1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libopus/silk/enc_API.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,556 @@
1.4 +/***********************************************************************
1.5 +Copyright (c) 2006-2011, Skype Limited. All rights reserved.
1.6 +Redistribution and use in source and binary forms, with or without
1.7 +modification, are permitted provided that the following conditions
1.8 +are met:
1.9 +- Redistributions of source code must retain the above copyright notice,
1.10 +this list of conditions and the following disclaimer.
1.11 +- Redistributions in binary form must reproduce the above copyright
1.12 +notice, this list of conditions and the following disclaimer in the
1.13 +documentation and/or other materials provided with the distribution.
1.14 +- Neither the name of Internet Society, IETF or IETF Trust, nor the
1.15 +names of specific contributors, may be used to endorse or promote
1.16 +products derived from this software without specific prior written
1.17 +permission.
1.18 +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
1.19 +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
1.20 +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
1.21 +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
1.22 +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
1.23 +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
1.24 +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
1.25 +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
1.26 +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
1.27 +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1.28 +POSSIBILITY OF SUCH DAMAGE.
1.29 +***********************************************************************/
1.30 +
1.31 +#ifdef HAVE_CONFIG_H
1.32 +#include "config.h"
1.33 +#endif
1.34 +#include "define.h"
1.35 +#include "API.h"
1.36 +#include "control.h"
1.37 +#include "typedef.h"
1.38 +#include "stack_alloc.h"
1.39 +#include "structs.h"
1.40 +#include "tuning_parameters.h"
1.41 +#ifdef FIXED_POINT
1.42 +#include "main_FIX.h"
1.43 +#else
1.44 +#include "main_FLP.h"
1.45 +#endif
1.46 +
1.47 +/***************************************/
1.48 +/* Read control structure from encoder */
1.49 +/***************************************/
1.50 +static opus_int silk_QueryEncoder( /* O Returns error code */
1.51 + const void *encState, /* I State */
1.52 + silk_EncControlStruct *encStatus /* O Encoder Status */
1.53 +);
1.54 +
1.55 +/****************************************/
1.56 +/* Encoder functions */
1.57 +/****************************************/
1.58 +
1.59 +opus_int silk_Get_Encoder_Size( /* O Returns error code */
1.60 + opus_int *encSizeBytes /* O Number of bytes in SILK encoder state */
1.61 +)
1.62 +{
1.63 + opus_int ret = SILK_NO_ERROR;
1.64 +
1.65 + *encSizeBytes = sizeof( silk_encoder );
1.66 +
1.67 + return ret;
1.68 +}
1.69 +
1.70 +/*************************/
1.71 +/* Init or Reset encoder */
1.72 +/*************************/
1.73 +opus_int silk_InitEncoder( /* O Returns error code */
1.74 + void *encState, /* I/O State */
1.75 + int arch, /* I Run-time architecture */
1.76 + silk_EncControlStruct *encStatus /* O Encoder Status */
1.77 +)
1.78 +{
1.79 + silk_encoder *psEnc;
1.80 + opus_int n, ret = SILK_NO_ERROR;
1.81 +
1.82 + psEnc = (silk_encoder *)encState;
1.83 +
1.84 + /* Reset encoder */
1.85 + silk_memset( psEnc, 0, sizeof( silk_encoder ) );
1.86 + for( n = 0; n < ENCODER_NUM_CHANNELS; n++ ) {
1.87 + if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ], arch ) ) {
1.88 + silk_assert( 0 );
1.89 + }
1.90 + }
1.91 +
1.92 + psEnc->nChannelsAPI = 1;
1.93 + psEnc->nChannelsInternal = 1;
1.94 +
1.95 + /* Read control structure */
1.96 + if( ret += silk_QueryEncoder( encState, encStatus ) ) {
1.97 + silk_assert( 0 );
1.98 + }
