static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, ec_dec *dec)
{
int i, curr, tf_select;
int tf_select_rsv;
int tf_changed;
int logp;
opus_uint32 budget;
opus_uint32 tell;
budget = dec->storage*8;
tell = ec_tell(dec);
logp = isTransient ? 2 : 4;
tf_select_rsv = LM>0 && tell+logp+1<=budget;
budget -= tf_select_rsv;
tf_changed = curr = 0;
for (i=start;i<end;i++)
{
if (tell+logp<=budget)
{
curr ^= ec_dec_bit_logp(dec, logp);
tell = ec_tell(dec);
tf_changed |= curr;
}
tf_res[i] = curr;
logp = isTransient ? 4 : 5;
}
tf_select = 0;
if (tf_select_rsv &&
tf_select_table[LM][4*isTransient+0+tf_changed] !=
tf_select_table[LM][4*isTransient+2+tf_changed])
{
tf_select = ec_dec_bit_logp(dec, 1);
}
for (i=start;i<end;i++)
{
tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]];
}
}
int main(int _argc,char **_argv){
ec_enc enc;
ec_dec dec;
long nbits;
long nbits2;
double entropy;
int ft;
int ftb;
int sz;
int i;
int ret;
unsigned int sym;
unsigned int seed;
unsigned char *ptr;
const char *env_seed;
ret=0;
entropy=0;
if (_argc > 2) {
fprintf(stderr, "Usage: %s [<seed>]\n", _argv[0]);
return 1;
}
env_seed = getenv("SEED");
if (_argc > 1)
seed = atoi(_argv[1]);
else if (env_seed)
seed = atoi(env_seed);
else
seed = time(NULL);
/*Testing encoding of raw bit values.*/
ptr = (unsigned char *)malloc(DATA_SIZE);
ec_enc_init(&enc,ptr, DATA_SIZE);
for(ft=2;ft<1024;ft++){
for(i=0;i<ft;i++){
entropy+=log(ft)*M_LOG2E;
ec_enc_uint(&enc,i,ft);
}
}
/*Testing encoding of raw bit values.*/
for(ftb=1;ftb<16;ftb++){
for(i=0;i<(1<<ftb);i++){
entropy+=ftb;
nbits=ec_tell(&enc);
ec_enc_bits(&enc,i,ftb);
nbits2=ec_tell(&enc);
if(nbits2-nbits!=ftb){
fprintf(stderr,"Used %li bits to encode %i bits directly.\n",
nbits2-nbits,ftb);
ret=-1;
}
}
}
nbits=ec_tell_frac(&enc);
ec_enc_done(&enc);
fprintf(stderr,
"Encoded %0.2lf bits of entropy to %0.2lf bits (%0.3lf%% wasted).\n",
entropy,ldexp(nbits,-3),100*(nbits-ldexp(entropy,3))/nbits);
fprintf(stderr,"Packed to %li bytes.\n",(long)ec_range_bytes(&enc));
ec_dec_init(&dec,ptr,DATA_SIZE);
for(ft=2;ft<1024;ft++){
for(i=0;i<ft;i++){
sym=ec_dec_uint(&dec,ft);
if(sym!=(unsigned)i){
fprintf(stderr,"Decoded %i instead of %i with ft of %i.\n",sym,i,ft);
ret=-1;
}
}
}
for(ftb=1;ftb<16;ftb++){
for(i=0;i<(1<<ftb);i++){
sym=ec_dec_bits(&dec,ftb);
if(sym!=(unsigned)i){
fprintf(stderr,"Decoded %i instead of %i with ftb of %i.\n",sym,i,ftb);
ret=-1;
}
}
}
nbits2=ec_tell_frac(&dec);
if(nbits!=nbits2){
fprintf(stderr,
"Reported number of bits used was %0.2lf, should be %0.2lf.\n",
ldexp(nbits2,-3),ldexp(nbits,-3));
ret=-1;
}
/*Testing an encoder bust prefers range coder data over raw bits.
This isn't a general guarantee, will only work for data that is buffered in
the encoder state and not yet stored in the user buffer, and should never
get used in practice.
It's mostly here for code coverage completeness.*/
/*Start with a 16-bit buffer.*/
ec_enc_init(&enc,ptr,2);
/*Write 7 raw bits.*/
ec_enc_bits(&enc,0x55,7);
/*Write 12.3 bits of range coder data.*/
ec_enc_uint(&enc,1,2);
ec_enc_uint(&enc,1,3);
ec_enc_uint(&enc,1,4);
ec_enc_uint(&enc,1,5);
ec_enc_uint(&enc,2,6);
ec_enc_uint(&enc,6,7);
ec_enc_done(&enc);
ec_dec_init(&dec,ptr,2);
if(!enc.error
/*The raw bits should have been overwritten by the range coder data.*/
||ec_dec_bits(&dec,7)!=0x05
/*And all the range coder data should have been encoded correctly.*/
||ec_dec_uint(&dec,2)!=1
||ec_dec_uint(&dec,3)!=1
||ec_dec_uint(&dec,4)!=1
||ec_dec_uint(&dec,5)!=1
||ec_dec_uint(&dec,6)!=2
||ec_dec_uint(&dec,7)!=6){
fprintf(stderr,"Encoder bust overwrote range coder data with raw bits.\n");
ret=-1;
}
srand(seed);
fprintf(stderr,"Testing random streams... Random seed: %u (%.4X)\n", seed, rand() % 65536);
for(i=0;i<409600;i++){
unsigned *data;
unsigned *tell;
unsigned tell_bits;
int j;
int zeros;
ft=rand()/((RAND_MAX>>(rand()%11U))+1U)+10;
sz=rand()/((RAND_MAX>>(rand()%9U))+1U);
data=(unsigned *)malloc(sz*sizeof(*data));
tell=(unsigned *)malloc((sz+1)*sizeof(*tell));
ec_enc_init(&enc,ptr,DATA_SIZE2);
zeros = rand()%13==0;
tell[0]=ec_tell_frac(&enc);
for(j=0;j<sz;j++){
if (zeros)
data[j]=0;
else
data[j]=rand()%ft;
ec_enc_uint(&enc,data[j],ft);
tell[j+1]=ec_tell_frac(&enc);
}
if (rand()%2==0)
while(ec_tell(&enc)%8 != 0)
ec_enc_uint(&enc, rand()%2, 2);
tell_bits = ec_tell(&enc);
ec_enc_done(&enc);
if(tell_bits!=(unsigned)ec_tell(&enc)){
fprintf(stderr,"ec_tell() changed after ec_enc_done(): %i instead of %i (Random seed: %u)\n",
ec_tell(&enc),tell_bits,seed);
ret=-1;
}
if ((tell_bits+7)/8 < ec_range_bytes(&enc))
{
fprintf (stderr, "ec_tell() lied, there's %i bytes instead of %d (Random seed: %u)\n",
ec_range_bytes(&enc), (tell_bits+7)/8,seed);
ret=-1;
}
ec_dec_init(&dec,ptr,DATA_SIZE2);
if(ec_tell_frac(&dec)!=tell[0]){
fprintf(stderr,
"Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
0,ec_tell_frac(&dec),tell[0],seed);
}
for(j=0;j<sz;j++){
sym=ec_dec_uint(&dec,ft);
if(sym!=data[j]){
fprintf(stderr,
"Decoded %i instead of %i with ft of %i at position %i of %i (Random seed: %u).\n",
sym,data[j],ft,j,sz,seed);
ret=-1;
}
if(ec_tell_frac(&dec)!=tell[j+1]){
fprintf(stderr,
"Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
j+1,ec_tell_frac(&dec),tell[j+1],seed);
}
}
free(tell);
free(data);
}
/*Test compatibility between multiple different encode/decode routines.*/
for(i=0;i<409600;i++){
unsigned *logp1;
unsigned *data;
unsigned *tell;
unsigned *enc_method;
int j;
sz=rand()/((RAND_MAX>>(rand()%9U))+1U);
logp1=(unsigned *)malloc(sz*sizeof(*logp1));
data=(unsigned *)malloc(sz*sizeof(*data));
tell=(unsigned *)malloc((sz+1)*sizeof(*tell));
enc_method=(unsigned *)malloc(sz*sizeof(*enc_method));
ec_enc_init(&enc,ptr,DATA_SIZE2);
tell[0]=ec_tell_frac(&enc);
for(j=0;j<sz;j++){
data[j]=rand()/((RAND_MAX>>1)+1);
logp1[j]=(rand()%15)+1;
enc_method[j]=rand()/((RAND_MAX>>2)+1);
switch(enc_method[j]){
case 0:{
ec_encode(&enc,data[j]?(1<<logp1[j])-1:0,
(1<<logp1[j])-(data[j]?0:1),1<<logp1[j]);
}break;
case 1:{
ec_encode_bin(&enc,data[j]?(1<<logp1[j])-1:0,
(1<<logp1[j])-(data[j]?0:1),logp1[j]);
}break;
case 2:{
ec_enc_bit_logp(&enc,data[j],logp1[j]);
}break;
case 3:{
unsigned char icdf[2];
icdf[0]=1;
icdf[1]=0;
ec_enc_icdf(&enc,data[j],icdf,logp1[j]);
}break;
}
tell[j+1]=ec_tell_frac(&enc);
}
ec_enc_done(&enc);
if((ec_tell(&enc)+7U)/8U<ec_range_bytes(&enc)){
fprintf(stderr,"tell() lied, there's %i bytes instead of %d (Random seed: %u)\n",
ec_range_bytes(&enc),(ec_tell(&enc)+7)/8,seed);
ret=-1;
}
ec_dec_init(&dec,ptr,DATA_SIZE2);
if(ec_tell_frac(&dec)!=tell[0]){
fprintf(stderr,
"Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
0,ec_tell_frac(&dec),tell[0],seed);
}
for(j=0;j<sz;j++){
int fs;
int dec_method;
dec_method=rand()/((RAND_MAX>>2)+1);
switch(dec_method){
case 0:{
fs=ec_decode(&dec,1<<logp1[j]);
sym=fs>=(1<<logp1[j])-1;
ec_dec_update(&dec,sym?(1<<logp1[j])-1:0,
(1<<logp1[j])-(sym?0:1),1<<logp1[j]);
}break;
case 1:{
fs=ec_decode_bin(&dec,logp1[j]);
sym=fs>=(1<<logp1[j])-1;
ec_dec_update(&dec,sym?(1<<logp1[j])-1:0,
(1<<logp1[j])-(sym?0:1),1<<logp1[j]);
}break;
case 2:{
sym=ec_dec_bit_logp(&dec,logp1[j]);
}break;
case 3:{
unsigned char icdf[2];
icdf[0]=1;
icdf[1]=0;
sym=ec_dec_icdf(&dec,icdf,logp1[j]);
}break;
}
if(sym!=data[j]){
fprintf(stderr,
