static void search_for_ms(AACEncContext *s, ChannelElement *cpe,
const float lambda)
{
int start = 0, i, w, w2, g;
float M[128], S[128];
float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3;
SingleChannelElement *sce0 = &cpe->ch[0];
SingleChannelElement *sce1 = &cpe->ch[1];
if (!cpe->common_window)
return;
for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
for (g = 0; g < sce0->ics.num_swb; g++) {
if (!cpe->ch[0].zeroes[w*16+g] && !cpe->ch[1].zeroes[w*16+g]) {
float dist1 = 0.0f, dist2 = 0.0f;
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];
float minthr = FFMIN(band0->threshold, band1->threshold);
float maxthr = FFMAX(band0->threshold, band1->threshold);
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
M[i] = (sce0->coeffs[start+w2*128+i]
+ sce1->coeffs[start+w2*128+i]) * 0.5;
S[i] = M[i]
- sce1->coeffs[start+w2*128+i];
}
abs_pow34_v(L34, sce0->coeffs+start+w2*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(R34, sce1->coeffs+start+w2*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, sce0->coeffs + start + w2*128,
L34,
sce0->ics.swb_sizes[g],
sce0->sf_idx[(w+w2)*16+g],
sce0->band_type[(w+w2)*16+g],
lambda / band0->threshold, INFINITY, NULL);
dist1 += quantize_band_cost(s, sce1->coeffs + start + w2*128,
R34,
sce1->ics.swb_sizes[g],
sce1->sf_idx[(w+w2)*16+g],
sce1->band_type[(w+w2)*16+g],
lambda / band1->threshold, INFINITY, NULL);
dist2 += quantize_band_cost(s, M,
M34,
sce0->ics.swb_sizes[g],
sce0->sf_idx[(w+w2)*16+g],
sce0->band_type[(w+w2)*16+g],
lambda / maxthr, INFINITY, NULL);
dist2 += quantize_band_cost(s, S,
S34,
sce1->ics.swb_sizes[g],
sce1->sf_idx[(w+w2)*16+g],
sce1->band_type[(w+w2)*16+g],
lambda / minthr, INFINITY, NULL);
}
cpe->ms_mask[w*16+g] = dist2 < dist1;
}
start += sce0->ics.swb_sizes[g];
}
}
}
struct AACISError ff_aac_is_encoding_err(AACEncContext *s, ChannelElement *cpe,
int start, int w, int g, float ener0,
float ener1, float ener01,
int use_pcoeffs, int phase)
{
int i, w2;
SingleChannelElement *sce0 = &cpe->ch[0];
SingleChannelElement *sce1 = &cpe->ch[1];
float *L = use_pcoeffs ? sce0->pcoeffs : sce0->coeffs;
float *R = use_pcoeffs ? sce1->pcoeffs : sce1->coeffs;
float *L34 = &s->scoefs[256*0], *R34 = &s->scoefs[256*1];
float *IS = &s->scoefs[256*2], *I34 = &s->scoefs[256*3];
float dist1 = 0.0f, dist2 = 0.0f;
struct AACISError is_error = {0};
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];
int is_band_type, is_sf_idx = FFMAX(1, sce0->sf_idx[(w+w2)*16+g]-4);
float e01_34 = phase*pow(ener1/ener0, 3.0/4.0);
float maxval, dist_spec_err = 0.0f;
float minthr = FFMIN(band0->threshold, band1->threshold);
for (i = 0; i < sce0->ics.swb_sizes[g]; i++)
IS[i] = (L[start+(w+w2)*128+i] + phase*R[start+(w+w2)*128+i])*sqrt(ener0/ener01);
abs_pow34_v(L34, &L[start+(w+w2)*128], sce0->ics.swb_sizes[g]);
abs_pow34_v(R34, &R[start+(w+w2)*128], sce0->ics.swb_sizes[g]);
abs_pow34_v(I34, IS, sce0->ics.swb_sizes[g]);
maxval = find_max_val(1, sce0->ics.swb_sizes[g], I34);
is_band_type = find_min_book(maxval, is_sf_idx);
dist1 += quantize_band_cost(s, &L[start + (w+w2)*128], L34,
sce0->ics.swb_sizes[g],
sce0->sf_idx[(w+w2)*16+g],
sce0->band_type[(w+w2)*16+g],
s->lambda / band0->threshold, INFINITY, NULL, 0);
dist1 += quantize_band_cost(s, &R[start + (w+w2)*128], R34,
sce1->ics.swb_sizes[g],
sce1->sf_idx[(w+w2)*16+g],
sce1->band_type[(w+w2)*16+g],
s->lambda / band1->threshold, INFINITY, NULL, 0);
dist2 += quantize_band_cost(s, IS, I34, sce0->ics.swb_sizes[g],
is_sf_idx, is_band_type,
s->lambda / minthr, INFINITY, NULL, 0);
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
dist_spec_err += (L34[i] - I34[i])*(L34[i] - I34[i]);
dist_spec_err += (R34[i] - I34[i]*e01_34)*(R34[i] - I34[i]*e01_34);
}
dist_spec_err *= s->lambda / minthr;
dist2 += dist_spec_err;
}
is_error.pass = dist2 <= dist1;
is_error.phase = phase;
is_error.error = fabsf(dist1 - dist2);
is_error.dist1 = dist1;
is_error.dist2 = dist2;
return is_error;
}
static void search_for_quantizers_faac(AVCodecContext *avctx, AACEncContext *s,
SingleChannelElement *sce,
const float lambda)
{
int start = 0, i, w, w2, g;
float uplim[128], maxq[128];
int minq, maxsf;
float distfact = ((sce->ics.num_windows > 1) ? 85.80 : 147.84) / lambda;
int last = 0, lastband = 0, curband = 0;
float avg_energy = 0.0;
if (sce->ics.num_windows == 1) {
start = 0;
for (i = 0; i < 1024; i++) {
if (i - start >= sce->ics.swb_sizes[curband]) {
start += sce->ics.swb_sizes[curband];
curband++;
}
if (sce->coeffs[i]) {
avg_energy += sce->coeffs[i] * sce->coeffs[i];
last = i;
lastband = curband;
}
}
} else {
for (w = 0; w < 8; w++) {
const float *coeffs = sce->coeffs + w*128;
start = 0;
for (i = 0; i < 128; i++) {
if (i - start >= sce->ics.swb_sizes[curband]) {
