|
1 /* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ |
|
2 /* This Source Code Form is subject to the terms of the Mozilla Public |
|
3 * License, v. 2.0. If a copy of the MPL was not distributed with this file, |
|
4 * You can obtain one at http://mozilla.org/MPL/2.0/. */ |
|
5 |
|
6 #ifndef MOZILLA_AUDIOSEGMENT_H_ |
|
7 #define MOZILLA_AUDIOSEGMENT_H_ |
|
8 |
|
9 #include "MediaSegment.h" |
|
10 #include "AudioSampleFormat.h" |
|
11 #include "SharedBuffer.h" |
|
12 #include "WebAudioUtils.h" |
|
13 #ifdef MOZILLA_INTERNAL_API |
|
14 #include "mozilla/TimeStamp.h" |
|
15 #endif |
|
16 |
|
17 namespace mozilla { |
|
18 |
|
19 template<typename T> |
|
20 class SharedChannelArrayBuffer : public ThreadSharedObject { |
|
21 public: |
|
22 SharedChannelArrayBuffer(nsTArray<nsTArray<T> >* aBuffers) |
|
23 { |
|
24 mBuffers.SwapElements(*aBuffers); |
|
25 } |
|
26 |
|
27 virtual size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const MOZ_OVERRIDE |
|
28 { |
|
29 size_t amount = 0; |
|
30 amount += mBuffers.SizeOfExcludingThis(aMallocSizeOf); |
|
31 for (size_t i = 0; i < mBuffers.Length(); i++) { |
|
32 amount += mBuffers[i].SizeOfExcludingThis(aMallocSizeOf); |
|
33 } |
|
34 |
|
35 return amount; |
|
36 } |
|
37 |
|
38 virtual size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const MOZ_OVERRIDE |
|
39 { |
|
40 return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); |
|
41 } |
|
42 |
|
43 nsTArray<nsTArray<T> > mBuffers; |
|
44 }; |
|
45 |
|
46 class AudioStream; |
|
47 class AudioMixer; |
|
48 |
|
49 /** |
|
50 * For auto-arrays etc, guess this as the common number of channels. |
|
51 */ |
|
52 const int GUESS_AUDIO_CHANNELS = 2; |
|
53 |
|
54 // We ensure that the graph advances in steps that are multiples of the Web |
|
55 // Audio block size |
|
56 const uint32_t WEBAUDIO_BLOCK_SIZE_BITS = 7; |
|
57 const uint32_t WEBAUDIO_BLOCK_SIZE = 1 << WEBAUDIO_BLOCK_SIZE_BITS; |
|
58 |
|
59 void InterleaveAndConvertBuffer(const void** aSourceChannels, |
|
60 AudioSampleFormat aSourceFormat, |
|
61 int32_t aLength, float aVolume, |
|
62 int32_t aChannels, |
|
63 AudioDataValue* aOutput); |
|
64 |
|
65 /** |
|
66 * Given an array of input channels (aChannelData), downmix to aOutputChannels, |
|
67 * interleave the channel data. A total of aOutputChannels*aDuration |
|
68 * interleaved samples will be copied to a channel buffer in aOutput. |
|
69 */ |
|
70 void DownmixAndInterleave(const nsTArray<const void*>& aChannelData, |
|
71 AudioSampleFormat aSourceFormat, int32_t aDuration, |
|
72 float aVolume, uint32_t aOutputChannels, |
|
73 AudioDataValue* aOutput); |
|
74 |
|
75 /** |
|
76 * An AudioChunk represents a multi-channel buffer of audio samples. |
|
77 * It references an underlying ThreadSharedObject which manages the lifetime |
|
78 * of the buffer. An AudioChunk maintains its own duration and channel data |
|
79 * pointers so it can represent a subinterval of a buffer without copying. |
|
80 * An AudioChunk can store its individual channels anywhere; it maintains |
|
81 * separate pointers to each channel's buffer. |
|
82 */ |
|
83 struct AudioChunk { |
|
84 typedef mozilla::AudioSampleFormat SampleFormat; |
|
85 |
|
86 // Generic methods |
|
87 void SliceTo(TrackTicks aStart, TrackTicks aEnd) |
|
88 { |
|
89 NS_ASSERTION(aStart >= 0 && aStart < aEnd && aEnd <= mDuration, |
|
90 "Slice out of bounds"); |
|
91 if (mBuffer) { |
|
92 MOZ_ASSERT(aStart < INT32_MAX, "Can't slice beyond 32-bit sample lengths"); |
|
93 for (uint32_t channel = 0; channel < mChannelData.Length(); ++channel) { |
|
94 mChannelData[channel] = AddAudioSampleOffset(mChannelData[channel], |
|
95 mBufferFormat, int32_t(aStart)); |
|
96 } |
|
97 } |
|
98 mDuration = aEnd - aStart; |
|
99 } |
|
100 TrackTicks GetDuration() const { return mDuration; } |
