A Swift library for hardware projects on Linux/ARM boards with support for GPIOs/SPI/I2C/PWM/UART/1Wire.
A Swift library for hardware projects on Linux/ARM boards with support for GPIOs/SPI/I2C/PWM/UART/1Wire.
This library provides an easy way to interact with external sensors and devices using the digital GPIOs, SPI/I2C interfaces, 1-Wire buses, PWM signals and serial ports that boards like the Raspberry Pi provide, on Linux using Swift.
Like Android Things or similar libraries in Python, SwiftyGPIO provides the basic functionalities you’ll need to control different devices: sensors, displays, input devices like joypads, RGB led strips and matrices.
You’ll be able to configure port attributes and read/write the current GPIOs value, use the SPI interfaces (via hardware if your board provides them or using software big-banging SPI), comunicate over a bus with I2C, generate a PWM to drive external displays, servos, leds and more complex sensors, interact with devices that expose UART serial connections using AT commands or custom protocols, and finally connect to 1-Wire devices.
While you’ll still be able to develop your project with Xcode or another IDE, the library is built to run exclusively on Linux ARM Boards (RaspberryPis, BeagleBones, ODROIDs, OrangePis, etc…).
Examples of device libraries and complete projects built using SwiftyGPIO that you can use as inspiration for your own DIY hardware projects are listed below, have fun!
The following boards are supported and have been tested with recent releases of Swift:
But basically everything that has an ARMv7/8+Ubuntu/Debian/Raspbian or an ARMv6+Raspbian/Debian should work if you can run Swift on it.
Please keep in mind that Swift on ARM is a completely community-driven effort, and that there are a multitude of possible board+OS configurations, don’t expect that everything will work right away on every configuration even if most of the times it does, especially if you are the first to try a new configuration or board.
To use this library, you’ll need a Linux ARM(ARMv7/8 or ARMv6) board with Swift 3.x/4.x/5.x.
If you have a RaspberryPi (A,B,A+,B+,Zero,ZeroW,2,3,4) with Ubuntu or Raspbian, get Swift 5.x from here or follow the instruction from buildSwiftOnARM to build it yourself in a few hours.
I always recommend to try one of the latest binaries available (either Ubuntu or Raspbian) before putting in the time to compile it yourself, those binaries could(and do most of the times) also work on other Debian-bases distibutions and on different boards.
An alternative way to get these Swift binaries on your Raspberry Pi is through the Swift on Balena project that provides well organized IoT focused Docker images.
You can also setup a cross-compiling toolchain and build ARM binaries (Ubuntu/Raspbian) from a Mac, thanks again to the work of Helge Heß (and Johannes Weiß for implementing it in SPM), read more about that here.
To start your project add SwiftyGPIO as a dependency in your Package.swift
:
// swift-tools-version:4.0
import PackageDescription
let package = Package(
name: "light",
dependencies: [
.package(url: "https://github.com/uraimo/SwiftyGPIO.git", from: "1.0.0")
]
)
And then build with swift build
.
The compiler will create an executable under .build/debug/MyProject
.
IMPORTANT: Like every library using GPIOs/SPI/I2C/etc…, if your OS does not come with a predefined user group to access these functionalities, you’ll need to run your application with root privileges using sudo
. If you are using a RaspberryPi with a Raspbian or a recent Ubuntu (from 16.04 Xenial onward) implementing /dev/gpiomem, sudo will be not required to use basic GPIOs, just launch your application calling the executable built by the compiler.
On misconfigured systems, features like the listeners may require root privileges too and advanced features like PWM sadly always require root privileges.
Alternatively, a specific user group for gpio access can be configured manually as shown here or in this answer on stackoverflow.
After following those instruction, remember to add your user (e.g. pi) to the gpio group with sudo usermod -aG gpio pi
and to reboot so that the changes you made are applied.
If you prefer starting with a real project instead of just reading documentation, you’ll find some ready to run examples under Examples/
and more than a few tutorials, full projects and videos available online:
Currently, SwiftyGPIO expose GPIOs, SPIs(if not available a bit-banging VirtualSPI can be created), I2Cs, PWMs, 1-Wire and UART ports, let’s see how to use them.
Let’s suppose we are using a Raspberry 3 board and have a led connected between the GPIO pin P2 (possibly with a resistance of 1K Ohm or so in between) and GND and we want to turn it on.
Note that SwiftyGPIO uses the raw Broadcom numbering scheme (described here) to assign a number to each pin.
First, we need to retrieve the list of GPIOs available on the board and get a reference to the one we want to modify:
import SwiftyGPIO
let gpios = SwiftyGPIO.GPIOs(for:.RaspberryPi3)
var gp = gpios[.P2]!