"Decoded %i instead of %i with logp1 of %i at position %i of %i (Random seed: %u).\n",
sym,data[j],logp1[j],j,sz,seed);
fprintf(stderr,"Encoding method: %i, decoding method: %i\n",
enc_method[j],dec_method);
ret=-1;
}
if(ec_tell_frac(&dec)!=tell[j+1]){
fprintf(stderr,
"Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
j+1,ec_tell_frac(&dec),tell[j+1],seed);
}
}
free(enc_method);
free(tell);
free(data);
free(logp1);
}
ec_enc_init(&enc,ptr,DATA_SIZE2);
ec_enc_bit_logp(&enc,0,1);
ec_enc_bit_logp(&enc,0,1);
ec_enc_bit_logp(&enc,0,1);
ec_enc_bit_logp(&enc,0,1);
ec_enc_bit_logp(&enc,0,2);
ec_enc_patch_initial_bits(&enc,3,2);
if(enc.error){
fprintf(stderr,"patch_initial_bits failed");
ret=-1;
}
ec_enc_patch_initial_bits(&enc,0,5);
if(!enc.error){
fprintf(stderr,"patch_initial_bits didn't fail when it should have");
ret=-1;
}
ec_enc_done(&enc);
if(ec_range_bytes(&enc)!=1||ptr[0]!=192){
fprintf(stderr,"Got %d when expecting 192 for patch_initial_bits",ptr[0]);
ret=-1;
}
ec_enc_init(&enc,ptr,DATA_SIZE2);
ec_enc_bit_logp(&enc,0,1);
ec_enc_bit_logp(&enc,0,1);
ec_enc_bit_logp(&enc,1,6);
ec_enc_bit_logp(&enc,0,2);
ec_enc_patch_initial_bits(&enc,0,2);
if(enc.error){
fprintf(stderr,"patch_initial_bits failed");
ret=-1;
}
ec_enc_done(&enc);
if(ec_range_bytes(&enc)!=2||ptr[0]!=63){
fprintf(stderr,"Got %d when expecting 63 for patch_initial_bits",ptr[0]);
ret=-1;
}
ec_enc_init(&enc,ptr,2);
ec_enc_bit_logp(&enc,0,2);
for(i=0;i<48;i++){
ec_enc_bits(&enc,0,1);
}
ec_enc_done(&enc);
if(!enc.error){
fprintf(stderr,"Raw bits overfill didn't fail when it should have");
ret=-1;
}
ec_enc_init(&enc,ptr,2);
for(i=0;i<17;i++){
ec_enc_bits(&enc,0,1);
}
ec_enc_done(&enc);
if(!enc.error){
fprintf(stderr,"17 raw bits encoded in two bytes");
ret=-1;
}
free(ptr);
return ret;
}
static int opus_decode_frame(OpusDecoder *st, const unsigned char *data,
opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec)
{
void *silk_dec;
CELTDecoder *celt_dec;
int i, silk_ret=0, celt_ret=0;
ec_dec dec;
opus_int32 silk_frame_size;
int pcm_silk_size;
VARDECL(opus_int16, pcm_silk);
int pcm_transition_silk_size;
VARDECL(opus_val16, pcm_transition_silk);
int pcm_transition_celt_size;
VARDECL(opus_val16, pcm_transition_celt);
opus_val16 *pcm_transition=NULL;
int redundant_audio_size;
VARDECL(opus_val16, redundant_audio);
int audiosize;
int mode;
int transition=0;
int start_band;
int redundancy=0;
int redundancy_bytes = 0;
int celt_to_silk=0;
int c;
int F2_5, F5, F10, F20;
const opus_val16 *window;
opus_uint32 redundant_rng = 0;
int celt_accum;
ALLOC_STACK;
silk_dec = (char*)st+st->silk_dec_offset;
celt_dec = (CELTDecoder*)((char*)st+st->celt_dec_offset);
F20 = st->Fs/50;
F10 = F20>>1;
F5 = F10>>1;
F2_5 = F5>>1;
if (frame_size < F2_5)
{
RESTORE_STACK;
return OPUS_BUFFER_TOO_SMALL;
}
/* Limit frame_size to avoid excessive stack allocations. */
frame_size = IMIN(frame_size, st->Fs/25*3);
/* Payloads of 1 (2 including ToC) or 0 trigger the PLC/DTX */
if (len<=1)
{
data = NULL;
/* In that case, don't conceal more than what the ToC says */
frame_size = IMIN(frame_size, st->frame_size);
}
if (data != NULL)
{
audiosize = st->frame_size;
mode = st->mode;
ec_dec_init(&dec,(unsigned char*)data,len);
} else {
audiosize = frame_size;
mode = st->prev_mode;
if (mode == 0)
{
/* If we haven't got any packet yet, all we can do is return zeros */
for (i=0; i<audiosize*st->channels; i++)
pcm[i] = 0;
RESTORE_STACK;
return audiosize;
}
/* Avoids trying to run the PLC on sizes other than 2.5 (CELT), 5 (CELT),
10, or 20 (e.g. 12.5 or 30 ms). */
if (audiosize > F20)
{
do {
int ret = opus_decode_frame(st, NULL, 0, pcm, IMIN(audiosize, F20), 0);
if (ret<0)
{
RESTORE_STACK;
return ret;
}
pcm += ret*st->channels;
audiosize -= ret;
} while (audiosize > 0);
RESTORE_STACK;
return frame_size;
} else if (audiosize < F20)
{
if (audiosize > F10)
audiosize = F10;
else if (mode != MODE_SILK_ONLY && audiosize > F5 && audiosize < F10)
audiosize = F5;
}
}
/* In fixed-point, we can tell CELT to do the accumulation on top of the
SILK PCM buffer. This saves some stack space. */
#ifdef FIXED_POINT
celt_accum = (mode != MODE_CELT_ONLY) && (frame_size >= F10);
#else
celt_accum = 0;
#endif
pcm_transition_silk_size = ALLOC_NONE;
pcm_transition_celt_size = ALLOC_NONE;
if (data!=NULL && st->prev_mode > 0 && (
(mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY && !st->prev_redundancy)
|| (mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) )
)
{
transition = 1;
/* Decide where to allocate the stack memory for pcm_transition */
if (mode == MODE_CELT_ONLY)
pcm_transition_celt_size = F5*st->channels;
else
pcm_transition_silk_size = F5*st->channels;
}
ALLOC(pcm_transition_celt, pcm_transition_celt_size, opus_val16);
if (transition && mode == MODE_CELT_ONLY)
{
pcm_transition = pcm_transition_celt;
opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0);
}
if (audiosize > frame_size)
{
/*fprintf(stderr, "PCM buffer too small: %d vs %d (mode = %d)\n", audiosize, frame_size, mode);*/
RESTORE_STACK;
return OPUS_BAD_ARG;
} else {
frame_size = audiosize;
}
/* Don't allocate any memory when in CELT-only mode */
pcm_silk_size = (mode != MODE_CELT_ONLY && !celt_accum) ? IMAX(F10, frame_size)*st->channels : ALLOC_NONE;
ALLOC(pcm_silk, pcm_silk_size, opus_int16);
/* SILK processing */
if (mode != MODE_CELT_ONLY)
{
int lost_flag, decoded_samples;
opus_int16 *pcm_ptr;
#ifdef FIXED_POINT
if (celt_accum)
pcm_ptr = pcm;
else
#endif
pcm_ptr = pcm_silk;
if (st->prev_mode==MODE_CELT_ONLY)
silk_InitDecoder( silk_dec );
/* The SILK PLC cannot produce frames of less than 10 ms */
st->DecControl.payloadSize_ms = IMAX(10, 1000 * audiosize / st->Fs);
if (data != NULL)
{
st->DecControl.nChannelsInternal = st->stream_channels;
if( mode == MODE_SILK_ONLY ) {
if( st->bandwidth == OPUS_BANDWIDTH_NARROWBAND ) {
st->DecControl.internalSampleRate = 8000;
} else if( st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND ) {
st->DecControl.internalSampleRate = 12000;
} else if( st->bandwidth == OPUS_BANDWIDTH_WIDEBAND ) {
st->DecControl.internalSampleRate = 16000;
} else {
st->DecControl.internalSampleRate = 16000;
silk_assert( 0 );
}
} else {
/* Hybrid mode */
st->DecControl.internalSampleRate = 16000;
}
}
lost_flag = data == NULL ? 1 : 2 * decode_fec;
decoded_samples = 0;
do {
/* Call SILK decoder */
int first_frame = decoded_samples == 0;
silk_ret = silk_Decode( silk_dec, &st->DecControl,
lost_flag, first_frame, &dec, pcm_ptr, &silk_frame_size );
if( silk_ret ) {
if (lost_flag) {
/* PLC failure should not be fatal */
silk_frame_size = frame_size;
for (i=0; i<frame_size*st->channels; i++)
pcm_ptr[i] = 0;
} else {
RESTORE_STACK;
return OPUS_INTERNAL_ERROR;
}
}
pcm_ptr += silk_frame_size * st->channels;
decoded_samples += silk_frame_size;
} while( decoded_samples < frame_size );
}
start_band = 0;
if (!decode_fec && mode != MODE_CELT_ONLY && data != NULL
&& ec_tell(&dec)+17+20*(st->mode == MODE_HYBRID) <= 8*len)
{
/* Check if we have a redundant 0-8 kHz band */
if (mode == MODE_HYBRID)
redundancy = ec_dec_bit_logp(&dec, 12);
else
redundancy = 1;
if (redundancy)
{
celt_to_silk = ec_dec_bit_logp(&dec, 1);
/* redundancy_bytes will be at least two, in the non-hybrid
case due to the ec_tell() check above */
redundancy_bytes = mode==MODE_HYBRID ?