1.99 +
1.100 + return ret;
1.101 +}
1.102 +
1.103 +/***************************************/
1.104 +/* Read control structure from encoder */
1.105 +/***************************************/
1.106 +static opus_int silk_QueryEncoder( /* O Returns error code */
1.107 + const void *encState, /* I State */
1.108 + silk_EncControlStruct *encStatus /* O Encoder Status */
1.109 +)
1.110 +{
1.111 + opus_int ret = SILK_NO_ERROR;
1.112 + silk_encoder_state_Fxx *state_Fxx;
1.113 + silk_encoder *psEnc = (silk_encoder *)encState;
1.114 +
1.115 + state_Fxx = psEnc->state_Fxx;
1.116 +
1.117 + encStatus->nChannelsAPI = psEnc->nChannelsAPI;
1.118 + encStatus->nChannelsInternal = psEnc->nChannelsInternal;
1.119 + encStatus->API_sampleRate = state_Fxx[ 0 ].sCmn.API_fs_Hz;
1.120 + encStatus->maxInternalSampleRate = state_Fxx[ 0 ].sCmn.maxInternal_fs_Hz;
1.121 + encStatus->minInternalSampleRate = state_Fxx[ 0 ].sCmn.minInternal_fs_Hz;
1.122 + encStatus->desiredInternalSampleRate = state_Fxx[ 0 ].sCmn.desiredInternal_fs_Hz;
1.123 + encStatus->payloadSize_ms = state_Fxx[ 0 ].sCmn.PacketSize_ms;
1.124 + encStatus->bitRate = state_Fxx[ 0 ].sCmn.TargetRate_bps;
1.125 + encStatus->packetLossPercentage = state_Fxx[ 0 ].sCmn.PacketLoss_perc;
1.126 + encStatus->complexity = state_Fxx[ 0 ].sCmn.Complexity;
1.127 + encStatus->useInBandFEC = state_Fxx[ 0 ].sCmn.useInBandFEC;
1.128 + encStatus->useDTX = state_Fxx[ 0 ].sCmn.useDTX;
1.129 + encStatus->useCBR = state_Fxx[ 0 ].sCmn.useCBR;
1.130 + encStatus->internalSampleRate = silk_SMULBB( state_Fxx[ 0 ].sCmn.fs_kHz, 1000 );
1.131 + encStatus->allowBandwidthSwitch = state_Fxx[ 0 ].sCmn.allow_bandwidth_switch;
1.132 + encStatus->inWBmodeWithoutVariableLP = state_Fxx[ 0 ].sCmn.fs_kHz == 16 && state_Fxx[ 0 ].sCmn.sLP.mode == 0;
1.133 +
1.134 + return ret;
1.135 +}
1.136 +
1.137 +
1.138 +/**************************/
1.139 +/* Encode frame with Silk */
1.140 +/**************************/
1.141 +/* Note: if prefillFlag is set, the input must contain 10 ms of audio, irrespective of what */
1.142 +/* encControl->payloadSize_ms is set to */
1.143 +opus_int silk_Encode( /* O Returns error code */
1.144 + void *encState, /* I/O State */
1.145 + silk_EncControlStruct *encControl, /* I Control status */
1.146 + const opus_int16 *samplesIn, /* I Speech sample input vector */
1.147 + opus_int nSamplesIn, /* I Number of samples in input vector */
1.148 + ec_enc *psRangeEnc, /* I/O Compressor data structure */
1.149 + opus_int32 *nBytesOut, /* I/O Number of bytes in payload (input: Max bytes) */
1.150 + const opus_int prefillFlag /* I Flag to indicate prefilling buffers no coding */
1.151 +)
1.152 +{
1.153 + opus_int n, i, nBits, flags, tmp_payloadSize_ms = 0, tmp_complexity = 0, ret = 0;
1.154 + opus_int nSamplesToBuffer, nSamplesToBufferMax, nBlocksOf10ms;
1.155 + opus_int nSamplesFromInput = 0, nSamplesFromInputMax;
1.156 + opus_int speech_act_thr_for_switch_Q8;
1.157 + opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol, sum;
1.158 + silk_encoder *psEnc = ( silk_encoder * )encState;
1.159 + VARDECL( opus_int16, buf );
1.160 + opus_int transition, curr_block, tot_blocks;
1.161 + SAVE_STACK;
1.162 +
1.163 + if (encControl->reducedDependency)
1.164 + {
1.165 + psEnc->state_Fxx[0].sCmn.first_frame_after_reset = 1;
1.166 + psEnc->state_Fxx[1].sCmn.first_frame_after_reset = 1;
1.167 + }
1.168 + psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded = psEnc->state_Fxx[ 1 ].sCmn.nFramesEncoded = 0;
1.169 +
1.170 + /* Check values in encoder control structure */
1.171 + if( ( ret = check_control_input( encControl ) != 0 ) ) {
1.172 + silk_assert( 0 );
1.173 + RESTORE_STACK;
1.174 + return ret;
1.175 + }
1.176 +
1.177 + encControl->switchReady = 0;
1.178 +
1.179 + if( encControl->nChannelsInternal > psEnc->nChannelsInternal ) {