The following are the possible values for the predefined boards:
The map returned by GPIOs(for:)
contains all the GPIOs of a specific board as described by these diagrams.
Alternatively, if our board is not supported, each single GPIO object can be instantiated manually, using its SysFS GPIO Id:
var gp = GPIO(name: "P2",id: 2) // User defined name and GPIO Id
The next step is configuring the port direction, that can be either GPIODirection.IN
or GPIODirection.OUT
, in this case we’ll choose .OUT:
gp.direction = .OUT
Then we’ll change the pin value to the HIGH value “1”:
gp.value = 1
That’s it, the led will turn on.
Now, suppose we have a switch or a button connected to P2 instead, to read the value coming in the P2 port, the direction must be configured as .IN
and the value can be read from the value
property:
gp.direction = .IN
let current = gp.value
Some boards like the RaspberryPi allow to enable a pull up/down resistance on some of the GPIO pins to connect a pin to 3.3V (.up), 0V (.down) or leave it floating (.neither) by default when external devices are disconnected, to enable it just set the pull
property:
gp.direction = .IN
gp.pull = .up
The pull state can only be set and not read back.
The other properties available on the GPIO object (edge,active low) refer to the additional attributes of the GPIO that can be configured but you will not need them most of the times. For a detailed description refer to the kernel sysfs documentation.
The GPIO object also supports the execution of closures when the value of the pin changes. Closures can be added with the methods onRaising
(the pin value changed from 0 to 1), onFalling
(the value changed from 1 to 0) and onChange
(the value simply changed from the previous one):
let gpios = SwiftyGPIO.GPIOs(for:.RaspberryPi3)
var gp = gpios[.P2]!
gp.onRaising{
gpio in
print("Transition to 1, current value:" + String(gpio.value))
}
gp.onFalling{
gpio in
print("Transition to 0, current value:" + String(gpio.value))
}
gp.onChange{
gpio in
gpio.clearListeners()
print("The value changed, current value:" + String(gpio.value))
}
The closure receives as its only parameter a reference to the GPIO object that has been updated so that you don’t need to use the external variable.
Calling clearListeners()
removes all the closures listening for changes and disables the changes handler.
While GPIOs are checked for updates, the direction
of the pin cannot be changed (and configured as .IN
), but once the listeners have been cleared, either inside the closure or somewhere else, you are free to modify it.
Setting the bounceTime
property will enable software debounce, that will limit the number of transitions notified to the closure allowing only one event in the specified time interval in seconds.
The following example allows only one transition every 500ms:
let gpios = SwiftyGPIO.GPIOs(for:.RaspberryPi3)
var gp = gpios[.P2]!
gp.bounceTime = 0.5
gp.onRaising{
gpio in
print("Transition to 1, current value:" + String(gpio.value))
}
This functionality is extremely useful when using switches, that tend to generate multiple value spikes when the switch is pressed due to the mechanical characteristics of the compoment.
If your board has a SPI connection and SwiftyGPIO has it among its presets, a list of the available SPI channels can be obtained by calling hardwareSPIs(for:)
with one of the predefined boards.
On RaspberryPi and other boards the hardware SPI SysFS interface is not enabled by default, check out the setup guide on wiki to enable it if needed using raspi-config
.
Let’s see some examples using a RaspberryPi 3 that has two bidirectional SPIs, managed by SwiftyGPIO as two SPIObjects:
let spis = SwiftyGPIO.hardwareSPIs(for:.RaspberryPi3)!
var spi = spis[0]
The interface is composed by 3 wire: a clock line (SCLK), an input line (MISO) and an output line (MOSI). One or more CS pins (with inverse logic) are available to enable or disable slave devices.
Alternatively, we can create a software SPI using four GPIOs, one that will serve as clock pin (SCLK), one as chip-select (CS or CE) and the other two will be used to send and receive the actual data (MOSI and MISO). This kind of bit-banging SPI is slower than the hardware one, so, the recommended approach is to use hardware SPIs when available.
To create a software SPI, just retrieve two pins and create a VirtualSPI
object:
let gpios = SwiftyGPIO.GPIOs(for:.RaspberryPi3)
var cs = gpios[.P27]!
var mosi = gpios[.P22]!
var miso = gpios[.P4]!
var clk = gpios[.P17]!
var spi = VirtualSPI(mosiGPIO: mosi, misoGPIO: miso, clockGPIO: clk, csGPIO: cs)
Both objects implement the same SPIObject
protocol and so provide the same methods.
To distinguish between hardware and software SPIObjects, use the isHardware
property.