(opus_int32)ec_dec_uint(&dec, 256)+2 :
len-((ec_tell(&dec)+7)>>3);
len -= redundancy_bytes;
/* This is a sanity check. It should never happen for a valid
packet, so the exact behaviour is not normative. */
if (len*8 < ec_tell(&dec))
{
len = 0;
redundancy_bytes = 0;
redundancy = 0;
}
/* Shrink decoder because of raw bits */
dec.storage -= redundancy_bytes;
}
}
if (mode != MODE_CELT_ONLY)
start_band = 17;
{
int endband=21;
switch(st->bandwidth)
{
case OPUS_BANDWIDTH_NARROWBAND:
endband = 13;
break;
case OPUS_BANDWIDTH_MEDIUMBAND:
case OPUS_BANDWIDTH_WIDEBAND:
endband = 17;
break;
case OPUS_BANDWIDTH_SUPERWIDEBAND:
endband = 19;
break;
case OPUS_BANDWIDTH_FULLBAND:
endband = 21;
break;
}
celt_decoder_ctl(celt_dec, CELT_SET_END_BAND(endband));
celt_decoder_ctl(celt_dec, CELT_SET_CHANNELS(st->stream_channels));
}
if (redundancy)
{
transition = 0;
pcm_transition_silk_size=ALLOC_NONE;
}
ALLOC(pcm_transition_silk, pcm_transition_silk_size, opus_val16);
if (transition && mode != MODE_CELT_ONLY)
{
pcm_transition = pcm_transition_silk;
opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0);
}
/* Only allocation memory for redundancy if/when needed */
redundant_audio_size = redundancy ? F5*st->channels : ALLOC_NONE;
ALLOC(redundant_audio, redundant_audio_size, opus_val16);
/* 5 ms redundant frame for CELT->SILK*/
if (redundancy && celt_to_silk)
{
celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0));
celt_decode_with_ec(celt_dec, data+len, redundancy_bytes,
redundant_audio, F5, NULL, 0);
celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng));
}
/* MUST be after PLC */
celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(start_band));
if (mode != MODE_SILK_ONLY)
{
int celt_frame_size = IMIN(F20, frame_size);
/* Make sure to discard any previous CELT state */
if (mode != st->prev_mode && st->prev_mode > 0 && !st->prev_redundancy)
celt_decoder_ctl(celt_dec, OPUS_RESET_STATE);
/* Decode CELT */
celt_ret = celt_decode_with_ec(celt_dec, decode_fec ? NULL : data,
len, pcm, celt_frame_size, &dec, celt_accum);
} else {
unsigned char silence[2] = {0xFF, 0xFF};
if (!celt_accum)
{
for (i=0; i<frame_size*st->channels; i++)
pcm[i] = 0;
}
/* For hybrid -> SILK transitions, we let the CELT MDCT
do a fade-out by decoding a silence frame */
if (st->prev_mode == MODE_HYBRID && !(redundancy && celt_to_silk && st->prev_redundancy) )
{
celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0));
celt_decode_with_ec(celt_dec, silence, 2, pcm, F2_5, NULL, celt_accum);
}
}
if (mode != MODE_CELT_ONLY && !celt_accum)
{
#ifdef FIXED_POINT
for (i=0; i<frame_size*st->channels; i++)
pcm[i] = SAT16(ADD32(pcm[i], pcm_silk[i]));
#else
for (i=0; i<frame_size*st->channels; i++)
pcm[i] = pcm[i] + (opus_val16)((1.f/32768.f)*pcm_silk[i]);
#endif
}
{
const CELTMode *celt_mode;
celt_decoder_ctl(celt_dec, CELT_GET_MODE(&celt_mode));
window = celt_mode->window;
}
/* 5 ms redundant frame for SILK->CELT */
if (redundancy && !celt_to_silk)
{
celt_decoder_ctl(celt_dec, OPUS_RESET_STATE);
celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0));
celt_decode_with_ec(celt_dec, data+len, redundancy_bytes, redundant_audio, F5, NULL, 0);
celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng));
smooth_fade(pcm+st->channels*(frame_size-F2_5), redundant_audio+st->channels*F2_5,
pcm+st->channels*(frame_size-F2_5), F2_5, st->channels, window, st->Fs);
}
if (redundancy && celt_to_silk)
{
for (c=0; c<st->channels; c++)
{
for (i=0; i<F2_5; i++)
pcm[st->channels*i+c] = redundant_audio[st->channels*i+c];
}
smooth_fade(redundant_audio+st->channels*F2_5, pcm+st->channels*F2_5,
pcm+st->channels*F2_5, F2_5, st->channels, window, st->Fs);
}
if (transition)
{
if (audiosize >= F5)
{
for (i=0; i<st->channels*F2_5; i++)
pcm[i] = pcm_transition[i];
smooth_fade(pcm_transition+st->channels*F2_5, pcm+st->channels*F2_5,
pcm+st->channels*F2_5, F2_5,
st->channels, window, st->Fs);
} else {
/* Not enough time to do a clean transition, but we do it anyway
This will not preserve amplitude perfectly and may introduce
a bit of temporal aliasing, but it shouldn't be too bad and
that's pretty much the best we can do. In any case, generating this
transition it pretty silly in the first place */
smooth_fade(pcm_transition, pcm,
pcm, F2_5,
st->channels, window, st->Fs);
}
}
if(st->decode_gain)
{
opus_val32 gain;
gain = celt_exp2(MULT16_16_P15(QCONST16(6.48814081e-4f, 25), st->decode_gain));
for (i=0; i<frame_size*st->channels; i++)
{
opus_val32 x;
x = MULT16_32_P16(pcm[i],gain);
pcm[i] = SATURATE(x, 32767);
}
}
if (len <= 1)
st->rangeFinal = 0;
else
st->rangeFinal = dec.rng ^ redundant_rng;
st->prev_mode = mode;
st->prev_redundancy = redundancy && !celt_to_silk;
if (celt_ret>=0)
{
if (OPUS_CHECK_ARRAY(pcm, audiosize*st->channels))
OPUS_PRINT_INT(audiosize);
}
RESTORE_STACK;
return celt_ret < 0 ? celt_ret : audiosize;
}
void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd,
const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget,
opus_val16 *error, ec_enc *enc, int C, int LM, int nbAvailableBytes,
int force_intra, opus_val32 *delayedIntra, int two_pass, int loss_rate)
{
int intra;
opus_val16 max_decay;
VARDECL(opus_val16, oldEBands_intra);
VARDECL(opus_val16, error_intra);
ec_enc enc_start_state;
opus_uint32 tell;
int badness1=0;
opus_int32 intra_bias;
opus_val32 new_distortion;
SAVE_STACK;
intra = force_intra || (!two_pass && *delayedIntra>2*C*(end-start) && nbAvailableBytes > (end-start)*C);
intra_bias = (opus_int32)((budget**delayedIntra*loss_rate)/(C*512));
new_distortion = loss_distortion(eBands, oldEBands, start, effEnd, m->nbEBands, C);
tell = ec_tell(enc);
if (tell+3 > budget)
two_pass = intra = 0;
/* Encode the global flags using a simple probability model
(first symbols in the stream) */
max_decay = QCONST16(16.f,DB_SHIFT);
if (end-start>10)
{
#ifdef FIXED_POINT
max_decay = MIN32(max_decay, SHL32(EXTEND32(nbAvailableBytes),DB_SHIFT-3));
#else
max_decay = MIN32(max_decay, .125f*nbAvailableBytes);
#endif
}
enc_start_state = *enc;
ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16);
ALLOC(error_intra, C*m->nbEBands, opus_val16);
OPUS_COPY(oldEBands_intra, oldEBands, C*m->nbEBands);
if (two_pass || intra)
{
badness1 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands_intra, budget,
tell, e_prob_model[LM][1], error_intra, enc, C, LM, 1, max_decay);
}
if (!intra)
{
unsigned char *intra_buf;
ec_enc enc_intra_state;
opus_int32 tell_intra;
opus_uint32 nstart_bytes;
opus_uint32 nintra_bytes;
int badness2;
VARDECL(unsigned char, intra_bits);
tell_intra = ec_tell_frac(enc);
enc_intra_state = *enc;
nstart_bytes = ec_range_bytes(&enc_start_state);
nintra_bytes = ec_range_bytes(&enc_intra_state);
intra_buf = ec_get_buffer(&enc_intra_state) + nstart_bytes;
ALLOC(intra_bits, nintra_bytes-nstart_bytes, unsigned char);
/* Copy bits from intra bit-stream */
OPUS_COPY(intra_bits, intra_buf, nintra_bytes - nstart_bytes);
*enc = enc_start_state;
badness2 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands, budget,
tell, e_prob_model[LM][intra], error, enc, C, LM, 0, max_decay);
if (two_pass && (badness1 < badness2 || (badness1 == badness2 && ((opus_int32)ec_tell_frac(enc))+intra_bias > tell_intra)))
{
*enc = enc_intra_state;
/* Copy intra bits to bit-stream */
OPUS_COPY(intra_buf, intra_bits, nintra_bytes - nstart_bytes);
OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands);
OPUS_COPY(error, error_intra, C*m->nbEBands);
intra = 1;
}
} else {
static int quant_coarse_energy_impl(const CELTMode *m, int start, int end,
const opus_val16 *eBands, opus_val16 *oldEBands,
opus_int32 budget, opus_int32 tell,
const unsigned char *prob_model, opus_val16 *error, ec_enc *enc,
int C, int LM, int intra, opus_val16 max_decay)
{
int i, c;
int badness = 0;
opus_val32 prev[2] = {0,0};
opus_val16 coef;
opus_val16 beta;
if (tell+3 <= budget)
ec_enc_bit_logp(enc, intra, 3);
if (intra)
{
coef = 0;
beta = beta_intra;
} else {
beta = beta_coef[LM];
coef = pred_coef[LM];
}
/* Encode at a fixed coarse resolution */
for (i=start;i<end;i++)
{
c=0;
do {
int bits_left;
int qi, qi0;
opus_val32 q;
opus_val16 x;
opus_val32 f, tmp;
opus_val16 oldE;
opus_val16 decay_bound;
x = eBands[i+c*m->nbEBands];
oldE = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]);
#ifdef FIXED_POINT
f = SHL32(EXTEND32(x),7) - PSHR32(MULT16_16(coef,oldE), 8) - prev[c];
/* Rounding to nearest integer here is really important! */
qi = (f+QCONST32(.5f,DB_SHIFT+7))>>(DB_SHIFT+7);
decay_bound = EXTRACT16(MAX32(-QCONST16(28.f,DB_SHIFT),
SUB32((opus_val32)oldEBands[i+c*m->nbEBands],max_decay)));
#else
f = x-coef*oldE-prev[c];
/* Rounding to nearest integer here is really important! */
qi = (int)floor(.5f+f);
decay_bound = MAX16(-QCONST16(28.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]) - max_decay;
#endif
/* Prevent the energy from going down too quickly (e.g. for bands
that have just one bin) */
if (qi < 0 && x < decay_bound)
{
qi += (int)SHR16(SUB16(decay_bound,x), DB_SHIFT);
if (qi > 0)
qi = 0;
}
qi0 = qi;
/* If we don't have enough bits to encode all the energy, just assume
something safe. */
tell = ec_tell(enc);
bits_left = budget-tell-3*C*(end-i);
if (i!=start && bits_left < 30)
{
if (bits_left < 24)
qi = IMIN(1, qi);
if (bits_left < 16)
qi = IMAX(-1, qi);
}
if (budget-tell >= 15)
{
int pi;
pi = 2*IMIN(i,20);
ec_laplace_encode(enc, &qi,
prob_model[pi]<<7, prob_model[pi+1]<<6);
}
else if(budget-tell >= 2)
{
qi = IMAX(-1, IMIN(qi, 1));
ec_enc_icdf(enc, 2*qi^-(qi<0), small_energy_icdf, 2);
}
else if(budget-tell >= 1)
{
qi = IMIN(0, qi);
ec_enc_bit_logp(enc, -qi, 1);
}
else
qi = -1;
error[i+c*m->nbEBands] = PSHR32(f,7) - SHL16(qi,DB_SHIFT);
badness += abs(qi0-qi);
q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT);
tmp = PSHR32(MULT16_16(coef,oldE),8) + prev[c] + SHL32(q,7);
#ifdef FIXED_POINT
tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp);
#endif
oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7);
prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8));
} while (++c < C);
}
return badness;
}
opus_int silk_encode_frame_FLP(
silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */
opus_int32 *pnBytesOut, /* O Number of payload bytes; */
ec_enc *psRangeEnc, /* I/O compressor data structure */
opus_int condCoding, /* I The type of conditional coding to use */
opus_int maxBits, /* I If > 0: maximum number of output bits */
opus_int useCBR /* I Flag to force constant-bitrate operation */
)
{
silk_encoder_control_FLP sEncCtrl;
opus_int i, iter, maxIter, found_upper, found_lower, ret = 0;
silk_float *x_frame, *res_pitch_frame;
silk_float xfw[ MAX_FRAME_LENGTH ];
silk_float res_pitch[ 2 * MAX_FRAME_LENGTH + LA_PITCH_MAX ];
ec_enc sRangeEnc_copy, sRangeEnc_copy2;
silk_nsq_state sNSQ_copy, sNSQ_copy2;
opus_int32 seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower, gainMult_upper;
opus_int32 gainsID, gainsID_lower, gainsID_upper;
opus_int16 gainMult_Q8;
