cyc

    ; set the standard top RAM and OS ROM on
    or #$81      ; 2 cyc
    sta PORTB    ; 4 cyc


    ; copy the entry into output
    ldy #0       ; 2 cyc
    lda (ptr),y  ; 5 cyc
    sta output   ; 3 cyc
    iny          ; 2 cyc
    lda (ptr),y  ; 5 cyc
    sta output+1 ; 3 cyc

    ; restore memory
    lda #$81     ; 2 cyc
    sta PORTB    ; 4 cyc

    ; restore interrupts
    pla          ; 3 cyc
    sta NMIEN    ; 4 cyc

    ; check that 1 bit we skipped to fit into space
    txa          ; 2 cyc
    and $#1      ; 2 cyc
    beq done     ; 2 cyc

    ; add the second param one last tie for the skipped bit
    clc          ; 2 cyc
    pla          ; 3 cyc
    adc output   ; 3 cyc
    sta output   ; 3 cyc
    lda #0       ; 2 cyc
    adc output+1 ; 3 cyc
    sta output+1 ; 3 cyc

done:
    pla
    rts          ; 6 cyc
endproc

proc imul8xe_init
    rts
endproc