start += sce->ics.swb_sizes[curband];
curband++;
}
if (coeffs[i]) {
avg_energy += coeffs[i] * coeffs[i];
last = FFMAX(last, i);
lastband = FFMAX(lastband, curband);
}
}
}
}
last++;
avg_energy /= last;
if (avg_energy == 0.0f) {
for (i = 0; i < FF_ARRAY_ELEMS(sce->sf_idx); i++)
sce->sf_idx[i] = SCALE_ONE_POS;
return;
}
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
float *coefs = sce->coeffs + start;
const int size = sce->ics.swb_sizes[g];
int start2 = start, end2 = start + size, peakpos = start;
float maxval = -1, thr = 0.0f, t;
maxq[w*16+g] = 0.0f;
if (g > lastband) {
maxq[w*16+g] = 0.0f;
start += size;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++)
memset(coefs + w2*128, 0, sizeof(coefs[0])*size);
continue;
}
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
for (i = 0; i < size; i++) {
float t = coefs[w2*128+i]*coefs[w2*128+i];
maxq[w*16+g] = FFMAX(maxq[w*16+g], fabsf(coefs[w2*128 + i]));
thr += t;
if (sce->ics.num_windows == 1 && maxval < t) {
maxval = t;
peakpos = start+i;
}
}
}
if (sce->ics.num_windows == 1) {
start2 = FFMAX(peakpos - 2, start2);
end2 = FFMIN(peakpos + 3, end2);
} else {
start2 -= start;
end2 -= start;
}
start += size;
thr = pow(thr / (avg_energy * (end2 - start2)), 0.3 + 0.1*(lastband - g) / lastband);
t = 1.0 - (1.0 * start2 / last);
uplim[w*16+g] = distfact / (1.4 * thr + t*t*t + 0.075);
}
}
memset(sce->sf_idx, 0, sizeof(sce->sf_idx));
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *coefs = sce->coeffs + start;
const float *scaled = s->scoefs + start;
const int size = sce->ics.swb_sizes[g];
int scf, prev_scf, step;
int min_scf = -1, max_scf = 256;
float curdiff;
if (maxq[w*16+g] < 21.544) {
sce->zeroes[w*16+g] = 1;
start += size;
continue;
}
sce->zeroes[w*16+g] = 0;
scf = prev_scf = av_clip(SCALE_ONE_POS - SCALE_DIV_512 - log2f(1/maxq[w*16+g])*16/3, 60, 218);
step = 16;
for (;;) {
float dist = 0.0f;
int quant_max;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
int b;
dist += quantize_band_cost(s, coefs + w2*128,
scaled + w2*128,
sce->ics.swb_sizes[g],
scf,
ESC_BT,
lambda,
INFINITY,
&b);
dist -= b;
}
dist *= 1.0f / 512.0f / lambda;
quant_max = quant(maxq[w*16+g], ff_aac_pow2sf_tab[POW_SF2_ZERO - scf + SCALE_ONE_POS - SCALE_DIV_512]);
if (quant_max >= 8191) { // too much, return to the previous quantizer
sce->sf_idx[w*16+g] = prev_scf;
break;
}
prev_scf = scf;
curdiff = fabsf(dist - uplim[w*16+g]);
if (curdiff <= 1.0f)
step = 0;
else
step = log2f(curdiff);
if (dist > uplim[w*16+g])
step = -step;
scf += step;
scf = av_clip_uint8(scf);
step = scf - prev_scf;
if (FFABS(step) <= 1 || (step > 0 && scf >= max_scf) || (step < 0 && scf <= min_scf)) {
sce->sf_idx[w*16+g] = av_clip(scf, min_scf, max_scf);
break;
}
if (step > 0)
min_scf = prev_scf;
else
max_scf = prev_scf;
}
start += size;
}
}
minq = sce->sf_idx[0] ? sce->sf_idx[0] : INT_MAX;
for (i = 1; i < 128; i++) {
if (!sce->sf_idx[i])
sce->sf_idx[i] = sce->sf_idx[i-1];
else
minq = FFMIN(minq, sce->sf_idx[i]);
}
if (minq == INT_MAX)
minq = 0;
minq = FFMIN(minq, SCALE_MAX_POS);
maxsf = FFMIN(minq + SCALE_MAX_DIFF, SCALE_MAX_POS);
for (i = 126; i >= 0; i--) {
if (!sce->sf_idx[i])
sce->sf_idx[i] = sce->sf_idx[i+1];
sce->sf_idx[i] = av_clip(sce->sf_idx[i], minq, maxsf);
}
}
/**
* two-loop quantizers search taken from ISO 13818-7 Appendix C
*/
static void search_for_quantizers_twoloop(AVCodecContext *avctx,
AACEncContext *s,
SingleChannelElement *sce,
const float lambda)
{
int start = 0, i, w, w2, g;
int destbits = avctx->bit_rate * 1024.0 / avctx->sample_rate / avctx->channels;
float dists[128], uplims[128];
float maxvals[128];
int fflag, minscaler;
int its = 0;
int allz = 0;
float minthr = INFINITY;
//XXX: some heuristic to determine initial quantizers will reduce search time
memset(dists, 0, sizeof(dists));
//determine zero bands and upper limits
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
for (g = 0; g < sce->ics.num_swb; g++) {
int nz = 0;
float uplim = 0.0f;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
uplim += band->threshold;
if (band->energy <= band->threshold || band->threshold == 0.0f) {
sce->zeroes[(w+w2)*16+g] = 1;
continue;
}
nz = 1;
}
uplims[w*16+g] = uplim *512;
sce->zeroes[w*16+g] = !nz;
if (nz)
minthr = FFMIN(minthr, uplim);
allz |= nz;
}
}
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
for (g = 0; g < sce->ics.num_swb; g++) {
if (sce->zeroes[w*16+g]) {
sce->sf_idx[w*16+g] = SCALE_ONE_POS;
continue;
}
sce->sf_idx[w*16+g] = SCALE_ONE_POS + FFMIN(log2f(uplims[w*16+g]/minthr)*4,59);
}
}
if (!allz)
return;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *scaled = s->scoefs + start;
maxvals[w*16+g] = find_max_val(sce->ics.group_len[w], sce->ics.swb_sizes[g], scaled);
start += sce->ics.swb_sizes[g];