opus_int16 ec_prevLagIndex_copy;
opus_int ec_prevSignalType_copy;
opus_int8 LastGainIndex_copy2;
opus_int32 pGains_Q16[ MAX_NB_SUBFR ];
opus_uint8 ec_buf_copy[ 1275 ];
/* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */
LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0;
psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3;
/**************************************************************/
/* Set up Input Pointers, and insert frame in input buffer */
/**************************************************************/
/* pointers aligned with start of frame to encode */
x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; /* start of frame to encode */
res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; /* start of pitch LPC residual frame */
/***************************************/
/* Ensure smooth bandwidth transitions */
/***************************************/
silk_LP_variable_cutoff( &psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length );
/*******************************************/
/* Copy new frame to front of input buffer */
/*******************************************/
silk_short2float_array( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length );
/* Add tiny signal to avoid high CPU load from denormalized floating point numbers */
for( i = 0; i < 8; i++ ) {
x_frame[ LA_SHAPE_MS * psEnc->sCmn.fs_kHz + i * ( psEnc->sCmn.frame_length >> 3 ) ] += ( 1 - ( i & 2 ) ) * 1e-6f;
}
if( !psEnc->sCmn.prefillFlag ) {
/*****************************************/
/* Find pitch lags, initial LPC analysis */
/*****************************************/
silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame );
/************************/
/* Noise shape analysis */
/************************/
silk_noise_shape_analysis_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_frame );
/***************************************************/
/* Find linear prediction coefficients (LPC + LTP) */
/***************************************************/
silk_find_pred_coefs_FLP( psEnc, &sEncCtrl, res_pitch, x_frame, condCoding );
/****************************************/
/* Process gains */
/****************************************/
silk_process_gains_FLP( psEnc, &sEncCtrl, condCoding );
/*****************************************/
/* Prefiltering for noise shaper */
/*****************************************/
silk_prefilter_FLP( psEnc, &sEncCtrl, xfw, x_frame );
/****************************************/
/* Low Bitrate Redundant Encoding */
/****************************************/
silk_LBRR_encode_FLP( psEnc, &sEncCtrl, xfw, condCoding );
/* Loop over quantizer and entroy coding to control bitrate */
maxIter = 6;
gainMult_Q8 = SILK_FIX_CONST( 1, 8 );
found_lower = 0;
found_upper = 0;
gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr );
gainsID_lower = -1;
gainsID_upper = -1;
/* Copy part of the input state */
silk_memcpy( &sRangeEnc_copy, psRangeEnc, sizeof( ec_enc ) );
silk_memcpy( &sNSQ_copy, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) );
seed_copy = psEnc->sCmn.indices.Seed;
ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex;
ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType;
for( iter = 0; ; iter++ ) {
if( gainsID == gainsID_lower ) {
nBits = nBits_lower;
} else if( gainsID == gainsID_upper ) {
nBits = nBits_upper;
} else {
/* Restore part of the input state */
if( iter > 0 ) {
silk_memcpy( psRangeEnc, &sRangeEnc_copy, sizeof( ec_enc ) );
silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy, sizeof( silk_nsq_state ) );
psEnc->sCmn.indices.Seed = seed_copy;
psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy;
psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy;
}
/*****************************************/
/* Noise shaping quantization */
/*****************************************/
silk_NSQ_wrapper_FLP( psEnc, &sEncCtrl, &psEnc->sCmn.indices, &psEnc->sCmn.sNSQ, psEnc->sCmn.pulses, xfw );
/****************************************/
/* Encode Parameters */
/****************************************/
silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding );
/****************************************/
/* Encode Excitation Signal */
/****************************************/
silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType,
psEnc->sCmn.pulses, psEnc->sCmn.frame_length );
nBits = ec_tell( psRangeEnc );
if( useCBR == 0 && iter == 0 && nBits <= maxBits ) {
break;
}
}
if( iter == maxIter ) {
if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits ) ) {
/* Restore output state from earlier iteration that did meet the bitrate budget */
silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) );
silk_assert( sRangeEnc_copy2.offs <= 1275 );
silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.offs );
silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy2, sizeof( silk_nsq_state ) );
psEnc->sShape.LastGainIndex = LastGainIndex_copy2;
}
break;
}
if( nBits > maxBits ) {
if( found_lower == 0 && iter >= 2 ) {
/* Adjust the quantizer's rate/distortion tradeoff and discard previous "upper" results */
sEncCtrl.Lambda *= 1.5f;
found_upper = 0;
gainsID_upper = -1;
} else {
found_upper = 1;
nBits_upper = nBits;
gainMult_upper = gainMult_Q8;
gainsID_upper = gainsID;
}
} else if( nBits < maxBits - 5 ) {
found_lower = 1;
nBits_lower = nBits;
gainMult_lower = gainMult_Q8;
if( gainsID != gainsID_lower ) {
gainsID_lower = gainsID;
/* Copy part of the output state */
silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) );
silk_assert( psRangeEnc->offs <= 1275 );
silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs );
silk_memcpy( &sNSQ_copy2, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) );
LastGainIndex_copy2 = psEnc->sShape.LastGainIndex;
}
} else {
/* Within 5 bits of budget: close enough */
break;
}
if( ( found_lower & found_upper ) == 0 ) {
/* Adjust gain according to high-rate rate/distortion curve */
opus_int32 gain_factor_Q16;
gain_factor_Q16 = silk_log2lin( silk_LSHIFT( nBits - maxBits, 7 ) / psEnc->sCmn.frame_length + SILK_FIX_CONST( 16, 7 ) );
gain_factor_Q16 = silk_min_32( gain_factor_Q16, SILK_FIX_CONST( 2, 16 ) );
if( nBits > maxBits ) {
gain_factor_Q16 = silk_max_32( gain_factor_Q16, SILK_FIX_CONST( 1.3, 16 ) );
}
gainMult_Q8 = silk_SMULWB( gain_factor_Q16, gainMult_Q8 );
} else {
/* Adjust gain by interpolating */
gainMult_Q8 = gainMult_lower + ( ( gainMult_upper - gainMult_lower ) * ( maxBits - nBits_lower ) ) / ( nBits_upper - nBits_lower );
/* New gain multplier must be between 25% and 75% of old range (note that gainMult_upper < gainMult_lower) */
if( gainMult_Q8 > silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ) ) {
gainMult_Q8 = silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 );
} else
if( gainMult_Q8 < silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ) ) {
gainMult_Q8 = silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 );
}
}
for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) {
pGains_Q16[ i ] = silk_LSHIFT_SAT32( silk_SMULWB( sEncCtrl.GainsUnq_Q16[ i ], gainMult_Q8 ), 8 );
}
/* Quantize gains */
psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev;
silk_gains_quant( psEnc->sCmn.indices.GainsIndices, pGains_Q16,
&psEnc->sShape.LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr );
/* Unique identifier of gains vector */
gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr );
/* Overwrite unquantized gains with quantized gains and convert back to Q0 from Q16 */
for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) {
sEncCtrl.Gains[ i ] = pGains_Q16[ i ] / 65536.0f;
}
}
}
/* Update input buffer */
silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ],
( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( silk_float ) );
/* Parameters needed for next frame */
psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ];
psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType;
/* Exit without entropy coding */
if( psEnc->sCmn.prefillFlag ) {
/* No payload */
*pnBytesOut = 0;
return ret;
}
/****************************************/
/* Finalize payload */
/****************************************/
psEnc->sCmn.first_frame_after_reset = 0;
/* Payload size */
*pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
return ret;
}
int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
unsigned char *data, int max_data_bytes)
{
void *silk_enc;
CELTEncoder *celt_enc;
int i;
int ret=0;
SKP_int32 nBytes;
ec_enc enc;
int framerate, period;
int silk_internal_bandwidth=-1;
int bytes_target;
int prefill=0;
int start_band = 0;
int redundancy = 0;
int redundancy_bytes = 0;
int celt_to_silk = 0;
/* TODO: This is 60 only so we can handle 60ms speech/audio switching
it shouldn't bee too hard to reduce to 20 ms if needed */
short pcm_buf[60*48*2];
int nb_compr_bytes;
int to_celt = 0;
celt_int32 mono_rate;
silk_enc = (char*)st+st->silk_enc_offset;
celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset);
if (st->user_bitrate_bps==OPUS_BITRATE_AUTO)
st->bitrate_bps = 60*st->Fs/frame_size + st->Fs*st->channels;
else
st->bitrate_bps = st->user_bitrate_bps;
/* Rate-dependent mono-stereo decision */
if (st->mode == MODE_CELT_ONLY && st->channels == 2)
{
celt_int32 decision_rate;
decision_rate = st->bitrate_bps + st->voice_ratio*st->voice_ratio;
/* Add some hysteresis */
if (st->stream_channels == 2)
decision_rate += 4000;
else
decision_rate -= 4000;
if (decision_rate>48000)
st->stream_channels = 2;
else
st->stream_channels = 1;
}
/* Equivalent bit-rate for mono */
mono_rate = st->bitrate_bps;
if (st->stream_channels==2)
mono_rate = (mono_rate+10000)/2;
/* Compensate for smaller frame sizes assuming an equivalent overhead
of 60 bits/frame */
mono_rate -= 60*(st->Fs/frame_size - 50);
/* Mode selection */
if (st->user_mode==OPUS_MODE_AUTO)
{
celt_int32 decision_rate;
/* SILK/CELT threshold is higher for voice than for music */
decision_rate = mono_rate - 3*st->voice_ratio*st->voice_ratio;
/* Hysteresis */
if (st->prev_mode == MODE_CELT_ONLY)
decision_rate += 4000;
else if (st->prev_mode>0)
decision_rate -= 4000;
if (decision_rate>24000)
st->mode = MODE_CELT_ONLY;
else
st->mode = MODE_SILK_ONLY;
} else if (st->user_mode==OPUS_MODE_VOICE)
{
st->mode = MODE_SILK_ONLY;
} else {/* OPUS_AUDIO_MODE */
st->mode = MODE_CELT_ONLY;
}
/* Automatic (rate-dependent) bandwidth selection */
if (st->mode == MODE_CELT_ONLY || st->first || st->silk_mode.allowBandwidthSwitch)
{
const int *bandwidth_thresholds;
int bandwidth = BANDWIDTH_FULLBAND;
bandwidth_thresholds = st->mode == MODE_CELT_ONLY ? audio_bandwidth_thresholds : voice_bandwidth_thresholds;
do {
int threshold, hysteresis;
threshold = bandwidth_thresholds[2*(bandwidth-BANDWIDTH_MEDIUMBAND)];
hysteresis = bandwidth_thresholds[2*(bandwidth-BANDWIDTH_MEDIUMBAND)+1];
if (!st->first)
{
if (st->bandwidth >= bandwidth)
threshold -= hysteresis;
else
threshold += hysteresis;
}
if (mono_rate >= threshold)
break;
