AudioFlinger: Split off audio processing library
Test: native AudioResampler test, general playback test
Bug: 31015569
Change-Id: Ifb248f4402a583438d756c014dcd7a4577aef713
diff --git a/include/media/AudioMixer.h b/include/media/AudioMixer.h
new file mode 100644
index 0000000..87ada76
--- /dev/null
+++ b/include/media/AudioMixer.h
@@ -0,0 +1,389 @@
+/*
+**
+** Copyright 2007, The Android Open Source Project
+**
+** Licensed under the Apache License, Version 2.0 (the "License");
+** you may not use this file except in compliance with the License.
+** You may obtain a copy of the License at
+**
+** http://www.apache.org/licenses/LICENSE-2.0
+**
+** Unless required by applicable law or agreed to in writing, software
+** distributed under the License is distributed on an "AS IS" BASIS,
+** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+** See the License for the specific language governing permissions and
+** limitations under the License.
+*/
+
+#ifndef ANDROID_AUDIO_MIXER_H
+#define ANDROID_AUDIO_MIXER_H
+
+#include <stdint.h>
+#include <sys/types.h>
+
+#include <media/AudioBufferProvider.h>
+#include <media/AudioResampler.h>
+#include <media/AudioResamplerPublic.h>
+#include <media/BufferProviders.h>
+#include <media/nbaio/NBLog.h>
+#include <system/audio.h>
+#include <utils/Compat.h>
+#include <utils/threads.h>
+
+// FIXME This is actually unity gain, which might not be max in future, expressed in U.12
+#define MAX_GAIN_INT AudioMixer::UNITY_GAIN_INT
+
+namespace android {
+
+// ----------------------------------------------------------------------------
+
+class AudioMixer
+{
+public:
+ AudioMixer(size_t frameCount, uint32_t sampleRate,
+ uint32_t maxNumTracks = MAX_NUM_TRACKS);
+
+ /*virtual*/ ~AudioMixer(); // non-virtual saves a v-table, restore if sub-classed
+
+
+ // This mixer has a hard-coded upper limit of 32 active track inputs.
+ // Adding support for > 32 tracks would require more than simply changing this value.
+ static const uint32_t MAX_NUM_TRACKS = 32;
+ // maximum number of channels supported by the mixer
+
+ // This mixer has a hard-coded upper limit of 8 channels for output.
+ static const uint32_t MAX_NUM_CHANNELS = 8;
+ static const uint32_t MAX_NUM_VOLUMES = 2; // stereo volume only
+ // maximum number of channels supported for the content
+ static const uint32_t MAX_NUM_CHANNELS_TO_DOWNMIX = AUDIO_CHANNEL_COUNT_MAX;
+
+ static const uint16_t UNITY_GAIN_INT = 0x1000;
+ static const CONSTEXPR float UNITY_GAIN_FLOAT = 1.0f;
+
+ enum { // names
+
+ // track names (MAX_NUM_TRACKS units)
+ TRACK0 = 0x1000,
+
+ // 0x2000 is unused
+
+ // setParameter targets
+ TRACK = 0x3000,
+ RESAMPLE = 0x3001,
+ RAMP_VOLUME = 0x3002, // ramp to new volume
+ VOLUME = 0x3003, // don't ramp
+ TIMESTRETCH = 0x3004,
+
+ // set Parameter names
+ // for target TRACK
+ CHANNEL_MASK = 0x4000,
+ FORMAT = 0x4001,
+ MAIN_BUFFER = 0x4002,
+ AUX_BUFFER = 0x4003,
+ DOWNMIX_TYPE = 0X4004,
+ MIXER_FORMAT = 0x4005, // AUDIO_FORMAT_PCM_(FLOAT|16_BIT)
+ MIXER_CHANNEL_MASK = 0x4006, // Channel mask for mixer output
+ // for target RESAMPLE
+ SAMPLE_RATE = 0x4100, // Configure sample rate conversion on this track name;
+ // parameter 'value' is the new sample rate in Hz.
+ // Only creates a sample rate converter the first time that
+ // the track sample rate is different from the mix sample rate.
+ // If the new sample rate is the same as the mix sample rate,
+ // and a sample rate converter already exists,
+ // then the sample rate converter remains present but is a no-op.
+ RESET = 0x4101, // Reset sample rate converter without changing sample rate.
+ // This clears out the resampler's input buffer.
+ REMOVE = 0x4102, // Remove the sample rate converter on this track name;
+ // the track is restored to the mix sample rate.
+ // for target RAMP_VOLUME and VOLUME (8 channels max)
+ // FIXME use float for these 3 to improve the dynamic range
+ VOLUME0 = 0x4200,
+ VOLUME1 = 0x4201,
+ AUXLEVEL = 0x4210,
+ // for target TIMESTRETCH
+ PLAYBACK_RATE = 0x4300, // Configure timestretch on this track name;
+ // parameter 'value' is a pointer to the new playback rate.
+ };
+
+
+ // For all APIs with "name": TRACK0 <= name < TRACK0 + MAX_NUM_TRACKS
+
+ // Allocate a track name. Returns new track name if successful, -1 on failure.
+ // The failure could be because of an invalid channelMask or format, or that
+ // the track capacity of the mixer is exceeded.
