Add audio player

Play the decoded audio using SDL.

The audio player frame sink receives the audio frames, resample them
and write them to a byte buffer (introduced by this commit).

On SDL audio callback (from an internal SDL thread), copy samples from
this byte buffer to the SDL audio buffer.

The byte buffer is protected by the SDL_AudioDeviceLock(), but it has
been designed so that the producer and the consumer may write and read
in parallel, provided that they don't access the same slices of the
ring-buffer buffer.

Co-authored-by: Simon Chan <1330321+yume-chan@users.noreply.github.com>

app/meson.build

@@ -4,6 +4,7 @@ src = [
     'src/adb/adb_device.c',
     'src/adb/adb_parser.c',
     'src/adb/adb_tunnel.c',
+    'src/audio_player.c',
     'src/cli.c',
     'src/clock.c',
     'src/compat.c',
@@ -30,6 +31,7 @@ src = [
     'src/version.c',
     'src/video_buffer.c',
     'src/util/acksync.c',
+    'src/util/average.c',
     'src/util/bytebuf.c',
     'src/util/file.c',
     'src/util/intmap.c',
@@ -100,6 +102,7 @@ if not crossbuild_windows
         dependency('libavformat', version: '>= 57.33'),
         dependency('libavcodec', version: '>= 57.37'),
         dependency('libavutil'),
+        dependency('libswresample'),
         dependency('sdl2', version: '>= 2.0.5'),
     ]
 
@@ -134,12 +137,14 @@ else
     ffmpeg_avcodec = meson.get_cross_property('ffmpeg_avcodec')
     ffmpeg_avformat = meson.get_cross_property('ffmpeg_avformat')
     ffmpeg_avutil = meson.get_cross_property('ffmpeg_avutil')
+    ffmpeg_swresample = meson.get_cross_property('ffmpeg_swresample')
 
     ffmpeg = declare_dependency(
         dependencies: [
             cc.find_library(ffmpeg_avcodec, dirs: ffmpeg_bin_dir),
             cc.find_library(ffmpeg_avformat, dirs: ffmpeg_bin_dir),
             cc.find_library(ffmpeg_avutil, dirs: ffmpeg_bin_dir),
+            cc.find_library(ffmpeg_swresample, dirs: ffmpeg_bin_dir),
         ],
         include_directories: include_directories(ffmpeg_include_dir)
     )