1.180 + /* Mono -> Stereo transition: init state of second channel and stereo state */
1.181 + ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ], psEnc->state_Fxx[ 0 ].sCmn.arch );
1.182 + silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) );
1.183 + silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) );
1.184 + psEnc->sStereo.mid_side_amp_Q0[ 0 ] = 0;
1.185 + psEnc->sStereo.mid_side_amp_Q0[ 1 ] = 1;
1.186 + psEnc->sStereo.mid_side_amp_Q0[ 2 ] = 0;
1.187 + psEnc->sStereo.mid_side_amp_Q0[ 3 ] = 1;
1.188 + psEnc->sStereo.width_prev_Q14 = 0;
1.189 + psEnc->sStereo.smth_width_Q14 = SILK_FIX_CONST( 1, 14 );
1.190 + if( psEnc->nChannelsAPI == 2 ) {
1.191 + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof( silk_resampler_state_struct ) );
1.192 + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.In_HP_State, &psEnc->state_Fxx[ 0 ].sCmn.In_HP_State, sizeof( psEnc->state_Fxx[ 1 ].sCmn.In_HP_State ) );
1.193 + }
1.194 + }
1.195 +
1.196 + transition = (encControl->payloadSize_ms != psEnc->state_Fxx[ 0 ].sCmn.PacketSize_ms) || (psEnc->nChannelsInternal != encControl->nChannelsInternal);
1.197 +
1.198 + psEnc->nChannelsAPI = encControl->nChannelsAPI;
1.199 + psEnc->nChannelsInternal = encControl->nChannelsInternal;
1.200 +
1.201 + nBlocksOf10ms = silk_DIV32( 100 * nSamplesIn, encControl->API_sampleRate );
1.202 + tot_blocks = ( nBlocksOf10ms > 1 ) ? nBlocksOf10ms >> 1 : 1;
1.203 + curr_block = 0;
1.204 + if( prefillFlag ) {
1.205 + /* Only accept input length of 10 ms */
1.206 + if( nBlocksOf10ms != 1 ) {
1.207 + silk_assert( 0 );
1.208 + RESTORE_STACK;
1.209 + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
1.210 + }
1.211 + /* Reset Encoder */
1.212 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.213 + ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch );
1.214 + silk_assert( !ret );
1.215 + }
1.216 + tmp_payloadSize_ms = encControl->payloadSize_ms;
1.217 + encControl->payloadSize_ms = 10;
1.218 + tmp_complexity = encControl->complexity;
1.219 + encControl->complexity = 0;
1.220 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.221 + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
1.222 + psEnc->state_Fxx[ n ].sCmn.prefillFlag = 1;
1.223 + }
1.224 + } else {
1.225 + /* Only accept input lengths that are a multiple of 10 ms */
1.226 + if( nBlocksOf10ms * encControl->API_sampleRate != 100 * nSamplesIn || nSamplesIn < 0 ) {
1.227 + silk_assert( 0 );
1.228 + RESTORE_STACK;
1.229 + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
1.230 + }
1.231 + /* Make sure no more than one packet can be produced */
1.232 + if( 1000 * (opus_int32)nSamplesIn > encControl->payloadSize_ms * encControl->API_sampleRate ) {
1.233 + silk_assert( 0 );
1.234 + RESTORE_STACK;
1.235 + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
1.236 + }
1.237 + }
1.238 +
1.239 + TargetRate_bps = silk_RSHIFT32( encControl->bitRate, encControl->nChannelsInternal - 1 );
1.240 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.241 + /* Force the side channel to the same rate as the mid */
1.242 + opus_int force_fs_kHz = (n==1) ? psEnc->state_Fxx[0].sCmn.fs_kHz : 0;
1.243 + if( ( ret = silk_control_encoder( &psEnc->state_Fxx[ n ], encControl, TargetRate_bps, psEnc->allowBandwidthSwitch, n, force_fs_kHz ) ) != 0 ) {
1.244 + silk_assert( 0 );
1.245 + RESTORE_STACK;
1.246 + return ret;
1.247 + }
1.248 + if( psEnc->state_Fxx[n].sCmn.first_frame_after_reset || transition ) {
1.249 + for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) {
1.250 + psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] = 0;
1.251 + }
1.252 + }
1.253 + psEnc->state_Fxx[ n ].sCmn.inDTX = psEnc->state_Fxx[ n ].sCmn.useDTX;
1.254 + }
1.255 + silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz );
1.256 +
1.257 + /* Input buffering/resampling and encoding */