} while (--bandwidth>BANDWIDTH_NARROWBAND);
st->bandwidth = bandwidth;
/* Prevents any transition to SWB/FB until the SILK layer has fully
switched to WB mode and turned the variable LP filter off */
if (st->mode != MODE_CELT_ONLY && !st->silk_mode.inWBmodeWithoutVariableLP && st->bandwidth > BANDWIDTH_WIDEBAND)
st->bandwidth = BANDWIDTH_WIDEBAND;
}
/* Prevents Opus from wasting bits on frequencies that are above
the Nyquist rate of the input signal */
if (st->Fs <= 24000 && st->bandwidth > BANDWIDTH_SUPERWIDEBAND)
st->bandwidth = BANDWIDTH_SUPERWIDEBAND;
if (st->Fs <= 16000 && st->bandwidth > BANDWIDTH_WIDEBAND)
st->bandwidth = BANDWIDTH_WIDEBAND;
if (st->Fs <= 12000 && st->bandwidth > BANDWIDTH_MEDIUMBAND)
st->bandwidth = BANDWIDTH_MEDIUMBAND;
if (st->Fs <= 8000 && st->bandwidth > BANDWIDTH_NARROWBAND)
st->bandwidth = BANDWIDTH_NARROWBAND;
if (st->user_bandwidth != BANDWIDTH_AUTO)
st->bandwidth = st->user_bandwidth;
/* Prevents nonsensical configurations, i.e. modes that don't exist */
if (frame_size < st->Fs/100 && st->mode != MODE_CELT_ONLY)
st->mode = MODE_CELT_ONLY;
if (frame_size > st->Fs/50 && st->mode != MODE_SILK_ONLY)
st->mode = MODE_SILK_ONLY;
if (st->mode == MODE_CELT_ONLY && st->bandwidth == BANDWIDTH_MEDIUMBAND)
st->bandwidth = BANDWIDTH_WIDEBAND;
if (st->mode == MODE_SILK_ONLY && st->bandwidth > BANDWIDTH_WIDEBAND)
st->mode = MODE_HYBRID;
if (st->mode == MODE_HYBRID && st->bandwidth <= BANDWIDTH_WIDEBAND)
st->mode = MODE_SILK_ONLY;
bytes_target = st->bitrate_bps * frame_size / (st->Fs * 8) - 1;
data += 1;
if (st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY)
{
silk_EncControlStruct dummy;
silk_InitEncoder( st->silk_enc, &dummy);
prefill=1;
}
if (st->prev_mode > 0 &&
((st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) ||
(st->mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY)))
{
redundancy = 1;
celt_to_silk = (st->mode != MODE_CELT_ONLY);
if (!celt_to_silk)
{
/* Switch to SILK/hybrid if frame size is 10 ms or more*/
if (frame_size >= st->Fs/100)
{
st->mode = st->prev_mode;
to_celt = 1;
} else {
redundancy=0;
}
}
}
ec_enc_init(&enc, data, max_data_bytes-1);
/* SILK processing */
if (st->mode != MODE_CELT_ONLY)
{
st->silk_mode.bitRate = st->bitrate_bps - 8*st->Fs/frame_size;
if( st->mode == MODE_HYBRID ) {
st->silk_mode.bitRate /= st->stream_channels;
if( st->bandwidth == BANDWIDTH_SUPERWIDEBAND ) {
if( st->Fs == 100 * frame_size ) {
/* 24 kHz, 10 ms */
st->silk_mode.bitRate = ( ( st->silk_mode.bitRate + 2000 + st->use_vbr * 1000 ) * 2 ) / 3;
} else {
/* 24 kHz, 20 ms */
st->silk_mode.bitRate = ( ( st->silk_mode.bitRate + 1000 + st->use_vbr * 1000 ) * 2 ) / 3;
}
} else {
if( st->Fs == 100 * frame_size ) {
/* 48 kHz, 10 ms */
st->silk_mode.bitRate = ( st->silk_mode.bitRate + 8000 + st->use_vbr * 3000 ) / 2;
} else {
/* 48 kHz, 20 ms */
st->silk_mode.bitRate = ( st->silk_mode.bitRate + 9000 + st->use_vbr * 1000 ) / 2;
}
}
st->silk_mode.bitRate *= st->stream_channels;
/* don't let SILK use more than 80% */
if( st->silk_mode.bitRate > ( st->bitrate_bps - 8*st->Fs/frame_size ) * 4/5 ) {
st->silk_mode.bitRate = ( st->bitrate_bps - 8*st->Fs/frame_size ) * 4/5;
}
}
st->silk_mode.payloadSize_ms = 1000 * frame_size / st->Fs;
st->silk_mode.nChannelsAPI = st->channels;
st->silk_mode.nChannelsInternal = st->stream_channels;
if (st->bandwidth == BANDWIDTH_NARROWBAND) {
st->silk_mode.desiredInternalSampleRate = 8000;
} else if (st->bandwidth == BANDWIDTH_MEDIUMBAND) {
st->silk_mode.desiredInternalSampleRate = 12000;
} else {
SKP_assert( st->mode == MODE_HYBRID || st->bandwidth == BANDWIDTH_WIDEBAND );
st->silk_mode.desiredInternalSampleRate = 16000;
}
if( st->mode == MODE_HYBRID ) {
/* Don't allow bandwidth reduction at lowest bitrates in hybrid mode */
st->silk_mode.minInternalSampleRate = 16000;
} else {
st->silk_mode.minInternalSampleRate = 8000;
}
st->silk_mode.maxInternalSampleRate = 16000;
/* Call SILK encoder for the low band */
nBytes = max_data_bytes-1;
if (prefill)
{
int zero=0;
silk_Encode( silk_enc, &st->silk_mode, st->delay_buffer, st->encoder_buffer, NULL, &zero, 1 );
}
ret = silk_Encode( silk_enc, &st->silk_mode, pcm, frame_size, &enc, &nBytes, 0 );
if( ret ) {
fprintf (stderr, "SILK encode error: %d\n", ret);
/* Handle error */
}
if (nBytes==0)
return 0;
/* Extract SILK internal bandwidth for signaling in first byte */
if( st->mode == MODE_SILK_ONLY ) {
if( st->silk_mode.internalSampleRate == 8000 ) {
silk_internal_bandwidth = BANDWIDTH_NARROWBAND;
} else if( st->silk_mode.internalSampleRate == 12000 ) {
silk_internal_bandwidth = BANDWIDTH_MEDIUMBAND;
} else if( st->silk_mode.internalSampleRate == 16000 ) {
silk_internal_bandwidth = BANDWIDTH_WIDEBAND;
}
} else {
SKP_assert( st->silk_mode.internalSampleRate == 16000 );
}
}
/* CELT processing */
{
int endband=21;
switch(st->bandwidth)
{
case BANDWIDTH_NARROWBAND:
endband = 13;
break;
case BANDWIDTH_WIDEBAND:
endband = 17;
break;
case BANDWIDTH_SUPERWIDEBAND:
endband = 19;
break;
case BANDWIDTH_FULLBAND:
endband = 21;
break;
}
celt_encoder_ctl(celt_enc, CELT_SET_END_BAND(endband));
celt_encoder_ctl(celt_enc, CELT_SET_CHANNELS(st->stream_channels));
}
if (st->mode != MODE_SILK_ONLY)
{
celt_encoder_ctl(celt_enc, CELT_SET_VBR(0));
celt_encoder_ctl(celt_enc, CELT_SET_BITRATE(510000));
if (st->prev_mode == MODE_SILK_ONLY)
{
unsigned char dummy[10];
celt_encoder_ctl(celt_enc, CELT_RESET_STATE);
celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0));
celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0));
/* TODO: This wastes CPU a bit compared to just prefilling the buffer */
celt_encode(celt_enc, &st->delay_buffer[(st->encoder_buffer-st->delay_compensation-st->Fs/400)*st->channels], st->Fs/400, dummy, 10);
} else {
celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(2));
}
if (st->mode == MODE_HYBRID)
{
int len;
len = (ec_tell(&enc)+7)>>3;
if( st->use_vbr ) {
nb_compr_bytes = len + bytes_target - (st->silk_mode.bitRate * frame_size) / (8 * st->Fs);
} else {
/* check if SILK used up too much */
nb_compr_bytes = len > bytes_target ? len : bytes_target;
}
} else {
int silk_encode_frame_FIX(silk_encoder_state_FIX * psEnc, /* I/O Pointer to Silk FIX encoder state */
int32_t * pnBytesOut, /* O Pointer to number of payload bytes; */
ec_enc * psRangeEnc, /* I/O compressor data structure */
int condCoding, /* I The type of conditional coding to use */
int maxBits, /* I If > 0: maximum number of output bits */
int useCBR /* I Flag to force constant-bitrate operation */
) {
silk_encoder_control_FIX sEncCtrl;
int i, iter, maxIter, found_upper, found_lower, ret = 0;
int16_t *x_frame;
ec_enc sRangeEnc_copy, sRangeEnc_copy2;
silk_nsq_state sNSQ_copy, sNSQ_copy2;
int32_t seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower,
gainMult_upper;
int32_t gainsID, gainsID_lower, gainsID_upper;
int16_t gainMult_Q8;
int16_t ec_prevLagIndex_copy;
int ec_prevSignalType_copy;
int8_t LastGainIndex_copy2;
/* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */
LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower =
gainMult_upper = 0;
psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3;
/**************************************************************/
/* Set up Input Pointers, and insert frame in input buffer */
/*************************************************************/
/* start of frame to encode */
x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length;
/***************************************/
/* Ensure smooth bandwidth transitions */
/***************************************/
silk_LP_variable_cutoff(&psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1,
psEnc->sCmn.frame_length);
/*******************************************/
/* Copy new frame to front of input buffer */
/*******************************************/
memcpy(x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz,
psEnc->sCmn.inputBuf + 1,
psEnc->sCmn.frame_length * sizeof(int16_t));
if (!psEnc->sCmn.prefillFlag) {
int16_t *res_pitch_frame;
int16_t res_pitch[psEnc->sCmn.la_pitch +
psEnc->sCmn.frame_length +
psEnc->sCmn.ltp_mem_length];
/* start of pitch LPC residual frame */
res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length;
/*****************************************/
/* Find pitch lags, initial LPC analysis */
/*****************************************/
silk_find_pitch_lags_FIX(psEnc, &sEncCtrl, res_pitch, x_frame,
psEnc->sCmn.arch);
/************************/
/* Noise shape analysis */
/************************/
silk_noise_shape_analysis_FIX(psEnc, &sEncCtrl, res_pitch_frame,
x_frame, psEnc->sCmn.arch);
/***************************************************/
/* Find linear prediction coefficients (LPC + LTP) */
/***************************************************/
silk_find_pred_coefs_FIX(psEnc, &sEncCtrl, res_pitch, x_frame,
condCoding);
/****************************************/
/* Process gains */
/****************************************/
silk_process_gains_FIX(psEnc, &sEncCtrl, condCoding);
/*****************************************/
/* Prefiltering for noise shaper */
/*****************************************/
int32_t xfw_Q3[psEnc->sCmn.frame_length];
silk_prefilter_FIX(psEnc, &sEncCtrl, xfw_Q3, x_frame);
/****************************************/
/* Low Bitrate Redundant Encoding */
/****************************************/
silk_LBRR_encode_FIX(psEnc, &sEncCtrl, xfw_Q3, condCoding);
/* Loop over quantizer and entropy coding to control bitrate */
maxIter = 6;
gainMult_Q8 = SILK_FIX_CONST(1, 8);
found_lower = 0;
found_upper = 0;
gainsID =
silk_gains_ID(psEnc->sCmn.indices.GainsIndices,
psEnc->sCmn.nb_subfr);
gainsID_lower = -1;
gainsID_upper = -1;
/* Copy part of the input state */
memcpy(&sRangeEnc_copy, psRangeEnc, sizeof(ec_enc));
memcpy(&sNSQ_copy, &psEnc->sCmn.sNSQ,
sizeof(silk_nsq_state));
seed_copy = psEnc->sCmn.indices.Seed;
ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex;
ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType;
uint8_t ec_buf_copy[1275];
for (iter = 0;; iter++) {
if (gainsID == gainsID_lower) {
nBits = nBits_lower;
} else if (gainsID == gainsID_upper) {
nBits = nBits_upper;
} else {
/* Restore part of the input state */
if (iter > 0) {
memcpy(psRangeEnc, &sRangeEnc_copy,
sizeof(ec_enc));
memcpy(&psEnc->sCmn.sNSQ,
&sNSQ_copy,
sizeof(silk_nsq_state));
psEnc->sCmn.indices.Seed = seed_copy;
psEnc->sCmn.ec_prevLagIndex =
ec_prevLagIndex_copy;
psEnc->sCmn.ec_prevSignalType =
ec_prevSignalType_copy;
}
/*****************************************/
/* Noise shaping quantization */
/*****************************************/
if (psEnc->sCmn.nStatesDelayedDecision > 1
|| psEnc->sCmn.warping_Q16 > 0) {
silk_NSQ_del_dec(&psEnc->sCmn,
&psEnc->sCmn.sNSQ,
&psEnc->sCmn.indices,
xfw_Q3,
psEnc->sCmn.pulses,
sEncCtrl.
PredCoef_Q12[0],
sEncCtrl.LTPCoef_Q14,
sEncCtrl.AR2_Q13,
sEncCtrl.
HarmShapeGain_Q14,
sEncCtrl.Tilt_Q14,
sEncCtrl.LF_shp_Q14,
sEncCtrl.Gains_Q16,
sEncCtrl.pitchL,
sEncCtrl.Lambda_Q10,
sEncCtrl.