app/src/audio_player.c

@@ -0,0 +1,289 @@
+#include "audio_player.h"
+
+#include <libavutil/opt.h>
+
+#include "util/log.h"
+
+#define SC_AUDIO_PLAYER_NDEBUG // comment to debug
+
+/** Downcast frame_sink to sc_audio_player */
+#define DOWNCAST(SINK) container_of(SINK, struct sc_audio_player, frame_sink)
+
+#define SC_AV_SAMPLE_FMT AV_SAMPLE_FMT_FLT
+#define SC_SDL_SAMPLE_FMT AUDIO_F32
+
+#define SC_AUDIO_OUTPUT_BUFFER_SAMPLES 480 // 10ms at 48000Hz
+
+// The target number of buffered samples between the producer and the consumer.
+// This value is directly use for compensation.
+#define SC_TARGET_BUFFERED_SAMPLES (3 * SC_AUDIO_OUTPUT_BUFFER_SAMPLES)
+
+// If the consumer is too late, skip samples to keep at most this value
+#define SC_BUFFERED_SAMPLES_THRESHOLD 2400 // 50ms at 48000Hz
+
+// Use a ring-buffer of 1 second (at 48000Hz) between the producer and the
+// consumer. It too big, but it guarantees that the producer and the consumer
+// will be able to access it in parallel without locking.
+#define SC_BYTEBUF_SIZE_IN_SAMPLES 48000
+
+void
+sc_audio_player_sdl_callback(void *userdata, uint8_t *stream, int len_int) {
+    struct sc_audio_player *ap = userdata;
+
+    // This callback is called with the lock used by SDL_AudioDeviceLock(), so
+    // the bytebuf is protected
+
+    assert(len_int > 0);
+    size_t len = len_int;
+
+#ifndef SC_AUDIO_PLAYER_NDEBUG
+    LOGD("[Audio] SDL callback requests %" SC_PRIsizet " samples",
+         len / (ap->nb_channels * ap->out_bytes_per_sample));
+#endif
+
+    size_t read = sc_bytebuf_read_remaining(&ap->buf);
+    size_t max_buffered_bytes = SC_BUFFERED_SAMPLES_THRESHOLD
+                              * ap->nb_channels * ap->out_bytes_per_sample;
+    if (read > max_buffered_bytes + len) {
+        size_t skip = read - (max_buffered_bytes + len);
+#ifndef SC_AUDIO_PLAYER_NDEBUG
+        LOGD("[Audio] Buffered samples threshold exceeded: %" SC_PRIsizet
+             " bytes, skipping %" SC_PRIsizet " bytes", read, skip);
+#endif
+        // After this callback, exactly max_buffered_bytes will remain
+        sc_bytebuf_skip(&ap->buf, skip);
+        read = max_buffered_bytes + len;
+    }
+
+    // Number of buffered samples (may be negative on underflow)
+    float buffered_samples = ((float) read - len_int)
+                           / (ap->nb_channels * ap->out_bytes_per_sample);
+    sc_average_push(&ap->avg_buffered_samples, buffered_samples);
+
+    if (read) {
+        if (read > len) {
+            read = len;
+        }
+        sc_bytebuf_read(&ap->buf, stream, read);
+    }
+
+    if (read < len) {
+        // Insert silence
+#ifndef SC_AUDIO_PLAYER_NDEBUG
+        LOGD("[Audio] Buffer underflow, inserting silence: %" SC_PRIsizet
+             " bytes", len - read);
+#endif
+        memset(stream + read, 0, len - read);
+    }
+}
+
+static size_t
+sc_audio_player_get_buf_size(struct sc_audio_player *ap, size_t samples) {
+    assert(ap->nb_channels);
+    assert(ap->out_bytes_per_sample);
+    return samples * ap->nb_channels * ap->out_bytes_per_sample;
+}
+
+static uint8_t *
+sc_audio_player_get_swr_buf(struct sc_audio_player *ap, size_t min_samples) {
+    size_t min_buf_size = sc_audio_player_get_buf_size(ap, min_samples);
+    if (min_buf_size < ap->swr_buf_alloc_size) {
+        size_t new_size = min_buf_size + 4096;
+        uint8_t *buf = realloc(ap->swr_buf, new_size);
+        if (!buf) {
+            LOG_OOM();
+            // Could not realloc to the requested size
+            return NULL;
+        }
+        ap->swr_buf = buf;
+        ap->swr_buf_alloc_size = new_size;
+    }
+
+    return ap->swr_buf;
+}
+
+static bool
+sc_audio_player_frame_sink_open(struct sc_frame_sink *sink,
+                                const AVCodecContext *ctx) {
+    struct sc_audio_player *ap = DOWNCAST(sink);
+
+    SDL_AudioSpec desired = {
+        .freq = ctx->sample_rate,
+        .format = SC_SDL_SAMPLE_FMT,
+        .channels = ctx->ch_layout.nb_channels,
+        .samples = SC_AUDIO_OUTPUT_BUFFER_SAMPLES,
+        .callback = sc_audio_player_sdl_callback,
+        .userdata = ap,
+    };
+    SDL_AudioSpec obtained;
+
+    ap->device = SDL_OpenAudioDevice(NULL, 0, &desired, &obtained, 0);
+    if (!ap->device) {
+        LOGE("Could not open audio device: %s", SDL_GetError());
+        return false;
+    }
+
+    SwrContext *swr_ctx = swr_alloc();
+    if (!swr_ctx) {
+        LOG_OOM();
+        goto error_close_audio_device;
+    }
+    ap->swr_ctx = swr_ctx;
+
+    assert(ctx->sample_rate > 0);
+    assert(ctx->ch_layout.nb_channels > 0);
+    assert(!av_sample_fmt_is_planar(SC_AV_SAMPLE_FMT));
+    int out_bytes_per_sample = av_get_bytes_per_sample(SC_AV_SAMPLE_FMT);
+    assert(out_bytes_per_sample > 0);
+
+    av_opt_set_chlayout(swr_ctx, "in_chlayout", &ctx->ch_layout, 0);
