1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
|
/*
* Copyright 2023 jacqueline <me@jacqueline.id.au>
*
* SPDX-License-Identifier: GPL-3.0-only
*/
#include "audio_task.hpp"
#include <stdlib.h>
#include <algorithm>
#include <cstddef>
#include <cstdint>
#include <deque>
#include <memory>
#include <variant>
#include "audio_events.hpp"
#include "audio_fsm.hpp"
#include "audio_sink.hpp"
#include "cbor.h"
#include "esp_err.h"
#include "esp_heap_caps.h"
#include "esp_log.h"
#include "event_queue.hpp"
#include "freertos/portmacro.h"
#include "freertos/projdefs.h"
#include "freertos/queue.h"
#include "pipeline.hpp"
#include "span.hpp"
#include "arena.hpp"
#include "audio_element.hpp"
#include "chunk.hpp"
#include "stream_event.hpp"
#include "stream_info.hpp"
#include "stream_message.hpp"
#include "sys/_stdint.h"
#include "tasks.hpp"
namespace audio {
namespace task {
static const char* kTag = "task";
// The default amount of time to wait between pipeline iterations for a single
// track.
static constexpr uint_fast16_t kDefaultDelayTicks = pdMS_TO_TICKS(5);
static constexpr uint_fast16_t kMaxDelayTicks = pdMS_TO_TICKS(10);
static constexpr uint_fast16_t kMinDelayTicks = pdMS_TO_TICKS(1);
void AudioTaskMain(std::unique_ptr<Pipeline> pipeline, IAudioSink* sink) {
// The stream format for bytes currently in the sink buffer.
std::optional<StreamInfo::Format> output_format;
// How long to wait between pipeline iterations. This is reset for each track,
// and readjusted on the fly to maintain a reasonable amount playback buffer.
// Buffering too much will mean we process samples inefficiently, wasting CPU
// time, whilst buffering too little will affect the quality of the output.
uint_fast16_t delay_ticks = kDefaultDelayTicks;
std::vector<Pipeline*> all_elements = pipeline->GetIterationOrder();
float current_sample_in_second = 0;
uint32_t previous_second = 0;
uint32_t current_second = 0;
bool previously_had_work = false;
events::EventQueue& event_queue = events::EventQueue::GetInstance();
while (1) {
// First, see if we actually have any pipeline work to do in this iteration.
bool has_work = false;
// We always have work to do if there's still bytes to be sunk.
has_work = all_elements.back()->OutStream().info->bytes_in_stream > 0;
if (!has_work) {
for (Pipeline* p : all_elements) {
has_work = p->OutputElement()->NeedsToProcess();
if (has_work) {
break;
}
}
}
if (!has_work) {
has_work = !xStreamBufferIsEmpty(sink->buffer());
}
if (previously_had_work && !has_work) {
events::Dispatch<AudioPipelineIdle, AudioState>({});
}
previously_had_work = has_work;
// See if there's any new events.
event_queue.ServiceAudio(has_work ? delay_ticks : portMAX_DELAY);
if (!has_work) {
// See if we've been given work by this event.
for (Pipeline* p : all_elements) {
has_work = p->OutputElement()->NeedsToProcess();
if (has_work) {
delay_ticks = kDefaultDelayTicks;
break;
}
}
if (!has_work) {
continue;
}
}
// We have work to do! Allow each element in the pipeline to process one
// chunk. We iterate from input nodes first, so this should result in
// samples in the output buffer.
for (int i = 0; i < all_elements.size(); i++) {
std::vector<RawStream> raw_in_streams;
all_elements.at(i)->InStreams(&raw_in_streams);
RawStream raw_out_stream = all_elements.at(i)->OutStream();
// 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<InputStream> in_streams;
std::for_each(raw_in_streams.begin(), raw_in_streams.end(),
[&](RawStream& s) { in_streams.emplace_back(&s); });
OutputStream out_stream(&raw_out_stream);
all_elements.at(i)->OutputElement()->Process(in_streams, &out_stream);
}
RawStream raw_sink_stream = all_elements.back()->OutStream();
InputStream sink_stream(&raw_sink_stream);
if (sink_stream.info().bytes_in_stream == 0) {
if (sink_stream.is_producer_finished()) {
sink_stream.mark_consumer_finished();
current_second = 0;
previous_second = 0;
current_sample_in_second = 0;
} else {
// The user is probably about to hear a skip :(
ESP_LOGW(kTag, "!! audio sink is underbuffered !!");
}
// No new bytes to sink, so skip sinking completely.
continue;
}
if (!output_format || output_format != sink_stream.info().format) {
// The format of the stream within the sink stream has changed. We
// need to reconfigure the sink, but shouldn't do so until we've fully
// drained the current buffer.
if (xStreamBufferIsEmpty(sink->buffer())) {
ESP_LOGI(kTag, "reconfiguring dac");
output_format = sink_stream.info().format;
sink->Configure(*output_format);
} else {
ESP_LOGI(kTag, "waiting to reconfigure");
continue;
}
}
// We've reconfigured the sink, or it was already configured correctly.
// Send through some data.
std::size_t bytes_sunk =
xStreamBufferSend(sink->buffer(), sink_stream.data().data(),
sink_stream.data().size_bytes(), 0);
if (std::holds_alternative<StreamInfo::Pcm>(*output_format)) {
StreamInfo::Pcm pcm = std::get<StreamInfo::Pcm>(*output_format);
float samples_sunk = bytes_sunk;
samples_sunk /= pcm.channels;
// Samples must be aligned to 16 bits. The number of actual bytes per
// sample is therefore the bps divided by 16, rounded up (align to word),
// times two (convert to bytes).
uint8_t bytes_per_sample = ((pcm.bits_per_sample + 16 - 1) / 16) * 2;
samples_sunk /= bytes_per_sample;
current_sample_in_second += samples_sunk;
while (current_sample_in_second >= pcm.sample_rate) {
current_second++;
current_sample_in_second -= pcm.sample_rate;
}
if (previous_second != current_second) {
events::Dispatch<PlaybackUpdate, AudioState>(
{.seconds_elapsed = current_second});
}
previous_second = current_second;
}
// Adjust how long we wait for the next iteration if we're getting too far
// ahead or behind.
float sunk_percent = static_cast<float>(bytes_sunk) /
static_cast<float>(sink_stream.info().bytes_in_stream);
if (sunk_percent > 0.66f) {
// We're sinking a lot of the output buffer per iteration, so we need to
// be running faster.
delay_ticks--;
} else if (sunk_percent < 0.33f) {
// We're not sinking much of the output buffer per iteration, so we can
// slow down to save some cycles.
delay_ticks++;
}
delay_ticks = std::clamp(delay_ticks, kMinDelayTicks, kMaxDelayTicks);
// Finally, actually mark the bytes we sunk as consumed.
if (bytes_sunk > 0) {
sink_stream.consume(bytes_sunk);
}
}
}
auto StartPipeline(Pipeline* pipeline, IAudioSink* sink) -> void {
ESP_LOGI(kTag, "starting audio pipeline task");
tasks::StartPersistent<tasks::Type::kAudio>(
[=]() { AudioTaskMain(std::unique_ptr<Pipeline>(pipeline), sink); });
}
} // namespace task
} // namespace audio
|
