From a0f4a4deeb5a597ef31411c6fe7be55cb7ef2fae Mon Sep 17 00:00:00 2001 From: Phil Taylor Date: Tue, 9 Mar 2021 17:22:16 +0000 Subject: [PATCH] Add audio resampler --- audiohandler.cpp | 223 +++++-- audiohandler.h | 12 +- packettypes.h | 3 +- resampler/arch.h | 219 +++++++ resampler/resample.c | 1240 +++++++++++++++++++++++++++++++++++ resampler/resample_sse.h | 128 ++++ resampler/speex_resampler.h | 344 ++++++++++ udphandler.cpp | 57 +- udphandler.h | 9 +- udpserver.cpp | 6 +- wfmain.cpp | 6 +- wfview.pro | 12 +- wfview.sln | 10 + wfview.vcxproj | 4 +- wfview.vcxproj.filters | 10 + 15 files changed, 2171 insertions(+), 112 deletions(-) create mode 100644 resampler/arch.h create mode 100644 resampler/resample.c create mode 100644 resampler/resample_sse.h create mode 100644 resampler/speex_resampler.h diff --git a/audiohandler.cpp b/audiohandler.cpp index 98d166d..6def10f 100644 --- a/audiohandler.cpp +++ b/audiohandler.cpp @@ -750,9 +750,13 @@ audioHandler::~audioHandler() if (audioInput != Q_NULLPTR) { delete audioInput; } + + if (resampler) { + speex_resampler_destroy(resampler); + } } -bool audioHandler::init(const quint8 bits, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool ulaw, const bool isinput, QString port) +bool audioHandler::init(const quint8 bits, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool ulaw, const bool isinput, QString port, quint8 resampleQuality) { if (isInitialized) { return false; @@ -760,7 +764,7 @@ bool audioHandler::init(const quint8 bits, const quint8 channels, const quint16 /* Always use 16 bit 48K samples internally*/ format.setSampleSize(16); format.setChannelCount(channels); - format.setSampleRate(48000); + format.setSampleRate(INTERNAL_SAMPLE_RATE); format.setCodec("audio/pcm"); format.setByteOrder(QAudioFormat::LittleEndian); format.setSampleType(QAudioFormat::SignedInt); @@ -770,7 +774,27 @@ bool audioHandler::init(const quint8 bits, const quint8 channels, const quint16 this->isInput = isinput; this->radioSampleBits = bits; this->radioSampleRate = samplerate; - this->chunkSize = this->radioSampleBits * 120; + this->radioChannels = channels; + + //this->chunkSize = (INTERNAL_SAMPLE_RATE / 25) * (radioSampleBits / 8)/2; + + this->chunkSize = 1920*radioChannels; + + qDebug(logAudio()) << "Audio chunkSize: " << this->chunkSize; + + int resample_error=0; + if (isinput) { + resampler = wf_resampler_init(radioChannels, INTERNAL_SAMPLE_RATE, samplerate, resampleQuality, &resample_error); + } + else + { + resampler = wf_resampler_init(radioChannels, samplerate, INTERNAL_SAMPLE_RATE, resampleQuality, &resample_error); + } + + + wf_resampler_get_ratio(resampler, &ratioNum, &ratioDen); + + qDebug(logAudio()) << "wf_resampler_init() returned: " << resample_error << " ratioNum" << ratioNum << " ratioDen" << ratioDen << " input " << isinput; qDebug(logAudio()) << "Got audio port name: " << port; @@ -869,8 +893,8 @@ void audioHandler::reinit() delete audioOutput; audioOutput = Q_NULLPTR; audioOutput = new QAudioOutput(deviceInfo, format, this); - audioOutput->setBufferSize((radioSampleRate/25)*(radioSampleBits/8)*2); - connect(audioOutput, SIGNAL(notify()), SLOT(notified())); + audioOutput->setBufferSize((radioSampleRate / 25) * (radioSampleBits / 8) * 2); + connect(audioOutput, SIGNAL(notify()), SLOT(notified())); connect(audioOutput, SIGNAL(stateChanged(QAudio::State)), SLOT(stateChanged(QAudio::State))); } @@ -929,11 +953,18 @@ void audioHandler::stop() } } +/// +/// This function processes the incoming audio FROM the radio and pushes it into the playback buffer *data +/// +/// +/// +/// qint64 audioHandler::readData(char* data, qint64 maxlen) { // Calculate output length, always full samples int sentlen = 0; + //qDebug(logAudio()) << "Looking for: " << maxlen << " bytes"; // We must lock the mutex for the entire time that the buffer may be modified. @@ -941,11 +972,8 @@ qint64 audioHandler::readData(char* data, qint64 maxlen) // Get next packet from buffer. if (!audioBuffer.isEmpty()) { - // Output buffer is ALWAYS 16 bit. - int divisor = 16 / radioSampleBits; - auto packet = audioBuffer.begin(); while (packet != audioBuffer.end() && sentlen < maxlen) { @@ -956,44 +984,18 @@ qint64 audioHandler::readData(char* data, qint64 maxlen) } else if (packet->seq == lastSeq+1 || packet->seq <= lastSeq) { + int send = qMin((int)maxlen-sentlen, packet->dataout.length() - packet->sent); lastSeq = packet->seq; //qDebug(logAudio()) << "Packet " << hex << packet->seq << " arrived on time " << dec << packet->time.msecsTo(QTime::currentTime()) << "ms"; - // Will this packet fit in the current buffer? - int send = qMin((int)((maxlen/divisor) - (sentlen/divisor)), packet->data.length() - packet->sent); - if (divisor == 2) - { - // Input buffer is 8bit and output buffer is 16bit - for (int f = 0; f < send; f++) - { - if (isUlaw) - qToLittleEndian(ulaw_decode[(quint8)packet->data[f+packet->sent]], data + (f * 2 + sentlen)); - else - qToLittleEndian((qint16)(packet->data[f+packet->sent] << 8) - 32640, data + (f * 2 + sentlen)); - } - } - else if (divisor == 1) - { - // 16 bit audio so just copy it in place. - //qDebug(logAudio()) << "Adding packet to buffer:" << (*packet).seq << ": " << (*packet).data.length()-(*packet).sent; - memcpy(data+sentlen, packet->data.constData()+packet->sent, send); - } - else - { - //qDebug(logAudio()) << "Invalid number of bits in audio " << radioSampleBits; - break; - } + memcpy(data + sentlen, packet->dataout.constData() + packet->sent, send); - sentlen = sentlen + (send * divisor); + sentlen = sentlen + send; - if (send == packet->data.length()) + if (send == packet->dataout.length()) { - lastSeq = packet->seq; + //qDebug(logAudio()) << "Get next packet"; packet = audioBuffer.erase(packet); // returns next packet - if (maxlen - sentlen == 0) - { - break; - } } else if (send == 0) { @@ -1019,9 +1021,7 @@ qint64 audioHandler::readData(char* data, qint64 maxlen) qint64 audioHandler::writeData(const char* data, qint64 len) { - int multiplier = (int)16 / radioSampleBits; qint64 sentlen = 0; - int tosend = 0; QMutexLocker locker(&mutex); audioPacket *current; @@ -1041,36 +1041,15 @@ qint64 audioHandler::writeData(const char* data, qint64 len) } current = &audioBuffer.last(); - tosend = qMin((int)((len - sentlen)/multiplier), (int)chunkSize-current->sent); + int send = qMin((int)(len - sentlen), (int)chunkSize-current->sent); - if (radioSampleBits == 8) { - int f = 0; - while (f < tosend) - { - quint8 outdata=0; - if (isUlaw) { - qint16 enc = qFromLittleEndian(data + ((f * multiplier) + sentlen)); - if (enc >= 0) - outdata=ulaw_encode[enc]; - else - outdata=0x7f & ulaw_encode[-enc]; - } - else { - outdata = (quint8)(((qFromLittleEndian((data + ((f * multiplier) + sentlen))) >> 8) ^ 0x80) & 0xff); - } - current->data.append((char)outdata); - f++; - } - } - else if (radioSampleBits == 16) - { - current->data.append(QByteArray::fromRawData(data + sentlen, tosend )); - } + current->datain.append(QByteArray::fromRawData(data + sentlen, send )); + + sentlen = sentlen + send; - sentlen = sentlen + (tosend * multiplier); current->seq = 0; // Not used in TX current->time = QTime::currentTime(); - current->sent = current->data.length(); + current->sent = current->datain.length(); if (current->sent == chunkSize) { @@ -1082,7 +1061,6 @@ qint64 audioHandler::writeData(const char* data, qint64 len) } - return (sentlen); // Always return the same number as we received } @@ -1139,10 +1117,59 @@ void audioHandler::stateChanged(QAudio::State state) -void audioHandler::incomingAudio(const audioPacket data) +void audioHandler::incomingAudio(audioPacket data) { if (audioOutput != Q_NULLPTR && audioOutput->state() != QAudio::StoppedState) { QMutexLocker locker(&mutex); + + // Incoming data is 8bits? + if (radioSampleBits == 8) + { + QByteArray inPacket((int)data.datain.length() * 2, (char)0xff); + qint16* in = (qint16*)inPacket.data(); + for (int f = 0; f < data.datain.length(); f++) + { + if (isUlaw) + { + in[f] = ulaw_decode[(quint8)data.datain[f]]; + } + else + { + // Convert 8-bit sample to 16-bit + in[f] = (qint16)(((quint8)data.datain[f] << 8) - 32640); + } + } + data.datain = inPacket; // Replace incoming data with converted. + } + + //qDebug(logAudio()) << "Adding packet to buffer:" << (*packet).seq << ": " << inPacket.length(); + + /* We