|
101 bool CanCombineWithFollowing(const AudioChunk& aOther) const |
|
102 { |
|
103 if (aOther.mBuffer != mBuffer) { |
|
104 return false; |
|
105 } |
|
106 if (mBuffer) { |
|
107 NS_ASSERTION(aOther.mBufferFormat == mBufferFormat, |
|
108 "Wrong metadata about buffer"); |
|
109 NS_ASSERTION(aOther.mChannelData.Length() == mChannelData.Length(), |
|
110 "Mismatched channel count"); |
|
111 if (mDuration > INT32_MAX) { |
|
112 return false; |
|
113 } |
|
114 for (uint32_t channel = 0; channel < mChannelData.Length(); ++channel) { |
|
115 if (aOther.mChannelData[channel] != AddAudioSampleOffset(mChannelData[channel], |
|
116 mBufferFormat, int32_t(mDuration))) { |
|
117 return false; |
|
118 } |
|
119 } |
|
120 } |
|
121 return true; |
|
122 } |
|
123 bool IsNull() const { return mBuffer == nullptr; } |
|
124 void SetNull(TrackTicks aDuration) |
|
125 { |
|
126 mBuffer = nullptr; |
|
127 mChannelData.Clear(); |
|
128 mDuration = aDuration; |
|
129 mVolume = 1.0f; |
|
130 mBufferFormat = AUDIO_FORMAT_SILENCE; |
|
131 } |
|
132 int ChannelCount() const { return mChannelData.Length(); } |
|
133 |
|
134 size_t SizeOfExcludingThisIfUnshared(MallocSizeOf aMallocSizeOf) const |
|
135 { |
|
136 return SizeOfExcludingThis(aMallocSizeOf, true); |
|
137 } |
|
138 |
|
139 size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf, bool aUnshared) const |
|
140 { |
|
141 size_t amount = 0; |
|
142 |
|
143 // Possibly owned: |
|
144 // - mBuffer - Can hold data that is also in the decoded audio queue. If it |
|
145 // is not shared, or unshared == false it gets counted. |
|
146 if (mBuffer && (!aUnshared || !mBuffer->IsShared())) { |
|
147 amount += mBuffer->SizeOfIncludingThis(aMallocSizeOf); |
|
148 } |
|
149 |
|
150 // Memory in the array is owned by mBuffer. |
|
151 amount += mChannelData.SizeOfExcludingThis(aMallocSizeOf); |
|
152 return amount; |
|
153 } |
|
154 |
|
155 TrackTicks mDuration; // in frames within the buffer |
|
156 nsRefPtr<ThreadSharedObject> mBuffer; // the buffer object whose lifetime is managed; null means data is all zeroes |
|
157 nsTArray<const void*> mChannelData; // one pointer per channel; empty if and only if mBuffer is null |
|
158 float mVolume; // volume multiplier to apply (1.0f if mBuffer is nonnull) |
|
159 SampleFormat mBufferFormat; // format of frames in mBuffer (only meaningful if mBuffer is nonnull) |
|
160 #ifdef MOZILLA_INTERNAL_API |
|
161 mozilla::TimeStamp mTimeStamp; // time at which this has been fetched from the MediaEngine |
|
162 #endif |
|
163 }; |
|
164 |
|
165 |
|
166 /** |
|
167 * A list of audio samples consisting of a sequence of slices of SharedBuffers. |
|
168 * The audio rate is determined by the track, not stored in this class. |
|
169 */ |
|
170 class AudioSegment : public MediaSegmentBase<AudioSegment, AudioChunk> { |
|
171 public: |
|
172 typedef mozilla::AudioSampleFormat SampleFormat; |
|
173 |
|
174 AudioSegment() : MediaSegmentBase<AudioSegment, AudioChunk>(AUDIO) {} |
|
175 |
|
176 // Resample the whole segment in place. |
|
177 template<typename T> |
|
178 void Resample(SpeexResamplerState* aResampler, uint32_t aInRate, uint32_t aOutRate) |
|
179 { |
|
180 mDuration = 0; |
|
181 |
|
182 for (ChunkIterator ci(*this); !ci.IsEnded(); ci.Next()) { |
|
183 nsAutoTArray<nsTArray<T>, GUESS_AUDIO_CHANNELS> output; |
|
184 nsAutoTArray<const T*, GUESS_AUDIO_CHANNELS> bufferPtrs; |
|
185 AudioChunk& c = *ci; |
|
186 // If this chunk is null, don't bother resampling, just alter its duration |
|
187 if (c.IsNull()) { |
|
188 c.mDuration = (c.mDuration * aOutRate) / aInRate; |
|
189 mDuration += c.mDuration; |
|
190 continue; |
|
191 } |
|
192 uint32_t channels = c.mChannelData.Length(); |
|
193 output.SetLength(channels); |
|
194 bufferPtrs.SetLength(channels); |
|
195 uint32_t inFrames = c.mDuration; |