+ int getTrackName(audio_channel_mask_t channelMask,
+ audio_format_t format, int sessionId);
+
+ // Free an allocated track by name
+ void deleteTrackName(int name);
+
+ // Enable or disable an allocated track by name
+ void enable(int name);
+ void disable(int name);
+
+ void setParameter(int name, int target, int param, void *value);
+
+ void setBufferProvider(int name, AudioBufferProvider* bufferProvider);
+ void process();
+
+ uint32_t trackNames() const { return mTrackNames; }
+
+ size_t getUnreleasedFrames(int name) const;
+
+ static inline bool isValidPcmTrackFormat(audio_format_t format) {
+ switch (format) {
+ case AUDIO_FORMAT_PCM_8_BIT:
+ case AUDIO_FORMAT_PCM_16_BIT:
+ case AUDIO_FORMAT_PCM_24_BIT_PACKED:
+ case AUDIO_FORMAT_PCM_32_BIT:
+ case AUDIO_FORMAT_PCM_FLOAT:
+ return true;
+ default:
+ return false;
+ }
+ }
+
+private:
+
+ enum {
+ // FIXME this representation permits up to 8 channels
+ NEEDS_CHANNEL_COUNT__MASK = 0x00000007,
+ };
+
+ enum {
+ NEEDS_CHANNEL_1 = 0x00000000, // mono
+ NEEDS_CHANNEL_2 = 0x00000001, // stereo
+
+ // sample format is not explicitly specified, and is assumed to be AUDIO_FORMAT_PCM_16_BIT
+
+ NEEDS_MUTE = 0x00000100,
+ NEEDS_RESAMPLE = 0x00001000,
+ NEEDS_AUX = 0x00010000,
+ };
+
+ struct state_t;
+ struct track_t;
+
+ typedef void (*hook_t)(track_t* t, int32_t* output, size_t numOutFrames, int32_t* temp,
+ int32_t* aux);
+ static const int BLOCKSIZE = 16; // 4 cache lines
+
+ struct track_t {
+ uint32_t needs;
+
+ // TODO: Eventually remove legacy integer volume settings
+ union {
+ int16_t volume[MAX_NUM_VOLUMES]; // U4.12 fixed point (top bit should be zero)
+ int32_t volumeRL;
+ };
+
+ int32_t prevVolume[MAX_NUM_VOLUMES];
+
+ // 16-byte boundary
+
+ int32_t volumeInc[MAX_NUM_VOLUMES];
+ int32_t auxInc;
+ int32_t prevAuxLevel;
+
+ // 16-byte boundary
+
+ int16_t auxLevel; // 0 <= auxLevel <= MAX_GAIN_INT, but signed for mul performance
+ uint16_t frameCount;
+
+ uint8_t channelCount; // 1 or 2, redundant with (needs & NEEDS_CHANNEL_COUNT__MASK)
+ uint8_t unused_padding; // formerly format, was always 16
+ uint16_t enabled; // actually bool
+ audio_channel_mask_t channelMask;
+
+ // actual buffer provider used by the track hooks, see DownmixerBufferProvider below
+ // for how the Track buffer provider is wrapped by another one when dowmixing is required
+ AudioBufferProvider* bufferProvider;
+
+ // 16-byte boundary
+
+ mutable AudioBufferProvider::Buffer buffer; // 8 bytes
+
+ hook_t hook;
+ const void* in; // current location in buffer
+
+ // 16-byte boundary
+
+ AudioResampler* resampler;
+ uint32_t sampleRate;
+ int32_t* mainBuffer;
+ int32_t* auxBuffer;
+
+ // 16-byte boundary
+
+ /* Buffer providers are constructed to translate the track input data as needed.
+ *
+ * TODO: perhaps make a single PlaybackConverterProvider class to move
+ * all pre-mixer track buffer conversions outside the AudioMixer class.
+ *
+ * 1) mInputBufferProvider: The AudioTrack buffer provider.
+ * 2) mReformatBufferProvider: If not NULL, performs the audio reformat to
+ * match either mMixerInFormat or mDownmixRequiresFormat, if the downmixer
+ * requires reformat. For example, it may convert floating point input to
+ * PCM_16_bit if that's required by the downmixer.
+ * 3) downmixerBufferProvider: If not NULL, performs the channel remixing to match
+ * the number of channels required by the mixer sink.
+ * 4) mPostDownmixReformatBufferProvider: If not NULL, performs reformatting from
+ * the downmixer requirements to the mixer engine input requirements.
+ * 5) mTimestretchBufferProvider: Adds timestretching for playback rate
+ */
+ AudioBufferProvider* mInputBufferProvider; // externally provided buffer provider.
+ PassthruBufferProvider* mReformatBufferProvider; // provider wrapper for reformatting.
+ PassthruBufferProvider* downmixerBufferProvider; // wrapper for channel conversion.
+ PassthruBufferProvider* mPostDownmixReformatBufferProvider;
+ PassthruBufferProvider* mTimestretchBufferProvider;
+
+ int32_t sessionId;
+
+ audio_format_t mMixerFormat; // output mix format: AUDIO_FORMAT_PCM_(FLOAT|16_BIT)
+ audio_format_t mFormat; // input track format
+ audio_format_t mMixerInFormat; // mix internal format AUDIO_FORMAT_PCM_(FLOAT|16_BIT)
+ // each track must be converted to this format.
+ audio_format_t mDownmixRequiresFormat; // required downmixer format
+ // AUDIO_FORMAT_PCM_16_BIT if 16 bit necessary
+ // AUDIO_FORMAT_INVALID if no required format
+
+ float mVolume[MAX_NUM_VOLUMES]; // floating point set volume
+ float mPrevVolume[MAX_NUM_VOLUMES]; // floating point previous volume
+ float mVolumeInc[MAX_NUM_VOLUMES]; // floating point volume increment
+
+ float mAuxLevel; // floating point set aux level
+ float mPrevAuxLevel; // floating point prev aux level
+ float mAuxInc; // floating point aux increment
+
+ audio_channel_mask_t mMixerChannelMask;
+ uint32_t mMixerChannelCount;
+
+ AudioPlaybackRate mPlaybackRate;
+
+ bool needsRamp() { return (volumeInc[0] | volumeInc[1] | auxInc) != 0; }
+ bool setResampler(uint32_t trackSampleRate, uint32_t devSampleRate);
+ bool doesResample() const { return resampler != NULL; }
+ void resetResampler() { if (resampler != NULL) resampler->reset(); }
+ void adjustVolumeRamp(bool aux, bool useFloat = false);
+ size_t getUnreleasedFrames() const { return resampler != NULL ?