now have an array of 16bit samples in the NATIVE samplerate of the radio + If the radio sample rate is below 48000, we need to resample. + */ + + if (ratioDen != 1) { + + // We need to resample + quint32 outFrames = ((data.datain.length() / 2) * ratioDen) / radioChannels; + quint32 inFrames = (data.datain.length() / 2) / radioChannels; + data.dataout.resize(outFrames * 2 * radioChannels); // Preset the output buffer size. + + int err = 0; + if (this->radioChannels == 1) { + err = wf_resampler_process_int(resampler, 0, (const qint16*)data.datain.constData(), &inFrames, (qint16*)data.dataout.data(), &outFrames); + } + else { + err = wf_resampler_process_interleaved_int(resampler, (const qint16*)data.datain.constData(), &inFrames, (qint16*)data.dataout.data(), &outFrames); + } + if (err) { + qDebug(logAudio()) << "Resampler error " << err << " inFrames:" << inFrames << " outFrames:" << outFrames; + } + } + else { + data.dataout = data.datain; + } + audioBuffer.push_back(data); // Sort the buffer by seq number. This is important and audio packets may have arrived out-of-order @@ -1193,9 +1220,63 @@ void audioHandler::getNextAudioChunk(QByteArray& ret) packet = audioBuffer.erase(packet); // returns next packet } else { - if (packet->data.length() == chunkSize && ret.length() == 0) + if (packet->datain.length() == chunkSize && ret.length() == 0) { - ret.append(packet->data); + /* We now have an array of samples in the computer native format (48000) + If the radio sample rate is below 48000, we need to resample. + */ + + if (ratioNum != 1) + { + // We need to resample (we are STILL 16 bit!) + quint32 outFrames = ((packet->datain.length() / 2) / ratioNum) / radioChannels; + quint32 inFrames = (packet->datain.length() / 2) / radioChannels; + packet->dataout.resize(outFrames * 2 * radioChannels); // Preset the output buffer size. + + int err = 0; + if (this->radioChannels == 1) { + err = wf_resampler_process_int(resampler, 0, (const qint16*)packet->datain.constData(), &inFrames, (qint16*)packet->dataout.data(), &outFrames); + } + else { + err = wf_resampler_process_interleaved_int(resampler, (const qint16*)packet->datain.constData(), &inFrames, (qint16*)packet->dataout.data(), &outFrames); + } + if (err) { + qDebug(logAudio()) << "Resampler error " << err << " inFrames:" << inFrames << " outFrames:" << outFrames; + } + //qDebug(logAudio()) << "Resampler run " << err << " inFrames:" << inFrames << " outFrames:" << outFrames; + //qDebug(logAudio()) << "Resampler run inLen:" << packet->datain.length() << " outLen:" << packet->dataout.length(); + if (radioSampleBits == 8) + { + packet->datain = packet->dataout; // Copy packet back to input buffer. + } + } + else if (radioSampleBits == 16 ){ + // Only copy buffer if radioSampleBits is 16, as it will be handled below otherwise. + packet->dataout = packet->datain; + } + + // Do we need to convert 16-bit to 8-bit? + if (radioSampleBits == 8) { + packet->dataout.resize(packet->datain.length() / 2); + qint16* in = (qint16*)packet->datain.data(); + for (int f = 0; f < packet->dataout.length(); f++) + { + quint8 outdata = 0; + if (isUlaw) { + qint16 enc = qFromLittleEndian(in + f); + if (enc >= 0) + outdata = ulaw_encode[enc]; + else + outdata = 0x7f & ulaw_encode[-enc]; + } + else { + outdata = (quint8)(((qFromLittleEndian(in + f) >> 8) ^ 0x80) & 0xff); + } + packet->dataout[f] = (char)outdata; + f++; + } + } + ret.append(packet->dataout); packet = audioBuffer.erase(packet); // returns next packet } else { diff --git a/audiohandler.h b/audiohandler.h index 4dae53e..47e436f 100644 --- a/audiohandler.h +++ b/audiohandler.h @@ -15,17 +15,20 @@ #include #include #include +#include "resampler/speex_resampler.h" #include //#define BUFFER_SIZE (32*1024) +#define INTERNAL_SAMPLE_RATE 48000 struct audioPacket { quint16 seq; QTime time; quint16 sent; - QByteArray data; + QByteArray datain; + QByteArray dataout; }; @@ -54,7 +57,7 @@ public: void getNextAudioChunk(QByteArray &data); bool isChunkAvailable(); public slots: - bool init(const quint8 bits, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool isulaw, const bool isinput, QString port); + bool init(const quint8 bits, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool isulaw, const bool isinput, QString port, quint8 resampleQuality); void incomingAudio(const audioPacket data); void changeLatency(const quint16 newSize); @@ -88,7 +91,12 @@ private: QAudioDeviceInfo deviceInfo; quint16 radioSampleRate; quint8 radioSampleBits; + quint8 radioChannels; QVector audioBuffer; + + SpeexResamplerState* resampler; + unsigned int ratioNum; + unsigned int ratioDen; }; #endif // AUDIOHANDLER_H diff --git a/packettypes.h b/packettypes.h index 2fcb110..17e270b 100644 --- a/packettypes.h +++ b/packettypes.h @@ -294,7 +294,8 @@ typedef union conninfo_packet { quint32 civport; // 0x7c quint32 audioport; // 0x80 quint32 txbuffer; // 0x84 - char unusedl[8]; // 0x88 + quint8 convert; // 0x88 + char unusedl[7]; // 0x89 }; }; }; diff --git a/resampler/arch.h b/resampler/arch.h new file mode 100644 index 0000000..225d727 --- /dev/null +++ b/resampler/arch.h @@ -0,0 +1,219 @@ +/* Copyright (C) 2003 Jean-Marc Valin */ +/** + @file arch.h + @brief Various architecture definitions Speex +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef ARCH_H +#define ARCH_H + +/* A couple test to catch stupid option combinations */ +#ifdef FIXED_POINT + +#if ((defined (ARM4_ASM)||defined (ARM4_ASM)) && defined(BFIN_ASM)) || (defined (ARM4_ASM)&&defined(ARM5E_ASM)) +#error Make up your mind. What CPU do you have? +#endif + +#else + +#if defined (ARM4_ASM) || defined(ARM5E_ASM) || defined(BFIN_ASM) +#error I suppose you can have a [ARM4/ARM5E/Blackfin] that has float instructions? +#endif + +#endif + +#ifndef OUTSIDE_SPEEX +#include "speex/speexdsp_types.h" +#endif + +#define ABS(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute integer value. */ +#define ABS16(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute 16-bit value. */ +#define MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Maximum 16-bit value. */ +#define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-bit value. */ +#define ABS32(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute 32-bit value. */ +#define MIN32(a,b) ((a) < (b) ? (a) : (b)) /**< Maximum 32-bit value. */ +#define MAX32(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 32-bit value. */ + +#ifdef FIXED_POINT + +typedef spx_int16_t spx_word16_t; +typedef spx_int32_t spx_word32_t; +typedef spx_word32_t spx_mem_t; +typedef spx_word16_t spx_coef_t; +typedef spx_word16_t spx_lsp_t; +typedef spx_word32_t spx_sig_t; + +#define Q15ONE 32767 + +#define LPC_SCALING 8192 +#define SIG_SCALING 16384 +#define LSP_SCALING 8192. +#define GAMMA_SCALING 32768. +#define GAIN_SCALING 64 +#define GAIN_SCALING_1 0.015625 + +#define LPC_SHIFT 13 +#define LSP_SHIFT 13 +#define SIG_SHIFT 14 +#define GAIN_SHIFT 6 + +#define WORD2INT(x) ((x) < -32767 ? -32768 : ((x) > 32766 ? 32767 : (x))) + +#define VERY_SMALL 0 +#define VERY_LARGE32 ((spx_word32_t)2147483647) +#define VERY_LARGE16 ((spx_word16_t)32767) +#define Q15_ONE ((spx_word16_t)32767) + + +#ifdef FIXED_DEBUG +#include "fixed_debug.h" +#else + +#include "fixed_generic.h" + +#ifdef ARM5E_ASM +#include "fixed_arm5e.h" +#elif defined (ARM4_ASM) +#include "fixed_arm4.h" +#elif defined (BFIN_ASM) +#include "fixed_bfin.h" +#endif + +#endif + + +#else + +typedef float spx_mem_t; +typedef float spx_coef_t; +typedef float spx_lsp_t; +typedef float spx_sig_t; +typedef float spx_word16_t; +typedef float spx_word32_t; + +#define Q15ONE 1.0f +#define LPC_SCALING 1.f +#define SIG_SCALING 1.f +#define LSP_SCALING 1.f +#define GAMMA_SCALING 1.f +#define GAIN_SCALING 1.f +#define GAIN_SCALING_1 1.f + + +#define VERY_SMALL 1e-15f +#define VERY_LARGE32 1e15f +#define VERY_LARGE16 1e15f +#define Q15_ONE ((spx_word16_t)1.f) + +#define QCONST16(x,bits) (x) +#define QCONST32(x,bits) (x) + +#define NEG16(x) (-(x)) +#define NEG32(x) (-(x)) +#define EXTRACT16(x) (x) +#define EXTEND32(x) (x) +#define SHR16(a,shift) (a) +#define SHL16(a,shift) (a) +#define SHR32(a,shift) (a) +#define SHL32(a,shift) (a) +#define PSHR16(a,shift) (a) +#define PSHR32(a,shift) (a) +#define VSHR32(a,shift) (a) +#define SATURATE16(x,a) (x) +#define SATURATE32(x,a) (x) +#define