1.258 + nSamplesToBufferMax =
1.259 + 10 * nBlocksOf10ms * psEnc->state_Fxx[ 0 ].sCmn.fs_kHz;
1.260 + nSamplesFromInputMax =
1.261 + silk_DIV32_16( nSamplesToBufferMax *
1.262 + psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz,
1.263 + psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 );
1.264 + ALLOC( buf, nSamplesFromInputMax, opus_int16 );
1.265 + while( 1 ) {
1.266 + nSamplesToBuffer = psEnc->state_Fxx[ 0 ].sCmn.frame_length - psEnc->state_Fxx[ 0 ].sCmn.inputBufIx;
1.267 + nSamplesToBuffer = silk_min( nSamplesToBuffer, nSamplesToBufferMax );
1.268 + nSamplesFromInput = silk_DIV32_16( nSamplesToBuffer * psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 );
1.269 + /* Resample and write to buffer */
1.270 + if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 2 ) {
1.271 + opus_int id = psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded;
1.272 + for( n = 0; n < nSamplesFromInput; n++ ) {
1.273 + buf[ n ] = samplesIn[ 2 * n ];
1.274 + }
1.275 + /* Making sure to start both resamplers from the same state when switching from mono to stereo */
1.276 + if( psEnc->nPrevChannelsInternal == 1 && id==0 ) {
1.277 + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof(psEnc->state_Fxx[ 1 ].sCmn.resampler_state));
1.278 + }
1.279 +
1.280 + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
1.281 + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
1.282 + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
1.283 +
1.284 + nSamplesToBuffer = psEnc->state_Fxx[ 1 ].sCmn.frame_length - psEnc->state_Fxx[ 1 ].sCmn.inputBufIx;
1.285 + nSamplesToBuffer = silk_min( nSamplesToBuffer, 10 * nBlocksOf10ms * psEnc->state_Fxx[ 1 ].sCmn.fs_kHz );
1.286 + for( n = 0; n < nSamplesFromInput; n++ ) {
1.287 + buf[ n ] = samplesIn[ 2 * n + 1 ];
1.288 + }
1.289 + ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state,
1.290 + &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
1.291 +
1.292 + psEnc->state_Fxx[ 1 ].sCmn.inputBufIx += nSamplesToBuffer;
1.293 + } else if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 1 ) {
1.294 + /* Combine left and right channels before resampling */
1.295 + for( n = 0; n < nSamplesFromInput; n++ ) {
1.296 + sum = samplesIn[ 2 * n ] + samplesIn[ 2 * n + 1 ];
1.297 + buf[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 );
1.298 + }
1.299 + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
1.300 + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
1.301 + /* On the first mono frame, average the results for the two resampler states */
1.302 + if( psEnc->nPrevChannelsInternal == 2 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 ) {
1.303 + ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state,
1.304 + &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
1.305 + for( n = 0; n < psEnc->state_Fxx[ 0 ].sCmn.frame_length; n++ ) {
1.306 + psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ] =
1.307 + silk_RSHIFT(psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ]
1.308 + + psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx+n+2 ], 1);
1.309 + }
1.310 + }
1.311 + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
1.312 + } else {
1.313 + silk_assert( encControl->nChannelsAPI == 1 && encControl->nChannelsInternal == 1 );
1.314 + silk_memcpy(buf, samplesIn, nSamplesFromInput*sizeof(opus_int16));
1.315 + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
1.316 + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
1.317 + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
1.318 + }
1.319 +
1.320 + samplesIn += nSamplesFromInput * encControl->nChannelsAPI;
1.321 + nSamplesIn -= nSamplesFromInput;
1.322 +
1.323 + /* Default */
1.324 + psEnc->allowBandwidthSwitch = 0;
1.325 +
1.326 + /* Silk encoder */
1.327 + if( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx >= psEnc->state_Fxx[ 0 ].sCmn.frame_length ) {
