New pipeline building, still needs proper control

src/audio/CMakeLists.txt

@@ -1,7 +1,7 @@
 idf_component_register(
   SRCS "audio_decoder.cpp" "audio_task.cpp" "chunk.cpp" "fatfs_audio_input.cpp"
   "stream_message.cpp" "i2s_audio_output.cpp" "stream_buffer.cpp"
-  "audio_playback.cpp" "stream_event.cpp" "audio_element.cpp" "pipeline.cpp"
+  "audio_playback.cpp" "stream_event.cpp" "pipeline.cpp"
   INCLUDE_DIRS "include"
   REQUIRES "codecs" "drivers" "cbor" "result" "tasks" "span" "memory")
 

src/audio/audio_decoder.cpp

@@ -26,22 +26,22 @@ static const char* kTag = "DEC";
 
 AudioDecoder::AudioDecoder()
     : IAudioElement(),
-      stream_info_({}),
+      current_codec_(),
-      has_samples_to_send_(false),
+      current_input_format_(),
-      needs_more_input_(true) {}
+      current_output_format_(),
+      has_samples_to_send_(false) {}
 
 AudioDecoder::~AudioDecoder() {}
 
 auto AudioDecoder::ProcessStreamInfo(const StreamInfo& info) -> bool {
-  if (!std::holds_alternative<StreamInfo::Encoded>(info.data)) {
+  if (!std::holds_alternative<StreamInfo::Encoded>(info.format)) {
     return false;
   }
-  const auto& encoded = std::get<StreamInfo::Encoded>(info.data);
+  const auto& encoded = std::get<StreamInfo::Encoded>(info.format);
 
   // Reuse the existing codec if we can. This will help with gapless playback,
   // since we can potentially just continue to decode as we were before,
   // without any setup overhead.
-  // TODO: use audio type from stream
   if (current_codec_ != nullptr &&
       current_codec_->CanHandleType(encoded.type)) {
     current_codec_->ResetForNewStream();
@@ -60,43 +60,47 @@ auto AudioDecoder::ProcessStreamInfo(const StreamInfo& info) -> bool {
   return true;
 }
 
-auto AudioDecoder::Process(std::vector<Stream>* inputs, MutableStream* output)
+auto AudioDecoder::Process(const std::vector<InputStream>& inputs,
-    -> void {
+                           OutputStream* output) -> void {
   // We don't really expect multiple inputs, so just pick the first that
   // contains data. If none of them contain data, then we can still flush
   // pending samples.
-  auto input =
+  auto input = std::find_if(
-      std::find_if(inputs->begin(), inputs->end(),
+      inputs.begin(), inputs.end(),
-                   [](const Stream& s) { return s.data.size_bytes() > 0; });
+      [](const InputStream& s) { return s.data().size_bytes() > 0; });
-
+  if (input == inputs.end()) {
-  if (input != inputs->end()) {
+    input = inputs.begin();
-    const StreamInfo* info = input->info;
-    if (!stream_info_ || *stream_info_ != *info) {
-      // The input stream has changed! Immediately throw everything away and
-      // start from scratch.
-      // TODO: special case gapless playback? needs thought.
-      stream_info_ = *info;
-      has_samples_to_send_ = false;
-      has_set_stream_info_ = false;
-
-      ProcessStreamInfo(*info);
-    }
-
-    current_codec_->SetInput(input->data);
   }
 
+  const StreamInfo& info = input->info();
+  if (!current_input_format_ || *current_input_format_ != info.format) {
+    // The input stream has changed! Immediately throw everything away and
+    // start from scratch.
+    current_input_format_ = info.format;
+    has_samples_to_send_ = false;
+
+    ProcessStreamInfo(info);
+  }
+
+  current_codec_->SetInput(input->data());
+
   while (true) {
     if (has_samples_to_send_) {
-      if (!has_set_stream_info_) {
+      if (!current_output_format_) {
-        has_set_stream_info_ = true;
         auto format = current_codec_->GetOutputFormat();
-        output->info->data.emplace<StreamInfo::Pcm>(
+        current_output_format_ = StreamInfo::Pcm{
-            format.bits_per_sample, format.sample_rate_hz, format.num_channels);
+            .channels = format.num_channels,
+            .bits_per_sample = format.bits_per_sample,
+            .sample_rate = format.sample_rate_hz,
+        };
       }
 
-      auto write_res = current_codec_->WriteOutputSamples(
+      if (!output->prepare(*current_output_format_)) {
-          output->data.subspan(output->info->bytes_in_stream));
+        break;
-      output->info->bytes_in_stream += write_res.first;
+      }
+
+      auto write_res = current_codec_->WriteOutputSamples(output->data());
+      output->add(write_res.first);
       has_samples_to_send_ = !write_res.second;
 
       if (has_samples_to_send_) {
@@ -106,26 +110,23 @@ auto AudioDecoder::Process(std::vector<Stream>* inputs, MutableStream* output)
       }
     }
 