SATURATE32PSHR(x,shift,a) (x) + +#define PSHR(a,shift) (a) +#define SHR(a,shift) (a) +#define SHL(a,shift) (a) +#define SATURATE(x,a) (x) + +#define ADD16(a,b) ((a)+(b)) +#define SUB16(a,b) ((a)-(b)) +#define ADD32(a,b) ((a)+(b)) +#define SUB32(a,b) ((a)-(b)) +#define MULT16_16_16(a,b) ((a)*(b)) +#define MULT16_16(a,b) ((spx_word32_t)(a)*(spx_word32_t)(b)) +#define MAC16_16(c,a,b) ((c)+(spx_word32_t)(a)*(spx_word32_t)(b)) + +#define MULT16_32_Q11(a,b) ((a)*(b)) +#define MULT16_32_Q13(a,b) ((a)*(b)) +#define MULT16_32_Q14(a,b) ((a)*(b)) +#define MULT16_32_Q15(a,b) ((a)*(b)) +#define MULT16_32_P15(a,b) ((a)*(b)) + +#define MAC16_32_Q11(c,a,b) ((c)+(a)*(b)) +#define MAC16_32_Q15(c,a,b) ((c)+(a)*(b)) + +#define MAC16_16_Q11(c,a,b) ((c)+(a)*(b)) +#define MAC16_16_Q13(c,a,b) ((c)+(a)*(b)) +#define MAC16_16_P13(c,a,b) ((c)+(a)*(b)) +#define MULT16_16_Q11_32(a,b) ((a)*(b)) +#define MULT16_16_Q13(a,b) ((a)*(b)) +#define MULT16_16_Q14(a,b) ((a)*(b)) +#define MULT16_16_Q15(a,b) ((a)*(b)) +#define MULT16_16_P15(a,b) ((a)*(b)) +#define MULT16_16_P13(a,b) ((a)*(b)) +#define MULT16_16_P14(a,b) ((a)*(b)) + +#define DIV32_16(a,b) (((spx_word32_t)(a))/(spx_word16_t)(b)) +#define PDIV32_16(a,b) (((spx_word32_t)(a))/(spx_word16_t)(b)) +#define DIV32(a,b) (((spx_word32_t)(a))/(spx_word32_t)(b)) +#define PDIV32(a,b) (((spx_word32_t)(a))/(spx_word32_t)(b)) + +#define WORD2INT(x) ((x) < -32767.5f ? -32768 : \ + ((x) > 32766.5f ? 32767 : (spx_int16_t)floor(.5 + (x)))) +#endif + + +#if defined (CONFIG_TI_C54X) || defined (CONFIG_TI_C55X) + +/* 2 on TI C5x DSP */ +#define BYTES_PER_CHAR 2 +#define BITS_PER_CHAR 16 +#define LOG2_BITS_PER_CHAR 4 + +#else + +#define BYTES_PER_CHAR 1 +#define BITS_PER_CHAR 8 +#define LOG2_BITS_PER_CHAR 3 + +#endif + + + +#ifdef FIXED_DEBUG +extern long long spx_mips; +#endif + + +#endif diff --git a/resampler/resample.c b/resampler/resample.c new file mode 100644 index 0000000..6c58f9f --- /dev/null +++ b/resampler/resample.c @@ -0,0 +1,1240 @@ +/* Copyright (C) 2007-2008 Jean-Marc Valin + Copyright (C) 2008 Thorvald Natvig + + File: resample.c + Arbitrary resampling code + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + 3. The name of the author may not be used to endorse or promote products + derived from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, + INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN + ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE. +*/ + +/* + The design goals of this code are: + - Very fast algorithm + - SIMD-friendly algorithm + - Low memory requirement + - Good *perceptual* quality (and not best SNR) + + Warning: This resampler is relatively new. Although I think I got rid of + all the major bugs and I don't expect the API to change anymore, there + may be something I've missed. So use with caution. + + This algorithm is based on this original resampling algorithm: + Smith, Julius O. Digital Audio Resampling Home Page + Center for Computer Research in Music and Acoustics (CCRMA), + Stanford University, 2007. + Web published at https://ccrma.stanford.edu/~jos/resample/. + + There is one main difference, though. This resampler uses cubic + interpolation instead of linear interpolation in the above paper. This + makes the table much smaller and makes it possible to compute that table + on a per-stream basis. In turn, being able to tweak the table for each + stream makes it possible to both reduce complexity on simple ratios + (e.g. 2/3), and get rid of the rounding operations in the inner loop. + The latter both reduces CPU time and makes the algorithm more SIMD-friendly. +*/ + + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifdef OUTSIDE_SPEEX +#include +static void *speex_alloc(int size) {return calloc(size,1);} +static void *speex_realloc(void *ptr, int size) {return realloc(ptr, size);} +static void speex_free(void *ptr) {free(ptr);} +#ifndef EXPORT +#define EXPORT +#endif +#include "speex_resampler.h" +#include "arch.h" +#else /* OUTSIDE_SPEEX */ + +#include "resampler/speex_resampler.h" +#include "resampler/arch.h" +#include "resampler/os_support.h" +#endif /* OUTSIDE_SPEEX */ + +#include +#include + +#ifndef M_PI +#define M_PI 3.14159265358979323846 +#endif + +#define IMAX(a,b) ((a) > (b) ? (a) : (b)) +#define IMIN(a,b) ((a) < (b) ? (a) : (b)) + +#ifndef NULL +#define NULL 0 +#endif + +#ifndef UINT32_MAX +#define UINT32_MAX 4294967295U +#endif + +#if defined(__SSE__) && !defined(FIXED_POINT) +#include "resample_sse.h" +#endif + +#ifdef USE_NEON +#include "resample_neon.h" +#endif + +/* Numer of elements to allocate on the stack */ +#ifdef VAR_ARRAYS +#define FIXED_STACK_ALLOC 8192 +#else +#define FIXED_STACK_ALLOC 1024 +#endif + +typedef int (*resampler_basic_func)(SpeexResamplerState *, spx_uint32_t , const spx_word16_t *, spx_uint32_t *, spx_word16_t *, spx_uint32_t *); + +struct SpeexResamplerState_ { + spx_uint32_t in_rate; + spx_uint32_t out_rate; + spx_uint32_t num_rate; + spx_uint32_t den_rate; + + int quality; + spx_uint32_t nb_channels; + spx_uint32_t filt_len; + spx_uint32_t mem_alloc_size; + spx_uint32_t buffer_size; + int int_advance; + int frac_advance; + float cutoff; + spx_uint32_t oversample; + int initialised; + int started; + + /* These are per-channel */ + spx_int32_t *last_sample; + spx_uint32_t *samp_frac_num; + spx_uint32_t *magic_samples; + + spx_word16_t *mem; + spx_word16_t *sinc_table; + spx_uint32_t sinc_table_length; + resampler_basic_func resampler_ptr; + + int in_stride; + int out_stride; +} ; + +static const double kaiser12_table[68] = { + 0.99859849, 1.00000000, 0.99859849, 0.99440475, 0.98745105, 0.97779076, + 0.96549770, 0.95066529, 0.93340547, 0.91384741, 0.89213598, 0.86843014, + 0.84290116, 0.81573067, 0.78710866, 0.75723148, 0.72629970, 0.69451601, + 0.66208321, 0.62920216, 0.59606986, 0.56287762, 0.52980938, 0.49704014, + 0.46473455, 0.43304576, 0.40211431, 0.37206735, 0.34301800, 0.31506490, + 0.28829195, 0.26276832, 0.23854851, 0.21567274, 0.19416736, 0.17404546, + 0.15530766, 0.13794294, 0.12192957, 0.10723616, 0.09382272, 0.08164178, + 0.07063950, 0.06075685, 0.05193064, 0.04409466, 0.03718069, 0.03111947, + 0.02584161, 0.02127838, 0.01736250, 0.01402878, 0.01121463, 0.00886058, + 0.00691064, 0.00531256, 0.00401805, 0.00298291, 0.00216702, 0.00153438, + 0.00105297, 0.00069463, 0.00043489, 0.00025272, 0.00013031, 0.0000527734, + 0.00001000, 0.00000000}; +/* +static const double kaiser12_table[36] = { + 0.99440475, 1.00000000, 0.99440475, 0.97779076, 0.95066529, 0.91384741, + 0.86843014, 0.81573067, 0.75723148, 0.69451601, 0.62920216, 0.56287762, + 0.49704014, 0.43304576, 0.37206735, 0.31506490, 0.26276832, 0.21567274, + 0.17404546, 0.13794294, 0.10723616, 0.08164178, 0.06075685, 0.04409466, + 0.03111947, 0.02127838, 0.01402878, 0.00886058, 0.00531256, 0.00298291, + 0.00153438, 0.00069463, 0.00025272, 0.0000527734, 0.00000500, 0.00000000}; +*/ +static const double kaiser10_table[36] = { + 0.99537781, 1.00000000, 0.99537781, 0.98162644, 0.95908712, 0.92831446, + 0.89005583, 0.84522401, 0.79486424, 0.74011713, 0.68217934, 0.62226347, + 0.56155915, 0.50119680, 0.44221549, 0.38553619, 0.33194107, 0.28205962, + 0.23636152, 0.19515633, 0.15859932, 0.12670280, 0.09935205, 0.07632451, + 0.05731132, 0.04193980, 0.02979584, 0.02044510, 0.01345224, 0.00839739, + 0.00488951, 0.00257636, 0.00115101, 0.00035515, 0.00000000, 0.00000000}; + +static const double kaiser8_table[36] = { + 0.99635258, 1.00000000, 0.99635258, 0.98548012, 0.96759014, 0.94302200, + 0.91223751, 0.87580811, 0.83439927, 0.78875245, 0.73966538, 0.68797126, + 0.63451750, 0.58014482, 0.52566725, 0.47185369, 0.41941150, 0.36897272, + 0.32108304, 0.27619388, 0.23465776, 0.19672670, 0.16255380, 0.13219758, + 0.10562887, 0.08273982, 0.06335451, 0.04724088, 0.03412321, 0.02369490, + 0.01563093, 0.00959968, 0.00527363, 0.00233883, 0.00050000, 0.00000000}; + +static const double kaiser6_table[36] = { + 0.99733006, 1.00000000, 0.99733006, 0.98935595, 0.97618418, 0.95799003, + 0.93501423, 0.90755855, 0.87598009, 0.84068475, 0.80211977, 0.76076565, + 0.71712752, 0.67172623, 0.62508937, 0.57774224, 0.53019925, 0.48295561, + 0.43647969, 0.39120616, 0.34752997, 0.30580127, 0.26632152, 0.22934058, + 0.19505503, 0.16360756, 0.13508755, 0.10953262, 0.08693120, 0.06722600, + 0.05031820, 0.03607231, 0.02432151, 0.01487334, 