+ resampler->getUnreleasedFrames() : 0; };
+
+ status_t prepareForDownmix();
+ void unprepareForDownmix();
+ status_t prepareForReformat();
+ void unprepareForReformat();
+ bool setPlaybackRate(const AudioPlaybackRate &playbackRate);
+ void reconfigureBufferProviders();
+ };
+
+ typedef void (*process_hook_t)(state_t* state);
+
+ // pad to 32-bytes to fill cache line
+ struct state_t {
+ uint32_t enabledTracks;
+ uint32_t needsChanged;
+ size_t frameCount;
+ process_hook_t hook; // one of process__*, never NULL
+ int32_t *outputTemp;
+ int32_t *resampleTemp;
+ NBLog::Writer* mLog;
+ int32_t reserved[1];
+ // FIXME allocate dynamically to save some memory when maxNumTracks < MAX_NUM_TRACKS
+ track_t tracks[MAX_NUM_TRACKS] __attribute__((aligned(32)));
+ };
+
+ // bitmask of allocated track names, where bit 0 corresponds to TRACK0 etc.
+ uint32_t mTrackNames;
+
+ // bitmask of configured track names; ~0 if maxNumTracks == MAX_NUM_TRACKS,
+ // but will have fewer bits set if maxNumTracks < MAX_NUM_TRACKS
+ const uint32_t mConfiguredNames;
+
+ const uint32_t mSampleRate;
+
+ NBLog::Writer mDummyLog;
+public:
+ void setLog(NBLog::Writer* log);
+private:
+ state_t mState __attribute__((aligned(32)));
+
+ // Call after changing either the enabled status of a track, or parameters of an enabled track.
+ // OK to call more often than that, but unnecessary.
+ void invalidateState(uint32_t mask);
+
+ bool setChannelMasks(int name,
+ audio_channel_mask_t trackChannelMask, audio_channel_mask_t mixerChannelMask);
+
+ static void track__genericResample(track_t* t, int32_t* out, size_t numFrames, int32_t* temp,
+ int32_t* aux);
+ static void track__nop(track_t* t, int32_t* out, size_t numFrames, int32_t* temp, int32_t* aux);
+ static void track__16BitsStereo(track_t* t, int32_t* out, size_t numFrames, int32_t* temp,
+ int32_t* aux);
+ static void track__16BitsMono(track_t* t, int32_t* out, size_t numFrames, int32_t* temp,
+ int32_t* aux);
+ static void volumeRampStereo(track_t* t, int32_t* out, size_t frameCount, int32_t* temp,
+ int32_t* aux);
+ static void volumeStereo(track_t* t, int32_t* out, size_t frameCount, int32_t* temp,
+ int32_t* aux);
+
+ static void process__validate(state_t* state);
+ static void process__nop(state_t* state);
+ static void process__genericNoResampling(state_t* state);
+ static void process__genericResampling(state_t* state);
+ static void process__OneTrack16BitsStereoNoResampling(state_t* state);
+
+ static pthread_once_t sOnceControl;
+ static void sInitRoutine();
+
+ /* multi-format volume mixing function (calls template functions
+ * in AudioMixerOps.h). The template parameters are as follows:
+ *
+ * MIXTYPE (see AudioMixerOps.h MIXTYPE_* enumeration)
+ * USEFLOATVOL (set to true if float volume is used)
+ * ADJUSTVOL (set to true if volume ramp parameters needs adjustment afterwards)
+ * TO: int32_t (Q4.27) or float
+ * TI: int32_t (Q4.27) or int16_t (Q0.15) or float
+ * TA: int32_t (Q4.27)
+ */
+ template <int MIXTYPE, bool USEFLOATVOL, bool ADJUSTVOL,
+ typename TO, typename TI, typename TA>
+ static void volumeMix(TO *out, size_t outFrames,
+ const TI *in, TA *aux, bool ramp, AudioMixer::track_t *t);
+
+ // multi-format process hooks
+ template <int MIXTYPE, typename TO, typename TI, typename TA>
+ static void process_NoResampleOneTrack(state_t* state);
+
+ // multi-format track hooks
+ template <int MIXTYPE, typename TO, typename TI, typename TA>
+ static void track__Resample(track_t* t, TO* out, size_t frameCount,
+ TO* temp __unused, TA* aux);
+ template <int MIXTYPE, typename TO, typename TI, typename TA>
+ static void track__NoResample(track_t* t, TO* out, size_t frameCount,
+ TO* temp __unused, TA* aux);
+
+ static void convertMixerFormat(void *out, audio_format_t mixerOutFormat,
+ void *in, audio_format_t mixerInFormat, size_t sampleCount);
+
+ // hook types
+ enum {
+ PROCESSTYPE_NORESAMPLEONETRACK,
+ };
+ enum {
+ TRACKTYPE_NOP,
+ TRACKTYPE_RESAMPLE,
+ TRACKTYPE_NORESAMPLE,
+ TRACKTYPE_NORESAMPLEMONO,
+ };
+
+ // functions for determining the proper process and track hooks.
+ static process_hook_t getProcessHook(int processType, uint32_t channelCount,
+ audio_format_t mixerInFormat, audio_format_t mixerOutFormat);
+ static hook_t getTrackHook(int trackType, uint32_t channelCount,
+ audio_format_t mixerInFormat, audio_format_t mixerOutFormat);
+};
+
+// ----------------------------------------------------------------------------
+} // namespace android
+
+#endif // ANDROID_AUDIO_MIXER_H
diff --git a/include/media/AudioResampler.h b/include/media/AudioResampler.h
new file mode 100644
index 0000000..c4627e8
--- /dev/null
+++ b/include/media/AudioResampler.h
@@ -0,0 +1,173 @@
+/*
+ * Copyright (C) 2007 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ANDROID_AUDIO_RESAMPLER_H
+#define ANDROID_AUDIO_RESAMPLER_H
+
+#include <stdint.h>
+#include <sys/types.h>
+
+#include <cutils/compiler.h>
+#include <utils/Compat.h>
+
+#include <media/AudioBufferProvider.h>
+#include <system/audio.h>
+
+namespace android {
+// ----------------------------------------------------------------------------
+
+class ANDROID_API AudioResampler {
+public:
+ // Determines quality of SRC.
