FR0    = $d4 ; float48
PORTB = $d301


EXTENDED_RAM = $4000 ; 16KiB bank on the XE

; lookup table for top byte -> PORTB value for bank-switch
.align 256
bankswitch:
    .repeat 256, i
        .byte ((i & $c0) >> 5) | $c1
    .endrepeat

; 58-77 cycles
; clobbers x, y, dest to dest + 3
.macro imul8xe dest, arg1, arg2
.local done
.local output
.local ptr

    output = dest
    ptr = dest + 2 ; scratch space assumed

    ; bottom 14 bits except the LSB are the per-bank table index
    ; add $4000 for the bank pointer
    lda arg1     ; 3 cyc
    and #$fe     ; 2 cyc
    sta ptr      ; 3 cyc
    lda arg2     ; 3 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
    ldx arg2          ; 3 cyc
    lda bank_switch,x ; 4 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

    ; note: we are not restoring memory to save 6 cycles!
    ; this means those 16kb have to be switched back to base RAM
    ; if we need to use them anywhere else
    ;;; restore memory
    ;;lda #$81     ; 2 cyc - disabled
    ;;sta PORTB    ; 4 cyc - disabled

    ; check that 1 bit we skipped to fit into space
    lda arg1     ; 3 cyc
    and #1       ; 2 cyc
    beq done     ; 2 cyc

    ; add the second param one last time for the skipped bit
    clc          ; 2 cyc
    lda arg2     ; 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:
.endmacro

proc imul8xe_init
    rts
endproc