0.00752000, 0.00000000}; + +struct FuncDef { + const double *table; + int oversample; +}; + +static const struct FuncDef kaiser12_funcdef = {kaiser12_table, 64}; +#define KAISER12 (&kaiser12_funcdef) +static const struct FuncDef kaiser10_funcdef = {kaiser10_table, 32}; +#define KAISER10 (&kaiser10_funcdef) +static const struct FuncDef kaiser8_funcdef = {kaiser8_table, 32}; +#define KAISER8 (&kaiser8_funcdef) +static const struct FuncDef kaiser6_funcdef = {kaiser6_table, 32}; +#define KAISER6 (&kaiser6_funcdef) + +struct QualityMapping { + int base_length; + int oversample; + float downsample_bandwidth; + float upsample_bandwidth; + const struct FuncDef *window_func; +}; + + +/* This table maps conversion quality to internal parameters. There are two + reasons that explain why the up-sampling bandwidth is larger than the + down-sampling bandwidth: + 1) When up-sampling, we can assume that the spectrum is already attenuated + close to the Nyquist rate (from an A/D or a previous resampling filter) + 2) Any aliasing that occurs very close to the Nyquist rate will be masked + by the sinusoids/noise just below the Nyquist rate (guaranteed only for + up-sampling). +*/ +static const struct QualityMapping quality_map[11] = { + { 8, 4, 0.830f, 0.860f, KAISER6 }, /* Q0 */ + { 16, 4, 0.850f, 0.880f, KAISER6 }, /* Q1 */ + { 32, 4, 0.882f, 0.910f, KAISER6 }, /* Q2 */ /* 82.3% cutoff ( ~60 dB stop) 6 */ + { 48, 8, 0.895f, 0.917f, KAISER8 }, /* Q3 */ /* 84.9% cutoff ( ~80 dB stop) 8 */ + { 64, 8, 0.921f, 0.940f, KAISER8 }, /* Q4 */ /* 88.7% cutoff ( ~80 dB stop) 8 */ + { 80, 16, 0.922f, 0.940f, KAISER10}, /* Q5 */ /* 89.1% cutoff (~100 dB stop) 10 */ + { 96, 16, 0.940f, 0.945f, KAISER10}, /* Q6 */ /* 91.5% cutoff (~100 dB stop) 10 */ + {128, 16, 0.950f, 0.950f, KAISER10}, /* Q7 */ /* 93.1% cutoff (~100 dB stop) 10 */ + {160, 16, 0.960f, 0.960f, KAISER10}, /* Q8 */ /* 94.5% cutoff (~100 dB stop) 10 */ + {192, 32, 0.968f, 0.968f, KAISER12}, /* Q9 */ /* 95.5% cutoff (~100 dB stop) 10 */ + {256, 32, 0.975f, 0.975f, KAISER12}, /* Q10 */ /* 96.6% cutoff (~100 dB stop) 10 */ +}; +/*8,24,40,56,80,104,128,160,200,256,320*/ +static double compute_func(float x, const struct FuncDef *func) +{ + float y, frac; + double interp[4]; + int ind; + y = x*func->oversample; + ind = (int)floor(y); + frac = (y-ind); + /* CSE with handle the repeated powers */ + interp[3] = -0.1666666667*frac + 0.1666666667*(frac*frac*frac); + interp[2] = frac + 0.5*(frac*frac) - 0.5*(frac*frac*frac); + /*interp[2] = 1.f - 0.5f*frac - frac*frac + 0.5f*frac*frac*frac;*/ + interp[0] = -0.3333333333*frac + 0.5*(frac*frac) - 0.1666666667*(frac*frac*frac); + /* Just to make sure we don't have rounding problems */ + interp[1] = 1.f-interp[3]-interp[2]-interp[0]; + + /*sum = frac*accum[1] + (1-frac)*accum[2];*/ + return interp[0]*func->table[ind] + interp[1]*func->table[ind+1] + interp[2]*func->table[ind+2] + interp[3]*func->table[ind+3]; +} + +#if 0 +#include +int main(int argc, char **argv) +{ + int i; + for (i=0;i<256;i++) + { + printf ("%f\n", compute_func(i/256., KAISER12)); + } + return 0; +} +#endif + +#ifdef FIXED_POINT +/* The slow way of computing a sinc for the table. Should improve that some day */ +static spx_word16_t sinc(float cutoff, float x, int N, const struct FuncDef *window_func) +{ + /*fprintf (stderr, "%f ", x);*/ + float xx = x * cutoff; + if (fabs(x)<1e-6f) + return WORD2INT(32768.*cutoff); + else if (fabs(x) > .5f*N) + return 0; + /*FIXME: Can it really be any slower than this? */ + return WORD2INT(32768.*cutoff*sin(M_PI*xx)/(M_PI*xx) * compute_func(fabs(2.*x/N), window_func)); +} +#else +/* The slow way of computing a sinc for the table. Should improve that some day */ +static spx_word16_t sinc(float cutoff, float x, int N, const struct FuncDef *window_func) +{ + /*fprintf (stderr, "%f ", x);*/ + float xx = x * cutoff; + if (fabs(x)<1e-6) + return cutoff; + else if (fabs(x) > .5*N) + return 0; + /*FIXME: Can it really be any slower than this? */ + return cutoff*sin(M_PI*xx)/(M_PI*xx) * compute_func(fabs(2.*x/N), window_func); +} +#endif + +#ifdef FIXED_POINT +static void cubic_coef(spx_word16_t x, spx_word16_t interp[4]) +{ + /* Compute interpolation coefficients. I'm not sure whether this corresponds to cubic interpolation + but I know it's MMSE-optimal on a sinc */ + spx_word16_t x2, x3; + x2 = MULT16_16_P15(x, x); + x3 = MULT16_16_P15(x, x2); + interp[0] = PSHR32(MULT16_16(QCONST16(-0.16667f, 15),x) + MULT16_16(QCONST16(0.16667f, 15),x3),15); + interp[1] = EXTRACT16(EXTEND32(x) + SHR32(SUB32(EXTEND32(x2),EXTEND32(x3)),1)); + interp[3] = PSHR32(MULT16_16(QCONST16(-0.33333f, 15),x) + MULT16_16(QCONST16(.5f,15),x2) - MULT16_16(QCONST16(0.16667f, 15),x3),15); + /* Just to make sure we don't have rounding problems */ + interp[2] = Q15_ONE-interp[0]-interp[1]-interp[3]; + if (interp[2]<32767) + interp[2]+=1; +} +#else +static void cubic_coef(spx_word16_t frac, spx_word16_t interp[4]) +{ + /* Compute interpolation coefficients. I'm not sure whether this corresponds to cubic interpolation + but I know it's MMSE-optimal on a sinc */ + interp[0] = -0.16667f*frac + 0.16667f*frac*frac*frac; + interp[1] = frac + 0.5f*frac*frac - 0.5f*frac*frac*frac; + /*interp[2] = 1.f - 0.5f*frac - frac*frac + 0.5f*frac*frac*frac;*/ + interp[3] = -0.33333f*frac + 0.5f*frac*frac - 0.16667f*frac*frac*frac; + /* Just to make sure we don't have rounding problems */ + interp[2] = 1.-interp[0]-interp[1]-interp[3]; +} +#endif + +static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + const int N = st->filt_len; + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const spx_word16_t *sinc_table = st->sinc_table; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + spx_word32_t sum; + + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + const spx_word16_t *sinct = & sinc_table[samp_frac_num*N]; + const spx_word16_t *iptr = & in[last_sample]; + +#ifndef OVERRIDE_INNER_PRODUCT_SINGLE + int j; + sum = 0; + for(j=0;j= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} + +#ifdef FIXED_POINT +#else +/* This is the same as the previous function, except with a double-precision accumulator */ +static int resampler_basic_direct_double(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + const int N = st->filt_len; + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const spx_word16_t *sinc_table = st->sinc_table; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + double sum; + + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + const spx_word16_t *sinct = & sinc_table[samp_frac_num*N]; + const spx_word16_t *iptr = & in[last_sample]; + +#ifndef OVERRIDE_INNER_PRODUCT_DOUBLE + int j; + double accum[4] = {0,0,0,0}; + + for(j=0;j= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} +#endif + +static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + const int N = st->filt_len; + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + spx_word32_t sum; + + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + const spx_word16_t *iptr = & in[last_sample]; + + const int offset = samp_frac_num*st->oversample/st->den_rate; +#ifdef FIXED_POINT + const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate); +#else + const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate; +#endif + spx_word16_t interp[4]; + + +#ifndef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE + int j; + spx_word32_t accum[4] = {0,0,0,0}; + + for(j=0;jsinc_table[4+(j+1)*st->oversample-offset-2]); + accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]); + accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]); + accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]); + } + + cubic_coef(frac, interp); + sum = MULT16_32_Q15(interp[0],SHR32(accum[0], 1)) + MULT16_32_Q15(interp[1],SHR32(accum[1], 1)) + MULT16_32_Q15(interp[2],SHR32(accum[2], 1)) + MULT16_32_Q15(interp[3],SHR32(accum[3], 1)); + sum = SATURATE32PSHR(sum, 15, 32767); +#else + cubic_coef(frac, interp); + sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp); +#endif + + out[out_stride * out_sample++] = sum; + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} + +#ifdef FIXED_POINT +#else +/* This is the same as the previous function, except with a double-precision accumulator */ +static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + const int N = st->filt_len; + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + spx_word32_t sum; + + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + const spx_word16_t *iptr = & in[last_sample]; + + const int offset = samp_frac_num*st->oversample/st->den_rate; +#ifdef FIXED_POINT + const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate); +#else + const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate; +#endif + spx_word16_t interp[4]; + + +#ifndef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE + int j; + double accum[4] = {0,0,0,0}; + + for(j=0;jsinc_table[4+(j+1)*st->oversample-offset-2]); + accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]); + accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]); + accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]); + } + + cubic_coef(frac, interp); + sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]); +#else + cubic_coef(frac, interp); + sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp); +#endif + + out[out_stride * out_sample++] = PSHR32(sum,15); + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} +#endif + +/* This resampler is used to produce zero output in situations where memory + for the filter could not be allocated. The expected numbers of input and + output samples are still processed so that callers failing to check error + codes are not surprised, possibly getting into infinite loops. */ +static int resampler_basic_zero(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + + (void)in; + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + out[out_stride * out_sample++] = 0; + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} + +static int multiply_frac(spx_uint32_t *result, spx_uint32_t value, spx_uint32_t num, spx_uint32_t den) +{ + spx_uint32_t major = value / den; + spx_uint32_t remain = value % den; + /* TODO: Could use 64 bits operation to check for overflow. But only guaranteed in C99+ */ + if (remain > UINT32_MAX / num || major > UINT32_MAX / num + || major * num > UINT32_MAX - remain * num / den) + return RESAMPLER_ERR_OVERFLOW; + *result = remain * num / den + major * num; + return RESAMPLER_ERR_SUCCESS; +} + +static int update_filter(SpeexResamplerState *st) +{ + spx_uint32_t old_length = st->filt_len; + spx_uint32_t old_alloc_size = st->mem_alloc_size; + int use_direct; + spx_uint32_t min_sinc_table_length; + spx_uint32_t min_alloc_size; + + st->int_advance = st->num_rate/st->den_rate; + st->frac_advance = st->num_rate%st->den_rate; + st->oversample = quality_map[st->quality].oversample; + st->filt_len = quality_map[st->quality].base_length; + + if (st->num_rate > st->den_rate) + { + /* down-sampling */ + st->cutoff = quality_map[st->quality].downsample_bandwidth * st->den_rate / st->num_rate; + if (multiply_frac(&st->filt_len,st->filt_len,st->num_rate,st->den_rate) != RESAMPLER_ERR_SUCCESS) + goto fail; + /* Round up to make sure we have a multiple of 8 for SSE */ + st->filt_len = ((st->filt_len-1)&(~0x7))+8; + if (2*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (4*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (8*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (16*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (st->oversample < 1) + st->oversample = 1; + } else { + /* up-sampling */ + st->cutoff = quality_map[st->quality].upsample_bandwidth; + } + +#ifdef RESAMPLE_FULL_SINC_TABLE + use_direct = 1; + if (INT_MAX/sizeof(spx_word16_t)/st->den_rate < st->filt_len) + goto fail; +#else + /* Choose the resampling type that requires the least amount of memory */ + use_direct = st->filt_len*st->den_rate <= st->filt_len*st->oversample+8 + && INT_MAX/sizeof(spx_word16_t)/st->den_rate >= st->filt_len; +#endif + if (use_direct) + { + min_sinc_table_length = st->filt_len*st->den_rate; + } else { + if ((INT_MAX/sizeof(spx_word16_t)-8)/st->oversample < st->filt_len) + goto fail; + + min_sinc_table_length = st->filt_len*st->oversample+8; + } + if (st->sinc_table_length < min_sinc_table_length) + { + spx_word16_t *sinc_table = (spx_word16_t *)speex_realloc(st->sinc_table,min_sinc_table_length*sizeof(spx_word16_t)); + if (!sinc_table) + goto fail; + + st->sinc_table = sinc_table; + st->sinc_table_length = min_sinc_table_length; + } + if (use_direct) + { + spx_uint32_t i; + for (i=0;iden_rate;i++) + { + spx_int32_t j; + for (j=0;jfilt_len;j++) + { + st->sinc_table[i*st->filt_len+j] = sinc(st->cutoff,((j-(spx_int32_t)st->filt_len/2+1)-((float)i)/st->den_rate), st->filt_len, quality_map[st->quality].window_func); + } + } +#ifdef FIXED_POINT + st->resampler_ptr = resampler_basic_direct_single; +#else + if (st->quality>8) + st->resampler_ptr = resampler_basic_direct_double; + else + st->resampler_ptr = resampler_basic_direct_single; +#endif + /*fprintf (stderr, "resampler uses direct sinc table and normalised cutoff %f\n", cutoff);*/ + } else { + spx_int32_t i; + for (i=-4;i<(spx_int32_t)(st->oversample*st->filt_len+4);i++) + st->sinc_table[i+4] = sinc(st->cutoff,(i/(float)st->oversample - st->filt_len/2), st->filt_len, quality_map[st->quality].window_func); +#ifdef FIXED_POINT + st->resampler_ptr = resampler_basic_interpolate_single; +#else + if (st->quality>8) + st->resampler_ptr = resampler_basic_interpolate_double; + else + st->resampler_ptr = resampler_basic_interpolate_single; +#endif + /*fprintf (stderr, "resampler uses interpolated sinc table and normalised cutoff %f\n", cutoff);*/ + } + + /* Here's the place where we update the filter memory to take into account + the change in filter length. It's probably the messiest part of the code + due to handling of lots of corner cases. */ + + /* Adding buffer_size to filt_len won't overflow here because filt_len + could be multiplied by sizeof(spx_word16_t) above. */ + min_alloc_size = st->filt_len-1 + st->buffer_size; + if (min_alloc_size > st->mem_alloc_size) + { + spx_word16_t *mem; + if (INT_MAX/sizeof(spx_word16_t)/st->nb_channels < min_alloc_size) + goto fail; + else if (!(mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*min_alloc_size * sizeof(*mem)))) + goto fail; + + st->mem = mem; + st->mem_alloc_size = min_alloc_size; + } + if (!st->started) + { + spx_uint32_t i; + for (i=0;inb_channels*st->mem_alloc_size;i++) + st->mem[i] = 0; + /*speex_warning("reinit filter");*/ + } else if (st->filt_len > old_length) + { + spx_uint32_t i; + /* Increase the filter length */ + /*speex_warning("increase filter size");*/ + for (i=st->nb_channels;i--;) + { + spx_uint32_t j; + spx_uint32_t olen = old_length; + /*if (st->magic_samples[i])*/ + { + /* Try and remove the magic samples as if nothing had happened */ + + /* FIXME: This is wrong but for now we need it to avoid going over the array bounds */ + olen = old_length + 2*st->magic_samples[i]; + for (j=old_length-1+st->magic_samples[i];j--;) + st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]] = st->mem[i*old_alloc_size+j]; + for (j=0;jmagic_samples[i];j++) + st->mem[i*st->mem_alloc_size+j] = 0; + st->magic_samples[i] = 0; + } + if (st->filt_len > olen) + { + /* If the new filter length is still bigger than the "augmented" length */ + /* Copy data going backward */ + for (j=0;jmem[i*st->mem_alloc_size+(st->filt_len-2-j)] = st->mem[i*st->mem_alloc_size+(olen-2-j)]; + /* Then put zeros for lack of anything better */ + for (;jfilt_len-1;j++) + st->mem[i*st->mem_alloc_size+(st->filt_len-2-j)] = 0; + /* Adjust last_sample */ + st->last_sample[i] += (st->filt_len - olen)/2; + } else { + /* Put back some of the magic! */ + st->magic_samples[i] = (olen - st->filt_len)/2; + for (j=0;jfilt_len-1+st->magic_samples[i];j++) + st->mem[i*st->mem_alloc_size+j] = st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]]; + } + } + } else if (st->filt_len < old_length) + { + spx_uint32_t i; + /* Reduce filter length, this a bit tricky. We need to store some of the memory as "magic" + samples so they can be used directly as input the next time(s) */ + for (i=0;inb_channels;i++) + { + spx_uint32_t j; + spx_uint32_t old_magic = st->magic_samples[i]; + st->magic_samples[i] = (old_length - st->filt_len)/2; + /* We must copy some of the memory that's no longer used */ + /* Copy data going backward */ + for (j=0;jfilt_len-1+st->magic_samples[i]+old_magic;j++) + st->mem[i*st->mem_alloc_size+j] = st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]]; + st->magic_samples[i] += old_magic; + } + } + return