To send one or more byte over a SPI, use the sendData
method.
In its simplest form it just needs an array of UInt8 as parameter:
spi?.sendData([UInt(42)], frequencyHz: 500_000)
The frequency at which the data will be sent can be specified if needed (alternatively the default will be used, that is 500khz for hardware SPIs and the best available speed for virtual SPIs).
Since the interface performs only full duplex transmissions, to read some data from the SPI you’ll need to write the same amount of bits. For most devices you’ll use this means that you’ll need to send some dummy data depending on the protocol used by your device. Check the device reference for more information.
Let’s see a simple example, that reads 32 bytes from a device sending just 32 empty bytes:
let data = [ UInt8 ](repeating: 0, count: 32)
let res = spi?.sendDataAndRead(data)
The res
array will contain the raw data received from the device. Again, what to send and how the received data should be interpreted depends from the device or IC you are using, always read the reference manual.
The I2C interface can be used to communicate using the SMBus protocol on a I2C bus, reading or writing registers on devices identified by a numerical address. This interface needs just two wires (clock and data) and unlike SPI, it does not need a dedicated chip select/enable wire to select which device will receive the signal being sent, since the address of the destination of the protocol’s messages is contained in the message itself, quite an improvement.
To obtain a reference to the I2CInterface
object, call the hardwareI2Cs(for:)
utility method of the SwiftyGPIO class:
let i2cs = SwiftyGPIO.hardwareI2Cs(for:.RaspberryPi3)!
let i2c = i2cs[1]
On Raspberry Pi and other boards this interface could not enabled by default, always verify its state checking the setup guide on the wiki to enable it if needed using raspi-config
.
This object provide methods to read and write registers of different sizes and to verify that a device at a certain address is reachable or to enable a CRC on the protocol’s messages:
func isReachable(_ address: Int) -> Bool
func setPEC(_ address: Int, enabled: Bool)
You should choose the read method to use depending on whatever of not your device supports multiple registers (command
in SMBus parlance) and depending of the size of the register you are going to read from:
func readByte(_ address: Int) -> UInt8
func readByte(_ address: Int, command: UInt8) -> UInt8
func readWord(_ address: Int, command: UInt8) -> UInt16
func readData(_ address: Int, command: UInt8) -> [UInt8]
func readI2CData(_ address: Int, command: UInt8) -> [UInt8]
Reading and writing data blocks supports two modes, a standard SMBus mode (readData
and writeData
) that prepends the length of the block before the actual data, and an old style I2C mode (readI2CData
and writeI2CData
) that just send the data without additional metadata. Depending on the device, only one of the two modes will be supported.
Let’s suppose that we want to read the seconds register (id 0) from a DS1307 RTC clock, that has an I2C address of 0x68:
print(i2c.readByte(0x68, command: 0)) //Prints the value of the 8bit register
You should choose the same way one of the write functions available, just note that writeQuick
is used to perform quick commands and does not perform a normal write. SMBus’s quick commands are usually used to turn on/off devices or perform similar tasks that don’t require additional parameters.
func writeQuick(_ address: Int)
func writeByte(_ address: Int, value: UInt8)
func writeByte(_ address: Int, command: UInt8, value: UInt8)
func writeWord(_ address: Int, command: UInt8, value: UInt16)
func writeData(_ address: Int, command: UInt8, values: [UInt8])
func writeI2CData(_ address: Int, command: UInt8, values: [UInt8])
While using the I2C functionality doesn’t require additional software to function, the tools contained in i2c-tools
are useful to perform I2C transactions manually to verify that everything is working correctly.
For example, I recommend to always check if your device has been connected correctly running i2cdetect -y 1
. More information on I2C, and configuration instruction for the Raspberry Pi, are available on Sparkfun.
The Example/
directory contains a Swift implementation of i2cdetect and could be a good place to start experimenting.
The docs/
directory contains instead a simple guide to debug communication issues with I2C devices.
PWM output signals can be used to drive servo motors, RGB leds and other devices, or more in general, to approximate analog output values (e.g. generate values as if they where between 0V and 3.3V) when you only have digital GPIO ports.
If your board has PWM ports and is supported (at the moment only RaspberryPi boards), retrieve the available PWMOutput
objects with the hardwarePWMs
factory method:
let pwms = SwiftyGPIO.hardwarePWMs(for:.RaspberryPi3)!
let pwm = (pwms[0]?[.P18])!
This method returns all the ports that support the PWM function, grouped by the PWM channel that controls them.
You’ll be able to use only one port per channel and considering that the Raspberries have two channels, you’ll be able to use two PWM outputs at the same time, for example GPIO12 and GPIO13 or GPIO18 and GPIO19.