-    if (input != inputs->end()) {
+    auto res = current_codec_->ProcessNextFrame();
-      auto res = current_codec_->ProcessNextFrame();
+    if (res.has_error()) {
-      if (res.has_error()) {
+      // TODO(jacqueline): Handle errors.
-        // TODO(jacqueline): Handle errors.
+      return;
-        return;
+    }
-      }
-      input->data = input->data.subspan(current_codec_->GetInputPosition());
 
-      if (res.value()) {
+    if (res.value()) {
-        // We're out of data in this buffer. Finish immediately; there's nothing
+      // We're out of useable data in this buffer. Finish immediately; there's
-        // to send.
+      // nothing to send.
-        break;
+      input->mark_incomplete();
-      } else {
-        has_samples_to_send_ = true;
-      }
-    } else {
-      // No input; nothing to do.
       break;
+    } else {
+      has_samples_to_send_ = true;
     }
   }
+
+  input->consume(current_codec_->GetInputPosition());
 }
 
 }  // namespace audio

src/audio/audio_element.cpp

@@ -1,56 +0,0 @@
-#include "audio_element.hpp"
-#include <memory>
-
-namespace audio {
-
-IAudioElement::IAudioElement()
-    : input_events_(xQueueCreate(kEventQueueSize, sizeof(void*))),
-      output_events_(nullptr),
-      buffered_output_() {}
-
-IAudioElement::~IAudioElement() {
-  // Ensure we don't leak any memory from events leftover in the queue.
-  while (uxQueueSpacesAvailable(input_events_) < kEventQueueSize) {
-    StreamEvent* event;
-    if (xQueueReceive(input_events_, &event, 0)) {
-      free(event);
-    } else {
-      break;
-    }
-  }
-  // Technically there's a race here if someone is still adding to the queue,
-  // but hopefully the whole pipeline is stopped if an element is being
-  // destroyed.
-  vQueueDelete(input_events_);
-}
-
-auto IAudioElement::SendOrBufferEvent(std::unique_ptr<StreamEvent> event)
-    -> bool {
-  if (!buffered_output_.empty()) {
-    // To ensure we send data in order, don't try to send if we've already
-    // failed to send something.
-    buffered_output_.push_back(std::move(event));
-    return false;
-  }
-  StreamEvent* raw_event = event.release();
-  if (!xQueueSend(output_events_, &raw_event, 0)) {
-    event.reset(raw_event);
-    buffered_output_.push_back(std::move(event));
-    return false;
-  }
-  return true;
-}
-
-auto IAudioElement::FlushBufferedOutput() -> bool {
-  while (!buffered_output_.empty()) {
-    StreamEvent* raw_event = buffered_output_.front().release();
-    buffered_output_.pop_front();
-    if (!xQueueSend(output_events_, &raw_event, 0)) {
-      buffered_output_.emplace_front(raw_event);
-      return false;
-    }
-  }
-  return true;
-}
-
-}  // namespace audio

src/audio/audio_playback.cpp

@@ -8,6 +8,7 @@
 #include "freertos/portmacro.h"
 
 #include "audio_decoder.hpp"
+#include "audio_element.hpp"
 #include "audio_task.hpp"
 #include "chunk.hpp"
 #include "fatfs_audio_input.hpp"
@@ -23,55 +24,30 @@ namespace audio {
 
 auto AudioPlayback::create(drivers::GpioExpander* expander)
     -> cpp::result<std::unique_ptr<AudioPlayback>, Error> {
-  // Create everything
-  auto source = std::make_shared<FatfsAudioInput>();
-  auto codec = std::make_shared<AudioDecoder>();
-
   auto sink_res = I2SAudioOutput::create(expander);
   if (sink_res.has_error()) {
     return cpp::fail(ERR_INIT_ELEMENT);
   }
-  auto sink = sink_res.value();
+  return std::make_unique<AudioPlayback>(std::move(sink_res.value()));
-
-  auto playback = std::make_unique<AudioPlayback>();
-
-  Pipeline *pipeline = new Pipeline(sink.get());
-  pipeline->AddInput(codec.get())->AddInput(source.get());
-
-  task::Start(pipeline);
-
-  return playback;
 }
 
-AudioPlayback::AudioPlayback() {
+AudioPlayback::AudioPlayback(std::unique_ptr<I2SAudioOutput> output)
-  // Create everything
+    : file_source_(), i2s_output_(std::move(output)) {
-  auto source = std::make_shared<FatfsAudioInput>();
+  AudioDecoder* codec = new AudioDecoder();
-  auto codec = std::make_shared<AudioDecoder>();
+  elements_.emplace_back(codec);
 
-  auto sink_res = I2SAudioOutput::create(expander);
+  Pipeline* pipeline = new Pipeline(elements_.front().get());
-  if (sink_res.has_error()) {
+  pipeline->AddInput(file_source_.get());
-    return cpp::fail(ERR_INIT_ELEMENT);
-  }
-  auto sink = sink_res.value();
 
-  auto playback = std::make_unique<AudioPlayback>();
+  task::StartPipeline(pipeline, i2s_output_.get());
-
+  task::StartDrain(i2s_output_.get());
-  Pipeline *pipeline = new Pipeline(sink.get());
-  pipeline->AddInput(codec.get())->AddInput(source.get());
-
-  task::Start(pipeline);
-
-  return playback;
 }
 