LTP_scale_Q14);
} else {
silk_NSQ(&psEnc->sCmn,
&psEnc->sCmn.sNSQ,
&psEnc->sCmn.indices, xfw_Q3,
psEnc->sCmn.pulses,
sEncCtrl.PredCoef_Q12[0],
sEncCtrl.LTPCoef_Q14,
sEncCtrl.AR2_Q13,
sEncCtrl.HarmShapeGain_Q14,
sEncCtrl.Tilt_Q14,
sEncCtrl.LF_shp_Q14,
sEncCtrl.Gains_Q16,
sEncCtrl.pitchL,
sEncCtrl.Lambda_Q10,
sEncCtrl.LTP_scale_Q14);
}
/****************************************/
/* Encode Parameters */
/****************************************/
silk_encode_indices(&psEnc->sCmn, psRangeEnc,
psEnc->sCmn.nFramesEncoded,
0, condCoding);
/****************************************/
/* Encode Excitation Signal */
/****************************************/
silk_encode_pulses(psRangeEnc,
psEnc->sCmn.indices.
signalType,
psEnc->sCmn.indices.
quantOffsetType,
psEnc->sCmn.pulses,
psEnc->sCmn.frame_length);
nBits = ec_tell(psRangeEnc);
if (useCBR == 0 && iter == 0
&& nBits <= maxBits) {
break;
}
}
if (iter == maxIter) {
if (found_lower
&& (gainsID == gainsID_lower
|| nBits > maxBits)) {
/* Restore output state from earlier iteration that did meet the bitrate budget */
memcpy(psRangeEnc,
&sRangeEnc_copy2,
sizeof(ec_enc));
assert(sRangeEnc_copy2.offs <=
1275);
memcpy(psRangeEnc->buf,
ec_buf_copy,
sRangeEnc_copy2.offs);
memcpy(&psEnc->sCmn.sNSQ,
&sNSQ_copy2,
sizeof(silk_nsq_state));
psEnc->sShape.LastGainIndex =
LastGainIndex_copy2;
}
break;
}
if (nBits > maxBits) {
if (found_lower == 0 && iter >= 2) {
/* Adjust the quantizer's rate/distortion tradeoff and discard previous "upper" results */
sEncCtrl.Lambda_Q10 =
silk_ADD_RSHIFT32(sEncCtrl.
Lambda_Q10,
sEncCtrl.
Lambda_Q10, 1);
found_upper = 0;
gainsID_upper = -1;
} else {
found_upper = 1;
nBits_upper = nBits;
gainMult_upper = gainMult_Q8;
gainsID_upper = gainsID;
}
} else if (nBits < maxBits - 5) {
found_lower = 1;
nBits_lower = nBits;
gainMult_lower = gainMult_Q8;
if (gainsID != gainsID_lower) {
gainsID_lower = gainsID;
/* Copy part of the output state */
memcpy(&sRangeEnc_copy2,
psRangeEnc, sizeof(ec_enc));
assert(psRangeEnc->offs <= 1275);
memcpy(ec_buf_copy,
psRangeEnc->buf,
psRangeEnc->offs);
memcpy(&sNSQ_copy2,
&psEnc->sCmn.sNSQ,
sizeof(silk_nsq_state));
LastGainIndex_copy2 =
psEnc->sShape.LastGainIndex;
}
} else {
/* Within 5 bits of budget: close enough */
break;
}
if ((found_lower & found_upper) == 0) {
/* Adjust gain according to high-rate rate/distortion curve */
int32_t gain_factor_Q16;
gain_factor_Q16 =
silk_log2lin(silk_LSHIFT(nBits - maxBits, 7)
/ psEnc->sCmn.frame_length +
SILK_FIX_CONST(16, 7));
gain_factor_Q16 =
silk_min_32(gain_factor_Q16,
SILK_FIX_CONST(2, 16));
if (nBits > maxBits) {
gain_factor_Q16 =
silk_max_32(gain_factor_Q16,
SILK_FIX_CONST(1.3,
16));
}
gainMult_Q8 =
silk_SMULWB(gain_factor_Q16, gainMult_Q8);
} else {
/* Adjust gain by interpolating */
assert(nBits_upper != nBits_lower);
gainMult_Q8 =
gainMult_lower +
silk_DIV32_16(silk_MUL
(gainMult_upper -
gainMult_lower,
maxBits - nBits_lower),
nBits_upper - nBits_lower);
/* New gain multplier must be between 25% and 75% of old range (note that gainMult_upper < gainMult_lower) */
if (gainMult_Q8 >
silk_ADD_RSHIFT32(gainMult_lower,
gainMult_upper -
gainMult_lower, 2)) {
gainMult_Q8 =
silk_ADD_RSHIFT32(gainMult_lower,
gainMult_upper -
gainMult_lower,
2);
} else if (gainMult_Q8 <
silk_SUB_RSHIFT32(gainMult_upper,
gainMult_upper -
gainMult_lower,
2)) {
gainMult_Q8 =
silk_SUB_RSHIFT32(gainMult_upper,
gainMult_upper -
gainMult_lower,
2);
}
}
for (i = 0; i < psEnc->sCmn.nb_subfr; i++) {
sEncCtrl.Gains_Q16[i] =
silk_LSHIFT_SAT32(silk_SMULWB
(sEncCtrl.GainsUnq_Q16[i],
gainMult_Q8), 8);
}
/* Quantize gains */
psEnc->sShape.LastGainIndex =
sEncCtrl.lastGainIndexPrev;
silk_gains_quant(psEnc->sCmn.indices.GainsIndices,
sEncCtrl.Gains_Q16,
&psEnc->sShape.LastGainIndex,
condCoding == CODE_CONDITIONALLY,
psEnc->sCmn.nb_subfr);
/* Unique identifier of gains vector */
gainsID =
silk_gains_ID(psEnc->sCmn.indices.GainsIndices,
psEnc->sCmn.nb_subfr);
}
}
/* Update input buffer */
memmove(psEnc->x_buf, &psEnc->x_buf[psEnc->sCmn.frame_length],
(psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz) * sizeof(int16_t));
/* Exit without entropy coding */
if (psEnc->sCmn.prefillFlag) {
/* No payload */
*pnBytesOut = 0;
return ret;
}
/* Parameters needed for next frame */
psEnc->sCmn.prevLag = sEncCtrl.pitchL[psEnc->sCmn.nb_subfr - 1];
psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType;
/****************************************/
/* Finalize payload */
/****************************************/
psEnc->sCmn.first_frame_after_reset = 0;
/* Payload size */
*pnBytesOut = silk_RSHIFT(ec_tell(psRangeEnc) + 7, 3);
return ret;
}
/* encControl->payloadSize_ms is set to */
opus_int silk_Encode( /* O Returns error code */
void *encState, /* I/O State */
silk_EncControlStruct *encControl, /* I Control status */
const opus_int16 *samplesIn, /* I Speech sample input vector */
opus_int nSamplesIn, /* I Number of samples in input vector */
ec_enc *psRangeEnc, /* I/O Compressor data structure */
opus_int32 *nBytesOut, /* I/O Number of bytes in payload (input: Max bytes) */
const opus_int prefillFlag /* I Flag to indicate prefilling buffers no coding */
)
{
opus_int n, i, nBits, flags, tmp_payloadSize_ms = 0, tmp_complexity = 0, ret = 0;
opus_int nSamplesToBuffer, nSamplesToBufferMax, nBlocksOf10ms;
opus_int nSamplesFromInput = 0, nSamplesFromInputMax;
opus_int speech_act_thr_for_switch_Q8;
opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol, sum;
silk_encoder *psEnc = ( silk_encoder * )encState;
VARDECL( opus_int16, buf );
opus_int transition, curr_block, tot_blocks;
SAVE_STACK;
if (encControl->reducedDependency)
{
psEnc->state_Fxx[0].sCmn.first_frame_after_reset = 1;
psEnc->state_Fxx[1].sCmn.first_frame_after_reset = 1;
}
psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded = psEnc->state_Fxx[ 1 ].sCmn.nFramesEncoded = 0;
/* Check values in encoder control structure */
if( ( ret = check_control_input( encControl ) ) != 0 ) {
silk_assert( 0 );
RESTORE_STACK;
return ret;
}
encControl->switchReady = 0;
if( encControl->nChannelsInternal > psEnc->nChannelsInternal ) {
/* Mono -> Stereo transition: init state of second channel and stereo state */
ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ], psEnc->state_Fxx[ 0 ].sCmn.arch );
silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) );
silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) );
psEnc->sStereo.mid_side_amp_Q0[ 0 ] = 0;
psEnc->sStereo.mid_side_amp_Q0[ 1 ] = 1;
psEnc->sStereo.mid_side_amp_Q0[ 2 ] = 0;
psEnc->sStereo.mid_side_amp_Q0[ 3 ] = 1;
psEnc->sStereo.width_prev_Q14 = 0;
psEnc->sStereo.smth_width_Q14 = SILK_FIX_CONST( 1, 14 );
if( psEnc->nChannelsAPI == 2 ) {
silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof( silk_resampler_state_struct ) );
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 ) );
}
}
transition = (encControl->payloadSize_ms != psEnc->state_Fxx[ 0 ].sCmn.PacketSize_ms) || (psEnc->nChannelsInternal != encControl->nChannelsInternal);
psEnc->nChannelsAPI = encControl->nChannelsAPI;
psEnc->nChannelsInternal = encControl->nChannelsInternal;
nBlocksOf10ms = silk_DIV32( 100 * nSamplesIn, encControl->API_sampleRate );
tot_blocks = ( nBlocksOf10ms > 1 ) ? nBlocksOf10ms >> 1 : 1;
curr_block = 0;
if( prefillFlag ) {
/* Only accept input length of 10 ms */
if( nBlocksOf10ms != 1 ) {
silk_assert( 0 );
RESTORE_STACK;
return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
}
/* Reset Encoder */
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch );
silk_assert( !ret );
}
tmp_payloadSize_ms = encControl->payloadSize_ms;
encControl->payloadSize_ms = 10;
tmp_complexity = encControl->complexity;
encControl->complexity = 0;
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
psEnc->state_Fxx[ n ].sCmn.prefillFlag = 1;
}
} else {
/* Only accept input lengths that are a multiple of 10 ms */
if( nBlocksOf10ms * encControl->API_sampleRate != 100 * nSamplesIn || nSamplesIn < 0 ) {
silk_assert( 0 );
RESTORE_STACK;
return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
}
/* Make sure no more than one packet can be produced */
if( 1000 * (opus_int32)nSamplesIn > encControl->payloadSize_ms * encControl->API_sampleRate ) {
silk_assert( 0 );
RESTORE_STACK;
return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
}
}
TargetRate_bps = silk_RSHIFT32( encControl->bitRate, encControl->nChannelsInternal - 1 );
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
/* Force the side channel to the same rate as the mid */
opus_int force_fs_kHz = (n==1) ? psEnc->state_Fxx[0].sCmn.fs_kHz : 0;
if( ( ret = silk_control_encoder( &psEnc->state_Fxx[ n ], encControl, TargetRate_bps, psEnc->allowBandwidthSwitch, n, force_fs_kHz ) ) != 0 ) {
silk_assert( 0 );
RESTORE_STACK;
return ret;
}
if( psEnc->state_Fxx[n].sCmn.first_frame_after_reset || transition ) {
for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) {
psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] = 0;
}
}
psEnc->state_Fxx[ n ].sCmn.inDTX = psEnc->state_Fxx[ n ].sCmn.useDTX;
}
silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz );
/* Input buffering/resampling and encoding */
nSamplesToBufferMax =
10 * nBlocksOf10ms * psEnc->state_Fxx[ 0 ].sCmn.fs_kHz;
nSamplesFromInputMax =
silk_DIV32_16( nSamplesToBufferMax *
psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz,
psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 );
ALLOC( buf, nSamplesFromInputMax, opus_int16 );
while( 1 ) {
nSamplesToBuffer = psEnc->state_Fxx[ 0 ].sCmn.frame_length - psEnc->state_Fxx[ 0 ].sCmn.inputBufIx;
nSamplesToBuffer = silk_min( nSamplesToBuffer, nSamplesToBufferMax );
nSamplesFromInput = silk_DIV32_16( nSamplesToBuffer * psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 );
/* Resample and write to buffer */
if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 2 ) {
opus_int id = psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded;
for( n = 0; n < nSamplesFromInput; n++ ) {
buf[ n ] = samplesIn[ 2 * n ];
}
/* Making sure to start both resamplers from the same state when switching from mono to stereo */
if( psEnc->nPrevChannelsInternal == 1 && id==0 ) {
silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof(psEnc->state_Fxx[ 1 ].sCmn.resampler_state));
}
ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
&psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