}
}
//perform two-loop search
//outer loop - improve quality
do {
int tbits, qstep;
minscaler = sce->sf_idx[0];
//inner loop - quantize spectrum to fit into given number of bits
qstep = its ? 1 : 32;
do {
int prev = -1;
tbits = 0;
fflag = 0;
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *coefs = sce->coeffs + start;
const float *scaled = s->scoefs + start;
int bits = 0;
int cb;
float dist = 0.0f;
if (sce->zeroes[w*16+g] || sce->sf_idx[w*16+g] >= 218) {
start += sce->ics.swb_sizes[g];
continue;
}
minscaler = FFMIN(minscaler, sce->sf_idx[w*16+g]);
cb = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
int b;
dist += quantize_band_cost(s, coefs + w2*128,
scaled + w2*128,
sce->ics.swb_sizes[g],
sce->sf_idx[w*16+g],
cb,
1.0f,
INFINITY,
&b);
bits += b;
}
dists[w*16+g] = dist - bits;
if (prev != -1) {
bits += ff_aac_scalefactor_bits[sce->sf_idx[w*16+g] - prev + SCALE_DIFF_ZERO];
}
tbits += bits;
start += sce->ics.swb_sizes[g];
prev = sce->sf_idx[w*16+g];
}
}
if (tbits > destbits) {
for (i = 0; i < 128; i++)
if (sce->sf_idx[i] < 218 - qstep)
sce->sf_idx[i] += qstep;
} else {
for (i = 0; i < 128; i++)
if (sce->sf_idx[i] > 60 - qstep)
sce->sf_idx[i] -= qstep;
}
qstep >>= 1;
if (!qstep && tbits > destbits*1.02 && sce->sf_idx[0] < 217)
qstep = 1;
} while (qstep);
fflag = 0;
minscaler = av_clip(minscaler, 60, 255 - SCALE_MAX_DIFF);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
for (g = 0; g < sce->ics.num_swb; g++) {
int prevsc = sce->sf_idx[w*16+g];
if (dists[w*16+g] > uplims[w*16+g] && sce->sf_idx[w*16+g] > 60) {
if (find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]-1))
sce->sf_idx[w*16+g]--;
else //Try to make sure there is some energy in every band
sce->sf_idx[w*16+g]-=2;
}
sce->sf_idx[w*16+g] = av_clip(sce->sf_idx[w*16+g], minscaler, minscaler + SCALE_MAX_DIFF);
sce->sf_idx[w*16+g] = FFMIN(sce->sf_idx[w*16+g], 219);
if (sce->sf_idx[w*16+g] != prevsc)
fflag = 1;
sce->band_type[w*16+g] = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);
}
}
its++;
} while (fflag && its < 10);
}
static void search_for_quantizers_anmr(AVCodecContext *avctx, AACEncContext *s,
SingleChannelElement *sce,
const float lambda)
{
int q, w, w2, g, start = 0;
int i, j;
int idx;
TrellisPath paths[TRELLIS_STAGES][TRELLIS_STATES];
int bandaddr[TRELLIS_STAGES];
int minq;
float mincost;
float q0f = FLT_MAX, q1f = 0.0f, qnrgf = 0.0f;
int q0, q1, qcnt = 0;
for (i = 0; i < 1024; i++) {
float t = fabsf(sce->coeffs[i]);
if (t > 0.0f) {
q0f = FFMIN(q0f, t);
q1f = FFMAX(q1f, t);
qnrgf += t*t;
qcnt++;
}
}
if (!qcnt) {
memset(sce->sf_idx, 0, sizeof(sce->sf_idx));
memset(sce->zeroes, 1, sizeof(sce->zeroes));
return;
}
//minimum scalefactor index is when minimum nonzero coefficient after quantizing is not clipped
q0 = coef2minsf(q0f);
//maximum scalefactor index is when maximum coefficient after quantizing is still not zero
q1 = coef2maxsf(q1f);
//av_log(NULL, AV_LOG_ERROR, "q0 %d, q1 %d\n", q0, q1);
if (q1 - q0 > 60) {
int q0low = q0;
int q1high = q1;
//minimum scalefactor index is when maximum nonzero coefficient after quantizing is not clipped
int qnrg = av_clip_uint8(log2f(sqrtf(qnrgf/qcnt))*4 - 31 + SCALE_ONE_POS - SCALE_DIV_512);
q1 = qnrg + 30;
q0 = qnrg - 30;
//av_log(NULL, AV_LOG_ERROR, "q0 %d, q1 %d\n", q0, q1);
if (q0 < q0low) {
q1 += q0low - q0;
q0 = q0low;
} else if (q1 > q1high) {
q0 -= q1 - q1high;
q1 = q1high;
}
}
//av_log(NULL, AV_LOG_ERROR, "q0 %d, q1 %d\n", q0, q1);
for (i = 0; i < TRELLIS_STATES; i++) {
paths[0][i].cost = 0.0f;
paths[0][i].prev = -1;
}
for (j = 1; j < TRELLIS_STAGES; j++) {
for (i = 0; i < TRELLIS_STATES; i++) {
paths[j][i].cost = INFINITY;
paths[j][i].prev = -2;
}
}
idx = 1;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *coefs = sce->coeffs + start;
float qmin, qmax;
int nz = 0;
bandaddr[idx] = w * 16 + g;
qmin = INT_MAX;
qmax = 0.0f;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
if (band->energy <= band->threshold || band->threshold == 0.0f) {
sce->zeroes[(w+w2)*16+g] = 1;
continue;
}
sce->zeroes[(w+w2)*16+g] = 0;
nz = 1;
for (i = 0; i < sce->ics.swb_sizes[g]; i++) {
float t = fabsf(coefs[w2*128+i]);
if (t > 0.0f)
qmin = FFMIN(qmin, t);
qmax = FFMAX(qmax, t);
}
}
if (nz) {
int minscale, maxscale;
float minrd = INFINITY;
float maxval;
//minimum scalefactor index is when minimum nonzero coefficient after quantizing is not clipped
minscale = coef2minsf(qmin);
//maximum scalefactor index is when maximum coefficient after quantizing is still not zero
maxscale = coef2maxsf(qmax);
minscale = av_clip(minscale - q0, 0, TRELLIS_STATES - 1);
maxscale = av_clip(maxscale - q0, 0, TRELLIS_STATES);
maxval = find_max_val(sce->ics.group_len[w], sce->ics.swb_sizes[g], s->scoefs+start);
for (q = minscale; q < maxscale; q++) {
