Programmable IO

The RP2040 has hardware support for standard communication protocols like I2C, SPI and UART. For protocols where there is no hardware support, or where there is a requirement of custom I/O behaviour, Programmable Input Output (PIO) comes into play. Also, some MicroPython applications make use of a technique called bit banging in which pins are rapidly turned on and off to transmit data. This can make the entire process slow as the processor concentrates on bit banging rather than executing other logic. However, PIO allows bit banging to happen in the background while the CPU is executing the main work.

Along with the two central Cortex-M0+ processing cores, the RP2040 has two PIO blocks each of which has four independent state machines. These state machines can transfer data to/from other entities using First-In-First-Out (FIFO) buffers, which allow the state machine and main processor to work independently yet also synchronise their data. Each FIFO has four words (each of 32 bits) which can be linked to the DMA to transfer larger amounts of data.

All PIO instructions follow a common pattern:

<instruction> .side(<side_set_value>) [<delay_value>]

The side-set .side(...) and delay [...] parts are both optional, and if specified allow the instruction to perform more than one operation. This keeps PIO programs small and efficient.

There are nine instructions which perform the following tasks:

  • jmp() transfers control to a different part of the code

  • wait() pauses until a particular action happens

  • in_() shifts the bits from a source (scratch register or set of pins) to the input shift register

  • out() shifts the bits from the output shift register to a destination

  • push() sends data to the RX FIFO

  • pull() receives data from the TX FIFO

  • mov() moves data from a source to a destination

  • irq() sets or clears an IRQ flag

  • set() writes a literal value to a destination

The instruction modifiers are:

  • .side() sets the side-set pins at the start of the instruction

  • [] delays for a certain number of cycles after execution of the instruction

There are also directives:

  • wrap_target() specifies where the program execution will get continued from

  • wrap() specifies the instruction where the control flow of the program will get wrapped from

  • label() sets a label for use with jmp() instructions

  • word() emits a raw 16-bit value which acts as an instruction in the program

An example

Take the example for a simple understanding of how to use the PIO and state machines. Below is the code for reference.

# Example using PIO to blink an LED and raise an IRQ at 1Hz.

import time
from machine import Pin
import rp2

def blink_1hz():
    # Cycles: 1 + 1 + 6 + 32 * (30 + 1) = 1000
    set(pins, 1)
    set(x, 31)                  [5]
    nop()                       [29]
    jmp(x_dec, "delay_high")

    # Cycles: 1 + 7 + 32 * (30 + 1) = 1000
    set(pins, 0)
    set(x, 31)                  [6]
    nop()                       [29]
    jmp(x_dec, "delay_low")

# Create the StateMachine with the blink_1hz program, outputting on Pin(25).
sm = rp2.StateMachine(0, blink_1hz, freq=2000, set_base=Pin(25))

# Set the IRQ handler to print the millisecond timestamp.
sm.irq(lambda p: print(time.ticks_ms()))

# Start the StateMachine.

This creates an instance of class rp2.StateMachine which runs the blink_1hz program at 2000Hz, and connects to pin 25. The blink_1hz program uses the PIO to blink an LED connected to this pin at 1Hz, and also raises an IRQ as the LED turns on. This IRQ then calls the lambda function which prints out a millisecond timestamp.

The blink_1hz program is a PIO assembler routine. It connects to a single pin which is configured as an output and starts out low. The instructions do the following:

  • irq(rel(0)) raises the IRQ associated with the state machine.

  • The LED is turned on via the set(pins, 1) instruction.

  • The value 31 is put into register X, and then there is a delay for 5 more cycles, specified by the [5].

  • The nop() [29] instruction waits for 30 cycles.

  • The jmp(x_dec, "delay_high") will keep looping to the delay_high label as long as the register X is non-zero, and will also post-decrement X. Since X starts with the value 31 this jump will happen 31 times, so the nop() [29] runs 32 times in total (note there is also one instruction cycle taken by the jmp for each of these 32 loops).

  • set(pins, 0) will turn the LED off by setting pin 25 low.

  • Another 32 loops of nop() [29] and jmp(...) will execute.

  • Because wrap_target() and wrap() are not specified, their default will be used and execution of the program will wrap around from the bottom to the top. This wrapping does not cost any execution cycles.

The entire routine takes exactly 2000 cycles of the state machine. Setting the frequency of the state machine to 2000Hz makes the LED blink at 1Hz.