+ // LOW_QUALITY: linear interpolator (1st order)
+ // MED_QUALITY: cubic interpolator (3rd order)
+ // HIGH_QUALITY: fixed multi-tap FIR (e.g. 48KHz->44.1KHz)
+ // NOTE: high quality SRC will only be supported for
+ // certain fixed rate conversions. Sample rate cannot be
+ // changed dynamically.
+ enum src_quality {
+ DEFAULT_QUALITY=0,
+ LOW_QUALITY=1,
+ MED_QUALITY=2,
+ HIGH_QUALITY=3,
+ VERY_HIGH_QUALITY=4,
+ DYN_LOW_QUALITY=5,
+ DYN_MED_QUALITY=6,
+ DYN_HIGH_QUALITY=7,
+ };
+
+ static const CONSTEXPR float UNITY_GAIN_FLOAT = 1.0f;
+
+ static AudioResampler* create(audio_format_t format, int inChannelCount,
+ int32_t sampleRate, src_quality quality=DEFAULT_QUALITY);
+
+ virtual ~AudioResampler();
+
+ virtual void init() = 0;
+ virtual void setSampleRate(int32_t inSampleRate);
+ virtual void setVolume(float left, float right);
+
+ // Resample int16_t samples from provider and accumulate into 'out'.
+ // A mono provider delivers a sequence of samples.
+ // A stereo provider delivers a sequence of interleaved pairs of samples.
+ //
+ // In either case, 'out' holds interleaved pairs of fixed-point Q4.27.
+ // That is, for a mono provider, there is an implicit up-channeling.
+ // Since this method accumulates, the caller is responsible for clearing 'out' initially.
+ //
+ // For a float resampler, 'out' holds interleaved pairs of float samples.
+ //
+ // Multichannel interleaved frames for n > 2 is supported for quality DYN_LOW_QUALITY,
+ // DYN_MED_QUALITY, and DYN_HIGH_QUALITY.
+ //
+ // Returns the number of frames resampled into the out buffer.
+ virtual size_t resample(int32_t* out, size_t outFrameCount,
+ AudioBufferProvider* provider) = 0;
+
+ virtual void reset();
+ virtual size_t getUnreleasedFrames() const { return mInputIndex; }
+
+ // called from destructor, so must not be virtual
+ src_quality getQuality() const { return mQuality; }
+
+protected:
+ // number of bits for phase fraction - 30 bits allows nearly 2x downsampling
+ static const int kNumPhaseBits = 30;
+
+ // phase mask for fraction
+ static const uint32_t kPhaseMask = (1LU<<kNumPhaseBits)-1;
+
+ // multiplier to calculate fixed point phase increment
+ static const double kPhaseMultiplier;
+
+ AudioResampler(int inChannelCount, int32_t sampleRate, src_quality quality);
+
+ // prevent copying
+ AudioResampler(const AudioResampler&);
+ AudioResampler& operator=(const AudioResampler&);
+
+ const int32_t mChannelCount;
+ const int32_t mSampleRate;
+ int32_t mInSampleRate;
+ AudioBufferProvider::Buffer mBuffer;
+ union {
+ int16_t mVolume[2];
+ uint32_t mVolumeRL;
+ };
+ int16_t mTargetVolume[2];
+ size_t mInputIndex;
+ int32_t mPhaseIncrement;
+ uint32_t mPhaseFraction;
+
+ // returns the inFrameCount required to generate outFrameCount frames.
+ //
+ // Placed here to be a consistent for all resamplers.
+ //
+ // Right now, we use the upper bound without regards to the current state of the
+ // input buffer using integer arithmetic, as follows:
+ //
+ // (static_cast<uint64_t>(outFrameCount)*mInSampleRate + (mSampleRate - 1))/mSampleRate;
+ //
+ // The double precision equivalent (float may not be precise enough):
+ // ceil(static_cast<double>(outFrameCount) * mInSampleRate / mSampleRate);
+ //
+ // this relies on the fact that the mPhaseIncrement is rounded down from
+ // #phases * mInSampleRate/mSampleRate and the fact that Sum(Floor(x)) <= Floor(Sum(x)).
+ // http://www.proofwiki.org/wiki/Sum_of_Floors_Not_Greater_Than_Floor_of_Sums
+ //
+ // (so long as double precision is computed accurately enough to be considered
+ // greater than or equal to the Floor(x) value in int32_t arithmetic; thus this
+ // will not necessarily hold for floats).
+ //
+ // TODO:
+ // Greater accuracy and a tight bound is obtained by:
+ // 1) subtract and adjust for the current state of the AudioBufferProvider buffer.
+ // 2) using the exact integer formula where (ignoring 64b casting)
+ // inFrameCount = (mPhaseIncrement * (outFrameCount - 1) + mPhaseFraction) / phaseWrapLimit;
+ // phaseWrapLimit is the wraparound (1 << kNumPhaseBits), if not specified explicitly.
+ //
+ inline size_t getInFrameCountRequired(size_t outFrameCount) {
+ return (static_cast<uint64_t>(outFrameCount)*mInSampleRate
+ + (mSampleRate - 1))/mSampleRate;
+ }
+
+ inline float clampFloatVol(float volume) {
+ if (volume > UNITY_GAIN_FLOAT) {
+ return UNITY_GAIN_FLOAT;
+ } else if (volume >= 0.) {
+ return volume;
+ }
+ return 0.; // NaN or negative volume maps to 0.