float dist = 0;
int cb = find_min_book(maxval, sce->sf_idx[w*16+g]);
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
dist += quantize_band_cost(s, coefs + w2*128, s->scoefs + start + w2*128, sce->ics.swb_sizes[g],
q + q0, cb, lambda / band->threshold, INFINITY, NULL);
}
minrd = FFMIN(minrd, dist);
for (i = 0; i < q1 - q0; i++) {
float cost;
cost = paths[idx - 1][i].cost + dist
+ ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO];
if (cost < paths[idx][q].cost) {
paths[idx][q].cost = cost;
paths[idx][q].prev = i;
}
}
}
} else {
for (q = 0; q < q1 - q0; q++) {
paths[idx][q].cost = paths[idx - 1][q].cost + 1;
paths[idx][q].prev = q;
}
}
sce->zeroes[w*16+g] = !nz;
start += sce->ics.swb_sizes[g];
idx++;
}
}
idx--;
mincost = paths[idx][0].cost;
minq = 0;
for (i = 1; i < TRELLIS_STATES; i++) {
if (paths[idx][i].cost < mincost) {
mincost = paths[idx][i].cost;
minq = i;
}
}
while (idx) {
sce->sf_idx[bandaddr[idx]] = minq + q0;
minq = paths[idx][minq].prev;
idx--;
}
//set the same quantizers inside window groups
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
for (g = 0; g < sce->ics.num_swb; g++)
for (w2 = 1; w2 < sce->ics.group_len[w]; w2++)
sce->sf_idx[(w+w2)*16+g] = sce->sf_idx[w*16+g];
}
static void codebook_trellis_rate(AACEncContext *s, SingleChannelElement *sce,
int win, int group_len, const float lambda)
{
BandCodingPath path[120][12];
int w, swb, cb, start, size;
int i, j;
const int max_sfb = sce->ics.max_sfb;
const int run_bits = sce->ics.num_windows == 1 ? 5 : 3;
const int run_esc = (1 << run_bits) - 1;
int idx, ppos, count;
int stackrun[120], stackcb[120], stack_len;
float next_minrd = INFINITY;
int next_mincb = 0;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
start = win*128;
for (cb = 0; cb < 12; cb++) {
path[0][cb].cost = run_bits+4;
path[0][cb].prev_idx = -1;
path[0][cb].run = 0;
}
for (swb = 0; swb < max_sfb; swb++) {
size = sce->ics.swb_sizes[swb];
if (sce->zeroes[win*16 + swb]) {
float cost_stay_here = path[swb][0].cost;
float cost_get_here = next_minrd + run_bits + 4;
if ( run_value_bits[sce->ics.num_windows == 8][path[swb][0].run]
!= run_value_bits[sce->ics.num_windows == 8][path[swb][0].run+1])
cost_stay_here += run_bits;
if (cost_get_here < cost_stay_here) {
path[swb+1][0].prev_idx = next_mincb;
path[swb+1][0].cost = cost_get_here;
path[swb+1][0].run = 1;
} else {
path[swb+1][0].prev_idx = 0;
path[swb+1][0].cost = cost_stay_here;
path[swb+1][0].run = path[swb][0].run + 1;
}
next_minrd = path[swb+1][0].cost;
next_mincb = 0;
for (cb = 1; cb < 12; cb++) {
path[swb+1][cb].cost = 61450;
path[swb+1][cb].prev_idx = -1;
path[swb+1][cb].run = 0;
}
} else {
float minrd = next_minrd;
int mincb = next_mincb;
int startcb = sce->band_type[win*16+swb];
next_minrd = INFINITY;
next_mincb = 0;
for (cb = 0; cb < startcb; cb++) {
path[swb+1][cb].cost = 61450;
path[swb+1][cb].prev_idx = -1;
path[swb+1][cb].run = 0;
}
for (cb = startcb; cb < 12; cb++) {
float cost_stay_here, cost_get_here;
float rd = 0.0f;
for (w = 0; w < group_len; w++) {
rd += quantize_band_cost(s, sce->coeffs + start + w*128,
s->scoefs + start + w*128, size,
sce->sf_idx[(win+w)*16+swb], cb,
0, INFINITY, NULL);
}
cost_stay_here = path[swb][cb].cost + rd;
cost_get_here = minrd + rd + run_bits + 4;
if ( run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run]
!= run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run+1])
cost_stay_here += run_bits;
if (cost_get_here < cost_stay_here) {
path[swb+1][cb].prev_idx = mincb;
path[swb+1][cb].cost = cost_get_here;
path[swb+1][cb].run = 1;
} else {
path[swb+1][cb].prev_idx = cb;
path[swb+1][cb].cost = cost_stay_here;
path[swb+1][cb].run = path[swb][cb].run + 1;
}
if (path[swb+1][cb].cost < next_minrd) {
next_minrd = path[swb+1][cb].cost;
next_mincb = cb;
}
}
}
start += sce->ics.swb_sizes[swb];
}
//convert resulting path from backward-linked list
stack_len = 0;
idx = 0;
for (cb = 1; cb < 12; cb++)
if (path[max_sfb][cb].cost < path[max_sfb][idx].cost)
idx = cb;
ppos = max_sfb;
while (ppos > 0) {
assert(idx >= 0);
cb = idx;
stackrun[stack_len] = path[ppos][cb].run;
stackcb [stack_len] = cb;
idx = path[ppos-path[ppos][cb].run+1][cb].prev_idx;
ppos -= path[ppos][cb].run;
stack_len++;
}
//perform actual band info encoding
start = 0;
for (i = stack_len - 1; i >= 0; i--) {
put_bits(&s->pb, 4, stackcb[i]);
count = stackrun[i];
memset(sce->zeroes + win*16 + start, !stackcb[i], count);
//XXX: memset when band_type is also uint8_t
for (j = 0; j < count; j++) {
sce->band_type[win*16 + start] = stackcb[i];
start++;
}
while (count >= run_esc) {
put_bits(&s->pb, run_bits, run_esc);
count -= run_esc;
}
put_bits(&s->pb, run_bits, count);
}
}
/**
* Encode band info for single window group bands.