-AudioPlayback::~AudioPlayback() {
+AudioPlayback::~AudioPlayback() {}
-  pipeline_->Quit();
-}
 
 auto AudioPlayback::Play(const std::string& filename) -> void {
   // TODO: concurrency, yo!
-  file_source->OpenFile(filename);
+  file_source_->OpenFile(filename);
-  pipeline_->Play();
 }
 
 auto AudioPlayback::LogStatus() -> void {

src/audio/audio_task.cpp

@@ -8,7 +8,9 @@
 #include <deque>
 #include <memory>
 
+#include "audio_sink.hpp"
 #include "cbor.h"
+#include "dac.hpp"
 #include "esp_err.h"
 #include "esp_heap_caps.h"
 #include "esp_log.h"
@@ -33,22 +35,29 @@ namespace task {
 
 static const char* kTag = "task";
 static const std::size_t kStackSize = 24 * 1024;
+static const std::size_t kDrainStackSize = 1024;
 static const uint8_t kAudioCore = 0;
 
-auto Start(Pipeline* pipeline) -> Handle* {
+auto StartPipeline(Pipeline* pipeline, IAudioSink* sink) -> void {
-  auto input_queue = xQueueCreate(8, 1);
-
   // Newly created task will free this.
-  AudioTaskArgs* args = new AudioTaskArgs{
+  AudioTaskArgs* args = new AudioTaskArgs{.pipeline = pipeline, .sink = sink};
-      .pipeline = pipeline,
-      .input = input_queue,
-  };
 
-  ESP_LOGI(kTag, "starting audio task");
+  ESP_LOGI(kTag, "starting audio pipeline task");
   xTaskCreatePinnedToCore(&AudioTaskMain, "pipeline", kStackSize, args,
                           kTaskPriorityAudio, NULL, kAudioCore);
+}
 
-  return new Handle(input_queue);
+auto StartDrain(IAudioSink* sink) -> void {
+  auto command = new std::atomic<Command>(PLAY);
+  // Newly created task will free this.
+  AudioDrainArgs* drain_args = new AudioDrainArgs{
+      .sink = sink,
+      .command = command,
+  };
+
+  ESP_LOGI(kTag, "starting audio drain task");
+  xTaskCreatePinnedToCore(&AudioDrainMain, "drain", kDrainStackSize, drain_args,
+                          kTaskPriorityAudio, NULL, kAudioCore);
 }
 
 void AudioTaskMain(void* args) {
@@ -57,10 +66,11 @@ void AudioTaskMain(void* args) {
   {
     AudioTaskArgs* real_args = reinterpret_cast<AudioTaskArgs*>(args);
     std::unique_ptr<Pipeline> pipeline(real_args->pipeline);
-    QueueHandle_t input;
+    IAudioSink* sink = real_args->sink;
-    StreamBufferHandle_t output;
     delete real_args;
 
+    std::optional<StreamInfo::Format> output_format;
+
     std::vector<Pipeline*> elements = pipeline->GetIterationOrder();
     std::size_t max_inputs =
         (*std::max_element(elements.begin(), elements.end(),
@@ -74,9 +84,7 @@ void AudioTaskMain(void* args) {
     std::vector<MappableRegion<kPipelineBufferSize>> in_regions(max_inputs);
     MappableRegion<kPipelineBufferSize> out_region;
     std::for_each(in_regions.begin(), in_regions.end(),
-                  [](const MappableRegion<kBufferSize>& region) {
+                  [](const auto& region) { assert(region.is_valid); });
-                    assert(region.is_valid);
-                  });
     assert(out_region.is_valid);
 
     // Each element has exactly one output buffer.
@@ -90,55 +98,68 @@ void AudioTaskMain(void* args) {
     bool playing = true;
     bool quit = false;
     while (!quit) {
-      // TODO: full event here?
-      Command cmd;
-      bool has_cmd = xQueueReceive(input, &cmd, 0);
-      if (has_cmd) {
-        switch (cmd) {
-          case PLAY:
-            playing = true;
-            break;
-          case PAUSE:
-            playing = false;
-            break;
-          case QUIT:
-            quit = true;
-            break;
-        }
-      }
-      if (quit) {
-        break;
-      }
-
       if (playing) {
         for (int i = 0; i < elements.size(); i++) {
-          std::vector<MutableStream> in_streams;
+          std::vector<RawStream> raw_in_streams;
-          elements.at(i)->InStreams(&in_regions, &in_streams);
+          elements.at(i)->InStreams(&in_regions, &raw_in_streams);
-          MutableStream out_stream = elements.at(i)->OutStream(&out_region);
+          RawStream raw_out_stream = elements.at(i)->OutStream(&out_region);
 
           // Crop the input and output streams to the ranges that are safe to
           // touch. For the input streams, this is the region that contains
           // data. For the output stream, this is the region that does *not*
           // already contain data.
-          std::vector<Stream> cropped_in_streams;
+          std::vector<InputStream> in_streams;
-          std::for_each(in_streams.begin(), in_streams.end(),
+          std::for_each(raw_in_streams.begin(), raw_in_streams.end(),
-                        [&](MutableStream& s) {
+                        [&](RawStream& s) { in_streams.emplace_back(&s); });
-                          cropped_in_streams.emplace_back(
+          OutputStream out_stream(&raw_out_stream);
-                              s.info, s.data.first(s.info->bytes_in_stream));
-                        });
 