|
196 // Round up to allocate; the last frame may not be used. |
|
197 NS_ASSERTION((UINT32_MAX - aInRate + 1) / c.mDuration >= aOutRate, |
|
198 "Dropping samples"); |
|
199 uint32_t outSize = (c.mDuration * aOutRate + aInRate - 1) / aInRate; |
|
200 for (uint32_t i = 0; i < channels; i++) { |
|
201 const T* in = static_cast<const T*>(c.mChannelData[i]); |
|
202 T* out = output[i].AppendElements(outSize); |
|
203 uint32_t outFrames = outSize; |
|
204 |
|
205 dom::WebAudioUtils::SpeexResamplerProcess(aResampler, i, |
|
206 in, &inFrames, |
|
207 out, &outFrames); |
|
208 MOZ_ASSERT(inFrames == c.mDuration); |
|
209 bufferPtrs[i] = out; |
|
210 output[i].SetLength(outFrames); |
|
211 } |
|
212 MOZ_ASSERT(channels > 0); |
|
213 c.mDuration = output[0].Length(); |
|
214 c.mBuffer = new mozilla::SharedChannelArrayBuffer<T>(&output); |
|
215 for (uint32_t i = 0; i < channels; i++) { |
|
216 c.mChannelData[i] = bufferPtrs[i]; |
|
217 } |
|
218 mDuration += c.mDuration; |
|
219 } |
|
220 } |
|
221 |
|
222 void ResampleChunks(SpeexResamplerState* aResampler); |
|
223 |
|
224 void AppendFrames(already_AddRefed<ThreadSharedObject> aBuffer, |
|
225 const nsTArray<const float*>& aChannelData, |
|
226 int32_t aDuration) |
|
227 { |
|
228 AudioChunk* chunk = AppendChunk(aDuration); |
|
229 chunk->mBuffer = aBuffer; |
|
230 for (uint32_t channel = 0; channel < aChannelData.Length(); ++channel) { |
|
231 chunk->mChannelData.AppendElement(aChannelData[channel]); |
|
232 } |
|
233 chunk->mVolume = 1.0f; |
|
234 chunk->mBufferFormat = AUDIO_FORMAT_FLOAT32; |
|
235 #ifdef MOZILLA_INTERNAL_API |
|
236 chunk->mTimeStamp = TimeStamp::Now(); |
|
237 #endif |
|
238 } |
|
239 void AppendFrames(already_AddRefed<ThreadSharedObject> aBuffer, |
|
240 const nsTArray<const int16_t*>& aChannelData, |
|
241 int32_t aDuration) |
|
242 { |
|
243 AudioChunk* chunk = AppendChunk(aDuration); |
|
244 chunk->mBuffer = aBuffer; |
|
245 for (uint32_t channel = 0; channel < aChannelData.Length(); ++channel) { |
|
246 chunk->mChannelData.AppendElement(aChannelData[channel]); |
|
247 } |
|
248 chunk->mVolume = 1.0f; |
|
249 chunk->mBufferFormat = AUDIO_FORMAT_S16; |
|
250 #ifdef MOZILLA_INTERNAL_API |
|
251 chunk->mTimeStamp = TimeStamp::Now(); |
|
252 #endif |
|
253 } |
|
254 // Consumes aChunk, and returns a pointer to the persistent copy of aChunk |
|
255 // in the segment. |
|
256 AudioChunk* AppendAndConsumeChunk(AudioChunk* aChunk) |
|
257 { |
|
258 AudioChunk* chunk = AppendChunk(aChunk->mDuration); |
|
259 chunk->mBuffer = aChunk->mBuffer.forget(); |
|
260 chunk->mChannelData.SwapElements(aChunk->mChannelData); |
|
261 chunk->mVolume = aChunk->mVolume; |
|
262 chunk->mBufferFormat = aChunk->mBufferFormat; |
|
263 #ifdef MOZILLA_INTERNAL_API |
|
264 chunk->mTimeStamp = TimeStamp::Now(); |
|
265 #endif |
|
266 return chunk; |
|
267 } |
|
268 void ApplyVolume(float aVolume); |
|
269 void WriteTo(uint64_t aID, AudioStream* aOutput, AudioMixer* aMixer = nullptr); |
|
270 |
|
271 int ChannelCount() { |
|
272 NS_WARN_IF_FALSE(!mChunks.IsEmpty(), |
|
273 "Cannot query channel count on a AudioSegment with no chunks."); |
|
274 // Find the first chunk that has non-zero channels. A chunk that hs zero |
|
275 // channels is just silence and we can simply discard it. |
|
276 for (ChunkIterator ci(*this); !ci.IsEnded(); ci.Next()) { |
|
277 if (ci->ChannelCount()) { |
|
278 return ci->ChannelCount(); |
|
279 } |
|
280 } |
|
281 return 0; |
|
282 } |
|
283 |
|
284 static Type StaticType() { return AUDIO; } |
|
285 |
|
286 virtual size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const MOZ_OVERRIDE |
|
287 { |
|
288 return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); |
|
289 } |
|
290 }; |
|
291 |
|
292 } |
|
293 |
|
294 #endif /* MOZILLA_AUDIOSEGMENT_H_ */ |