+    av_opt_set_chlayout(swr_ctx, "out_chlayout", &ctx->ch_layout, 0);
+
+    av_opt_set_int(swr_ctx, "in_sample_rate", ctx->sample_rate, 0);
+    av_opt_set_int(swr_ctx, "out_sample_rate", ctx->sample_rate, 0);
+
+    av_opt_set_sample_fmt(swr_ctx, "in_sample_fmt", ctx->sample_fmt, 0);
+    av_opt_set_sample_fmt(swr_ctx, "out_sample_fmt", SC_AV_SAMPLE_FMT, 0);
+
+    int ret = swr_init(swr_ctx);
+    if (ret) {
+        LOGE("Failed to initialize the resampling context");
+        goto error_free_swr_ctx;
+    }
+
+    ap->sample_rate = ctx->sample_rate;
+    ap->nb_channels = ctx->ch_layout.nb_channels;
+    ap->out_bytes_per_sample = out_bytes_per_sample;
+
+    size_t bytebuf_size =
+        sc_audio_player_get_buf_size(ap, SC_BYTEBUF_SIZE_IN_SAMPLES);
+
+    bool ok = sc_bytebuf_init(&ap->buf, bytebuf_size);
+    if (!ok) {
+        goto error_free_swr_ctx;
+    }
+
+    ap->safe_empty_buffer = sc_bytebuf_write_remaining(&ap->buf);
+
+    size_t initial_swr_buf_size = sc_audio_player_get_buf_size(ap, 4096);
+    ap->swr_buf = malloc(initial_swr_buf_size);
+    if (!ap->swr_buf) {
+        LOG_OOM();
+        goto error_destroy_bytebuf;
+    }
+    ap->swr_buf_alloc_size = initial_swr_buf_size;
+
+    sc_average_init(&ap->avg_buffered_samples, 32);
+    ap->samples_since_resync = 0;
+
+    SDL_PauseAudioDevice(ap->device, 0);
+
+    return true;
+
+error_destroy_bytebuf:
+    sc_bytebuf_destroy(&ap->buf);
+error_free_swr_ctx:
+    swr_free(&ap->swr_ctx);
+error_close_audio_device:
+    SDL_CloseAudioDevice(ap->device);
+
+    return false;
+}
+
+static void
+sc_audio_player_frame_sink_close(struct sc_frame_sink *sink) {
+    struct sc_audio_player *ap = DOWNCAST(sink);
+
+    assert(ap->device);
+    SDL_PauseAudioDevice(ap->device, 1);
+    SDL_CloseAudioDevice(ap->device);
+
+    free(ap->swr_buf);
+    sc_bytebuf_destroy(&ap->buf);
+    swr_free(&ap->swr_ctx);
+}
+
+static bool
+sc_audio_player_frame_sink_push(struct sc_frame_sink *sink, const AVFrame *frame) {
+    struct sc_audio_player *ap = DOWNCAST(sink);
+
+    SwrContext *swr_ctx = ap->swr_ctx;
+
+    int64_t delay = swr_get_delay(swr_ctx, ap->sample_rate);
+    // No need to av_rescale_rnd(), input and output sample rates are the same
+    int dst_nb_samples = delay + frame->nb_samples;
+
+    uint8_t *swr_buf = sc_audio_player_get_swr_buf(ap, frame->nb_samples);
+    if (!swr_buf) {
+        return false;
+    }
+
+    int ret = swr_convert(swr_ctx, &swr_buf, dst_nb_samples,
+                          (const uint8_t **) frame->data, frame->nb_samples);
+    if (ret < 0) {
+        LOGE("Resampling failed: %d", ret);
+        return false;
+    }
+
+    size_t samples_written = ret;
+    size_t swr_buf_size = sc_audio_player_get_buf_size(ap, samples_written);
+#ifndef SC_AUDIO_PLAYER_NDEBUG
+    LOGI("[Audio] %" SC_PRIsizet " samples written to buffer", samples_written);
+#endif
+
+    // It should almost always be possible to write without lock
+    bool can_write_without_lock = swr_buf_size <= ap->safe_empty_buffer;
+    if (can_write_without_lock) {
+        sc_bytebuf_prepare_write(&ap->buf, swr_buf, swr_buf_size);
+    }
+
+    SDL_LockAudioDevice(ap->device);
+    if (can_write_without_lock) {
+        sc_bytebuf_commit_write(&ap->buf, swr_buf_size);
+    } else {
+        sc_bytebuf_write(&ap->buf, swr_buf, swr_buf_size);
+    }
+
+    // The next time, it will remain at least the current empty space
+    ap->safe_empty_buffer = sc_bytebuf_write_remaining(&ap->buf);
+
+    // Read the value written by the SDL thread under lock
+    float avg;
+    bool has_avg = sc_average_get(&ap->avg_buffered_samples, &avg);
+
+    SDL_UnlockAudioDevice(ap->device);
+
+    if (has_avg) {
+        ap->samples_since_resync += samples_written;
+        if (ap->samples_since_resync >= ap->sample_rate) {
+            // Resync every second
+            ap->samples_since_resync = 0;
+
+            int diff = SC_TARGET_BUFFERED_SAMPLES - avg;
+#ifndef SC_AUDIO_PLAYER_NDEBUG
+            LOGI("[Audio] Average buffered samples = %f, compensation %d",
+                 avg, diff);
+#endif
+            // Compensate the diff over 3 seconds (but will be recomputed after
+            // 1 second)
+            int ret = swr_set_compensation(swr_ctx, diff, 3 * ap->sample_rate);
+            if (ret < 0) {
+                LOGW("Resampling compensation failed: %d", ret);
+                // not fatal
+            }
+        }
+    }
+
+    return true;
+}
+
+void
+sc_audio_player_init(struct sc_audio_player *ap) {
+    static const struct sc_frame_sink_ops ops = {
+        .open = sc_audio_player_frame_sink_open,
+        .close = sc_audio_player_frame_sink_close,
+        .push = sc_audio_player_frame_sink_push,
+    };
+
+    ap->frame_sink.ops = &ops;
+}