-          elements.at(i)->OutputElement()->Process(&cropped_in_streams,
+          elements.at(i)->OutputElement()->Process(in_streams, &out_stream);
-                                                   &out_stream);
+        }
 
-          for (int stream = 0; stream < in_streams.size(); stream++) {
+        RawStream raw_sink_stream = elements.back()->OutStream(&out_region);
-            MutableStream& orig_stream = in_streams.at(stream);
+        InputStream sink_stream(&raw_sink_stream);
-            Stream& cropped_stream = cropped_in_streams.at(stream);
+
-            std::move(cropped_stream.data.begin(), cropped_stream.data.end(),
+        if (!output_format || output_format != sink_stream.info().format) {
-                      orig_stream.data.begin());
+          // The format of the stream within the sink stream has changed. We
-            orig_stream.info->bytes_in_stream =
+          // need to reconfigure the sink, but shouldn't do so until we've fully
-                cropped_stream.data.size_bytes();
+          // drained the current buffer.
+          if (xStreamBufferIsEmpty(sink->buffer())) {
+            output_format = sink_stream.info().format;
+            sink->Configure(*output_format);
           }
         }
+
+        // We've reconfigured the sink, or it was already configured correctly.
+        // Send through some data.
+        if (output_format == sink_stream.info().format) {
+          // TODO: tune the delay on this, as it's currently the only way to
+          // throttle this task's CPU time. Maybe also hold off on the pipeline
+          // if the buffer is already close to full?
+          std::size_t sent = xStreamBufferSend(
+              sink->buffer(), sink_stream.data().data(),
+              sink_stream.data().size_bytes(), pdMS_TO_TICKS(10));
+          sink_stream.consume(sent);
+        }
+      }
+    }
+  }
+  vTaskDelete(NULL);
+}
+
+void AudioDrainMain(void* args) {
+  {
+    AudioDrainArgs* real_args = reinterpret_cast<AudioDrainArgs*>(args);
+    IAudioSink* sink = real_args->sink;
+    std::atomic<Command>* command = real_args->command;
+    delete real_args;
+
+    // TODO(jacqueline): implement PAUSE without busy-waiting.
+    while (*command != QUIT) {
+      std::byte buf[64];
+      std::size_t len =
+          xStreamBufferReceive(sink->buffer(), buf, sizeof(buf), portMAX_DELAY);
+      if (len > 0) {
+        sink->Send({buf, len});
       }
     }
   }

src/audio/fatfs_audio_input.cpp

@@ -44,26 +44,30 @@ auto FatfsAudioInput::OpenFile(const std::string& path) -> void {
   is_file_open_ = true;
 }
 
-auto FatfsAudioInput::Process(std::vector<Stream>* inputs,
+auto FatfsAudioInput::Process(const std::vector<InputStream>& inputs,
-                              MutableStream* output) -> void {
+                              OutputStream* output) -> void {
   if (!is_file_open_) {
     return;
   }
 
+  StreamInfo::Format format = StreamInfo::Encoded{codecs::STREAM_MP3};
+  if (!output->prepare(format)) {
+    return;
+  }
+
+  std::size_t max_size = output->data().size_bytes();
+  std::size_t size = 0;
   FRESULT result =
-      f_read(&current_file_, output->data.data(), output->data.size_bytes(),
+      f_read(&current_file_, output->data().data(), max_size, &size);
-             &output->info->bytes_in_stream);
   if (result != FR_OK) {
     ESP_LOGE(kTag, "file I/O error %d", result);
     // TODO(jacqueline): Handle errors.
     return;
   }
 
-  // TODO: read from filename?
+  output->add(size);
-  output->info->data = StreamInfo::Encoded{codecs::STREAM_MP3};
 
-  if (output->info->bytes_in_stream < output->data.size_bytes() ||
+  if (size < max_size || f_eof(&current_file_)) {
-      f_eof(&current_file_)) {
     f_close(&current_file_);
     is_file_open_ = false;
   }

src/audio/i2s_audio_output.cpp

@@ -1,6 +1,7 @@
 #include "i2s_audio_output.hpp"
 
 #include <algorithm>
+#include <cstddef>
 #include <variant>
 
 #include "esp_err.h"
@@ -18,7 +19,7 @@ static const char* kTag = "I2SOUT";
 namespace audio {
 
 auto I2SAudioOutput::create(drivers::GpioExpander* expander)
-    -> cpp::result<std::shared_ptr<I2SAudioOutput>, Error> {
+    -> cpp::result<std::unique_ptr<I2SAudioOutput>, Error> {
   // First, we need to perform initial configuration of the DAC chip.
   auto dac_result = drivers::AudioDac::create(expander);
   if (dac_result.has_error()) {
@@ -32,7 +33,7 @@ auto I2SAudioOutput::create(drivers::GpioExpander* expander)
   // dac->WriteVolume(255);
   dac->WriteVolume(120);  // for testing
 