+ }
+
+private:
+ const src_quality mQuality;
+
+ // Return 'true' if the quality level is supported without explicit request
+ static bool qualityIsSupported(src_quality quality);
+
+ // For pthread_once()
+ static void init_routine();
+
+ // Return the estimated CPU load for specific resampler in MHz.
+ // The absolute number is irrelevant, it's the relative values that matter.
+ static uint32_t qualityMHz(src_quality quality);
+};
+
+// ----------------------------------------------------------------------------
+} // namespace android
+
+#endif // ANDROID_AUDIO_RESAMPLER_H
diff --git a/include/media/BufferProviders.h b/include/media/BufferProviders.h
new file mode 100644
index 0000000..68b3f23
--- /dev/null
+++ b/include/media/BufferProviders.h
@@ -0,0 +1,208 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ANDROID_BUFFER_PROVIDERS_H
+#define ANDROID_BUFFER_PROVIDERS_H
+
+#include <stdint.h>
+#include <sys/types.h>
+
+#include <media/AudioBufferProvider.h>
+#include <system/audio.h>
+#include <system/audio_effect.h>
+#include <utils/StrongPointer.h>
+
+// external forward declaration from external/sonic/sonic.h
+struct sonicStreamStruct;
+typedef struct sonicStreamStruct *sonicStream;
+
+namespace android {
+
+class EffectBufferHalInterface;
+class EffectHalInterface;
+class EffectsFactoryHalInterface;
+
+// ----------------------------------------------------------------------------
+
+class PassthruBufferProvider : public AudioBufferProvider {
+public:
+ PassthruBufferProvider() : mTrackBufferProvider(NULL) { }
+
+ virtual ~PassthruBufferProvider() { }
+
+ // call this to release the buffer to the upstream provider.
+ // treat it as an audio discontinuity for future samples.
+ virtual void reset() { }
+
+ // set the upstream buffer provider. Consider calling "reset" before this function.
+ virtual void setBufferProvider(AudioBufferProvider *p) {
+ mTrackBufferProvider = p;
+ }
+
+protected:
+ AudioBufferProvider *mTrackBufferProvider;
+};
+
+// Base AudioBufferProvider class used for DownMixerBufferProvider, RemixBufferProvider,
+// and ReformatBufferProvider.
+// It handles a private buffer for use in converting format or channel masks from the
+// input data to a form acceptable by the mixer.
+// TODO: Make a ResamplerBufferProvider when integers are entirely removed from the
+// processing pipeline.
+class CopyBufferProvider : public PassthruBufferProvider {
+public:
+ // Use a private buffer of bufferFrameCount frames (each frame is outputFrameSize bytes).
+ // If bufferFrameCount is 0, no private buffer is created and in-place modification of
+ // the upstream buffer provider's buffers is performed by copyFrames().
+ CopyBufferProvider(size_t inputFrameSize, size_t outputFrameSize,
+ size_t bufferFrameCount);
+ virtual ~CopyBufferProvider();
+
+ // Overrides AudioBufferProvider methods
+ virtual status_t getNextBuffer(Buffer *buffer);
+ virtual void releaseBuffer(Buffer *buffer);
+
+ // Overrides PassthruBufferProvider
+ virtual void reset();
+
+ // this function should be supplied by the derived class. It converts
+ // #frames in the *src pointer to the *dst pointer. It is public because
+ // some providers will allow this to work on arbitrary buffers outside
+ // of the internal buffers.
+ virtual void copyFrames(void *dst, const void *src, size_t frames) = 0;
+
+protected:
+ const size_t mInputFrameSize;
+ const size_t mOutputFrameSize;
+private:
+ AudioBufferProvider::Buffer mBuffer;
+ const size_t mLocalBufferFrameCount;
+ void *mLocalBufferData;
+ size_t mConsumed;
+};
+
+// DownmixerBufferProvider derives from CopyBufferProvider to provide
+// position dependent downmixing by an Audio Effect.
+class DownmixerBufferProvider : public CopyBufferProvider {
+public:
+ DownmixerBufferProvider(audio_channel_mask_t inputChannelMask,
+ audio_channel_mask_t outputChannelMask, audio_format_t format,
+ uint32_t sampleRate, int32_t sessionId, size_t bufferFrameCount);
+ virtual ~DownmixerBufferProvider();
+ //Overrides
+ virtual void copyFrames(void *dst, const void *src, size_t frames);
+
+ bool isValid() const { return mDownmixInterface.get() != NULL; }
+ static status_t init();
+ static bool isMultichannelCapable() { return sIsMultichannelCapable; }
+
+protected:
+ sp<EffectsFactoryHalInterface> mEffectsFactory;
+ sp<EffectHalInterface> mDownmixInterface;
+ sp<EffectBufferHalInterface> mInBuffer;
+ sp<EffectBufferHalInterface> mOutBuffer;
+ effect_config_t mDownmixConfig;
+
+ // effect descriptor for the downmixer used by the mixer
+ static effect_descriptor_t sDwnmFxDesc;
+ // indicates whether a downmix effect has been found and is usable by this mixer
+ static bool sIsMultichannelCapable;
+ // FIXME: should we allow effects outside of the framework?
+ // We need to here. A special ioId that must be <= -2 so it does not map to a session.
+ static const int32_t SESSION_ID_INVALID_AND_IGNORED = -2;
+};
+
+// RemixBufferProvider derives from CopyBufferProvider to perform an
+// upmix or downmix to the proper channel count and mask.
+class RemixBufferProvider : public CopyBufferProvider {
+public:
+ RemixBufferProvider(audio_channel_mask_t inputChannelMask,
+ audio_channel_mask_t outputChannelMask, audio_format_t format,
+ size_t bufferFrameCount);
+ //Overrides
+ virtual void copyFrames(void *dst, const void *src, size_t frames);
+
+protected:
+ const audio_format_t mFormat;
+ const size_t mSampleSize;
+ const size_t mInputChannels;
+ const size_t mOutputChannels;
+ int8_t mIdxAry[sizeof(uint32_t) * 8]; // 32 bits => channel indices
+};
+
+// ReformatBufferProvider derives from CopyBufferProvider to convert the input data
+// to an acceptable mixer input format type.