*/
static void encode_window_bands_info(AACEncContext *s, SingleChannelElement *sce,
int win, int group_len, const float lambda)
{
BandCodingPath path[120][CB_TOT_ALL];
int w, swb, cb, start, size;
int i, j;
const int max_sfb = sce->ics.max_sfb;
const int run_bits = sce->ics.num_windows == 1 ? 5 : 3;
const int run_esc = (1 << run_bits) - 1;
int idx, ppos, count;
int stackrun[120], stackcb[120], stack_len;
float next_minrd = INFINITY;
int next_mincb = 0;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
start = win*128;
for (cb = 0; cb < CB_TOT_ALL; cb++) {
path[0][cb].cost = 0.0f;
path[0][cb].prev_idx = -1;
path[0][cb].run = 0;
}
for (swb = 0; swb < max_sfb; swb++) {
size = sce->ics.swb_sizes[swb];
if (sce->zeroes[win*16 + swb]) {
for (cb = 0; cb < CB_TOT_ALL; cb++) {
path[swb+1][cb].prev_idx = cb;
path[swb+1][cb].cost = path[swb][cb].cost;
path[swb+1][cb].run = path[swb][cb].run + 1;
}
} else {
float minrd = next_minrd;
int mincb = next_mincb;
next_minrd = INFINITY;
next_mincb = 0;
for (cb = 0; cb < CB_TOT_ALL; cb++) {
float cost_stay_here, cost_get_here;
float rd = 0.0f;
if (cb >= 12 && sce->band_type[win*16+swb] < aac_cb_out_map[cb] ||
cb < aac_cb_in_map[sce->band_type[win*16+swb]] && sce->band_type[win*16+swb] > aac_cb_out_map[cb]) {
path[swb+1][cb].prev_idx = -1;
path[swb+1][cb].cost = INFINITY;
path[swb+1][cb].run = path[swb][cb].run + 1;
continue;
}
for (w = 0; w < group_len; w++) {
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(win+w)*16+swb];
rd += quantize_band_cost(s, &sce->coeffs[start + w*128],
&s->scoefs[start + w*128], size,
sce->sf_idx[(win+w)*16+swb], aac_cb_out_map[cb],
lambda / band->threshold, INFINITY, NULL, NULL, 0);
}
cost_stay_here = path[swb][cb].cost + rd;
cost_get_here = minrd + rd + run_bits + 4;
if ( run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run]
!= run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run+1])
cost_stay_here += run_bits;
if (cost_get_here < cost_stay_here) {
path[swb+1][cb].prev_idx = mincb;
path[swb+1][cb].cost = cost_get_here;
path[swb+1][cb].run = 1;
} else {
path[swb+1][cb].prev_idx = cb;
path[swb+1][cb].cost = cost_stay_here;
path[swb+1][cb].run = path[swb][cb].run + 1;
}
if (path[swb+1][cb].cost < next_minrd) {
next_minrd = path[swb+1][cb].cost;
next_mincb = cb;
}
}
}
start += sce->ics.swb_sizes[swb];
}
//convert resulting path from backward-linked list
stack_len = 0;
idx = 0;
for (cb = 1; cb < CB_TOT_ALL; cb++)
if (path[max_sfb][cb].cost < path[max_sfb][idx].cost)
idx = cb;
ppos = max_sfb;
while (ppos > 0) {
av_assert1(idx >= 0);
cb = idx;
stackrun[stack_len] = path[ppos][cb].run;
stackcb [stack_len] = cb;
idx = path[ppos-path[ppos][cb].run+1][cb].prev_idx;
ppos -= path[ppos][cb].run;
stack_len++;
}
//perform actual band info encoding
start = 0;
for (i = stack_len - 1; i >= 0; i--) {
cb = aac_cb_out_map[stackcb[i]];
put_bits(&s->pb, 4, cb);
count = stackrun[i];
memset(sce->zeroes + win*16 + start, !cb, count);
//XXX: memset when band_type is also uint8_t
for (j = 0; j < count; j++) {
sce->band_type[win*16 + start] = cb;
start++;
}
while (count >= run_esc) {
put_bits(&s->pb, run_bits, run_esc);
count -= run_esc;
}
put_bits(&s->pb, run_bits, count);
}
}
static void search_for_ms(AACEncContext *s, ChannelElement *cpe)
{
int start = 0, i, w, w2, g, sid_sf_boost, prev_mid, prev_side;
uint8_t nextband0[128], nextband1[128];
float M[128], S[128];
float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3;
const float lambda = s->lambda;
const float mslambda = FFMIN(1.0f, lambda / 120.f);
SingleChannelElement *sce0 = &cpe->ch[0];
SingleChannelElement *sce1 = &cpe->ch[1];
if (!cpe->common_window)
return;
/** Scout out next nonzero bands */
ff_init_nextband_map(sce0, nextband0);
ff_init_nextband_map(sce1, nextband1);
prev_mid = sce0->sf_idx[0];
prev_side = sce1->sf_idx[0];
for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
start = 0;
for (g = 0; g < sce0->ics.num_swb; g++) {
float bmax = bval2bmax(g * 17.0f / sce0->ics.num_swb) / 0.0045f;
if (!cpe->is_mask[w*16+g])
cpe->ms_mask[w*16+g] = 0;
if (!sce0->zeroes[w*16+g] && !sce1->zeroes[w*16+g] && !cpe->is_mask[w*16+g]) {
float Mmax = 0.0f, Smax = 0.0f;
/* Must compute mid/side SF and book for the whole window group */
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
M[i] = (sce0->coeffs[start+(w+w2)*128+i]
+ sce1->coeffs[start+(w+w2)*128+i]) * 0.5;
S[i] = M[i]
- sce1->coeffs[start+(w+w2)*128+i];
}
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
for (i = 0; i < sce0->ics.swb_sizes[g]; i++ ) {
Mmax = FFMAX(Mmax, M34[i]);
Smax = FFMAX(Smax, S34[i]);
}
}