-  return std::make_shared<I2SAudioOutput>(expander, std::move(dac));
+  return std::make_unique<I2SAudioOutput>(expander, std::move(dac));
 }
 
 I2SAudioOutput::I2SAudioOutput(drivers::GpioExpander* expander,
@@ -41,18 +42,18 @@ I2SAudioOutput::I2SAudioOutput(drivers::GpioExpander* expander,
 
 I2SAudioOutput::~I2SAudioOutput() {}
 
-auto I2SAudioOutput::ProcessStreamInfo(const StreamInfo& info) -> bool {
+auto I2SAudioOutput::Configure(const StreamInfo::Format& format) -> bool {
-  if (!std::holds_alternative<StreamInfo::Pcm>(info.data)) {
+  if (!std::holds_alternative<StreamInfo::Pcm>(format)) {
     return false;
   }
 
-  StreamInfo::Pcm pcm = std::get<StreamInfo::Pcm>(info.data);
+  StreamInfo::Pcm pcm = std::get<StreamInfo::Pcm>(format);
 
   if (current_config_ && pcm == *current_config_) {
     return true;
   }
 
-  ESP_LOGI(kTag, "incoming audio stream: %u bpp @ %u Hz", pcm.bits_per_sample,
+  ESP_LOGI(kTag, "incoming audio stream: %u bpp @ %lu Hz", pcm.bits_per_sample,
            pcm.sample_rate);
 
   drivers::AudioDac::BitsPerSample bps;
@@ -92,14 +93,8 @@ auto I2SAudioOutput::ProcessStreamInfo(const StreamInfo& info) -> bool {
   return true;
 }
 
-auto I2SAudioOutput::Process(std::vector<Stream>* inputs, MutableStream* output)
+auto I2SAudioOutput::Send(const cpp::span<std::byte>& data) -> void {
-    -> void {
+  dac_->WriteData(data);
-  std::for_each(inputs->begin(), inputs->end(), [&](Stream& s) {
-    if (ProcessStreamInfo(s.info)) {
-      std::size_t bytes_written = dac_->WriteData(s.data);
-      s.data = s.data.subspan(bytes_written);
-    }
-  });
 }
 
 auto I2SAudioOutput::SetVolume(uint8_t volume) -> void {

src/audio/include/audio_decoder.hpp

@@ -24,7 +24,7 @@ class AudioDecoder : public IAudioElement {
   AudioDecoder();
   ~AudioDecoder();
 
-  auto Process(std::vector<Stream>* inputs, MutableStream* output)
+  auto Process(const std::vector<InputStream>& inputs, OutputStream* output)
       -> void override;
 
   AudioDecoder(const AudioDecoder&) = delete;
@@ -32,11 +32,9 @@ class AudioDecoder : public IAudioElement {
 
  private:
   std::unique_ptr<codecs::ICodec> current_codec_;
-  std::optional<StreamInfo> stream_info_;
+  std::optional<StreamInfo::Format> current_input_format_;
-
+  std::optional<StreamInfo::Format> current_output_format_;
-  bool has_set_stream_info_;
   bool has_samples_to_send_;
-  bool needs_more_input_;
 
   auto ProcessStreamInfo(const StreamInfo& info) -> bool;
 };

src/audio/include/audio_element.hpp

@@ -37,11 +37,11 @@ static const size_t kEventQueueSize = 8;
  */
 class IAudioElement {
  public:
-  IAudioElement();
+  IAudioElement() {}
-  virtual ~IAudioElement();
+  virtual ~IAudioElement() {}
 
-  virtual auto Process(std::vector<Stream>* inputs, MutableStream* output)
+  virtual auto Process(const std::vector<InputStream>& inputs,
-      -> void = 0;
+                       OutputStream* output) -> void = 0;
 };
 
 }  // namespace audio

src/audio/include/audio_playback.hpp

@@ -8,6 +8,7 @@
 #include "audio_task.hpp"
 #include "esp_err.h"
 #include "fatfs_audio_input.hpp"
+#include "i2s_audio_output.hpp"
 #include "result.hpp"
 #include "span.hpp"
 
@@ -28,7 +29,7 @@ class AudioPlayback {
   static auto create(drivers::GpioExpander* expander)
       -> cpp::result<std::unique_ptr<AudioPlayback>, Error>;
 
-  AudioPlayback(FatfsAudioInput *file_input);
+  explicit AudioPlayback(std::unique_ptr<I2SAudioOutput> output);
   ~AudioPlayback();
 
   /*
@@ -44,10 +45,9 @@ class AudioPlayback {
   AudioPlayback& operator=(const AudioPlayback&) = delete;
 
  private:
-  FatfsAudioInput *file_source;
+  std::unique_ptr<FatfsAudioInput> file_source_;
-
+  std::unique_ptr<I2SAudioOutput> i2s_output_;
-  std::vector<std::unique_ptr<IAudioElement>> all_elements_;
+  std::vector<std::unique_ptr<IAudioElement>> elements_;
-  std::unique_ptr<task::Handle> pipeline_;
 };
 