+class ReformatBufferProvider : public CopyBufferProvider {
+public:
+ ReformatBufferProvider(int32_t channelCount,
+ audio_format_t inputFormat, audio_format_t outputFormat,
+ size_t bufferFrameCount);
+ virtual void copyFrames(void *dst, const void *src, size_t frames);
+
+protected:
+ const uint32_t mChannelCount;
+ const audio_format_t mInputFormat;
+ const audio_format_t mOutputFormat;
+};
+
+// TimestretchBufferProvider derives from PassthruBufferProvider for time stretching
+class TimestretchBufferProvider : public PassthruBufferProvider {
+public:
+ TimestretchBufferProvider(int32_t channelCount,
+ audio_format_t format, uint32_t sampleRate,
+ const AudioPlaybackRate &playbackRate);
+ virtual ~TimestretchBufferProvider();
+
+ // Overrides AudioBufferProvider methods
+ virtual status_t getNextBuffer(Buffer* buffer);
+ virtual void releaseBuffer(Buffer* buffer);
+
+ // Overrides PassthruBufferProvider
+ virtual void reset();
+
+ virtual status_t setPlaybackRate(const AudioPlaybackRate &playbackRate);
+
+ // processes frames
+ // dstBuffer is where to place the data
+ // dstFrames [in/out] is the desired frames (return with actual placed in buffer)
+ // srcBuffer is the source data
+ // srcFrames [in/out] is the available source frames (return with consumed)
+ virtual void processFrames(void *dstBuffer, size_t *dstFrames,
+ const void *srcBuffer, size_t *srcFrames);
+
+protected:
+ const uint32_t mChannelCount;
+ const audio_format_t mFormat;
+ const uint32_t mSampleRate; // const for now (TODO change this)
+ const size_t mFrameSize;
+ AudioPlaybackRate mPlaybackRate;
+
+private:
+ AudioBufferProvider::Buffer mBuffer; // for upstream request
+ size_t mLocalBufferFrameCount; // size of local buffer
+ void *mLocalBufferData; // internally allocated buffer for data returned
+ // to caller
+ size_t mRemaining; // remaining data in local buffer
+ sonicStream mSonicStream; // handle to sonic timestretch object
+ //FIXME: this dependency should be abstracted out
+ bool mFallbackFailErrorShown; // log fallback error only once
+ bool mAudioPlaybackRateValid; // flag for current parameters validity
+};
+
+// ----------------------------------------------------------------------------
+} // namespace android
+
+#endif // ANDROID_BUFFER_PROVIDERS_H
diff --git a/include/media/LinearMap.h b/include/media/LinearMap.h
new file mode 100644
index 0000000..2220a0c
--- /dev/null
+++ b/include/media/LinearMap.h
@@ -0,0 +1,366 @@
+/*
+ * Copyright 2015 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ANDROID_LINEAR_MAP_H
+#define ANDROID_LINEAR_MAP_H
+
+#include <stdint.h>
+
+namespace android {
+
+/*
+A general purpose lookup utility that defines a mapping between X and Y as a
+continuous set of line segments with shared (x, y) end-points.
+The (x, y) points must be added in order, monotonically increasing in both x and y;
+a log warning is emitted if this does not happen (See general usage notes below).
+
+A limited history of (x, y) points is kept for space reasons (See general usage notes).
+
+In AudioFlinger, we use the LinearMap to associate track frames to
+sink frames. When we want to obtain a client track timestamp, we first
+get a timestamp from the sink. The sink timestamp's position (mPosition)
+corresponds to the sink frames written. We use LinearMap to figure out which track frame
+the sink frame corresponds to. This allows us to substitute a track frame for the
+the sink frame (keeping the mTime identical) and return that timestamp back to the client.
+
+The method findX() can be used to retrieve an x value from a given y value and is
+used for timestamps, similarly for findY() which is provided for completeness.
+
+We update the (track frame, sink frame) points in the LinearMap each time we write data
+to the sink by the AudioFlinger PlaybackThread (MixerThread).
+
+
+AudioFlinger Timestamp Notes:
+
+1) Example: Obtaining a track timestamp during playback. In this case, the LinearMap
+looks something like this:
+
+Track Frame Sink Frame
+(track start)
+0 50000 (track starts here, the sink may already be running)
+1000 51000
+2000 52000
+
+When we request a track timestamp, we call the sink getTimestamp() and get for example
+mPosition = 51020. Using the LinearMap, we find we have played to track frame 1020.
+We substitute the sink mPosition of 51020 with the track position 1020,
+and return that timestamp to the app.
+
+2) Example: Obtaining a track timestamp duing pause. In this case, the LinearMap
+looks something like this:
+
+Track Frame Sink Frame
+... (some time has gone by)
+15000 30000
+16000 31000
+17000 32000
+(pause here)
+(suppose we call sink getTimestamp() here and get sink mPosition = 31100; that means
+ we have played to track frame 16100. The track timestamp mPosition will
+ continue to advance until the sink timestamp returns a value of mPosition
+ greater than 32000, corresponding to track frame 17000 when the pause was called).
+17000 33000
+17000 34000
+...
+
+3) If the track underruns, it appears as if a pause was called on that track.
+
+4) If there is an underrun in the HAL layer, then it may be possible that
+the sink getTimestamp() will return a value greater than the number of frames written
+(it should always be less). This should be rare, if not impossible by some
+HAL implementations of the sink getTimestamp. In that case, timing is lost
+and we will return the most recent track frame written.
+
+5) When called with no points in the map, findX() returns the start value (default 0).
