Tutorial - Part 3: Forcing A Roundtrip | Index | Tutorial - Part 5: Capturing Video Frames

In this tutorial we show how to use a stream to play a tone.

Let's take a look at the code before we break it down:

#include <math.h>
 
#include <spa/param/audio/format-utils.h>
 
#include <pipewire/pipewire.h>
 
#define M_PI_M2 ( M_PI + M_PI )
 
#define DEFAULT_RATE            44100
#define DEFAULT_CHANNELS        2
#define DEFAULT_VOLUME          0.7
 
struct data {
        struct pw_main_loop *loop;
        struct pw_stream *stream;
        double accumulator;
};
 
/* [on_process] */
static void on_process(void *userdata)
{
        struct data *data = userdata;
        struct pw_buffer *b;
        struct spa_buffer *buf;
        int i, c, n_frames, stride;
        int16_t *dst, val;
 
        if ((b = pw_stream_dequeue_buffer(data->stream)) == NULL) {
                pw_log_warn("out of buffers: %m");
                return;
        }
 
        buf = b->buffer;
        if ((dst = buf->datas[0].data) == NULL)
                return;
 
        stride = sizeof(int16_t) * DEFAULT_CHANNELS;
        n_frames = buf->datas[0].maxsize / stride;
        if (b->requested)
                n_frames = SPA_MIN(b->requested, n_frames);
 
        for (i = 0; i < n_frames; i++) {
                data->accumulator += M_PI_M2 * 440 / DEFAULT_RATE;
                if (data->accumulator >= M_PI_M2)
                        data->accumulator -= M_PI_M2;
 
                /* sin() gives a value between -1.0 and 1.0, we first apply
                 * the volume and then scale with 32767.0 to get a 16 bits value
                 * between [-32767 32767].
                 * Another common method to convert a double to
                 * 16 bits is to multiple by 32768.0 and then clamp to
                 * [-32768 32767] to get the full 16 bits range. */
                val = sin(data->accumulator) * DEFAULT_VOLUME * 32767.0;
                for (c = 0; c < DEFAULT_CHANNELS; c++)
                        *dst++ = val;
        }
 
        buf->datas[0].chunk->offset = 0;
        buf->datas[0].chunk->stride = stride;
        buf->datas[0].chunk->size = n_frames * stride;
 
        pw_stream_queue_buffer(data->stream, b);
}
/* [on_process] */
 
static const struct pw_stream_events stream_events = {
        PW_VERSION_STREAM_EVENTS,
        .process = on_process,
};
 
int main(int argc, char *argv[])
{
        struct data data = { 0, };
        const struct spa_pod *params[1];
        uint8_t buffer[1024];
        struct spa_pod_builder b = SPA_POD_BUILDER_INIT(buffer, sizeof(buffer));
 
        pw_init(&argc, &argv);
 
        data.loop = pw_main_loop_new(NULL);
 
        data.stream = pw_stream_new_simple(
                        pw_main_loop_get_loop(data.loop),
                        "audio-src",
                        pw_properties_new(
                                PW_KEY_MEDIA_TYPE, "Audio",
                                PW_KEY_MEDIA_CATEGORY, "Playback",
                                PW_KEY_MEDIA_ROLE, "Music",
                                NULL),
                        &stream_events,
                        &data);
 
        params[0] = spa_format_audio_raw_build(&b, SPA_PARAM_EnumFormat,
                        &SPA_AUDIO_INFO_RAW_INIT(
                                .format = SPA_AUDIO_FORMAT_S16,
                                .channels = DEFAULT_CHANNELS,
                                .rate = DEFAULT_RATE ));
 
        pw_stream_connect(data.stream,
                          PW_DIRECTION_OUTPUT,
                          PW_ID_ANY,
                          PW_STREAM_FLAG_AUTOCONNECT |
                          PW_STREAM_FLAG_MAP_BUFFERS |
                          PW_STREAM_FLAG_RT_PROCESS,
                          params, 1);
 
        pw_main_loop_run(data.loop);
 
        pw_stream_destroy(data.stream);
        pw_main_loop_destroy(data.loop);
 
        return 0;
}

Save as tutorial4.c and compile with:

gcc -Wall tutorial4.c -o tutorial4 -lm $(pkg-config --cflags --libs libpipewire-0.3)

We start with the usual boilerplate, pw_init() and a pw_main_loop_new(). We're going to store our objects in a structure so that we can pass them around in callbacks later.

struct data {
        struct pw_main_loop *loop;
        struct pw_stream *stream;
        double accumulator;
};
 
int main(int argc, char *argv[])
{
        struct data data = { 0, };
 
        pw_init(&argc, &argv);
 
        data.loop = pw_main_loop_new(NULL);

Next we create a stream object. It takes the mainloop as first argument and a stream name as the second. Next we provide some properties for the stream and a callback + data.

data.stream = pw_stream_new_simple(
                pw_main_loop_get_loop(data.loop),
                "audio-src",
                pw_properties_new(
                        PW_KEY_MEDIA_TYPE, "Audio",
                        PW_KEY_MEDIA_CATEGORY, "Playback",
                        PW_KEY_MEDIA_ROLE, "Music",
                        NULL),
                &stream_events,
                &data);

We are using pw_stream_new_simple() but there is also a pw_stream_new() that takes an existing struct pw_core as the first argument and that requires you to add the event handle manually, for more control. The pw_stream_new_simple() is, as the name implies, easier to use because it creates a struct pw_context and struct pw_core automatically.