nSamplesToBuffer = psEnc->state_Fxx[ 1 ].sCmn.frame_length - psEnc->state_Fxx[ 1 ].sCmn.inputBufIx;
nSamplesToBuffer = silk_min( nSamplesToBuffer, 10 * nBlocksOf10ms * psEnc->state_Fxx[ 1 ].sCmn.fs_kHz );
for( n = 0; n < nSamplesFromInput; n++ ) {
buf[ n ] = samplesIn[ 2 * n + 1 ];
}
ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state,
&psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
psEnc->state_Fxx[ 1 ].sCmn.inputBufIx += nSamplesToBuffer;
} else if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 1 ) {
/* Combine left and right channels before resampling */
for( n = 0; n < nSamplesFromInput; n++ ) {
sum = samplesIn[ 2 * n ] + samplesIn[ 2 * n + 1 ];
buf[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 );
}
ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
&psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
/* On the first mono frame, average the results for the two resampler states */
if( psEnc->nPrevChannelsInternal == 2 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 ) {
ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state,
&psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
for( n = 0; n < psEnc->state_Fxx[ 0 ].sCmn.frame_length; n++ ) {
psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ] =
silk_RSHIFT(psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ]
+ psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx+n+2 ], 1);
}
}
psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
} else {
silk_assert( encControl->nChannelsAPI == 1 && encControl->nChannelsInternal == 1 );
silk_memcpy(buf, samplesIn, nSamplesFromInput*sizeof(opus_int16));
ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
&psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
}
samplesIn += nSamplesFromInput * encControl->nChannelsAPI;
nSamplesIn -= nSamplesFromInput;
/* Default */
psEnc->allowBandwidthSwitch = 0;
/* Silk encoder */
if( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx >= psEnc->state_Fxx[ 0 ].sCmn.frame_length ) {
/* Enough data in input buffer, so encode */
silk_assert( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx == psEnc->state_Fxx[ 0 ].sCmn.frame_length );
silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inputBufIx == psEnc->state_Fxx[ 1 ].sCmn.frame_length );
/* Deal with LBRR data */
if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 && !prefillFlag ) {
/* Create space at start of payload for VAD and FEC flags */
opus_uint8 iCDF[ 2 ] = { 0, 0 };
iCDF[ 0 ] = 256 - silk_RSHIFT( 256, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal );
ec_enc_icdf( psRangeEnc, 0, iCDF, 8 );
/* Encode any LBRR data from previous packet */
/* Encode LBRR flags */
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
LBRR_symbol = 0;
for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) {
LBRR_symbol |= silk_LSHIFT( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ], i );
}
psEnc->state_Fxx[ n ].sCmn.LBRR_flag = LBRR_symbol > 0 ? 1 : 0;
if( LBRR_symbol && psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket > 1 ) {
ec_enc_icdf( psRangeEnc, LBRR_symbol - 1, silk_LBRR_flags_iCDF_ptr[ psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket - 2 ], 8 );
}
}
/* Code LBRR indices and excitation signals */
for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) {
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
if( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] ) {
opus_int condCoding;
if( encControl->nChannelsInternal == 2 && n == 0 ) {
silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ i ] );
/* For LBRR data there's no need to code the mid-only flag if the side-channel LBRR flag is set */
if( psEnc->state_Fxx[ 1 ].sCmn.LBRR_flags[ i ] == 0 ) {
silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ i ] );
}
}
/* Use conditional coding if previous frame available */
if( i > 0 && psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i - 1 ] ) {
condCoding = CODE_CONDITIONALLY;
} else {
condCoding = CODE_INDEPENDENTLY;
}
silk_encode_indices( &psEnc->state_Fxx[ n ].sCmn, psRangeEnc, i, 1, condCoding );
silk_encode_pulses( psRangeEnc, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].signalType, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].quantOffsetType,
psEnc->state_Fxx[ n ].sCmn.pulses_LBRR[ i ], psEnc->state_Fxx[ n ].sCmn.frame_length );
}
}
}
/* Reset LBRR flags */
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
silk_memset( psEnc->state_Fxx[ n ].sCmn.LBRR_flags, 0, sizeof( psEnc->state_Fxx[ n ].sCmn.LBRR_flags ) );
}
psEnc->nBitsUsedLBRR = ec_tell( psRangeEnc );
}
silk_HP_variable_cutoff( psEnc->state_Fxx );
/* Total target bits for packet */
nBits = silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 );
/* Subtract bits used for LBRR */
if( !prefillFlag ) {
nBits -= psEnc->nBitsUsedLBRR;
}
/* Divide by number of uncoded frames left in packet */
nBits = silk_DIV32_16( nBits, psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket );
/* Convert to bits/second */
if( encControl->payloadSize_ms == 10 ) {
TargetRate_bps = silk_SMULBB( nBits, 100 );
} else {
TargetRate_bps = silk_SMULBB( nBits, 50 );
}
/* Subtract fraction of bits in excess of target in previous frames and packets */
TargetRate_bps -= silk_DIV32_16( silk_MUL( psEnc->nBitsExceeded, 1000 ), BITRESERVOIR_DECAY_TIME_MS );
if( !prefillFlag && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded > 0 ) {
/* Compare actual vs target bits so far in this packet */
opus_int32 bitsBalance = ec_tell( psRangeEnc ) - psEnc->nBitsUsedLBRR - nBits * psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded;
TargetRate_bps -= silk_DIV32_16( silk_MUL( bitsBalance, 1000 ), BITRESERVOIR_DECAY_TIME_MS );
}
/* Never exceed input bitrate */
TargetRate_bps = silk_LIMIT( TargetRate_bps, encControl->bitRate, 5000 );
/* Convert Left/Right to Mid/Side */
if( encControl->nChannelsInternal == 2 ) {
silk_stereo_LR_to_MS( &psEnc->sStereo, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ 2 ], &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ 2 ],
psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ], &psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ],
MStargetRates_bps, TargetRate_bps, psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8, encControl->toMono,
psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, psEnc->state_Fxx[ 0 ].sCmn.frame_length );
if( psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) {
/* Reset side channel encoder memory for first frame with side coding */
if( psEnc->prev_decode_only_middle == 1 ) {
silk_memset( &psEnc->state_Fxx[ 1 ].sShape, 0, sizeof( psEnc->state_Fxx[ 1 ].sShape ) );
silk_memset( &psEnc->state_Fxx[ 1 ].sPrefilt, 0, sizeof( psEnc->state_Fxx[ 1 ].sPrefilt ) );
silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sNSQ, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sNSQ ) );
silk_memset( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15 ) );
silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State ) );
psEnc->state_Fxx[ 1 ].sCmn.prevLag = 100;
psEnc->state_Fxx[ 1 ].sCmn.sNSQ.lagPrev = 100;
psEnc->state_Fxx[ 1 ].sShape.LastGainIndex = 10;
psEnc->state_Fxx[ 1 ].sCmn.prevSignalType = TYPE_NO_VOICE_ACTIVITY;
psEnc->state_Fxx[ 1 ].sCmn.sNSQ.prev_gain_Q16 = 65536;
psEnc->state_Fxx[ 1 ].sCmn.first_frame_after_reset = 1;
}
silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 1 ] );
} else {
psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] = 0;
}
if( !prefillFlag ) {
silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] );
if( psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) {
silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] );
}
}
} else {
/* Buffering */
silk_memcpy( psEnc->state_Fxx[ 0 ].sCmn.inputBuf, psEnc->sStereo.sMid, 2 * sizeof( opus_int16 ) );
silk_memcpy( psEnc->sStereo.sMid, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.frame_length ], 2 * sizeof( opus_int16 ) );
}
silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 0 ] );
/* Encode */
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
opus_int maxBits, useCBR;
/* Handling rate constraints */
maxBits = encControl->maxBits;
if( tot_blocks == 2 && curr_block == 0 ) {
maxBits = maxBits * 3 / 5;
} else if( tot_blocks == 3 ) {
if( curr_block == 0 ) {
maxBits = maxBits * 2 / 5;
} else if( curr_block == 1 ) {
maxBits = maxBits * 3 / 4;
}
}
useCBR = encControl->useCBR && curr_block == tot_blocks - 1;
if( encControl->nChannelsInternal == 1 ) {
channelRate_bps = TargetRate_bps;
} else {
channelRate_bps = MStargetRates_bps[ n ];
if( n == 0 && MStargetRates_bps[ 1 ] > 0 ) {
useCBR = 0;
/* Give mid up to 1/2 of the max bits for that frame */
maxBits -= encControl->maxBits / ( tot_blocks * 2 );
}
}
if( channelRate_bps > 0 ) {
opus_int condCoding;
silk_control_SNR( &psEnc->state_Fxx[ n ].sCmn, channelRate_bps );
/* Use independent coding if no previous frame available */
if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - n <= 0 ) {
condCoding = CODE_INDEPENDENTLY;
} else if( n > 0 && psEnc->prev_decode_only_middle ) {
/* If we skipped a side frame in this packet, we don't
need LTP scaling; the LTP state is well-defined. */
condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING;
} else {
condCoding = CODE_CONDITIONALLY;
}
if( ( ret = silk_encode_frame_Fxx( &psEnc->state_Fxx[ n ], nBytesOut, psRangeEnc, condCoding, maxBits, useCBR ) ) != 0 ) {
silk_assert( 0 );
}
}
psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
psEnc->state_Fxx[ n ].sCmn.inputBufIx = 0;
psEnc->state_Fxx[ n ].sCmn.nFramesEncoded++;
}
psEnc->prev_decode_only_middle = psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - 1 ];
/* Insert VAD and FEC flags at beginning of bitstream */
if( *nBytesOut > 0 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket) {
flags = 0;
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) {
flags = silk_LSHIFT( flags, 1 );
flags |= psEnc->state_Fxx[ n ].sCmn.VAD_flags[ i ];
}
flags = silk_LSHIFT( flags, 1 );
flags |= psEnc->state_Fxx[ n ].sCmn.LBRR_flag;
}
if( !prefillFlag ) {
ec_enc_patch_initial_bits( psRangeEnc, flags, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal );
}
/* Return zero bytes if all channels DTXed */
if( psEnc->state_Fxx[ 0 ].sCmn.inDTX && ( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inDTX ) ) {
*nBytesOut = 0;
}
psEnc->nBitsExceeded += *nBytesOut * 8;
psEnc->nBitsExceeded -= silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 );