app/src/audio_player.h

@@ -0,0 +1,54 @@
+#ifndef SC_AUDIO_PLAYER_H
+#define SC_AUDIO_PLAYER_H
+
+#include "common.h"
+
+#include <stdbool.h>
+#include "trait/frame_sink.h"
+#include <util/average.h>
+#include <util/bytebuf.h>
+#include <util/thread.h>
+
+#include <libavformat/avformat.h>
+#include <libswresample/swresample.h>
+#include <SDL2/SDL.h>
+
+struct sc_audio_player {
+    struct sc_frame_sink frame_sink;
+
+    SDL_AudioDeviceID device;
+
+    // protected by SDL_AudioDeviceLock()
+    struct sc_bytebuf buf;
+    // Number of bytes which could be written without locking
+    size_t safe_empty_buffer;
+
+    struct SwrContext *swr_ctx;
+
+    // The sample rate is the same for input and output
+    unsigned sample_rate;
+    // The number of channels is the same for input and output
+    unsigned nb_channels;
+
+    unsigned out_bytes_per_sample;
+
+    // Target buffer for resampling
+    uint8_t *swr_buf;
+    size_t swr_buf_alloc_size;
+
+    // Number of buffered samples (may be negative on underflow)
+    struct sc_average avg_buffered_samples;
+    unsigned samples_since_resync;
+
+    const struct sc_audio_player_callbacks *cbs;
+    void *cbs_userdata;
+};
+
+struct sc_audio_player_callbacks {
+    void (*on_ended)(struct sc_audio_player *ap, bool success, void *userdata);
+};
+
+void
+sc_audio_player_init(struct sc_audio_player *ap);
+
+#endif

app/src/decoder.c

@@ -2,6 +2,7 @@
 
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
+#include <libavutil/channel_layout.h>
 
 #include "events.h"
 #include "video_buffer.h"
@@ -50,6 +51,11 @@ sc_decoder_open(struct sc_decoder *decoder, const AVCodec *codec) {
     if (codec->type == AVMEDIA_TYPE_VIDEO) {
         // Hardcoded video properties
         decoder->codec_ctx->pix_fmt = AV_PIX_FMT_YUV420P;
+    } else {
+        // Hardcoded audio properties
+        decoder->codec_ctx->ch_layout =
+            (AVChannelLayout) AV_CHANNEL_LAYOUT_STEREO;
+        decoder->codec_ctx->sample_rate = 48000;
     }
 
     if (avcodec_open2(decoder->codec_ctx, codec, NULL) < 0) {

app/src/scrcpy.c

@@ -13,6 +13,7 @@
 # include <windows.h>
 #endif
 
+#include "audio_player.h"
 #include "controller.h"
 #include "decoder.h"
 #include "demuxer.h"
@@ -40,6 +41,7 @@
 struct scrcpy {
     struct sc_server server;
     struct sc_screen screen;
+    struct sc_audio_player audio_player;
     struct sc_demuxer video_demuxer;
     struct sc_demuxer audio_demuxer;
     struct sc_decoder video_decoder;
@@ -383,9 +385,16 @@ scrcpy(struct scrcpy_options *options) {
     }
 