Once you’ve retrieved the PWMOutput
for the port you plan to use you need to initialize it to select the PWM function. On this kind of boards, each port can have more than one function (simple GPIO, SPI, PWM, etc…) and you can choose the function you want configuring dedicated registers.
pwm.initPWM()
To start the PWM signal call startPWM
providing the period in nanoseconds (if you have the frequency convert it with 1/frequency) and the duty cycle as a percentage:
print("PWM from GPIO18 with 500ns period and 50% duty cycle")
pwm.startPWM(period: 500, duty: 50)
Once you call this method, the PWM subsystem of the ARM SoC will start generating the signal, you don’t need to do anything else and your program will continue to execute, you could insert a sleep(seconds)
here if you just want to wait.
And when you want to stop the PWM signal call the stopPWM()
method:
pwm.stopPWM()
If you want to change the signal being generated, you don’t need to stop the previous one, just call startPWM
with different parameters.
This feature uses the M/S algorithm and has been tested with signals with a period in a range from 300ns to 200ms, generating a signal outside of this range could lead to excessive jitter that could not be acceptable for some applications. If you need to generate a signal near to the extremes of that range and have an oscilloscope at hand, always verify if the resulting signal is good enough for what you need.
This functionality leverages the PWM to generate digital signals based on two patterns representing a 0 or a 1 value through a variation of the duty cycle. Let’s look at a practical example to better understand the use case and how to use this signal generator:
Let’s consider for example the WS2812/NeoPixel (see the dedicated library), a led with integrated driver used in many led strips.
This led is activated with a signal between 400Khz and 800Khz containing a series of encoded 3 byte values representing respectively the Green,Blue and Red color components, one for each led. Each bit of the color component byte will have to be encoded this way:
And once the whole sequence of colors for your strip of leds has been sent, you’ll need to keep the voltage at 0 for 50us, before you’ll be able to transmit a new sequence. The bytes sent will configure the leds of the strip starting from the last one, going backwards to the first one.
This diagram from the official documentation gives you a better idea of what those signals look like, based on the T0H,T0L,T1H,T1L defined earlier:
You could think to just send this signal based on those 0 and 1 pattern changing the values of a GPIO, but it’s actually impossible for an ARM board to keep up with the rate required by devices like the WS2812 leds and trying to generate these signals in software introduces significant jitter too.
Once the period of the pattern is lower than 100us or so you need another way to send these signals.
And this is the problem that the pattern-based signal generator solves, leveraging PWM-capable output pins.
You’ll find a complete example under Examples/PWMPattern
, but let’s describe each one of the steps needed to use this feature.
In this brief guide I’m using an 8x8 led matrix with 64 WS2812 leds (these matrices are usually marketed as NeoPixel matrix, Nulsom Rainbow matrix, etc… and you can find one of these in some Pimoroni products like the UnicornHat).
First of all let’s retrieve a PWMOutput
object and then initialize it:
let pwms = SwiftyGPIO.hardwarePWMs(for:.RaspberryPi3)!
let pwm = (pwms[0]?[.P18])!
// Initialize PWM
pwm.initPWM()
We’ll then configure the signal generator specifying the frequency we need (800KHz for a 1250ns pattern period), the number of leds in the sequence (I’m using and 8x8 led matrix here), and the duration of the reset time (55us). We’ll call the initPWMPattern
to configure these parameters. We specify the duty cycle (percentage of the period at which the pattern should have a high value) for the 0 and 1 values.
let NUM_ELEMENTS = 64
let WS2812_FREQ = 800000 // 800Khz
let WS2812_RESETDELAY = 55 // 55us reset
pwm.initPWMPattern(bytes: NUM_ELEMENTS*3,
at: WS2812_FREQ,
with: WS2812_RESETDELAY,
dutyzero: 33, dutyone: 66)
Once this is done, we can start sending data, this time we are using a function that sets the colors and another function that turn them in a series of UInt8
in the GBR
format:
func toByteStream(_ values: [UInt32]) -> [UInt8]{
var byteStream = [UInt8]()
for led in values {
// Add as GRB, converted from RGB+0x00
byteStream.append(UInt8((led >> UInt32(16)) & 0xff))
byteStream.append(UInt8((led >> UInt32(24)) & 0xff))
byteStream.append(UInt8((led >> UInt32(8)) & 0xff))
}
return byteStream
}
var initial = [UInt32](repeating:0x50000000, count:NUM_ELEMENTS)
var byteStream: [UInt8] = toByteStream(initial)
pwm.sendDataWithPattern(values: byteStream)
The method sendDataWithPatter
will use the sequence of UInt8
to produce a signal composed by the patterns described above.
