Controller reading code

From NESdev Wiki
Revision as of 01:19, 11 January 2024 by Rainwarrior (talk | contribs) (→‎DPCM Safety using Repeated Reads: finishing note about SMB3, will resolve edit conflict in the next edit)
Jump to navigationJump to search

This page contains example code for reading the NES controller.

See also: Controller reading

Basic Example

This is a tutorial example of the bare minimum needed to read the controller. It will explain the basic principles in detail, but once understood, you may wish to continue to the Standard Read example that follows, as a more complete and ready-to-use code example.

This code describes an efficient method of reading the standard controller using ca65 syntax.

The result byte buttons should be placed in zero page to save a cycle each time through the loop.

; we reserve one byte for storing the data that is read from controller
.zeropage
buttons .res 1

When reading from JOYPAD* what is read might be different from $01/$00 for various reasons. (See Controller reading.) In this code the only concern is bit 0 read from JOYPAD*..

JOYPAD1 = $4016
JOYPAD2 = $4017

This is the end result that will be stored in buttons. 1 if the button was pressed, 0 otherwise.

bit     7         6         5         4         3         2         1         0    
button A B Select Start Up Down Left Right

This subroutine takes 132 cycles to execute but ignores the Famicom expansion controller. Many controller reading subroutines use the X or Y register to count 8 times through the loop. But this one uses a more clever ring counter technique: $01 is loaded into the result first, and once eight bits are shifted in, the 1 bit will be shifted out, terminating the loop.

; At the same time that we strobe bit 0, we initialize the ring counter
; so we're hitting two birds with one stone here
readjoy:
    lda #$01
    ; While the strobe bit is set, buttons will be continuously reloaded.
    ; This means that reading from JOYPAD1 will only return the state of the
    ; first button: button A.
    sta JOYPAD1
    sta buttons
    lsr a        ; now A is 0
    ; By storing 0 into JOYPAD1, the strobe bit is cleared and the reloading stops.
    ; This allows all 8 buttons (newly reloaded) to be read from JOYPAD1.
    sta JOYPAD1
loop:
    lda JOYPAD1
    lsr a	       ; bit 0 -> Carry
    rol buttons  ; Carry -> bit 0; bit 7 -> Carry
    bcc loop
    rts

Continue to the next example for a more complete read routine that handles both controllers and the standard Famicom expansion controllers.

Standard Read for 2 Controllers and Famicom

Adding support for controllers on the Famicom's DA15 expansion port and for player 2's controller is straightforward. Something similar to the following routine is used in most Famicom games. Even though the expansion port is unused on the NES, the unconnected bit will read as 0, so this solution works safely with both Famicom and NES hardware.

.zeropage
buttons: .res 2     ; space for 2 reads

.code
readjoyx2:
    ldx #$00
    jsr readjoyx    ; X=0: read controller 1
    inx
    ; fall through to readjoyx below, X=1: read controller 2

readjoyx:           ; X register = 0 for controller 1, 1 for controller 2
    lda #$01
    sta JOYPAD1
    sta buttons, X
    lsr a
    sta JOYPAD1
loop:
    lda JOYPAD1, X
    and #%00000011  ; ignore bits other than controller
    cmp #$01        ; Set carry if and only if nonzero
    rol buttons, X  ; Carry -> bit 0; but 7 -> Carry
    bcc loop
    rts

If playing DPCM samples, there is an additional reread step to prevent errors ((see below).

Alternative 2 Controllers Read

Alternatively, we could combine both controller reads into 1 loop with a single strobe, though this routine is not safe to use with DPCM samples playing (see below).

.zeropage
buttons: .res 2

.code
readjoy2:
    lda #$01
    sta JOYPAD1
    sta buttons+1 ; player 2's buttons double as a ring counter
    lsr a
    sta JOYPAD1
loop:
    lda JOYPAD1
    and #%00000011
    cmp #$01
    rol buttons+9
    lda JOYPAD2
    and #%00000011
    cmp #$01
    rol buttons+1
    bcc loop
    rts

DPCM Safety using Repeated Reads

If your code is intended to be used with APU DMC playback, this code will need to be altered. The NES occasionally glitches the controller port twice in a row if sample playback is enabled, and games using samples need to work around this. The most common technique, as seen in Super Mario Bros. 3[1] and other games, will read each controller twice, but in the event of a mismatch it will keep re-reading the controller until two results in a row are the same.

readjoyx_safe:
    jsr readjoyx
reread:
    lda buttons, X
    pha
    jsr readjoyx
    pla
    cmp buttons, X
    bne reread
    rts

readjoy2_safe:
    ldx #$00
    jsr readjoyx_safe  ; X=0: safe read controller 1
    inx
    jmp readjoyx_safe  ; X=1: safe read controller 2

Note that the time between the start of one read and the end of the next read must be less than the length of the fastest DMC fetch period (432 cycles). For this reason, it is normal to read controllers one at a time with this method, rather than attempting both at once. (Note: readjoy2 above takes too long to be suitable.) Gimmick! has such a bug resulting from trying to read both at once.

