.program hub75_row

; side-set pin 0 is LATCH
; side-set pin 1 is OEn
; OUT pins are row select A-E
;
; Each FIFO record consists of:
; - 5-bit row select (LSBs)
; - Pulse width - 1 (27 MSBs)
;
; Repeatedly select a row, pulse LATCH, and generate a pulse of a certain
; width on OEn.

.side_set 2

.wrap_target
    out pins, 5 [7]    side 0x2 ; Deassert OEn, output row select
    out x, 27   [7]    side 0x3 ; Pulse LATCH, get OEn pulse width
pulse_loop:
    jmp x-- pulse_loop side 0x0 ; Assert OEn for x+1 cycles
.wrap

% c-sdk {
static inline void hub75_row_program_init(PIO pio, uint sm, uint offset, uint row_base_pin, uint n_row_pins, uint latch_base_pin) {
    pio_sm_set_consecutive_pindirs(pio, sm, row_base_pin, n_row_pins, true);
    pio_sm_set_consecutive_pindirs(pio, sm, latch_base_pin, 2, true);
    for (uint i = row_base_pin; i < row_base_pin + n_row_pins; ++i)
        pio_gpio_init(pio, i);
    pio_gpio_init(pio, latch_base_pin);
    pio_gpio_init(pio, latch_base_pin + 1);

    pio_sm_config c = hub75_row_program_get_default_config(offset);
    sm_config_set_out_pins(&c, row_base_pin, n_row_pins);
    sm_config_set_sideset_pins(&c, latch_base_pin);
    sm_config_set_out_shift(&c, true, true, 32);
    pio_sm_init(pio, sm, offset, &c);
    pio_sm_set_enabled(pio, sm, true);
}

static inline void hub75_wait_tx_stall(PIO pio, uint sm) {
    uint32_t txstall_mask = 1u << (PIO_FDEBUG_TXSTALL_LSB + sm);
    pio->fdebug = txstall_mask;
    while (!(pio->fdebug & txstall_mask))
        tight_loop_contents();
}
%}

.program hub75_data_rgb888
.side_set 1

; Each FIFO record consists of a RGB888 pixel. (This is ok for e.g. an RGB565
; source which has been gamma-corrected)
;
; Even pixels are sent on R0, G0, B0 and odd pixels on R1, G1, B1 (typically
; these are for different parts of the screen, NOT for adjacent pixels, so the
; frame buffer must be interleaved before passing to PIO.)
;
; Each pass through, we take bit n, n + 8 and n + 16 from each pixel, for n in
; {0...7}. Therefore the pixels need to be transmitted 8 times (ouch) to build
; up the full 8 bit value for each channel, and perform bit-planed PWM by
; varying pulse widths on the other state machine, in ascending powers of 2.
; This avoids a lot of bit shuffling on the processors, at the cost of DMA
; bandwidth (which we have loads of).

; Might want to close your eyes before you read this
public entry_point:
.wrap_target
public shift0:
    pull             side 0 ; gets patched to `out null, n` if n nonzero (otherwise the PULL is required for fencing)
    in osr, 1        side 0 ; shuffle shuffle shuffle
    out null, 8      side 0
    in osr, 1        side 0
    out null, 8      side 0
    in osr, 1        side 0
    out null, 32     side 0 ; Discard remainder of OSR contents
public shift1:
    pull             side 0 ; gets patched to out null, n if n is nonzero (otherwise PULL required)
    in osr, 1        side 1 ; Note this posedge clocks in the data from the previous iteration
    out null, 8      side 1
    in osr, 1        side 1
    out null, 8      side 1
    in osr, 1        side 1
    out null, 32     side 1
    in null, 26      side 1 ; Note we are just doing this little manoeuvre here to get GPIOs in the order
    mov pins, ::isr  side 1 ; R0, G0, B0, R1, G1, B1. Can go 1 cycle faster if reversed
.wrap
; Note that because the clock edge for pixel n is in the middle of pixel n +
; 1, a dummy pixel at the end is required to clock the last piece of genuine
; data. (Also 1 pixel of garbage is clocked out at the start, but this is
; harmless)

% c-sdk {
static inline void hub75_data_rgb888_program_init(PIO pio, uint sm, uint offset, uint rgb_base_pin, uint clock_pin) {
    pio_sm_set_consecutive_pindirs(pio, sm, rgb_base_pin, 6, true);
    pio_sm_set_consecutive_pindirs(pio, sm, clock_pin, 1, true);
    for (uint i = rgb_base_pin; i < rgb_base_pin + 6; ++i)
        pio_gpio_init(pio, i);
    pio_gpio_init(pio, clock_pin);

    pio_sm_config c = hub75_data_rgb888_program_get_default_config(offset);
    sm_config_set_out_pins(&c, rgb_base_pin, 6);
    sm_config_set_sideset_pins(&c, clock_pin);
    sm_config_set_out_shift(&c, true, true, 24);
    // ISR shift to left. R0 ends up at bit 5. We push it up to MSB and then flip the register.
    sm_config_set_in_shift(&c, false, false, 32);
    sm_config_set_fifo_join(&c, PIO_FIFO_JOIN_TX);
    pio_sm_init(pio, sm, offset, &c);
    pio_sm_exec(pio, sm, offset + hub75_data_rgb888_offset_entry_point);
    pio_sm_set_enabled(pio, sm, true);
}

// Patch a data program at `offset` to preshift pixels by `shamt`
static inline void hub75_data_rgb888_set_shift(PIO pio, uint sm, uint offset, uint shamt) {
    uint16_t instr;
    if (shamt == 0)
        instr = pio_encode_pull(false, true); // blocking PULL
    else
        instr = pio_encode_out(pio_null, shamt);
    pio->instr_mem[offset + hub75_data_rgb888_offset_shift0] = instr;
    pio->instr_mem[offset + hub75_data_rgb888_offset_shift1] = instr;
}
%}