In the properties we need to give as much information about the stream as we can so that the session manager can make good decisions about how and where to route this stream. There are three important properties to configure:

PW_KEY_MEDIA_TYPE: The media type; like Audio, Video, MIDI.
PW_KEY_MEDIA_CATEGORY: The category; like Playback, Capture, Duplex, Monitor.
PW_KEY_MEDIA_ROLE: The media role; like Movie, Music, Camera, Screen, Communication, Game, Notification, DSP, Production, Accessibility, Test.

The properties are owned by the stream and freed when the stream is destroyed later.

This is the event structure that we use to listen for events:

static const struct pw_stream_events stream_events = {
        PW_VERSION_STREAM_EVENTS,
        .process = on_process,
};

We are for the moment only interested now in the process event. This event is called whenever we need to produce more data. We'll see how that function is implemented but first we need to setup the format of the stream:

        const struct spa_pod *params[1];
        uint8_t buffer[1024];
        struct spa_pod_builder b = SPA_POD_BUILDER_INIT(buffer, sizeof(buffer));
 
#define DEFAULT_RATE            44100
#define DEFAULT_CHANNELS        2
 
        params[0] = spa_format_audio_raw_build(&b, SPA_PARAM_EnumFormat,
                        &SPA_AUDIO_INFO_RAW_INIT(
                                .format = SPA_AUDIO_FORMAT_S16,
                                .channels = DEFAULT_CHANNELS,
                                .rate = DEFAULT_RATE ));

This is using a struct spa_pod_builder to make a struct spa_pod * object in the buffer array on the stack. The parameter is of type SPA_PARAM_EnumFormat which means that it enumerates the possible formats for this stream. We have only one, a Signed 16 bit stereo format at 44.1KHz.

We use spa_format_audio_raw_build() which is a helper function to make the param with the builder. See SPA POD for more information about how to make these POD objects.

Now we're ready to connect the stream and run the main loop:

pw_stream_connect(data.stream,
                  PW_DIRECTION_OUTPUT,
                  PW_ID_ANY,
                  PW_STREAM_FLAG_AUTOCONNECT |
                  PW_STREAM_FLAG_MAP_BUFFERS |
                  PW_STREAM_FLAG_RT_PROCESS,
                  params, 1);
 
pw_main_loop_run(data.loop);

To connect we specify that we have a PW_DIRECTION_OUTPUT stream. The third argument is always PW_ID_ANY. Next we set some flags:

PW_STREAM_FLAG_AUTOCONNECT: Automatically connect this stream. This instructs the session manager to link us to some consumer.
PW_STREAM_FLAG_MAP_BUFFERS: mmap the buffers for us so we can access the memory. If you don't set these flags you have either work with the fd or mmap yourself.
PW_STREAM_FLAG_RT_PROCESS: Run the process function in the realtime thread. Only use this if the process function only uses functions that are realtime safe, this means no allocation or file access or any locking.

And last we pass the extra parameters for our stream. Here we only have the allowed formats (SPA_PARAM_EnumFormat).

Running the mainloop will then start processing and will result in our process callback to be called. Let's have a look at that function now.

The main program flow of the process function is:

pw_stream_dequeue_buffer() to obtain a buffer to write into.
Get pointers in buffer memory to write to.
Write data into buffer.
Adjust buffer with number of written bytes, offset, stride.
pw_stream_queue_buffer() to queue the buffer for playback.

static void on_process(void *userdata)
{
        struct data *data = userdata;
        struct pw_buffer *b;
        struct spa_buffer *buf;
        int i, c, n_frames, stride;
        int16_t *dst, val;
 
        if ((b = pw_stream_dequeue_buffer(data->stream)) == NULL) {
                pw_log_warn("out of buffers: %m");
                return;
        }
 
        buf = b->buffer;
        if ((dst = buf->datas[0].data) == NULL)
                return;
 
        stride = sizeof(int16_t) * DEFAULT_CHANNELS;
        n_frames = buf->datas[0].maxsize / stride;
        if (b->requested)
                n_frames = SPA_MIN(b->requested, n_frames);
 
        for (i = 0; i < n_frames; i++) {
                data->accumulator += M_PI_M2 * 440 / DEFAULT_RATE;
                if (data->accumulator >= M_PI_M2)
                        data->accumulator -= M_PI_M2;
 
                /* sin() gives a value between -1.0 and 1.0, we first apply
                 * the volume and then scale with 32767.0 to get a 16 bits value
                 * between [-32767 32767].
                 * Another common method to convert a double to
                 * 16 bits is to multiple by 32768.0 and then clamp to
                 * [-32768 32767] to get the full 16 bits range. */
                val = sin(data->accumulator) * DEFAULT_VOLUME * 32767.0;
                for (c = 0; c < DEFAULT_CHANNELS; c++)
                        *dst++ = val;
        }
 
        buf->datas[0].chunk->offset = 0;
        buf->datas[0].chunk->stride = stride;
        buf->datas[0].chunk->size = n_frames * stride;
 
        pw_stream_queue_buffer(data->stream, b);
}

Check out the docs for SPA Buffers for more information about how to work with buffers.

Try to change the number of channels, samplerate or format; the stream will automatically convert to the format on the server.

Tutorial - Part 3: Forcing A Roundtrip | Index | Tutorial - Part 5: Capturing Video Frames