psEnc->nBitsExceeded = silk_LIMIT( psEnc->nBitsExceeded, 0, 10000 );
/* Update flag indicating if bandwidth switching is allowed */
speech_act_thr_for_switch_Q8 = (opus_int) silk_SMLAWB( SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ),
SILK_FIX_CONST( ( 1 - SPEECH_ACTIVITY_DTX_THRES ) / MAX_BANDWIDTH_SWITCH_DELAY_MS, 16 + 8 ), psEnc->timeSinceSwitchAllowed_ms );
if( psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8 < speech_act_thr_for_switch_Q8 ) {
psEnc->allowBandwidthSwitch = 1;
psEnc->timeSinceSwitchAllowed_ms = 0;
} else {
psEnc->allowBandwidthSwitch = 0;
psEnc->timeSinceSwitchAllowed_ms += encControl->payloadSize_ms;
}
}
if( nSamplesIn == 0 ) {
break;
}
} else {
break;
}
curr_block++;
}
psEnc->nPrevChannelsInternal = encControl->nChannelsInternal;
encControl->allowBandwidthSwitch = psEnc->allowBandwidthSwitch;
encControl->inWBmodeWithoutVariableLP = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == 16 && psEnc->state_Fxx[ 0 ].sCmn.sLP.mode == 0;
encControl->internalSampleRate = silk_SMULBB( psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, 1000 );
encControl->stereoWidth_Q14 = encControl->toMono ? 0 : psEnc->sStereo.smth_width_Q14;
if( prefillFlag ) {
encControl->payloadSize_ms = tmp_payloadSize_ms;
encControl->complexity = tmp_complexity;
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
psEnc->state_Fxx[ n ].sCmn.prefillFlag = 0;
}
}
RESTORE_STACK;
return ret;
}
/* Encode side-information parameters to payload */
void silk_encode_indices(
silk_encoder_state *psEncC, /* I/O Encoder state */
ec_enc *psRangeEnc, /* I/O Compressor data structure */
opus_int FrameIndex, /* I Frame number */
opus_int encode_LBRR /* I Flag indicating LBRR data is being encoded */
)
{
opus_int i, k, condCoding, typeOffset;
opus_int encode_absolute_lagIndex, delta_lagIndex;
opus_int16 ec_ix[ MAX_LPC_ORDER ];
opus_uint8 pred_Q8[ MAX_LPC_ORDER ];
const SideInfoIndices *psIndices;
#if SAVE_ALL_INTERNAL_DATA
opus_int nBytes_lagIndex, nBytes_contourIndex, nBytes_LTP;
opus_int nBytes_after, nBytes_before;
#endif
/* Use conditional coding if previous frame available */
if( FrameIndex > 0 && ( encode_LBRR == 0 || psEncC->LBRR_flags[ FrameIndex - 1 ] == 1 ) ) {
condCoding = 1;
} else {
condCoding = 0;
}
if( encode_LBRR ) {
psIndices = &psEncC->indices_LBRR[ FrameIndex ];
} else {
psIndices = &psEncC->indices;
}
/*******************************************/
/* Encode signal type and quantizer offset */
/*******************************************/
typeOffset = 2 * psIndices->signalType + psIndices->quantOffsetType;
SKP_assert( typeOffset >= 0 && typeOffset < 6 );
SKP_assert( encode_LBRR == 0 || typeOffset >= 2 );
if( encode_LBRR || typeOffset >= 2 ) {
ec_enc_icdf( psRangeEnc, typeOffset - 2, silk_type_offset_VAD_iCDF, 8 );
} else {
ec_enc_icdf( psRangeEnc, typeOffset, silk_type_offset_no_VAD_iCDF, 8 );
}
/****************/
/* Encode gains */
/****************/
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_before = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
#endif
/* first subframe */
if( condCoding ) {
/* conditional coding */
SKP_assert( psIndices->GainsIndices[ 0 ] >= 0 && psIndices->GainsIndices[ 0 ] < MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 );
ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ 0 ], silk_delta_gain_iCDF, 8 );
} else {
/* independent coding, in two stages: MSB bits followed by 3 LSBs */
SKP_assert( psIndices->GainsIndices[ 0 ] >= 0 && psIndices->GainsIndices[ 0 ] < N_LEVELS_QGAIN );
ec_enc_icdf( psRangeEnc, SKP_RSHIFT( psIndices->GainsIndices[ 0 ], 3 ), silk_gain_iCDF[ psIndices->signalType ], 8 );
ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ 0 ] & 7, silk_uniform8_iCDF, 8 );
}
/* remaining subframes */
for( i = 1; i < psEncC->nb_subfr; i++ ) {
SKP_assert( psIndices->GainsIndices[ i ] >= 0 && psIndices->GainsIndices[ i ] < MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 );
ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ i ], silk_delta_gain_iCDF, 8 );
}
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_after = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
nBytes_after -= nBytes_before; // bytes just added
DEBUG_STORE_DATA( nBytes_gains.dat, &nBytes_after, sizeof( opus_int ) );
#endif
/****************/
/* Encode NLSFs */
/****************/
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_before = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
#endif
ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ 0 ], &psEncC->psNLSF_CB->CB1_iCDF[ ( psIndices->signalType >> 1 ) * psEncC->psNLSF_CB->nVectors ], 8 );
silk_NLSF_unpack( ec_ix, pred_Q8, psEncC->psNLSF_CB, psIndices->NLSFIndices[ 0 ] );
SKP_assert( psEncC->psNLSF_CB->order == psEncC->predictLPCOrder );
for( i = 0; i < psEncC->psNLSF_CB->order; i++ ) {
if( psIndices->NLSFIndices[ i+1 ] >= NLSF_QUANT_MAX_AMPLITUDE ) {
ec_enc_icdf( psRangeEnc, 2 * NLSF_QUANT_MAX_AMPLITUDE, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 );
ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ i+1 ] - NLSF_QUANT_MAX_AMPLITUDE, silk_NLSF_EXT_iCDF, 8 );
} else if( psIndices->NLSFIndices[ i+1 ] <= -NLSF_QUANT_MAX_AMPLITUDE ) {
ec_enc_icdf( psRangeEnc, 0, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 );
ec_enc_icdf( psRangeEnc, -psIndices->NLSFIndices[ i+1 ] - NLSF_QUANT_MAX_AMPLITUDE, silk_NLSF_EXT_iCDF, 8 );
} else {
ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ i+1 ] + NLSF_QUANT_MAX_AMPLITUDE, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 );
}
}
/* Encode NLSF interpolation factor */
if( psEncC->nb_subfr == MAX_NB_SUBFR ) {
SKP_assert( psEncC->useInterpolatedNLSFs == 1 || psIndices->NLSFInterpCoef_Q2 == ( 1 << 2 ) );
SKP_assert( psIndices->NLSFInterpCoef_Q2 >= 0 && psIndices->NLSFInterpCoef_Q2 < 5 );
ec_enc_icdf( psRangeEnc, psIndices->NLSFInterpCoef_Q2, silk_NLSF_interpolation_factor_iCDF, 8 );
}
#ifdef SAVE_ALL_INTERNAL_DATA
DEBUG_STORE_DATA( lsf_interpol.dat, &psIndices->NLSFInterpCoef_Q2, sizeof(int) );
nBytes_after = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
nBytes_after -= nBytes_before; // bytes just added
DEBUG_STORE_DATA( nBytes_LSF.dat, &nBytes_after, sizeof( opus_int ) );
#endif
if( psIndices->signalType == TYPE_VOICED )
{
/*********************/
/* Encode pitch lags */
/*********************/
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_before = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
#endif
/* lag index */
encode_absolute_lagIndex = 1;
if( condCoding && psEncC->ec_prevSignalType == TYPE_VOICED ) {
/* Delta Encoding */
delta_lagIndex = psIndices->lagIndex - psEncC->ec_prevLagIndex;
if( delta_lagIndex < -8 || delta_lagIndex > 11 ) {
delta_lagIndex = 0;
} else {
delta_lagIndex = delta_lagIndex + 9;
encode_absolute_lagIndex = 0; /* Only use delta */
}
SKP_assert( delta_lagIndex >= 0 && delta_lagIndex < 21 );
ec_enc_icdf( psRangeEnc, delta_lagIndex, silk_pitch_delta_iCDF, 8 );
}
if( encode_absolute_lagIndex ) {
/* Absolute encoding */
opus_int32 pitch_high_bits, pitch_low_bits;
pitch_high_bits = SKP_DIV32_16( psIndices->lagIndex, SKP_RSHIFT( psEncC->fs_kHz, 1 ) );
pitch_low_bits = psIndices->lagIndex - SKP_SMULBB( pitch_high_bits, SKP_RSHIFT( psEncC->fs_kHz, 1 ) );
SKP_assert( pitch_low_bits < psEncC->fs_kHz / 2 );
SKP_assert( pitch_high_bits < 32 );
ec_enc_icdf( psRangeEnc, pitch_high_bits, silk_pitch_lag_iCDF, 8 );
ec_enc_icdf( psRangeEnc, pitch_low_bits, psEncC->pitch_lag_low_bits_iCDF, 8 );
}
psEncC->ec_prevLagIndex = psIndices->lagIndex;
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_after = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
nBytes_lagIndex = nBytes_after - nBytes_before; // bytes just added
#endif
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_before = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
#endif
/* Countour index */
SKP_assert( psIndices->contourIndex >= 0 );
SKP_assert( ( psIndices->contourIndex < 34 && psEncC->fs_kHz > 8 && psEncC->nb_subfr == 4 ) ||
( psIndices->contourIndex < 11 && psEncC->fs_kHz == 8 && psEncC->nb_subfr == 4 ) ||
( psIndices->contourIndex < 12 && psEncC->fs_kHz > 8 && psEncC->nb_subfr == 2 ) ||
( psIndices->contourIndex < 3 && psEncC->fs_kHz == 8 && psEncC->nb_subfr == 2 ) );
ec_enc_icdf( psRangeEnc, psIndices->contourIndex, psEncC->pitch_contour_iCDF, 8 );
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_after = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
nBytes_contourIndex = nBytes_after - nBytes_before; // bytes just added
#endif
/********************/
/* Encode LTP gains */
/********************/
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_before = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
#endif
/* PERIndex value */
SKP_assert( psIndices->PERIndex >= 0 && psIndices->PERIndex < 3 );
ec_enc_icdf( psRangeEnc, psIndices->PERIndex, silk_LTP_per_index_iCDF, 8 );
/* Codebook Indices */
for( k = 0; k < psEncC->nb_subfr; k++ ) {
SKP_assert( psIndices->LTPIndex[ k ] >= 0 && psIndices->LTPIndex[ k ] < ( 8 << psIndices->PERIndex ) );
ec_enc_icdf( psRangeEnc, psIndices->LTPIndex[ k ], silk_LTP_gain_iCDF_ptrs[ psIndices->PERIndex ], 8 );
}
/**********************/
/* Encode LTP scaling */
/**********************/
if( !condCoding ) {
SKP_assert( psIndices->LTP_scaleIndex >= 0 && psIndices->LTP_scaleIndex < 3 );
ec_enc_icdf( psRangeEnc, psIndices->LTP_scaleIndex, silk_LTPscale_iCDF, 8 );
}
SKP_assert( !condCoding || psIndices->LTP_scaleIndex == 0 );
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_after = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
nBytes_LTP = nBytes_after - nBytes_before; // bytes just added
#endif
}
#ifdef SAVE_ALL_INTERNAL_DATA
else {
// Unvoiced speech
nBytes_lagIndex = 0;
nBytes_contourIndex = 0;
nBytes_LTP = 0;
}
DEBUG_STORE_DATA( nBytes_lagIndex.dat, &nBytes_lagIndex, sizeof( opus_int ) );
DEBUG_STORE_DATA( nBytes_contourIndex.dat, &nBytes_contourIndex, sizeof( opus_int ) );
DEBUG_STORE_DATA( nBytes_LTP.dat, &nBytes_LTP, sizeof( opus_int ) );
#endif
psEncC->ec_prevSignalType = psIndices->signalType;
#ifdef SAVE_ALL_INTERNAL_DATA
nBytes_before = SKP_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
#endif
/***************/
/* Encode seed */
/***************/
SKP_assert( psIndices->Seed >= 0 && psIndices->Seed < 4 );
ec_enc_icdf( psRangeEnc, psIndices->Seed, silk_uniform4_iCDF, 8 );
}