for (sid_sf_boost = 0; sid_sf_boost < 4; sid_sf_boost++) {
float dist1 = 0.0f, dist2 = 0.0f;
int B0 = 0, B1 = 0;
int minidx;
int mididx, sididx;
int midcb, sidcb;
minidx = FFMIN(sce0->sf_idx[w*16+g], sce1->sf_idx[w*16+g]);
mididx = av_clip(minidx, 0, SCALE_MAX_POS - SCALE_DIV_512);
sididx = av_clip(minidx - sid_sf_boost * 3, 0, SCALE_MAX_POS - SCALE_DIV_512);
if (sce0->band_type[w*16+g] != NOISE_BT && sce1->band_type[w*16+g] != NOISE_BT
&& ( !ff_sfdelta_can_replace(sce0, nextband0, prev_mid, mididx, w*16+g)
|| !ff_sfdelta_can_replace(sce1, nextband1, prev_side, sididx, w*16+g))) {
/* scalefactor range violation, bad stuff, will decrease quality unacceptably */
continue;
}
midcb = find_min_book(Mmax, mididx);
sidcb = find_min_book(Smax, sididx);
/* No CB can be zero */
midcb = FFMAX(1,midcb);
sidcb = FFMAX(1,sidcb);
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];
float minthr = FFMIN(band0->threshold, band1->threshold);
int b1,b2,b3,b4;
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
M[i] = (sce0->coeffs[start+(w+w2)*128+i]
+ sce1->coeffs[start+(w+w2)*128+i]) * 0.5;
S[i] = M[i]
- sce1->coeffs[start+(w+w2)*128+i];
}
abs_pow34_v(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128],
L34,
sce0->ics.swb_sizes[g],
sce0->sf_idx[w*16+g],
sce0->band_type[w*16+g],
lambda / band0->threshold, INFINITY, &b1, NULL, 0);
dist1 += quantize_band_cost(s, &sce1->coeffs[start + (w+w2)*128],
R34,
sce1->ics.swb_sizes[g],
sce1->sf_idx[w*16+g],
sce1->band_type[w*16+g],
lambda / band1->threshold, INFINITY, &b2, NULL, 0);
dist2 += quantize_band_cost(s, M,
M34,
sce0->ics.swb_sizes[g],
mididx,
midcb,
lambda / minthr, INFINITY, &b3, NULL, 0);
dist2 += quantize_band_cost(s, S,
S34,
sce1->ics.swb_sizes[g],
sididx,
sidcb,
mslambda / (minthr * bmax), INFINITY, &b4, NULL, 0);
B0 += b1+b2;
B1 += b3+b4;
dist1 -= b1+b2;
dist2 -= b3+b4;
}
cpe->ms_mask[w*16+g] = dist2 <= dist1 && B1 < B0;
if (cpe->ms_mask[w*16+g]) {
if (sce0->band_type[w*16+g] != NOISE_BT && sce1->band_type[w*16+g] != NOISE_BT) {
sce0->sf_idx[w*16+g] = mididx;
sce1->sf_idx[w*16+g] = sididx;
sce0->band_type[w*16+g] = midcb;
sce1->band_type[w*16+g] = sidcb;
} else if ((sce0->band_type[w*16+g] != NOISE_BT) ^ (sce1->band_type[w*16+g] != NOISE_BT)) {
/* ms_mask unneeded, and it confuses some decoders */
cpe->ms_mask[w*16+g] = 0;
}
break;
} else if (B1 > B0) {
/* More boost won't fix this */
break;
}
}
}
if (!sce0->zeroes[w*16+g] && sce0->band_type[w*16+g] < RESERVED_BT)
prev_mid = sce0->sf_idx[w*16+g];
if (!sce1->zeroes[w*16+g] && !cpe->is_mask[w*16+g] && sce1->band_type[w*16+g] < RESERVED_BT)
prev_side = sce1->sf_idx[w*16+g];
start += sce0->ics.swb_sizes[g];
}
}
}
static void search_for_pns(AACEncContext *s, AVCodecContext *avctx, SingleChannelElement *sce)
{
FFPsyBand *band;
int w, g, w2, i;
int wlen = 1024 / sce->ics.num_windows;
int bandwidth, cutoff;
float *PNS = &s->scoefs[0*128], *PNS34 = &s->scoefs[1*128];
float *NOR34 = &s->scoefs[3*128];
uint8_t nextband[128];
const float lambda = s->lambda;
const float freq_mult = avctx->sample_rate*0.5f/wlen;
const float thr_mult = NOISE_LAMBDA_REPLACE*(100.0f/lambda);
const float spread_threshold = FFMIN(0.75f, NOISE_SPREAD_THRESHOLD*FFMAX(0.5f, lambda/100.f));
const float dist_bias = av_clipf(4.f * 120 / lambda, 0.25f, 4.0f);
const float pns_transient_energy_r = FFMIN(0.7f, lambda / 140.f);
int refbits = avctx->bit_rate * 1024.0 / avctx->sample_rate
/ ((avctx->flags & CODEC_FLAG_QSCALE) ? 2.0f : avctx->channels)
* (lambda / 120.f);
/** Keep this in sync with twoloop's cutoff selection */
float rate_bandwidth_multiplier = 1.5f;
int prev = -1000, prev_sf = -1;
int frame_bit_rate = (avctx->flags & CODEC_FLAG_QSCALE)
? (refbits * rate_bandwidth_multiplier * avctx->sample_rate / 1024)
: (avctx->bit_rate / avctx->channels);
frame_bit_rate *= 1.15f;
if (avctx->cutoff > 0) {
bandwidth = avctx->cutoff;
} else {
bandwidth = FFMAX(3000, AAC_CUTOFF_FROM_BITRATE(frame_bit_rate, 1, avctx->sample_rate));
}
cutoff = bandwidth * 2 * wlen / avctx->sample_rate;
memcpy(sce->band_alt, sce->band_type, sizeof(sce->band_type));
ff_init_nextband_map(sce, nextband);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
int wstart = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
int noise_sfi;
float dist1 = 0.0f, dist2 = 0.0f, noise_amp;
float pns_energy = 0.0f, pns_tgt_energy, energy_ratio, dist_thresh;
float sfb_energy = 0.0f, threshold = 0.0f, spread = 2.0f;
float min_energy = -1.0f, max_energy = 0.0f;
const int start = wstart+sce->ics.swb_offset[g];
const float freq = (start-wstart)*freq_mult;