+This is consistent with starting after a stop() or flush().
+
+6) Resuming after Track standby will be similar to coming out of pause, as the HAL ensures
+framesWritten() and getTimestamp() are contiguous for non-offloaded/direct tracks.
+
+7) LinearMap works for different speeds and sample rates as it uses
+linear interpolation. Since AudioFlinger only updates speed and sample rate
+exactly at the sample points pushed into the LinearMap, the returned values
+from findX() and findY() are accurate regardless of how many speed or sample
+rate changes are made, so long as the coordinate looked up is within the
+sample history.
+
+General usage notes:
+
+1) In order for the LinearMap to work reliably, you cannot look backwards more
+than the size of its circular buffer history, set upon creation (typically 16).
+If you look back further, the position is extrapolated either from a passed in
+extrapolation parameter or from the oldest line segment.
+
+2) Points must monotonically increase in x and y. The increment between adjacent
+points cannot be greater than signed 32 bits. Wrap in the x, y coordinates are supported,
+since we use differences in our computation.
+
+3) If the frame data is discontinuous (due to stop or flush) call reset() to clear
+the sample counter.
+
+4) If (x, y) are not strictly monotonic increasing, i.e. (x2 > x1) and (y2 > y1),
+then one or both of the inverses y = f(x) or x = g(y) may have multiple solutions.
+In that case, the most recent solution is returned by findX() or findY(). We
+do not warn if (x2 == x1) or (y2 == y1), but we do logcat warn if (x2 < x1) or
+(y2 < y1).
+
+5) Due to rounding it is possible x != findX(findY(x)) or y != findY(findX(y))
+even when the inverse exists. Nevertheless, the values should be close.
+
+*/
+
+template <typename T>
+class LinearMap {
+public:
+ // This enumeration describes the reliability of the findX() or findY() estimation
+ // in descending order.
+ enum FindMethod {
+ FIND_METHOD_INTERPOLATION, // High reliability (errors due to rounding)
+ FIND_METHOD_FORWARD_EXTRAPOLATION, // Reliability based on no future speed changes
+ FIND_METHOD_BACKWARD_EXTRAPOLATION, // Reliability based on prior estimated speed
+ FIND_METHOD_START_VALUE, // No samples in history, using start value
+ };
+
+ explicit LinearMap(size_t size)
+ : mSize(size),
+ mPos(0), // a circular buffer, so could start anywhere. the first sample is at 1.
+ mSamples(0),
+ // mStepValid(false), // only valid if mSamples > 1
+ // mExtrapolateTail(false), // only valid if mSamples > 0
+ mX(new T[size]),
+ mY(new T[size]) { }
+
+ ~LinearMap() {
+ delete[] mX;
+ delete[] mY;
+ }
+
+ // Add a new sample point to the linear map.
+ //
+ // The difference between the new sample and the previous sample
+ // in the x or y coordinate must be less than INT32_MAX for purposes
+ // of the linear interpolation or extrapolation.
+ //
+ // The value should be monotonic increasing (e.g. diff >= 0);
+ // logcat warnings are issued if they are not.
+ __attribute__((no_sanitize("integer")))
+ void push(T x, T y) {
+ // Assumption: we assume x, y are monotonic increasing values,
+ // which (can) wrap in precision no less than 32 bits and have
+ // "step" or differences between adjacent points less than 32 bits.
+
+ if (mSamples > 0) {
+ const bool lastStepValid = mStepValid;
+ int32_t xdiff;
+ int32_t ydiff;
+ // check difference assumption here
+ mStepValid = checkedDiff(&xdiff, x, mX[mPos], "x")
+ & /* bitwise AND to always warn for ydiff, though logical AND is also OK */
+ checkedDiff(&ydiff, y, mY[mPos], "y");
+
+ // Optimization: do not add a new sample if the line segment would
+ // simply extend the previous line segment. This extends the useful
+ // history by removing redundant points.
+ if (mSamples > 1 && mStepValid && lastStepValid) {
+ const size_t prev = previousPosition();
+ const int32_t xdiff2 = x - mX[prev];
+ const int32_t ydiff2 = y - mY[prev];
+
+ // if both current step and previous step are valid (non-negative and
+ // less than INT32_MAX for precision greater than 4 bytes)
+ // then the sum of the two steps is valid when the
+ // int32_t difference is non-negative.
+ if (xdiff2 >= 0 && ydiff2 >= 0
+ && (int64_t)xdiff2 * ydiff == (int64_t)ydiff2 * xdiff) {
+ // ALOGD("reusing sample! (%u, %u) sample depth %zd", x, y, mSamples);
+ mX[mPos] = x;
+ mY[mPos] = y;
+ return;
+ }
+ }
+ }
+ if (++mPos >= mSize) {
+ mPos = 0;
+ }
+ if (mSamples < mSize) {
+ mExtrapolateTail = false;
+ ++mSamples;
+ } else {
+ // we enable extrapolation beyond the oldest sample
+ // if the sample buffers are completely full and we
+ // no longer know the full history.
+ mExtrapolateTail = true;
+ }
+ mX[mPos] = x;
+ mY[mPos] = y;
+ }
+
+ // clear all samples from the circular array
+ void reset() {
+ // no need to reset mPos, we use a circular buffer.
+ // computed values such as mStepValid are set after a subsequent push().
+ mSamples = 0;
+ }
+
+ // returns true if LinearMap contains at least one sample.
+ bool hasData() const {
+ return mSamples != 0;
+ }
+
+ // find the corresponding X point from a Y point.
+ // See findU for details.
+ __attribute__((no_sanitize("integer")))
+ T findX(T y, FindMethod *method = NULL, double extrapolation = 0.0, T startValue = 0) const {
+ return findU(y, mX, mY, method, extrapolation, startValue);
+ }
+
+ // find the corresponding Y point from a X point.
+ // See findU for details.