RESAMPLER_ERR_SUCCESS; + +fail: + st->resampler_ptr = resampler_basic_zero; + /* st->mem may still contain consumed input samples for the filter. + Restore filt_len so that filt_len - 1 still points to the position after + the last of these samples. */ + st->filt_len = old_length; + return RESAMPLER_ERR_ALLOC_FAILED; +} + +EXPORT SpeexResamplerState *speex_resampler_init(spx_uint32_t nb_channels, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) +{ + return speex_resampler_init_frac(nb_channels, in_rate, out_rate, in_rate, out_rate, quality, err); +} + +EXPORT SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) +{ + SpeexResamplerState *st; + int filter_err; + + if (nb_channels == 0 || ratio_num == 0 || ratio_den == 0 || quality > 10 || quality < 0) + { + if (err) + *err = RESAMPLER_ERR_INVALID_ARG; + return NULL; + } + st = (SpeexResamplerState *)speex_alloc(sizeof(SpeexResamplerState)); + if (!st) + { + if (err) + *err = RESAMPLER_ERR_ALLOC_FAILED; + return NULL; + } + st->initialised = 0; + st->started = 0; + st->in_rate = 0; + st->out_rate = 0; + st->num_rate = 0; + st->den_rate = 0; + st->quality = -1; + st->sinc_table_length = 0; + st->mem_alloc_size = 0; + st->filt_len = 0; + st->mem = 0; + st->resampler_ptr = 0; + + st->cutoff = 1.f; + st->nb_channels = nb_channels; + st->in_stride = 1; + st->out_stride = 1; + + st->buffer_size = 160; + + /* Per channel data */ + if (!(st->last_sample = (spx_int32_t*)speex_alloc(nb_channels*sizeof(spx_int32_t)))) + goto fail; + if (!(st->magic_samples = (spx_uint32_t*)speex_alloc(nb_channels*sizeof(spx_uint32_t)))) + goto fail; + if (!(st->samp_frac_num = (spx_uint32_t*)speex_alloc(nb_channels*sizeof(spx_uint32_t)))) + goto fail; + + speex_resampler_set_quality(st, quality); + speex_resampler_set_rate_frac(st, ratio_num, ratio_den, in_rate, out_rate); + + filter_err = update_filter(st); + if (filter_err == RESAMPLER_ERR_SUCCESS) + { + st->initialised = 1; + } else { + speex_resampler_destroy(st); + st = NULL; + } + if (err) + *err = filter_err; + + return st; + +fail: + if (err) + *err = RESAMPLER_ERR_ALLOC_FAILED; + speex_resampler_destroy(st); + return NULL; +} + +EXPORT void speex_resampler_destroy(SpeexResamplerState *st) +{ + speex_free(st->mem); + speex_free(st->sinc_table); + speex_free(st->last_sample); + speex_free(st->magic_samples); + speex_free(st->samp_frac_num); + speex_free(st); +} + +static int speex_resampler_process_native(SpeexResamplerState *st, spx_uint32_t channel_index, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + int j=0; + const int N = st->filt_len; + int out_sample = 0; + spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size; + spx_uint32_t ilen; + + st->started = 1; + + /* Call the right resampler through the function ptr */ + out_sample = st->resampler_ptr(st, channel_index, mem, in_len, out, out_len); + + if (st->last_sample[channel_index] < (spx_int32_t)*in_len) + *in_len = st->last_sample[channel_index]; + *out_len = out_sample; + st->last_sample[channel_index] -= *in_len; + + ilen = *in_len; + + for(j=0;jmagic_samples[channel_index]; + spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size; + const int N = st->filt_len; + + speex_resampler_process_native(st, channel_index, &tmp_in_len, *out, &out_len); + + st->magic_samples[channel_index] -= tmp_in_len; + + /* If we couldn't process all "magic" input samples, save the rest for next time */ + if (st->magic_samples[channel_index]) + { + spx_uint32_t i; + for (i=0;imagic_samples[channel_index];i++) + mem[N-1+i]=mem[N-1+i+tmp_in_len]; + } + *out += out_len*st->out_stride; + return out_len; +} + +#ifdef FIXED_POINT +EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +#else +EXPORT int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) +#endif +{ + int j; + spx_uint32_t ilen = *in_len; + spx_uint32_t olen = *out_len; + spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size; + const int filt_offs = st->filt_len - 1; + const spx_uint32_t xlen = st->mem_alloc_size - filt_offs; + const int istride = st->in_stride; + + if (st->magic_samples[channel_index]) + olen -= speex_resampler_magic(st, channel_index, &out, olen); + if (! st->magic_samples[channel_index]) { + while (ilen && olen) { + spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen; + spx_uint32_t ochunk = olen; + + if (in) { + for(j=0;jout_stride; + if (in) + in += ichunk * istride; + } + } + *in_len -= ilen; + *out_len -= olen; + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +#ifdef FIXED_POINT +EXPORT int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) +#else +EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +#endif +{ + int j; + const int istride_save = st->in_stride; + const int ostride_save = st->out_stride; + spx_uint32_t ilen = *in_len; + spx_uint32_t olen = *out_len; + spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size; + const spx_uint32_t xlen = st->mem_alloc_size - (st->filt_len - 1); +#ifdef VAR_ARRAYS + const unsigned int ylen = (olen < FIXED_STACK_ALLOC) ? olen : FIXED_STACK_ALLOC; + spx_word16_t ystack[ylen]; +#else + const unsigned int ylen = FIXED_STACK_ALLOC; + spx_word16_t ystack[FIXED_STACK_ALLOC]; +#endif + + st->out_stride = 1; + + while (ilen && olen) { + spx_word16_t *y = ystack; + spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen; + spx_uint32_t ochunk = (olen > ylen) ? ylen : olen; + spx_uint32_t omagic = 0; + + if (st->magic_samples[channel_index]) { + omagic = speex_resampler_magic(st, channel_index, &y, ochunk); + ochunk -= omagic; + olen -= omagic; + } + if (! st->magic_samples[channel_index]) { + if (in) { + for(j=0;jfilt_len-1]=WORD2INT(in[j*istride_save]); +#else + x[j+st->filt_len-1]=in[j*istride_save]; +#endif + } else { + for(j=0;jfilt_len-1]=0; + } + + speex_resampler_process_native(st, channel_index, &ichunk, y, &ochunk); + } else { + ichunk = 0; + ochunk = 0; + } + + for (j=0;jout_stride = ostride_save; + *in_len -= ilen; + *out_len -= olen; + + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_process_interleaved_float(SpeexResamplerState *st, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) +{ + spx_uint32_t i; + int istride_save, ostride_save; + spx_uint32_t bak_out_len = *out_len; + spx_uint32_t bak_in_len = *in_len; + istride_save = st->in_stride; + ostride_save = st->out_stride; + st->in_stride = st->out_stride = st->nb_channels; + for (i=0;inb_channels;i++) + { + *out_len = bak_out_len; + *in_len = bak_in_len; + if (in != NULL) + speex_resampler_process_float(st, i, in+i, in_len, out+i, out_len); + else + speex_resampler_process_float(st, i, NULL, in_len, out+i, out_len); + } + st->in_stride = istride_save; + st->out_stride = ostride_save; + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_process_interleaved_int(SpeexResamplerState *st, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +{ + spx_uint32_t i; + int istride_save, ostride_save; + spx_uint32_t bak_out_len = *out_len; + spx_uint32_t bak_in_len = *in_len; + istride_save = st->in_stride; + ostride_save = st->out_stride; + st->in_stride = st->out_stride = st->nb_channels; + for (i=0;inb_channels;i++) + { + *out_len = bak_out_len; + *in_len = bak_in_len; + if (in != NULL) + speex_resampler_process_int(st, i, in+i, in_len, out+i, out_len); + else + speex_resampler_process_int(st, i, NULL, in_len, out+i, out_len); + } + st->in_stride = istride_save; + st->out_stride = ostride_save; + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_set_rate(SpeexResamplerState *st, spx_uint32_t in_rate, spx_uint32_t out_rate) +{ + return speex_resampler_set_rate_frac(st, in_rate, out_rate, in_rate, out_rate); +} + +EXPORT void speex_resampler_get_rate(SpeexResamplerState *st, spx_uint32_t *in_rate, spx_uint32_t *out_rate) +{ + *in_rate = st->in_rate; + *out_rate = st->out_rate; +} + +static inline spx_uint32_t compute_gcd(spx_uint32_t a, spx_uint32_t b) +{ + while (b != 0) + { + spx_uint32_t temp = a; + + a = b; + b = temp % b; + } + return a; +} + +EXPORT int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate) +{ + spx_uint32_t fact; + spx_uint32_t old_den; + spx_uint32_t i; + + if (ratio_num == 0 || ratio_den == 0) + return RESAMPLER_ERR_INVALID_ARG; + + if (st->in_rate == in_rate && st->out_rate == out_rate && st->num_rate == ratio_num && st->den_rate == ratio_den) + return