const float freq_boost = FFMAX(0.88f*freq/NOISE_LOW_LIMIT, 1.0f);
if (freq < NOISE_LOW_LIMIT || (start-wstart) >= cutoff) {
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
continue;
}
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
sfb_energy += band->energy;
spread = FFMIN(spread, band->spread);
threshold += band->threshold;
if (!w2) {
min_energy = max_energy = band->energy;
} else {
min_energy = FFMIN(min_energy, band->energy);
max_energy = FFMAX(max_energy, band->energy);
}
}
/* Ramps down at ~8000Hz and loosens the dist threshold */
dist_thresh = av_clipf(2.5f*NOISE_LOW_LIMIT/freq, 0.5f, 2.5f) * dist_bias;
/* PNS is acceptable when all of these are true:
* 1. high spread energy (noise-like band)
* 2. near-threshold energy (high PE means the random nature of PNS content will be noticed)
* 3. on short window groups, all windows have similar energy (variations in energy would be destroyed by PNS)
*
* At this stage, point 2 is relaxed for zeroed bands near the noise threshold (hole avoidance is more important)
*/
if ((!sce->zeroes[w*16+g] && !ff_sfdelta_can_remove_band(sce, nextband, prev_sf, w*16+g)) ||
((sce->zeroes[w*16+g] || !sce->band_alt[w*16+g]) && sfb_energy < threshold*sqrtf(1.0f/freq_boost)) || spread < spread_threshold ||
(!sce->zeroes[w*16+g] && sce->band_alt[w*16+g] && sfb_energy > threshold*thr_mult*freq_boost) ||
min_energy < pns_transient_energy_r * max_energy ) {
sce->pns_ener[w*16+g] = sfb_energy;
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
continue;
}
pns_tgt_energy = sfb_energy*FFMIN(1.0f, spread*spread);
noise_sfi = av_clip(roundf(log2f(pns_tgt_energy)*2), -100, 155); /* Quantize */
noise_amp = -ff_aac_pow2sf_tab[noise_sfi + POW_SF2_ZERO]; /* Dequantize */
if (prev != -1000) {
int noise_sfdiff = noise_sfi - prev + SCALE_DIFF_ZERO;
if (noise_sfdiff < 0 || noise_sfdiff > 2*SCALE_MAX_DIFF) {
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
continue;
}
}
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
float band_energy, scale, pns_senergy;
const int start_c = (w+w2)*128+sce->ics.swb_offset[g];
band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
for (i = 0; i < sce->ics.swb_sizes[g]; i+=2) {
double rnd[2];
av_bmg_get(&s->lfg, rnd);
PNS[i+0] = (float)rnd[0];
PNS[i+1] = (float)rnd[1];
}
band_energy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]);
scale = noise_amp/sqrtf(band_energy);
s->fdsp->vector_fmul_scalar(PNS, PNS, scale, sce->ics.swb_sizes[g]);
pns_senergy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]);
pns_energy += pns_senergy;
abs_pow34_v(NOR34, &sce->coeffs[start_c], sce->ics.swb_sizes[g]);
abs_pow34_v(PNS34, PNS, sce->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, &sce->coeffs[start_c],
NOR34,
sce->ics.swb_sizes[g],
sce->sf_idx[(w+w2)*16+g],
sce->band_alt[(w+w2)*16+g],
lambda/band->threshold, INFINITY, NULL, NULL, 0);
/* Estimate rd on average as 5 bits for SF, 4 for the CB, plus spread energy * lambda/thr */
dist2 += band->energy/(band->spread*band->spread)*lambda*dist_thresh/band->threshold;
}
if (g && sce->band_type[w*16+g-1] == NOISE_BT) {
dist2 += 5;
} else {
dist2 += 9;
}
energy_ratio = pns_tgt_energy/pns_energy; /* Compensates for quantization error */
sce->pns_ener[w*16+g] = energy_ratio*pns_tgt_energy;
if (sce->zeroes[w*16+g] || !sce->band_alt[w*16+g] || (energy_ratio > 0.85f && energy_ratio < 1.25f && dist2 < dist1)) {
sce->band_type[w*16+g] = NOISE_BT;
sce->zeroes[w*16+g] = 0;
prev = noise_sfi;
} else {
if (!sce->zeroes[w*16+g])
prev_sf = sce->sf_idx[w*16+g];
}
}
}
}
static void search_for_quantizers_anmr(AVCodecContext *avctx, AACEncContext *s,
SingleChannelElement *sce,
const float lambda)
{
int q, w, w2, g, start = 0;
int i, j;
int idx;
TrellisPath paths[TRELLIS_STAGES][TRELLIS_STATES];
int bandaddr[TRELLIS_STAGES];
int minq;
float mincost;
for (i = 0; i < TRELLIS_STATES; i++) {
paths[0][i].cost = 0.0f;
paths[0][i].prev = -1;
paths[0][i].min_val = i;
paths[0][i].max_val = i;
}
for (j = 1; j < TRELLIS_STAGES; j++) {
for (i = 0; i < TRELLIS_STATES; i++) {
paths[j][i].cost = INFINITY;
paths[j][i].prev = -2;
paths[j][i].min_val = INT_MAX;
paths[j][i].max_val = 0;
}
}
idx = 1;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *coefs = sce->coeffs + start;
float qmin, qmax;
int nz = 0;
bandaddr[idx] = w * 16 + g;
qmin = INT_MAX;
qmax = 0.0f;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g];
if (band->energy <= band->threshold || band->threshold == 0.0f) {
sce->zeroes[(w+w2)*16+g] = 1;
continue;
}
sce->zeroes[(w+w2)*16+g] = 0;
nz = 1;
for (i = 0; i < sce->ics.swb_sizes[g]; i++) {
float t = fabsf(coefs[w2*128+i]);