+ __attribute__((no_sanitize("integer")))
+ T findY(T x, FindMethod *method = NULL, double extrapolation = 0.0, T startValue = 0) const {
+ return findU(x, mY, mX, method, extrapolation, startValue);
+ }
+
+protected:
+
+ // returns false if the diff is out of int32_t bounds or negative.
+ __attribute__((no_sanitize("integer")))
+ static inline bool checkedDiff(int32_t *diff, T x2, T x1, const char *coord) {
+ if (sizeof(T) >= 8) {
+ const int64_t diff64 = x2 - x1;
+ *diff = (int32_t)diff64; // intentionally lose precision
+ if (diff64 > INT32_MAX) {
+ ALOGW("LinearMap: %s overflow diff(%lld) from %llu - %llu exceeds INT32_MAX",
+ coord, (long long)diff64,
+ (unsigned long long)x2, (unsigned long long)x1);
+ return false;
+ } else if (diff64 < 0) {
+ ALOGW("LinearMap: %s negative diff(%lld) from %llu - %llu",
+ coord, (long long)diff64,
+ (unsigned long long)x2, (unsigned long long)x1);
+ return false;
+ }
+ return true;
+ }
+ // for 32 bit integers we cannot detect overflow (it
+ // shows up as a negative difference).
+ *diff = x2 - x1;
+ if (*diff < 0) {
+ ALOGW("LinearMap: %s negative diff(%d) from %u - %u",
+ coord, *diff, (unsigned)x2, (unsigned)x1);
+ return false;
+ }
+ return true;
+ }
+
+ // Returns the previous position in the mSamples array
+ // going backwards back steps.
+ //
+ // Parameters:
+ // back: number of backward steps, cannot be less than zero or greater than mSamples.
+ //
+ __attribute__((no_sanitize("integer")))
+ size_t previousPosition(ssize_t back = 1) const {
+ LOG_ALWAYS_FATAL_IF(back < 0 || (size_t)back > mSamples, "Invalid back(%zd)", back);
+ ssize_t position = mPos - back;
+ if (position < 0) position += mSize;
+ return (size_t)position;
+ }
+
+ // A generic implementation of finding the "other coordinate" with coordinates
+ // (u, v) = (x, y) or (u, v) = (y, x).
+ //
+ // Parameters:
+ // uArray: the u axis samples.
+ // vArray: the v axis samples.
+ // method: [out] how the returned value was computed.
+ // extrapolation: the slope used when extrapolating from the
+ // first sample value or the last sample value in the history.
+ // If mExtrapolateTail is set, the slope of the last line segment
+ // is used if the extrapolation parameter is zero to continue the tail of history.
+ // At this time, we do not use a different value for forward extrapolation from the
+ // head of history from backward extrapolation from the tail of history.
+ // TODO: back extrapolation value could be stored along with mX, mY in history.
+ // startValue: used only when there are no samples in history. One can detect
+ // whether there are samples in history by the method hasData().
+ //
+ __attribute__((no_sanitize("integer")))
+ T findU(T v, T *uArray, T *vArray, FindMethod *method,
+ double extrapolation, T startValue) const {
+ if (mSamples == 0) {
+ if (method != NULL) {
+ *method = FIND_METHOD_START_VALUE;
+ }
+ return startValue; // nothing yet
+ }
+ ssize_t previous = 0;
+ int32_t diff = 0;
+ for (ssize_t i = 0; i < (ssize_t)mSamples; ++i) {
+ size_t current = previousPosition(i);
+
+ // Assumption: even though the type "T" may have precision greater
+ // than 32 bits, the difference between adjacent points is limited to 32 bits.
+ diff = v - vArray[current];
+ if (diff >= 0 ||
+ (i == (ssize_t)mSamples - 1 && mExtrapolateTail && extrapolation == 0.0)) {
+ // ALOGD("depth = %zd out of %zd", i, limit);
+ if (i == 0) {
+ if (method != NULL) {
+ *method = FIND_METHOD_FORWARD_EXTRAPOLATION;
+ }
+ return uArray[current] + diff * extrapolation;
+ }
+ // interpolate / extrapolate: For this computation, we
+ // must use differentials here otherwise we have inconsistent
+ // values on modulo wrap. previous is always valid here since
+ // i > 0. we also perform rounding with the assumption
+ // that uStep, vStep, and diff are non-negative.
+ int32_t uStep = uArray[previous] - uArray[current]; // non-negative
+ int32_t vStep = vArray[previous] - vArray[current]; // positive
+ T u = uStep <= 0 || vStep <= 0 ? // we do not permit negative ustep or vstep
+ uArray[current]
+ : ((int64_t)diff * uStep + (vStep >> 1)) / vStep + uArray[current];
+ // ALOGD("u:%u diff:%d uStep:%d vStep:%d u_current:%d",
+ // u, diff, uStep, vStep, uArray[current]);
+ if (method != NULL) {
+ *method = (diff >= 0) ?
+ FIND_METHOD_INTERPOLATION : FIND_METHOD_BACKWARD_EXTRAPOLATION;
+ }
+ return u;
+ }
+ previous = current;
+ }
+ // previous is always valid here.
+ if (method != NULL) {
+ *method = FIND_METHOD_BACKWARD_EXTRAPOLATION;
+ }
+ return uArray[previous] + diff * extrapolation;
+ }
+
+private:
+ const size_t mSize; // Size of mX and mY arrays (history).
+ size_t mPos; // Index in mX and mY of last pushed data;
+ // (incremented after push) [0, mSize - 1].
+ size_t mSamples; // Number of valid samples in the array [0, mSize].
+ bool mStepValid; // Last sample step was valid (non-negative)
+ bool mExtrapolateTail; // extrapolate tail using oldest line segment
+ T * const mX; // History of X values as a circular array.
+ T * const mY; // History of Y values as a circular array.
+};
+
+} // namespace android
+
+#endif // ANDROID_LINEAR_MAP_H