1.328 + /* Enough data in input buffer, so encode */
1.329 + silk_assert( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx == psEnc->state_Fxx[ 0 ].sCmn.frame_length );
1.330 + silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inputBufIx == psEnc->state_Fxx[ 1 ].sCmn.frame_length );
1.331 +
1.332 + /* Deal with LBRR data */
1.333 + if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 && !prefillFlag ) {
1.334 + /* Create space at start of payload for VAD and FEC flags */
1.335 + opus_uint8 iCDF[ 2 ] = { 0, 0 };
1.336 + iCDF[ 0 ] = 256 - silk_RSHIFT( 256, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal );
1.337 + ec_enc_icdf( psRangeEnc, 0, iCDF, 8 );
1.338 +
1.339 + /* Encode any LBRR data from previous packet */
1.340 + /* Encode LBRR flags */
1.341 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.342 + LBRR_symbol = 0;
1.343 + for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) {
1.344 + LBRR_symbol |= silk_LSHIFT( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ], i );
1.345 + }
1.346 + psEnc->state_Fxx[ n ].sCmn.LBRR_flag = LBRR_symbol > 0 ? 1 : 0;
1.347 + if( LBRR_symbol && psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket > 1 ) {
1.348 + ec_enc_icdf( psRangeEnc, LBRR_symbol - 1, silk_LBRR_flags_iCDF_ptr[ psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket - 2 ], 8 );
1.349 + }
1.350 + }
1.351 +
1.352 + /* Code LBRR indices and excitation signals */
1.353 + for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) {
1.354 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.355 + if( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] ) {
1.356 + opus_int condCoding;
1.357 +
1.358 + if( encControl->nChannelsInternal == 2 && n == 0 ) {
1.359 + silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ i ] );
1.360 + /* For LBRR data there's no need to code the mid-only flag if the side-channel LBRR flag is set */
1.361 + if( psEnc->state_Fxx[ 1 ].sCmn.LBRR_flags[ i ] == 0 ) {
1.362 + silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ i ] );
1.363 + }
1.364 + }
1.365 + /* Use conditional coding if previous frame available */
1.366 + if( i > 0 && psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i - 1 ] ) {
1.367 + condCoding = CODE_CONDITIONALLY;
1.368 + } else {
1.369 + condCoding = CODE_INDEPENDENTLY;
1.370 + }
1.371 + silk_encode_indices( &psEnc->state_Fxx[ n ].sCmn, psRangeEnc, i, 1, condCoding );
1.372 + silk_encode_pulses( psRangeEnc, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].signalType, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].quantOffsetType,
1.373 + psEnc->state_Fxx[ n ].sCmn.pulses_LBRR[ i ], psEnc->state_Fxx[ n ].sCmn.frame_length );
1.374 + }
1.375 + }
1.376 + }
1.377 +
1.378 + /* Reset LBRR flags */
1.379 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.380 + silk_memset( psEnc->state_Fxx[ n ].sCmn.LBRR_flags, 0, sizeof( psEnc->state_Fxx[ n ].sCmn.LBRR_flags ) );
1.381 + }
1.382 + }
1.383 +
1.384 + silk_HP_variable_cutoff( psEnc->state_Fxx );
1.385 +
1.386 + /* Total target bits for packet */
1.387 + nBits = silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 );
1.388 + /* Subtract half of the bits already used */
1.389 + if( !prefillFlag ) {
1.390 + nBits -= ec_tell( psRangeEnc ) >> 1;
1.391 + }
1.392 + /* Divide by number of uncoded frames left in packet */
1.393 + nBits = silk_DIV32_16( nBits, psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket - psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded );
1.394 + /* Convert to bits/second */
1.395 + if( encControl->payloadSize_ms == 10 ) {
1.396 + TargetRate_bps = silk_SMULBB( nBits, 100 );
1.397 + } else {
1.398 + TargetRate_bps = silk_SMULBB( nBits, 50 );
1.399 + }
1.400 + /* Subtract fraction of bits in excess of target in previous packets */