     // Initialize SDL video in addition if display is enabled
-    if (options->display && SDL_Init(SDL_INIT_VIDEO)) {
-        LOGE("Could not initialize SDL: %s", SDL_GetError());
-        goto end;
+    if (options->display) {
+        if (SDL_Init(SDL_INIT_VIDEO)) {
+            LOGE("Could not initialize SDL video: %s", SDL_GetError());
+            goto end;
+        }
+
+        if (options->audio && SDL_Init(SDL_INIT_AUDIO)) {
+            LOGE("Could not initialize SDL audio: %s", SDL_GetError());
+            goto end;
+        }
     }
 
     sdl_configure(options->display, options->disable_screensaver);
@@ -663,6 +672,11 @@ aoa_hid_end:
         screen_initialized = true;
 
         sc_decoder_add_sink(&s->video_decoder, &s->screen.frame_sink);
+
+        if (options->audio) {
+            sc_audio_player_init(&s->audio_player);
+            sc_decoder_add_sink(&s->audio_decoder, &s->audio_player.frame_sink);
+        }
     }
 
 #ifdef HAVE_V4L2

app/src/util/average.c

@@ -0,0 +1,26 @@
+#include "average.h"
+
+#include <assert.h>
+
+void
+sc_average_init(struct sc_average *avg, unsigned range) {
+    avg->range = range;
+    avg->avg = 0;
+    avg->count = 0;
+}
+
+void
+sc_average_push(struct sc_average *avg, float value) {
+    if (avg->count < avg->range) {
+        ++avg->count;
+    }
+
+    assert(avg->count);
+    avg->avg = ((avg->count - 1) * avg->avg + value) / avg->count;
+}
+
+bool
+sc_average_get(struct sc_average *avg, float *value) {
+    *value = avg->avg;
+    return avg->count;
+}

app/src/util/average.h

@@ -0,0 +1,36 @@
+#ifndef SC_AVERAGE
+#define SC_AVERAGE
+
+#include "common.h"
+
+#include <stdbool.h>
+#include <stdint.h>
+
+struct sc_average {
+    // Current average value
+    float avg;
+
+    // Target range, to update the average as follow:
+    //     avg = ((range - 1) * avg + new_value) / range
+    unsigned range;
+
+    // Number of values pushed when less than range (count <= range).
+    // The purpose is to handle the first (range - 1) values properly.
+    unsigned count;
+};
+
+void
+sc_average_init(struct sc_average *avg, unsigned range);
+
+/* Push a new value to update the "rolling" average */
+void
+sc_average_push(struct sc_average *avg, float value);
+
+/* Get the current average value (if available)
+ *
+ * An average is available if sc_average_push() has been called at least once.
+ */
+bool
+sc_average_get(struct sc_average *avg, float *value);
+
+#endif

cross_win32.txt

@@ -19,6 +19,7 @@ endian = 'little'
 ffmpeg_avcodec = 'avcodec-58'
 ffmpeg_avformat = 'avformat-58'
 ffmpeg_avutil = 'avutil-56'
+ffmpeg_swresample = 'swresample-3'
 prebuilt_ffmpeg = 'ffmpeg-win32-4.3.1'
 prebuilt_sdl2 = 'SDL2-2.26.1/i686-w64-mingw32'
 prebuilt_libusb_root = 'libusb-1.0.26'

cross_win64.txt

@@ -19,6 +19,7 @@ endian = 'little'
 ffmpeg_avcodec = 'avcodec-59'
 ffmpeg_avformat = 'avformat-59'
 ffmpeg_avutil = 'avutil-57'
+ffmpeg_swresample = 'swresample-4'
 prebuilt_ffmpeg = 'ffmpeg-win64-5.1.2'
 prebuilt_sdl2 = 'SDL2-2.26.1/x86_64-w64-mingw32'
 prebuilt_libusb_root = 'libusb-1.0.26'