-
-
Notifications
You must be signed in to change notification settings - Fork 237
Multicore Processing
So far I never needed to use any of the ESP32 multicore functionality because all the examples are very simple and everything is fitting on a single core. However if you write some more complex scenarios which are e.g. updating a screen, process user input w/o impacting the audio, you should consider to use multiple tasks. Then you need to use a save way to communicate data between the tasks.
- Task: a simple C++ class which represents a FreeRTPS task.
- BufferRTOS: a FreeRTOS StreamBuffer implementation using the BaseBuffer API
- SynchronizedNBuffer: a thread save NBuffer.
- Mutex: with different implementations per platform.
- LockGuard: a RAII class for managing the Mutex
- SynchronizedBuffer: which allows you to use any BaseBuffer implementation in a save way
- QueueRTOS You can also use the C++ API of an RTOS Queue to comminicate between processes.
You can use any of the buffers above in a QueueStream to use them as copy source and copy targets.
If you use an processor that supports FreeRTOS, you can start multiple tasks
#include "AudioTools.h"
#include "AudioTools/Concurrency/All.h"
Task task("write", 3000, 10, 0);
void method(){}
task.begin(method);
If you use an ESP32, it is much more efficient to use a FreeRTOS StreamBuffer. Just write the data with writeArray() on one Task and read it back with readArray() from the other task. It is the same API you would use with any other class which is based on BaseBuffer
#include "AudioTools.h"
#include "AudioTools/Concurrency/All.h"
BufferRTOS<int16_t> buffer(2048, 512);
In my test sketch I was measuring a thruput of 36.57-37.93 Mbytes per second.
I have extended the NBuffer to use a FreeRTOS queue to manage the allocated buffers.
#include "AudioTools.h"
#include "AudioTools/Concurrency/All.h"
SynchronizedNBuffer<int16_t> buffer(2048, 512);
Here is an example how to use the SynchronizedBuffer class. Just wrap one of the existing buffer classes. Here is an example of a thread-save Double Buffer:
#include "AudioTools.h"
#include "AudioTools/Concurrency/All.h"
audio_tools::Mutex mutex;
NBuffer<int16_t> nbuffer(512,2);
SynchronizedBuffer<int16_t> buffer(nbuffer, mutex);
In my test sketch this was giving a thuput of only 1.68 Mbytes per second.
Here is a thread-save RingBuffer
#include "AudioTools.h"
#include "AudioTools/Concurrency/All.h"
audio_tools::Mutex mutex;
RingBuffer<int16_t> nbuffer(512*4);
SynchronizedBuffer<int16_t> buffer(nbuffer, mutex);
and I was measuring 2.35 Mbytes per second.
Usually it is good enought to just move the copy operation to a separate task, so that we can use the Arduino loop w/o impacting any audio.
In the following example we just decode the mp3 from an URL and send the output to I2S in a separate task.
#include "AudioTools.h"
#include "AudioTools/AudioCodecs/CodecMP3Helix.h"
#include "AudioTools/AudioLibs/AudioBoardStream.h"
#include "AudioTools/Concurrency/All.h"
URLStream url("ssid","password"); // or replace with ICYStream to get metadata
AudioBoardStream i2s(AudioKitEs8388V1); // final output of decoded stream
EncodedAudioStream dec(&i2s, new MP3DecoderHelix()); // Decoding stream
StreamCopy copier(dec, url); // copy url to decoder
Task task("mp3-copy", 10000, 1, 0);
void setup(){
Serial.begin(115200);
AudioLogger::instance().begin(Serial, AudioLogger::Info);
// setup i2s
auto config = i2s.defaultConfig(TX_MODE);
i2s.begin(config);
// setup I2S based on sampling rate provided by decoder
dec.begin();
// mp3 radio
url.begin("http://stream.srg-ssr.ch/m/rsj/mp3_128","audio/mp3");
// start copy task
task.begin([](){copier.copy();});
}
void loop(){
delay(1000);
Serial.println("ping...");
}
Please note that in this sceanrio the loop processing can still starve the audio task if you execute some long lasting operations that do not yield: in this case add some short delay() calls (to your looop porcess), to yield control to the audio task!
Here is a maximum example where we use a separate task to generate the audio and a task to output the audio. Both tasks communicate via a queue that is created from a synchronized buffer implementation.
Both tasks can use a StreamCopy to manage the data copying so they just need to call the copy operation. Because this is quite short we do not write separate methods, but just implment this with the help of a lambda expression!
We still have the Arduino loop task available for other processing!
#include "AudioTools.h"
#include "AudioTools/AudioLibs/AudioBoardStream.h"
#include "AudioTools/Concurrency/All.h"
AudioInfo info(48000, 2, 16);
// source and sink
SineWaveGenerator<int16_t> sineWave(32000);
GeneratedSoundStream<int16_t> sound(sineWave);
AudioBoardStream out(AudioKitEs8388V1);
// queue
BufferRTOS<uint8_t> buffer(1024 * 10);
QueueStream<uint8_t> queue(buffer);
// copy
StreamCopy copierSource(queue, sound);
StreamCopy copierSink(out, queue);
// tasks
Task writeTask("write", 3000, 10, 0);
Task readTask("read", 3000, 10, 1);
void setup() {
Serial.begin(115200);
AudioLogger::instance().begin(Serial, AudioLogger::Warning);
// start Queue
queue.begin();
// start I2S Output
Serial.println("starting I2S...");
auto config = out.defaultConfig(TX_MODE);
config.copyFrom(info);
out.begin(config);
// Setup sine wave Data Source
sineWave.begin(info, N_B4);
Serial.println("started...");
// Start audio tasks
writeTask.begin([]() {
copierSource.copy();
});
readTask.begin([]() {
copierSink.copy();
});
Serial.println("started...");
}
void loop() { delay(1000); }
I am providing the most important multi core functionality as C++ classes. Alternatively I recommend to use the arduino-freertos-addons which provides all the functionality you need in an elegant way: Just read the Readme of the project. However, if you prefer you can also use the official FreeRTOS C API.