1.401 + TargetRate_bps -= silk_DIV32_16( silk_MUL( psEnc->nBitsExceeded, 1000 ), BITRESERVOIR_DECAY_TIME_MS );
1.402 + /* Never exceed input bitrate */
1.403 + TargetRate_bps = silk_LIMIT( TargetRate_bps, encControl->bitRate, 5000 );
1.404 +
1.405 + /* Convert Left/Right to Mid/Side */
1.406 + if( encControl->nChannelsInternal == 2 ) {
1.407 + silk_stereo_LR_to_MS( &psEnc->sStereo, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ 2 ], &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ 2 ],
1.408 + psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ], &psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ],
1.409 + MStargetRates_bps, TargetRate_bps, psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8, encControl->toMono,
1.410 + psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, psEnc->state_Fxx[ 0 ].sCmn.frame_length );
1.411 + if( psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) {
1.412 + /* Reset side channel encoder memory for first frame with side coding */
1.413 + if( psEnc->prev_decode_only_middle == 1 ) {
1.414 + silk_memset( &psEnc->state_Fxx[ 1 ].sShape, 0, sizeof( psEnc->state_Fxx[ 1 ].sShape ) );
1.415 + silk_memset( &psEnc->state_Fxx[ 1 ].sPrefilt, 0, sizeof( psEnc->state_Fxx[ 1 ].sPrefilt ) );
1.416 + silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sNSQ, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sNSQ ) );
1.417 + silk_memset( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15 ) );
1.418 + silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State ) );
1.419 + psEnc->state_Fxx[ 1 ].sCmn.prevLag = 100;
1.420 + psEnc->state_Fxx[ 1 ].sCmn.sNSQ.lagPrev = 100;
1.421 + psEnc->state_Fxx[ 1 ].sShape.LastGainIndex = 10;
1.422 + psEnc->state_Fxx[ 1 ].sCmn.prevSignalType = TYPE_NO_VOICE_ACTIVITY;
1.423 + psEnc->state_Fxx[ 1 ].sCmn.sNSQ.prev_gain_Q16 = 65536;
1.424 + psEnc->state_Fxx[ 1 ].sCmn.first_frame_after_reset = 1;
1.425 + }
1.426 + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 1 ] );
1.427 + } else {
1.428 + psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] = 0;
1.429 + }
1.430 + if( !prefillFlag ) {
1.431 + silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] );
1.432 + if( psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) {
1.433 + silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] );
1.434 + }
1.435 + }
1.436 + } else {
1.437 + /* Buffering */
1.438 + silk_memcpy( psEnc->state_Fxx[ 0 ].sCmn.inputBuf, psEnc->sStereo.sMid, 2 * sizeof( opus_int16 ) );
1.439 + silk_memcpy( psEnc->sStereo.sMid, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.frame_length ], 2 * sizeof( opus_int16 ) );
1.440 + }
1.441 + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 0 ] );
1.442 +
1.443 + /* Encode */
1.444 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.445 + opus_int maxBits, useCBR;
1.446 +
1.447 + /* Handling rate constraints */
1.448 + maxBits = encControl->maxBits;
1.449 + if( tot_blocks == 2 && curr_block == 0 ) {
1.450 + maxBits = maxBits * 3 / 5;
1.451 + } else if( tot_blocks == 3 ) {
1.452 + if( curr_block == 0 ) {
1.453 + maxBits = maxBits * 2 / 5;
1.454 + } else if( curr_block == 1 ) {
1.455 + maxBits = maxBits * 3 / 4;
1.456 + }
1.457 + }
1.458 + useCBR = encControl->useCBR && curr_block == tot_blocks - 1;
1.459 +
1.460 + if( encControl->nChannelsInternal == 1 ) {
1.461 + channelRate_bps = TargetRate_bps;
1.462 + } else {
1.463 + channelRate_bps = MStargetRates_bps[ n ];
1.464 + if( n == 0 && MStargetRates_bps[ 1 ] > 0 ) {
1.465 + useCBR = 0;
1.466 + /* Give mid up to 1/2 of the max bits for that frame */
1.467 + maxBits -= encControl->maxBits / ( tot_blocks * 2 );
1.468 + }
1.469 + }
1.470 +
1.471 + if( channelRate_bps > 0 ) {
1.472 + opus_int condCoding;
1.473 +
1.474 + silk_control_SNR( &psEnc->state_Fxx[ n ].sCmn, channelRate_bps );