RESAMPLER_ERR_SUCCESS; + + old_den = st->den_rate; + st->in_rate = in_rate; + st->out_rate = out_rate; + st->num_rate = ratio_num; + st->den_rate = ratio_den; + + fact = compute_gcd(st->num_rate, st->den_rate); + + st->num_rate /= fact; + st->den_rate /= fact; + + if (old_den > 0) + { + for (i=0;inb_channels;i++) + { + if (multiply_frac(&st->samp_frac_num[i],st->samp_frac_num[i],st->den_rate,old_den) != RESAMPLER_ERR_SUCCESS) + return RESAMPLER_ERR_OVERFLOW; + /* Safety net */ + if (st->samp_frac_num[i] >= st->den_rate) + st->samp_frac_num[i] = st->den_rate-1; + } + } + + if (st->initialised) + return update_filter(st); + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT void speex_resampler_get_ratio(SpeexResamplerState *st, spx_uint32_t *ratio_num, spx_uint32_t *ratio_den) +{ + *ratio_num = st->num_rate; + *ratio_den = st->den_rate; +} + +EXPORT int speex_resampler_set_quality(SpeexResamplerState *st, int quality) +{ + if (quality > 10 || quality < 0) + return RESAMPLER_ERR_INVALID_ARG; + if (st->quality == quality) + return RESAMPLER_ERR_SUCCESS; + st->quality = quality; + if (st->initialised) + return update_filter(st); + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT void speex_resampler_get_quality(SpeexResamplerState *st, int *quality) +{ + *quality = st->quality; +} + +EXPORT void speex_resampler_set_input_stride(SpeexResamplerState *st, spx_uint32_t stride) +{ + st->in_stride = stride; +} + +EXPORT void speex_resampler_get_input_stride(SpeexResamplerState *st, spx_uint32_t *stride) +{ + *stride = st->in_stride; +} + +EXPORT void speex_resampler_set_output_stride(SpeexResamplerState *st, spx_uint32_t stride) +{ + st->out_stride = stride; +} + +EXPORT void speex_resampler_get_output_stride(SpeexResamplerState *st, spx_uint32_t *stride) +{ + *stride = st->out_stride; +} + +EXPORT int speex_resampler_get_input_latency(SpeexResamplerState *st) +{ + return st->filt_len / 2; +} + +EXPORT int speex_resampler_get_output_latency(SpeexResamplerState *st) +{ + return ((st->filt_len / 2) * st->den_rate + (st->num_rate >> 1)) / st->num_rate; +} + +EXPORT int speex_resampler_skip_zeros(SpeexResamplerState *st) +{ + spx_uint32_t i; + for (i=0;inb_channels;i++) + st->last_sample[i] = st->filt_len/2; + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_reset_mem(SpeexResamplerState *st) +{ + spx_uint32_t i; + for (i=0;inb_channels;i++) + { + st->last_sample[i] = 0; + st->magic_samples[i] = 0; + st->samp_frac_num[i] = 0; + } + for (i=0;inb_channels*(st->filt_len-1);i++) + st->mem[i] = 0; + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT const char *speex_resampler_strerror(int err) +{ + switch (err) + { + case RESAMPLER_ERR_SUCCESS: + return "Success."; + case RESAMPLER_ERR_ALLOC_FAILED: + return "Memory allocation failed."; + case RESAMPLER_ERR_BAD_STATE: + return "Bad resampler state."; + case RESAMPLER_ERR_INVALID_ARG: + return "Invalid argument."; + case RESAMPLER_ERR_PTR_OVERLAP: + return "Input and output buffers overlap."; + default: + return "Unknown error. Bad error code or strange version mismatch."; + } +} diff --git a/resampler/resample_sse.h b/resampler/resample_sse.h new file mode 100644 index 0000000..a0c7a20 --- /dev/null +++ b/resampler/resample_sse.h @@ -0,0 +1,128 @@ +/* Copyright (C) 2007-2008 Jean-Marc Valin + * Copyright (C) 2008 Thorvald Natvig + */ +/** + @file resample_sse.h + @brief Resampler functions (SSE version) +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include + +#define OVERRIDE_INNER_PRODUCT_SINGLE +static inline float inner_product_single(const float *a, const float *b, unsigned int len) +{ + int i; + float ret; + __m128 sum = _mm_setzero_ps(); + for (i=0;i +#define OVERRIDE_INNER_PRODUCT_DOUBLE + +static inline double inner_product_double(const float *a, const float *b, unsigned int len) +{ + int i; + double ret; + __m128d sum = _mm_setzero_pd(); + __m128 t; + for (i=0;iport = this->controlPort; @@ -150,6 +151,8 @@ void udpHandler::dataReceived() control_packet_t in = (control_packet_t)r.constData(); if (in->type == 0x04) { // If timer is active, stop it as they are obviously there! + qDebug(logUdp()) << this->metaObject()->className() << ": Received I am here from: " <isActive()) { // send ping packets every second areYouThereTimer->stop(); @@ -305,7 +308,7 @@ void udpHandler::dataReceived() } else { civ = new udpCivData(localIP, radioIP, civPort); - audio = new udpAudio(localIP, radioIP, audioPort, rxLatency, txLatency, rxSampleRate, rxCodec, txSampleRate, txCodec, audioOutputPort, audioInputPort); + audio = new udpAudio(localIP, radioIP, audioPort, rxLatency, txLatency, rxSampleRate, rxCodec, txSampleRate, txCodec, audioOutputPort, audioInputPort,resampleQuality); QObject::connect(civ, SIGNAL(receive(QByteArray)), this, SLOT(receiveFromCivStream(QByteArray))); QObject::connect(audio, SIGNAL(haveAudioData(audioPacket)), this, SLOT(receiveAudioData(audioPacket))); @@ -399,7 +402,7 @@ void udpHandler::sendRequestStream() p.civport = qToBigEndian((quint32)civPort); p.audioport = qToBigEndian((quint32)audioPort); p.txbuffer = qToBigEndian((quint32)txLatency); - + p.convert = 1; sendTrackedPacket(QByteArray::fromRawData((const char*)p.packet, sizeof(p))); return; } @@ -642,7 +645,7 @@ void udpCivData::dataReceived() // Audio stream -udpAudio::udpAudio(QHostAddress local, QHostAddress ip, quint16 audioPort, quint16 rxlatency, quint16 txlatency, quint16 rxsample, quint8 rxcodec, quint16 txsample, quint8 txcodec, QString outputPort, QString inputPort) +udpAudio::udpAudio(QHostAddress local, QHostAddress ip, quint16 audioPort, quint16 rxlatency, quint16 txlatency, quint16 rxsample, quint8 rxcodec, quint16 txsample, quint8 txcodec, QString outputPort, QString inputPort,quint8 resampleQuality) { qDebug(logUdp()) << "Starting udpAudio"; this->localIP = local; @@ -685,7 +688,7 @@ udpAudio::udpAudio(QHostAddress local, QHostAddress ip, quint16 audioPort, quint rxaudio->moveToThread(rxAudioThread); - connect(this, SIGNAL(setupRxAudio(quint8, quint8, quint16, quint16, bool, bool, QString)), rxaudio, SLOT(init(quint8, quint8, quint16, quint16, bool, bool,QString))); + connect(this, SIGNAL(setupRxAudio(quint8, quint8, quint16, quint16, bool, bool, QString, quint8)), rxaudio, SLOT(init(quint8, quint8, quint16, quint16, bool, bool,QString, quint8))); qRegisterMetaType(); connect(this, SIGNAL(haveAudioData(audioPacket)), rxaudio, SLOT(incomingAudio(audioPacket))); @@ -704,7 +707,7 @@ udpAudio::udpAudio(QHostAddress local, QHostAddress ip, quint16 audioPort, quint txaudio->moveToThread(txAudioThread); - connect(this, SIGNAL(setupTxAudio(quint8, quint8, quint16, quint16, bool, bool,QString)), txaudio, SLOT(init(quint8, quint8, quint16, quint16, bool, bool,QString))); + connect(this, SIGNAL(setupTxAudio(quint8, quint8, quint16, quint16, bool, bool,QString,quint8)), txaudio, SLOT(init(quint8, quint8, quint16, quint16, bool, bool,QString,quint8))); connect(txAudioThread, SIGNAL(finished()), txaudio, SLOT(deleteLater())); rxAudioThread->start(); @@ -717,8 +720,8 @@ udpAudio::udpAudio(QHostAddress local, QHostAddress ip, quint16 audioPort, quint connect(pingTimer, &QTimer::timeout, this, &udpBase::sendPing); pingTimer->start(PING_PERIOD); // send ping packets every 100ms - emit setupTxAudio(txNumSamples, txChannelCount, txSampleRate, txLatency, txIsUlawCodec, true, inputPort); - emit setupRxAudio(rxNumSamples, rxChannelCount, rxSampleRate, txLatency, rxIsUlawCodec, false, outputPort); + emit setupTxAudio(txNumSamples, txChannelCount, txSampleRate, txLatency, txIsUlawCodec, true, inputPort,resampleQuality); + emit setupRxAudio(rxNumSamples, rxChannelCount, rxSampleRate, txLatency, rxIsUlawCodec, false, outputPort,resampleQuality); watchdogTimer = new QTimer(); connect(watchdogTimer, &QTimer::timeout, this, &udpAudio::watchdog); @@ -788,7 +791,12 @@ void udpAudio::sendTxAudio() p.len = sizeof(p) + partial.length(); p.sentid = myId; p.rcvdid = remoteId; - p.ident = 0x0080; // TX audio is always this? + if (partial.length() == 0xa0) { + p.ident = 0x9781; + } + else { + p.ident = 0x0080; // TX audio is always this? + } p.datalen = (quint16)qToBigEndian((quint16)partial.length()); p.sendseq = (quint16)qToBigEndian((quint16)sendAudioSeq); // THIS IS BIG ENDIAN! QByteArray tx = QByteArray::fromRawData((const char*)p.packet, sizeof(p)); @@ -838,23 +846,18 @@ void udpAudio::dataReceived() */ control_packet_t in = (control_packet_t)r.constData(); - if (in->type != 0x01) { - if (r.mid(0, 2) == QByteArrayLiteral("\x6c\x05") || - r.mid(0, 2) == QByteArrayLiteral("\x44\x02") || - r.mid(0, 2) == QByteArrayLiteral("\xd8\x03") || - r.mid(0, 2) == QByteArrayLiteral("\x70\x04")) - { - lastReceived = QTime::currentTime(); - audioPacket tempAudio; - tempAudio.seq = in->seq; - tempAudio.time = lastReceived; - tempAudio.sent = 0; - tempAudio.data = r.mid(0x18); - // Prefer signal/slot to forward audio as it is thread/safe - // Need to do more testing but latency appears fine. - emit haveAudioData(tempAudio); - //rxaudio->incomingAudio(tempAudio); - } + if (in->type != 0x01 && in->len >= 0xAC) { + // 0xac is the smallest possible audio packet. + lastReceived = QTime::currentTime(); + audioPacket tempAudio; + tempAudio.seq = in->seq; + tempAudio.time = lastReceived; + tempAudio.sent = 0; + tempAudio.datain = r.mid(0x18); + // Prefer signal/slot to forward audio as it is thread/safe + // Need to do more testing but latency appears fine. + emit haveAudioData(tempAudio); + //rxaudio->incomingAudio(tempAudio); } break; } @@ -956,7 +959,7 @@ void udpBase::dataReceived(QByteArray r) } } if (in->type == 0x04) { - qDebug(logUdp()) << this->metaObject()->className() << ": Received I am here"; + qDebug(logUdp()) << this->metaObject()->className() << ": Received I am here "; areYouThereCounter = 0; // I don't think that we will ever receive an "I am here" other than in response to "Are you there?" remoteId = in->sentid; diff --git a/udphandler.h b/udphandler.h index 00010ef..a511b25 100644 --- a/udphandler.h +++ b/udphandler.h @@ -48,6 +48,7 @@ struct udpPreferences { quint8 audioRXCodec; quint16 audioTXSampleRate; quint8 audioTXCodec; + quint8 resampleQuality; }; void passcode(QString in, QByteArray& out); @@ -166,14 +167,14 @@ class udpAudio : public udpBase Q_OBJECT public: - udpAudio(QHostAddress local, QHostAddress ip, quint16 aport, quint16 rxlatency, quint16 txlatency, quint16 rxsample, quint8 rxcodec, quint16 txsample, quint8 txcodec, QString outputPort, QString inputPort); + udpAudio(QHostAddress local, QHostAddress ip, quint16 aport, quint16 rxlatency, quint16 txlatency, quint16 rxsample, quint8 rxcodec, quint16 txsample, quint8 txcodec, QString outputPort, QString inputPort,quint8 resampleQuality); ~udpAudio(); signals: void haveAudioData(audioPacket data); - void setupTxAudio(const quint8 samples, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool isUlaw, const bool isInput, QString port); - void setupRxAudio(const quint8 samples, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool isUlaw, const bool isInput, QString port); + void setupTxAudio(const quint8 samples, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool isUlaw, const bool isInput, QString port,quint8 resampleQuality); + void setupRxAudio(const quint8 samples, const quint8 channels, const quint16 samplerate, const quint16 latency, const bool isUlaw, const bool isInput, QString port,quint8 resampleQuality); void haveChangeLatency(quint16 value); @@ -276,6 +277,8 @@ private: QString audioInputPort; QString audioOutputPort; + + quint8 resampleQuality; quint16 reauthInterval = 60000; QString devName; diff --git a/udpserver.cpp b/udpserver.cpp index f2d7fe6..609739b 100644 --- a/udpserver.cpp +++ b/udpserver.cpp @@ -1062,14 +1062,14 @@ void udpServer::receiveAudioData(const audioPacket &d) if (client != Q_NULLPTR && client->connected) { audio_packet p; memset(p.packet, 0x0, sizeof(p)); // We can't be sure it is initialized with 0x00! - p.len = sizeof(p) + d.data.length(); + p.len = sizeof(p) + d.datain.length(); p.sentid = client->myId; p.rcvdid = client->remoteId; p.ident = 0x0080; // audio is always this? - p.datalen = (quint16)qToBigEndian((quint16)d.data.length()); + p.datalen = (quint16)qToBigEndian((quint16)d.datain.length()); p.sendseq = (quint16)qToBigEndian((quint16)client->sendAudioSeq); // THIS IS BIG ENDIAN! QByteArray t = QByteArray::fromRawData((const char*)p.packet, sizeof(p)); - t.append(d.data); + t.append(d.datain); QMutexLocker locker(&mutex); client->txSeqBuf.append(SEQBUFENTRY()); client->txSeqBuf.last().seqNum = p.seq; diff --git a/wfmain.cpp b/wfmain.cpp index ddab534..fcd6fad 100644 --- a/wfmain.cpp +++ b/wfmain.cpp @@ -761,8 +761,7 @@ void wfmain::setDefPrefs() udpDefPrefs.audioRXCodec = 4; udpDefPrefs.audioTXSampleRate = 48000; udpDefPrefs.audioTXCodec = 4; - - + udpDefPrefs.resampleQuality = 4; } void wfmain::loadSettings() @@ -878,6 +877,8 @@ void wfmain::loadSettings() ui->audioInputCombo->setCurrentIndex(audioInputIndex); } + udpPrefs.resampleQuality = settings.value("ResampleQuality", udpDefPrefs.resampleQuality).toInt(); + settings.endGroup(); settings.beginGroup("Server"); @@ -985,6 +986,7 @@ void wfmain::saveSettings() settings.setValue("AudioTXCodec", udpPrefs.audioTXCodec); settings.setValue("AudioOutput", udpPrefs.audioOutput); settings.setValue("AudioInput", udpPrefs.audioInput); + settings.setValue("ResampleQuality", udpPrefs.resampleQuality); settings.endGroup(); // Memory channels diff --git a/wfview.pro b/wfview.pro index 96e2251..fc1825a 100644 --- a/wfview.pro +++ b/wfview.pro @@ -31,6 +31,10 @@ QMAKE_LFLAGS += -O2 -march=native -s DEFINES += QT_DEPRECATED_WARNINGS DEFINES += QCUSTOMPLOT_COMPILE_LIBRARY +# These defines are used for the resampler +DEFINES += OUTSIDE_SPEEX +DEFINES += RANDOM_PREFIX=wf + linux:DEFINES += HOST=\\\"`hostname`\\\" UNAME=\\\"`whoami`\\\" linux:DEFINES += GITSHORT="\\\"$(shell git -C $$PWD rev-parse --short HEAD)\\\"" @@ -88,7 +92,8 @@ SOURCES += main.cpp\ udpserver.cpp \ meter.cpp \ qledlabel.cpp \ - pttyhandler.cpp + pttyhandler.cpp \ + resampler/resample.cpp HEADERS += wfmain.h \ commhandler.h \ @@ -105,7 +110,10 @@ HEADERS += wfmain.h \ packettypes.h \ meter.h \ qledlabel.h \ - pttyhandler.h + pttyhandler.h \ + resampler/speex_resampler.h \ + resampler/arch.h \ + resampler/resample_sse.h FORMS += wfmain.ui \ diff --git a/wfview.sln b/wfview.sln index 477a4b4..9148927 100644 --- a/wfview.sln +++ b/wfview.sln @@ -8,13 +8,23 @@ EndProject Global GlobalSection(SolutionConfigurationPlatforms) = preSolution Debug|x64 = Debug|x64 + Debug|x86 = Debug|x86 Release|x64 = Release|x64 + Release|x86 = Release|x86 + Template|x64 = Template|x64 + Template|x86 = Template|x86 EndGlobalSection GlobalSection(ProjectConfigurationPlatforms) = postSolution {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Debug|x64.ActiveCfg = Debug|x64 {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Debug|x64.Build.0 = Debug|x64 + {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Debug|x86.ActiveCfg = Debug|x64 {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Release|x64.ActiveCfg = Release|x64 {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Release|x64.Build.0 = Release|x64 + {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Release|x86.ActiveCfg = Release|x64 + {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Template|x64.ActiveCfg = Release|x64 + {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Template|x64.Build.0 = Release|x64 + {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Template|x86.ActiveCfg = Release|x64 + {326108AD-FA9D-3AAF-8D3E-062C4DDC34E2}.Template|x86.Build.0 = Release|x64 EndGlobalSection GlobalSection(SolutionProperties) = preSolution HideSolutionNode = FALSE diff --git a/wfview.vcxproj b/wfview.vcxproj index cf67190..1a65ae5 100644 --- a/wfview.vcxproj +++ b/wfview.vcxproj @@ -152,7 +152,7 @@ Sync debug\ Disabled - _WINDOWS;UNICODE;_UNICODE;WIN32;_ENABLE_EXTENDED_ALIGNED_STORAGE;WIN64;QT_DEPRECATED_WARNINGS;QCUSTOMPLOT_USE_OPENGL;HOST=1;UNAME=1;GITSHORT=1;%(PreprocessorDefinitions) + _WINDOWS;UNICODE;_UNICODE;WIN32;_ENABLE_EXTENDED_ALIGNED_STORAGE;WIN64;QT_DEPRECATED_WARNINGS;QCUSTOMPLOT_USE_OPENGL;HOST=1;UNAME=1;GITSHORT=1;OUTSIDE_SPEEX;RANDOM_PREFIX=wf;%(PreprocessorDefinitions) false MultiThreadedDebugDLL true @@ -210,6 +210,7 @@ + @@ -237,6 +238,7 @@ + diff --git a/wfview.vcxproj.filters b/wfview.vcxproj.filters index 158fab7..8114315 100644 --- a/wfview.vcxproj.filters +++ b/wfview.vcxproj.filters @@ -108,6 +108,9 @@ Source Files + + Source Files + @@ -323,6 +326,10 @@ + + + + @@ -339,5 +346,8 @@ Header Files + + Header Files + \ No newline at end of file