Most often a controller will be read 2 times, and 3 or 4 in the case of a DPCM corruption, or the player pressing a button during the read. With the assistance of tools, a malicious controller input could change the buttons on every read, holding it in this reread loop indefinitely[2], but this is generally not an important edge case to account for.

DPCM Safety using OAM DMA

Because halts for DPCM fetches normally only occur on an put cycles, it is possible to get glitch-free controller reads by timing all $4016 and $4017 reads to fall on get cycles. This is made possible by the behavior of OAM DMA: the first cycle after an OAM DMA is normally guaranteed to be a get cycle.[3] This is a relatively new technique and is not supported by some emulators.[4] In the following example code, the controller1 and controller2 labels must be in zeropage for the timing to work.

    lda #OAM
    sta $4014          ; ------ OAM DMA ------
    ldx #1             ; get put          <- strobe code must take an odd number of cycles total
    stx controller1    ; get put get      <- controller1 and controller2 must be in the zeropage
    stx $4016          ; put get put get
    dex                ; put get
    stx $4016          ; put get put get
read_loop:
    lda $4017          ; put get put GET  <- loop code must take an even number of cycles total
    and #3             ; put get
    cmp #1             ; put get
    rol controller2, x ; put get put get put get (X = 0; waste 1 cycle and 0 bytes for alignment)
    lda $4016          ; put get put GET
    and #3             ; put get
    cmp #1             ; put get
    rol controller1    ; put get put get put
    bcc read_loop      ; get put [get]    <- this branch must not be allowed to cross a page

Note that this example routine only reads two 8-bit controllers and does not take enough time to span more than one DPCM fetch. Routines longer than this must contend with two additional constraints:

  • When DMC DMA is delayed by an odd number of cycles, it takes 3 cycles instead of 4, changing the cycle parity. If extending this function to read more bits, care must be taken so that all CPU write cycles are aligned. Instructions with a single write cycle must align the write to avoid conflict with the DPCM fetch, and double-write instructions like ROL need to align both writes so that the DPCM fetch falls on the first write.[5] If an interrupt can occur during the routine, it must be aligned so the fetch can only fall on the second of the three automatic stack writes.
  • When DMC DMA occurs near the end of OAM DMA, it only steals 1 or 3 cycles, inverting the cycle parity. Every DMC period after that, a misaligned DPCM fetch will occur. Care must be taken to ensure this does not land on a joypad read.

See DMA for detailed information on DMA timing.

Directional Safety

To reject opposing presses (Up+Down and Left+Right), which are possible on a worn Control Pad:

    lda buttons1,x
    and #%00001010    ; Compare Up and Left...
    lsr a
    and buttons1,x    ; to Down and Right
    beq not_updown
        ; Use previous frame's directions
        lda buttons1,x
        eor last_frame_buttons1,x
        and #%11110000
        eor last_frame_buttons1,x
        sta buttons1,x
    not_updown:

To instead reject all diagonal presses, simulating a 4-way joystick:

    lda buttons1,x
    and #%00001111    ; If A & (A - 1) is nonzero, A has more than one bit set
    beq not_diagonal
    sec
    sbc #1
    and buttons1,x
    beq not_diagonal
        ; Use previous frame's directions
        lda buttons1,x
        eor last_frame_buttons1,x
        and #%11110000
        eor last_frame_buttons1,x
        sta buttons1,x
    not_diagonal:

Calculating Presses and Releases

To calculate newly pressed and newly released buttons:

    lda buttons1,x
    eor #%11111111
    and last_frame_buttons1,x
    sta released_buttons1,x
    lda last_frame_buttons1,x
    eor #%11111111
    and buttons1,x
    sta pressed_buttons1,x

Button Flags

It is helpful to define the buttons as a series of bit flags:

BUTTON_A      = 1 << 7
BUTTON_B      = 1 << 6
BUTTON_SELECT = 1 << 5
BUTTON_START  = 1 << 4
BUTTON_UP     = 1 << 3
BUTTON_DOWN   = 1 << 2
BUTTON_LEFT   = 1 << 1
BUTTON_RIGHT  = 1 << 0

And then buttons can be checked as follows:

    lda buttons
    and #BUTTON_A | BUTTON_B
    beq notPressingAorB
    ; Handle presses.
notPressingAorB:

External Examples

  • Forum post: Blargg's DMC-fortified controller read routine
  • Forum post: Rahsennor's OAM-synchronized controller read
  • Forum post: Drag's bitwise DMC-safe controller reading

References

  1. Super Mario Bros. 3 controller reread method
  2. [https://arstechnica.com/gaming/2016/07/how-to-beat-super-mario-bros-3-in-less-than-a-second/ Ars Technica: How to beat Super Mario Bros. 3 in less than a second
  3. Forum post: Rahsennor's OAM-synchronized controller read
  4. Forum post: as of May 2016, Nintendulator and Nestopia do not accurately emulate OAM-synchronized controller reading.
  5. Forum post: demonstration of how ROL instruction affects alignment for OAM DMA synchronized controller reading.