magrob/firmware/base/pio/quadrature.pio

;
; Copyright (c) 2023 Raspberry Pi (Trading) Ltd.
;
; SPDX-License-Identifier: BSD-3-Clause
;
; quadrature_encoder_substep: reads a quadrature encoder with no CPU
; intervention and provides the current position on request.
;
; the "substep" version uses not only the step counts, but also the timing of 
; the steps to compute the current speed. See README.md for details


.program quadrature_encoder_substep

.origin 0

; the PIO code counts steps like the standard quadrature encoder code, but also
; keeps track of the time passed since the last transition. That allows the C
; code to build a good estimate of a fractional step position based on the
; latest speed and time passed
;
; since it needs to push two values, it only pushes new data when the FIFO has
; enough space to hold both values. Otherwise it could either stall or go out
; of sync
;
; because we need to count the time passed, all loops must take the same number
; of cycles and there are delays added to the fastest branches to make sure it
; always takes 13 cycles per loop (e.g., sysclk 133MHz, max step rate = ~10.2
; Msteps/sec)

	; push the step count and transition clock count to the RX FIFO (using
	; auto push). This is reached by the "MOV PC, ~STATUS" instruction when
	; status is all 1 (meaning fifo has space for this push). It also may
	; execute once at program start, but that has little effect
	IN X, 32
	IN Y, 32

update_state:
	; build the state by using 2 bits from the negated previous state of the
	; pins and the new 2 bit state of the pins
	OUT ISR, 2
	IN PINS, 2
	MOV OSR, ~ISR
	; use the jump table to update the step count accordingly
	MOV PC, OSR

decrement:
	; decrement the step count
	JMP Y--, decrement_cont
decrement_cont:
	; when decrementing, X is set to 2^31, when incrementing it is set to
	; zero. That way the C code can infer in which direction the last
	; transition was taken and how long ago
	SET X, 1
	MOV X, ::X
	; after incrementing or decrementing, continue to "check_fifo"
check_fifo:
.wrap_target
	; on each iteration we decrement X to count the number of loops since
	; the last transition
	JMP X--, check_fifo_cont
check_fifo_cont:
	; push data or continue, depending on the state of the fifo
	MOV PC, ~STATUS

increment:
	; the PIO does not have a increment instruction, so to do that we do a
	; negate, decrement, negate sequence
	MOV Y, ~Y
	JMP Y--, increment_cont
increment_cont:
	MOV Y, ~Y
	; reset X to zero when incrementing
	SET X, 0
	; wrap above to check the fifo state
	.wrap

invalid:
	; this is just a placeholder to document what the code does on invalid
	; transitions, where the two phases change at the same time. We don't do
	; anything special, but the encoder should note generate these invalid
	; transitions anyway
	JMP update_state

	; this jump table starts at address 16 and is accessed by the
	; "MOV PC, OSR" instruction above, that loads the PC with the state on
	; the lower 4 bits and the 5th bit on. The delays here extend the faster
	; branches to take the same time as the slower branches
	JMP invalid
	JMP increment		[0]
	JMP decrement		[1]
	JMP check_fifo		[4]

	JMP decrement		[1]
	JMP invalid
	JMP check_fifo		[4]
	JMP increment		[0]

	JMP increment		[0]
	JMP check_fifo		[4]
	JMP invalid
	JMP decrement		[1]

	JMP check_fifo		[4]
	JMP decrement		[1]
	JMP increment		[0]
	; this instruction should be usually reached by the "MOV PC, ~STATUS"
	; instruction above when the status is zero, which means that the fifo
	; has data and we don't want to push more data. This can also be reached
	; on an invalid state transition, which should not happen. Even if it 
	; happens, it should be a transient state and the only side effect is
	; that we'll call update_state twice in a row
	JMP update_state	[1]





% c-sdk {

#include "hardware/clocks.h"
#include "hardware/timer.h"
#include "hardware/gpio.h"
#include "hardware/sync.h"


// "substep" version low-level interface
//
// note: user code should use the high level functions in quadrature_encoder.c
// and not call these directly

// initialize the PIO state and the substep_state_t structure that keeps track
// of the encoder state
static inline void quadrature_encoder_substep_program_init(PIO pio, uint sm, uint pin_A)
{
	uint pin_state, position, ints;
	pio_gpio_init(pio, pin_A);
	pio_gpio_init(pio, pin_A + 1);

	pio_sm_set_consecutive_pindirs(pio, sm, pin_A, 2, false);
	gpio_pull_up(pin_A);
	gpio_pull_up(pin_A + 1);

	pio_sm_config c = quadrature_encoder_substep_program_get_default_config(0);
	sm_config_set_in_pins(&c, pin_A); // for WAIT, IN
	// shift to left, auto-push at 32 bits
	sm_config_set_in_shift(&c, false, true, 32);
	sm_config_set_out_shift(&c, true, false, 32);
	// don't join FIFO's
	sm_config_set_fifo_join(&c, PIO_FIFO_JOIN_NONE);

	// always run at sysclk, to have the maximum possible time resolution
	sm_config_set_clkdiv(&c, 1.0);

	pio_sm_init(pio, sm, 0, &c);

	// set up status to be rx_fifo < 1
	pio->sm[sm].execctrl = ((pio->sm[sm].execctrl & 0xFFFFFF80) | 0x12);

	// init the state machine according to the current phase. Since we are
	// setting the state running PIO instructions from C state, the encoder may
	// step during this initialization. This should not be a problem though,
	// because as long as it is just one step, the state machine will update
	// correctly when it starts. We disable interrupts anyway, to be safe
	ints = save_and_disable_interrupts();

	pin_state = (gpio_get_all() >> pin_A) & 3;

	// to setup the state machine, we need to set the lower 2 bits of OSR to be
	// the negated pin state
	pio_sm_exec(pio, sm, pio_encode_set(pio_y, ~pin_state));
	pio_sm_exec(pio, sm, pio_encode_mov(pio_osr, pio_y));

	// also set the Y (current step) so that the lower 2 bits of Y have a 1:1
	// mapping to the current phase (input pin state). That simplifies the code
	// to compensate for differences in encoder phase sizes:
	switch (pin_state) {
		case 0: position = 0; break;
		case 1: position = 3; break;
		case 2: position = 1; break;
		case 3: position = 2; break;
	} 
	pio_sm_exec(pio, sm, pio_encode_set(pio_y, position));

	pio_sm_set_enabled(pio, sm, true);
	
	restore_interrupts(ints);
}

static inline void quadrature_encoder_substep_get_counts(PIO pio, uint sm, uint *step, int *cycles, uint *us)
{
	int i, pairs;
	uint ints;
	
	pairs = pio_sm_get_rx_fifo_level(pio, sm) >> 1;

	// read all data with interrupts disabled, so that there can not be a
	// big time gap between reading the PIO data and the current us
	ints = save_and_disable_interrupts();
	for (i = 0; i < pairs + 1; i++) {
		*cycles = pio_sm_get_blocking(pio, sm);
		*step = pio_sm_get_blocking(pio, sm);
	}
	*us = time_us_32();
	restore_interrupts(ints);
}

%}