We can then wait until the signal is completely sent and then perform the necessary final cleanup:
// Wait for the transmission to end
pwm.waitOnSendData()
// Clean up once you are done with the generator
pwm.cleanupPattern()
At this point you could configure a different signal calling again initPWMPattern
if you want to.
If your board support the UART serial ports feature (disable the login on serial with raspi-config
for RaspberryPi boards), you can retrieve the list of available UARTInterface
with SwiftyGPIO.UARTs(for:)
:
let uarts = SwiftyGPIO.UARTs(for:.RaspberryPi3)!
var uart = uarts[0]
On Raspberry Pi and other boards this interface could not enabled by default, always verify its state checking the setup guide on the wiki to enable it if needed using raspi-config
.
Before we can start trasmitting data, you need to configure the serial port, specifying: the speed (from 9600bps to 115200bps), the character size (6,7 or 8 bits per character), the number of stop bits (1 or 2) and the parity of your signal (no parity, odd or even). Software and hardware flow control are both disabled when using this library.
uart.configureInterface(speed: .S9600, bitsPerChar: .Eight, stopBits: .One, parity: .None)
Once the port is configured you can start reading or writing strings of sequence of UInt8
with one of the specific methods of UARTInterface
:
func readString() -> String
func readData() -> [CChar]
func writeString(_ value: String)
func writeData(_ values: [CChar])
func hasAvailableData() throws -> Bool
func readLine() -> String
A method to know if there is available data on the UART serial port and a specific method that reads lines of text (\n
is used as line terminator, the serial read is still non-canonical) are also provided.
If your board provides a 1-Wire port (right now only RaspberryPi boards), you can retrieve the list of available OneWireInterface
with SwiftyGPIO.hardware1Wires(for:)
:
let onewires = SwiftyGPIO.hardware1Wires(for:.RaspberryPi3)!
var onewire = onewires[0]
To retrieve the string identifiers associated to the devices connected to the 1-Wire bus, call getSlaves()
.
The data provided by these devices can then be retrieved via readData(slaveId:)
using one of the identifiers obtained at the previous step.
The data coming from the device is returned by the Linux driver as a series of lines, most of which will be just protocol data you should just ignore. Check out the reference of your sensor to know how to interpret the formatted information.
Examples for different boards and functionalities are available in the Examples directory, you can just start from there modifying one of those.
The following example, built to run on the C.H.I.P. board, shows the current value of all the attributes of a single GPIO port, changes direction and value and then shows again a recap of the attributes:
let gpios = SwiftyGPIO.GPIOs(for:.CHIP)
var gp0 = gpios[.P0]!
print("Current Status")
print("Direction: "+gp0.direction.rawValue)
print("Edge: "+gp0.edge.rawValue)
print("Active Low: "+String(gp0.activeLow))
print("Value: "+String(gp0.value))
gp0.direction = .OUT
gp0.value = 1
print("New Status")
print("Direction: "+gp0.direction.rawValue)
print("Edge: "+gp0.edge.rawValue)
print("Active Low: "+String(gp0.activeLow))
print("Value: "+String(gp0.value))
This second example makes a led blink with a frequency of 150ms:
import Glibc
let gpios = SwiftyGPIO.GPIOs(for:.CHIP)
var gp0 = gpios[.P0]!
gp0.direction = .OUT
repeat{
gp0.value = (gp0.value == 0) ? 1 : 0
usleep(150*1000)
}while(true)
We can’t test the hardware SPI with the CHIP but SwiftyGPIO also provide a bit banging software implementation of a SPI interface, you just need two GPIOs to initialize it:
let gpios = SwiftyGPIO.GPIOs(for:.CHIP)
var sclk = gpios[.P0]!
var dnmosi = gpios[.P1]!
var spi = VirtualSPI(dataGPIO:dnmosi,clockGPIO:sclk)
pi.sendData([UInt8(truncatingBitPattern:0x9F)])
Notice that we are converting the 0x9F Int
using the constructor UInt8(truncatingBitPattern:)
, that in this case it’s not actually needed, but it’s recommended for every user-provided or calculated integer because Swift does not support implicit truncation for conversion to smaller integer types, it will just crash if the Int
you are trying to convert does not fit in a UInt8
.
A few projects and libraries built using SwiftyGPIO. Have you built something that you want to share? Let me know!
Libraries for specific devices.
Complete IoT projects.
Additional libraries that could be useful for your IoT projects.
Additional documentation, mostly implementation details, can be found in the docs
directory.