 }  // namespace audio

src/audio/include/audio_sink.hpp

@@ -0,0 +1,22 @@
+#pragma once
+
+#include "audio_element.hpp"
+#include "stream_info.hpp"
+namespace audio {
+
+class IAudioSink {
+ private:
+  static const std::size_t kDrainBufferSize = 8 * 1024;
+  StreamBufferHandle_t buffer_;
+
+ public:
+  IAudioSink() : buffer_(xStreamBufferCreate(kDrainBufferSize, 1)) {}
+  virtual ~IAudioSink() { vStreamBufferDelete(buffer_); }
+
+  virtual auto Configure(const StreamInfo::Format& format) -> bool = 0;
+  virtual auto Send(const cpp::span<std::byte>& data) -> void = 0;
+
+  auto buffer() const -> StreamBufferHandle_t { return buffer_; }
+};
+
+}  // namespace audio

src/audio/include/audio_task.hpp

@@ -5,38 +5,32 @@
 #include <string>
 
 #include "audio_element.hpp"
+#include "audio_sink.hpp"
+#include "dac.hpp"
 #include "freertos/portmacro.h"
 #include "pipeline.hpp"
+#include "stream_buffer.hpp"
 
 namespace audio {
 
 namespace task {
-struct AudioTaskArgs {
-  Pipeline* pipeline;
-  QueueHandle_t input;
-};
-
-extern "C" void AudioTaskMain(void* args);
 
 enum Command { PLAY, PAUSE, QUIT };
 
-class Handle {
+struct AudioTaskArgs {
- public:
+  Pipeline* pipeline;
-  explicit Handle(QueueHandle_t input);
+  IAudioSink* sink;
-  ~Handle();
+};
-
+struct AudioDrainArgs {
-  auto SetStreamInfo() -> void;
+  IAudioSink* sink;
-  auto Play() -> void;
+  std::atomic<Command>* command;
-  auto Pause() -> void;
-  auto Quit() -> void;
-
-  auto OutputBuffer() -> StreamBufferHandle_t;
-
- private:
-  QueueHandle_t input;
 };
 
-auto Start(Pipeline* pipeline) -> Handle*;
+extern "C" void AudioTaskMain(void* args);
+extern "C" void AudioDrainMain(void* args);
+
+auto StartPipeline(Pipeline* pipeline, IAudioSink* sink) -> void;
+auto StartDrain(IAudioSink* sink) -> void;
 
 }  // namespace task
 

src/audio/include/fatfs_audio_input.hpp

@@ -16,6 +16,7 @@
 
 #include "audio_element.hpp"
 #include "stream_buffer.hpp"
+#include "stream_info.hpp"
 
 namespace audio {
 
@@ -26,7 +27,7 @@ class FatfsAudioInput : public IAudioElement {
 
   auto OpenFile(const std::string& path) -> void;
 
-  auto Process(std::vector<Stream>* inputs, MutableStream* output)
+  auto Process(const std::vector<InputStream>& inputs, OutputStream* output)
       -> void override;
 
   FatfsAudioInput(const FatfsAudioInput&) = delete;

src/audio/include/i2s_audio_output.hpp

@@ -5,6 +5,7 @@
 #include <vector>
 
 #include "audio_element.hpp"
+#include "audio_sink.hpp"
 #include "chunk.hpp"
 #include "result.hpp"
 
@@ -14,18 +15,18 @@
 
 namespace audio {
 
-class I2SAudioOutput : public IAudioElement {
+class I2SAudioOutput : public IAudioSink {
  public:
   enum Error { DAC_CONFIG, I2S_CONFIG, STREAM_INIT };
   static auto create(drivers::GpioExpander* expander)
-      -> cpp::result<std::shared_ptr<I2SAudioOutput>, Error>;
+      -> cpp::result<std::unique_ptr<I2SAudioOutput>, Error>;
 
   I2SAudioOutput(drivers::GpioExpander* expander,
                  std::unique_ptr<drivers::AudioDac> dac);
   ~I2SAudioOutput();
 
-  auto Process(std::vector<Stream>* inputs, MutableStream* output)
+  auto Configure(const StreamInfo::Format& format) -> bool override;
-      -> void override;
+  auto Send(const cpp::span<std::byte>& data) -> void override;
 
   I2SAudioOutput(const I2SAudioOutput&) = delete;
   I2SAudioOutput& operator=(const I2SAudioOutput&) = delete;
@@ -37,8 +38,6 @@ class I2SAudioOutput : public IAudioElement {
   std::unique_ptr<drivers::AudioDac> dac_;
 
   std::optional<StreamInfo::Pcm> current_config_;
-
-  auto ProcessStreamInfo(const StreamInfo& info) -> bool;
 };
 
 }  // namespace audio

src/audio/include/pipeline.hpp

@@ -3,6 +3,7 @@
 #include <memory>
 #include <optional>
 #include <string>
+#include <vector>
 
 #include "freertos/portmacro.h"
 
@@ -16,7 +17,7 @@ static const std::size_t kPipelineBufferSize = 32 * 1024;
 
 class Pipeline {
  public:
-  Pipeline(IAudioElement* output);
+  explicit Pipeline(IAudioElement* output);
   ~Pipeline();
   auto AddInput(IAudioElement* input) -> Pipeline*;
 