1.475 +
1.476 + /* Use independent coding if no previous frame available */
1.477 + if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - n <= 0 ) {
1.478 + condCoding = CODE_INDEPENDENTLY;
1.479 + } else if( n > 0 && psEnc->prev_decode_only_middle ) {
1.480 + /* If we skipped a side frame in this packet, we don't
1.481 + need LTP scaling; the LTP state is well-defined. */
1.482 + condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING;
1.483 + } else {
1.484 + condCoding = CODE_CONDITIONALLY;
1.485 + }
1.486 + if( ( ret = silk_encode_frame_Fxx( &psEnc->state_Fxx[ n ], nBytesOut, psRangeEnc, condCoding, maxBits, useCBR ) ) != 0 ) {
1.487 + silk_assert( 0 );
1.488 + }
1.489 + }
1.490 + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
1.491 + psEnc->state_Fxx[ n ].sCmn.inputBufIx = 0;
1.492 + psEnc->state_Fxx[ n ].sCmn.nFramesEncoded++;
1.493 + }
1.494 + psEnc->prev_decode_only_middle = psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - 1 ];
1.495 +
1.496 + /* Insert VAD and FEC flags at beginning of bitstream */
1.497 + if( *nBytesOut > 0 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket) {
1.498 + flags = 0;
1.499 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.500 + for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) {
1.501 + flags = silk_LSHIFT( flags, 1 );
1.502 + flags |= psEnc->state_Fxx[ n ].sCmn.VAD_flags[ i ];
1.503 + }
1.504 + flags = silk_LSHIFT( flags, 1 );
1.505 + flags |= psEnc->state_Fxx[ n ].sCmn.LBRR_flag;
1.506 + }
1.507 + if( !prefillFlag ) {
1.508 + ec_enc_patch_initial_bits( psRangeEnc, flags, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal );
1.509 + }
1.510 +
1.511 + /* Return zero bytes if all channels DTXed */
1.512 + if( psEnc->state_Fxx[ 0 ].sCmn.inDTX && ( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inDTX ) ) {
1.513 + *nBytesOut = 0;
1.514 + }
1.515 +
1.516 + psEnc->nBitsExceeded += *nBytesOut * 8;
1.517 + psEnc->nBitsExceeded -= silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 );
1.518 + psEnc->nBitsExceeded = silk_LIMIT( psEnc->nBitsExceeded, 0, 10000 );
1.519 +
1.520 + /* Update flag indicating if bandwidth switching is allowed */
1.521 + speech_act_thr_for_switch_Q8 = silk_SMLAWB( SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ),
1.522 + SILK_FIX_CONST( ( 1 - SPEECH_ACTIVITY_DTX_THRES ) / MAX_BANDWIDTH_SWITCH_DELAY_MS, 16 + 8 ), psEnc->timeSinceSwitchAllowed_ms );
1.523 + if( psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8 < speech_act_thr_for_switch_Q8 ) {
1.524 + psEnc->allowBandwidthSwitch = 1;
1.525 + psEnc->timeSinceSwitchAllowed_ms = 0;
1.526 + } else {
1.527 + psEnc->allowBandwidthSwitch = 0;
1.528 + psEnc->timeSinceSwitchAllowed_ms += encControl->payloadSize_ms;
1.529 + }
1.530 + }
1.531 +
1.532 + if( nSamplesIn == 0 ) {
1.533 + break;
1.534 + }
1.535 + } else {
1.536 + break;
1.537 + }
1.538 + curr_block++;
1.539 + }
1.540 +
1.541 + psEnc->nPrevChannelsInternal = encControl->nChannelsInternal;
1.542 +
1.543 + encControl->allowBandwidthSwitch = psEnc->allowBandwidthSwitch;
1.544 + encControl->inWBmodeWithoutVariableLP = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == 16 && psEnc->state_Fxx[ 0 ].sCmn.sLP.mode == 0;
1.545 + encControl->internalSampleRate = silk_SMULBB( psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, 1000 );
1.546 + encControl->stereoWidth_Q14 = encControl->toMono ? 0 : psEnc->sStereo.smth_width_Q14;
1.547 + if( prefillFlag ) {
1.548 + encControl->payloadSize_ms = tmp_payloadSize_ms;
1.549 + encControl->complexity = tmp_complexity;
1.550 + for( n = 0; n < encControl->nChannelsInternal; n++ ) {
1.551 + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
1.552 + psEnc->state_Fxx[ n ].sCmn.prefillFlag = 0;
1.553 + }
1.554 + }
1.555 +
1.556 + RESTORE_STACK;
1.557 + return ret;
1.558 +}
1.559 +