@@ -25,9 +26,9 @@ class Pipeline {
   auto NumInputs() const -> std::size_t;
 
   auto InStreams(std::vector<MappableRegion<kPipelineBufferSize>>*,
-                 std::vector<MutableStream>*) -> void;
+                 std::vector<RawStream>*) -> void;
 
-  auto OutStream(MappableRegion<kPipelineBufferSize>*) -> MutableStream;
+  auto OutStream(MappableRegion<kPipelineBufferSize>*) -> RawStream;
 
   auto GetIterationOrder() -> std::vector<Pipeline*>;
 

src/audio/include/stream_info.hpp

@@ -4,6 +4,8 @@
 #include <optional>
 #include <string>
 #include <string_view>
+#include <type_traits>
+#include <utility>
 #include <variant>
 
 #include "result.hpp"
@@ -35,37 +37,82 @@ struct StreamInfo {
     // Number of bits per sample.
     uint8_t bits_per_sample;
     // The sample rate.
-    uint16_t sample_rate;
+    uint32_t sample_rate;
 
     bool operator==(const Pcm&) const = default;
   };
 
-  std::variant<Encoded, Pcm> data;
+  typedef std::variant<Encoded, Pcm> Format;
+  Format format;
 
   bool operator==(const StreamInfo&) const = default;
 };
 
-class MutableStream {
+class RawStream {
  public:
   StreamInfo* info;
   cpp::span<std::byte> data;
+  bool is_incomplete;
 
-  MutableStream(StreamInfo* i, cpp::span<std::byte> d)
+  RawStream(StreamInfo* i, cpp::span<std::byte> d)
-      : info(i), data(d) {}
+      : info(i), data(d), is_incomplete(false) {}
 };
 
 /*
  * A byte buffer + associated metadata, which is not allowed to modify any of
  * the underlying data.
  */
-class Stream {
+class InputStream {
  public:
-  explicit Stream(const MutableStream& s) : info(*s.info), data(s.data) {}
+  explicit InputStream(RawStream* s) : raw_(s) {}
 
-  const StreamInfo& info;
+  void consume(std::size_t bytes) const {
-  // `data` itself left mutable for signalling how much of the stream was
+    auto new_data = raw_->data.subspan(bytes);
-  // consumed
+    std::move(new_data.begin(), new_data.end(), raw_->data.begin());
-  cpp::span<const std::byte> data;
+    raw_->info->bytes_in_stream = new_data.size_bytes();
+  }
+
+  void mark_incomplete() const { raw_->is_incomplete = true; }
+
+  const StreamInfo& info() const { return *raw_->info; }
+
+  cpp::span<const std::byte> data() const {
+    return raw_->data.first(raw_->info->bytes_in_stream);
+  }
+
+ private:
+  RawStream* raw_;
+};
+
+class OutputStream {
+ public:
+  explicit OutputStream(RawStream* s) : raw_(s) {}
+
+  void add(std::size_t bytes) const { raw_->info->bytes_in_stream += bytes; }
+
+  bool prepare(const StreamInfo::Format& new_format) {
+    if (new_format == raw_->info->format) {
+      raw_->info->format = new_format;
+      return true;
+    }
+    if (raw_->is_incomplete) {
+      raw_->info->format = new_format;
+      raw_->info->bytes_in_stream = 0;
+      return true;
+    }
+    return false;
+  }
+
+  const StreamInfo& info() const { return *raw_->info; }
+
+  cpp::span<std::byte> data() const {
+    return raw_->data.subspan(raw_->info->bytes_in_stream);
+  }
+
+  bool is_incomplete() const { return raw_->is_incomplete; }
+
+ private:
+  RawStream* raw_;
 };
 
 }  // namespace audio

src/audio/pipeline.cpp

@@ -1,4 +1,5 @@
 #include "pipeline.hpp"
+#include <memory>
 #include "stream_info.hpp"
 
 namespace audio {
@@ -7,7 +8,7 @@ Pipeline::Pipeline(IAudioElement* output) : root_(output), subtrees_() {}
 Pipeline::~Pipeline() {}
 
 auto Pipeline::AddInput(IAudioElement* input) -> Pipeline* {
-  subtrees_.emplace_back(input);
+  subtrees_.push_back(std::make_unique<Pipeline>(input));
   return subtrees_.back().get();
 }
 
@@ -21,15 +22,15 @@ auto Pipeline::NumInputs() const -> std::size_t {
 
 auto Pipeline::InStreams(
     std::vector<MappableRegion<kPipelineBufferSize>>* regions,
-    std::vector<MutableStream>* out) -> void {
+    std::vector<RawStream>* out) -> void {
   for (int i = 0; i < subtrees_.size(); i++) {
-    MutableStream s = subtrees_[i]->OutStream(&regions->at(i));
+    RawStream s = subtrees_[i]->OutStream(&regions->at(i));
     out->push_back(s);
   }
 }
 
 auto Pipeline::OutStream(MappableRegion<kPipelineBufferSize>* region)
-    -> MutableStream {
+    -> RawStream {
   return {&output_info_, region->Map(output_buffer_)};
 }
 
@@ -42,8 +43,9 @@ auto Pipeline::GetIterationOrder() -> std::vector<Pipeline*> {
     to_search.pop_back();
     found.push_back(current);
 
-    to_search.insert(to_search.end(), current->subtrees_.begin(),
+    for (const auto& i : current->subtrees_) {
-                     current->subtrees_.end());
+      to_search.push_back(i.get());
+    }
   }
 
   return found;

src/codecs/codec.cpp

@@ -5,7 +5,7 @@
 
 namespace codecs {
 
-auto CreateCodecForFile(const std::string& file)
+auto CreateCodecForType(StreamType type)
     -> cpp::result<std::unique_ptr<ICodec>, CreateCodecError> {
   return std::make_unique<MadMp3Decoder>();  // TODO.
 }

src/codecs/include/mad.hpp

@@ -17,10 +17,10 @@ class MadMp3Decoder : public ICodec {
   MadMp3Decoder();
   ~MadMp3Decoder();
 
-  auto CanHandleFile(const std::string& path) -> bool override;
+  auto CanHandleType(StreamType type) -> bool override;
   auto GetOutputFormat() -> OutputFormat override;
   auto ResetForNewStream() -> void override;
-  auto SetInput(cpp::span<std::byte> input) -> void override;
+  auto SetInput(cpp::span<const std::byte> input) -> void override;
   auto GetInputPosition() -> std::size_t override;
   auto ProcessNextFrame() -> cpp::result<bool, ProcessingError> override;
   auto WriteOutputSamples(cpp::span<std::byte> output)

src/codecs/mad.cpp

@@ -5,6 +5,7 @@
 #include "mad.h"
 
 #include "codec.hpp"
+#include "types.hpp"
 
 namespace codecs {
 
@@ -35,8 +36,8 @@ MadMp3Decoder::~MadMp3Decoder() {
   mad_header_finish(&header_);
 }
 
-auto MadMp3Decoder::CanHandleFile(const std::string& path) -> bool {
+auto MadMp3Decoder::CanHandleType(StreamType type) -> bool {
-  return true;  // TODO.
+  return type == STREAM_MP3;
 }
 
 auto MadMp3Decoder::GetOutputFormat() -> OutputFormat {
@@ -52,7 +53,7 @@ auto MadMp3Decoder::ResetForNewStream() -> void {
   has_decoded_header_ = false;
 }
 
-auto MadMp3Decoder::SetInput(cpp::span<std::byte> input) -> void {
+auto MadMp3Decoder::SetInput(cpp::span<const std::byte> input) -> void {
   mad_stream_buffer(&stream_,
                     reinterpret_cast<const unsigned char*>(input.data()),
                     input.size());

src/drivers/dac.cpp

@@ -192,15 +192,13 @@ auto AudioDac::Reconfigure(BitsPerSample bps, SampleRate rate) -> void {
   WriteRegister(Register::POWER_MODE, 0);
 }
 
-auto AudioDac::WriteData(const cpp::span<const std::byte>& data)
+auto AudioDac::WriteData(const cpp::span<const std::byte>& data) -> void {
-    -> std::size_t {
   std::size_t bytes_written = 0;
   esp_err_t err = i2s_channel_write(i2s_handle_, data.data(), data.size_bytes(),
-                                    &bytes_written, 0);
+                                    &bytes_written, portMAX_DELAY);
   if (err != ESP_ERR_TIMEOUT) {
     ESP_ERROR_CHECK(err);
   }
-  return bytes_written;
 }
 
 auto AudioDac::Stop() -> void {

src/drivers/include/dac.hpp

@@ -71,7 +71,7 @@ class AudioDac {
   // TODO(jacqueline): worth supporting channels here as well?
   auto Reconfigure(BitsPerSample bps, SampleRate rate) -> void;
 
-  auto WriteData(const cpp::span<const std::byte>& data) -> std::size_t;
+  auto WriteData(const cpp::span<const std::byte>& data) -> void;
 
   auto Stop() -> void;
   auto LogStatus() -> void;

src/main/main.cpp

@@ -133,10 +133,10 @@ extern "C" void app_main(void) {
                                 (void*)lvglArgs, 1, sLvglStack,
                                 &sLvglTaskBuffer, 1);
 
-  std::shared_ptr<audio::AudioPlayback> playback;
+  std::unique_ptr<audio::AudioPlayback> playback;
   if (storage) {
     ESP_LOGI(TAG, "Init audio pipeline");
-    auto playback_res = audio::AudioPlayback::create(expander, storage);
+    auto playback_res = audio::AudioPlayback::create(expander);
     if (playback_res.has_error()) {
       ESP_LOGE(TAG, "Failed! Playback will not work.");
     } else {

src/memory/include/himem.hpp

@@ -63,7 +63,7 @@ class MappableRegion {
     return {bytes_, size};
   }
 
-  auto Map(const HimemAlloc<size> &alloc) -> cpp::span<std::byte> {
+  auto Map(const HimemAlloc<size>& alloc) -> cpp::span<std::byte> {
     if (bytes_ != nullptr) {
       ESP_ERROR_CHECK(esp_himem_unmap(range_handle, bytes_, size));
     }