if (t > 0.0f)
qmin = FFMIN(qmin, t);
qmax = FFMAX(qmax, t);
}
}
if (nz) {
int minscale, maxscale;
float minrd = INFINITY;
//minimum scalefactor index is when minimum nonzero coefficient after quantizing is not clipped
minscale = av_clip_uint8(log2(qmin)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512);
//maximum scalefactor index is when maximum coefficient after quantizing is still not zero
maxscale = av_clip_uint8(log2(qmax)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512);
for (q = minscale; q < maxscale; q++) {
float dists[12], dist;
memset(dists, 0, sizeof(dists));
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g];
int cb;
for (cb = 0; cb <= ESC_BT; cb++)
dists[cb] += quantize_band_cost(s, coefs + w2*128, s->scoefs + start + w2*128, sce->ics.swb_sizes[g],
q, cb, lambda / band->threshold, INFINITY, NULL);
}
dist = dists[0];
for (i = 1; i <= ESC_BT; i++)
dist = FFMIN(dist, dists[i]);
minrd = FFMIN(minrd, dist);
for (i = FFMAX(q - SCALE_MAX_DIFF, 0); i < FFMIN(q + SCALE_MAX_DIFF, TRELLIS_STATES); i++) {
float cost;
int minv, maxv;
if (isinf(paths[idx - 1][i].cost))
continue;
cost = paths[idx - 1][i].cost + dist
+ ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO];
minv = FFMIN(paths[idx - 1][i].min_val, q);
maxv = FFMAX(paths[idx - 1][i].max_val, q);
if (cost < paths[idx][q].cost && maxv-minv < SCALE_MAX_DIFF) {
paths[idx][q].cost = cost;
paths[idx][q].prev = i;
paths[idx][q].min_val = minv;
paths[idx][q].max_val = maxv;
}
}
}
} else {
for (q = 0; q < TRELLIS_STATES; q++) {
if (!isinf(paths[idx - 1][q].cost)) {
paths[idx][q].cost = paths[idx - 1][q].cost + 1;
paths[idx][q].prev = q;
paths[idx][q].min_val = FFMIN(paths[idx - 1][q].min_val, q);
paths[idx][q].max_val = FFMAX(paths[idx - 1][q].max_val, q);
continue;
}
for (i = FFMAX(q - SCALE_MAX_DIFF, 0); i < FFMIN(q + SCALE_MAX_DIFF, TRELLIS_STATES); i++) {
float cost;
int minv, maxv;
if (isinf(paths[idx - 1][i].cost))
continue;
cost = paths[idx - 1][i].cost + ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO];
minv = FFMIN(paths[idx - 1][i].min_val, q);
maxv = FFMAX(paths[idx - 1][i].max_val, q);
if (cost < paths[idx][q].cost && maxv-minv < SCALE_MAX_DIFF) {
paths[idx][q].cost = cost;
paths[idx][q].prev = i;
paths[idx][q].min_val = minv;
paths[idx][q].max_val = maxv;
}
}
}
}
sce->zeroes[w*16+g] = !nz;
start += sce->ics.swb_sizes[g];
idx++;
}
}
idx--;
mincost = paths[idx][0].cost;
minq = 0;
for (i = 1; i < TRELLIS_STATES; i++) {
if (paths[idx][i].cost < mincost) {
mincost = paths[idx][i].cost;
minq = i;
}
}
while (idx) {
sce->sf_idx[bandaddr[idx]] = minq;
minq = paths[idx][minq].prev;
idx--;
}
//set the same quantizers inside window groups
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
for (g = 0; g < sce->ics.num_swb; g++)
for (w2 = 1; w2 < sce->ics.group_len[w]; w2++)
sce->sf_idx[(w+w2)*16+g] = sce->sf_idx[w*16+g];
}
/**
* Calculate rate distortion cost for quantizing with given codebook
*
* @return quantization distortion
*/
static float quantize_band_cost(struct AACEncContext *s, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
const float IQ = ff_aac_pow2sf_tab[200 + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
const float Q = ff_aac_pow2sf_tab[200 - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float CLIPPED_ESCAPE = 165140.0f*IQ;
int i, j, k;
float cost = 0;
const int dim = cb < FIRST_PAIR_BT ? 4 : 2;
int resbits = 0;
#ifndef USE_REALLY_FULL_SEARCH
const float Q34 = sqrtf(Q * sqrtf(Q));
const int range = aac_cb_range[cb];
const int maxval = aac_cb_maxval[cb];
int offs[4];
#endif /* USE_REALLY_FULL_SEARCH */
if (!cb) {
for (i = 0; i < size; i++)
cost += in[i]*in[i];
if (bits)
*bits = 0;
return cost * lambda;
}
#ifndef USE_REALLY_FULL_SEARCH
offs[0] = 1;
for (i = 1; i < dim; i++)
offs[i] = offs[i-1]*range;
quantize_bands(s->qcoefs, in, scaled, size, Q34, !IS_CODEBOOK_UNSIGNED(cb), maxval);
#endif /* USE_REALLY_FULL_SEARCH */
for (i = 0; i < size; i += dim) {
float mincost;
int minidx = 0;
int minbits = 0;
const float *vec;
#ifndef USE_REALLY_FULL_SEARCH
int (*quants)[2] = &s->qcoefs[i];
mincost = 0.0f;
for (j = 0; j < dim; j++)
mincost += in[i+j]*in[i+j];
minidx = IS_CODEBOOK_UNSIGNED(cb) ? 0 : 40;
minbits = ff_aac_spectral_bits[cb-1][minidx];
mincost = mincost * lambda + minbits;
for (j = 0; j < (1<<dim); j++) {
float rd = 0.0f;
int curbits;
int curidx = IS_CODEBOOK_UNSIGNED(cb) ? 0 : 40;
int same = 0;
for (k = 0; k < dim; k++) {
if ((j & (1 << k)) && quants[k][0] == quants[k][1]) {
same = 1;
break;
}
}
if (same)
continue;
for (k = 0; k < dim; k++)
curidx += quants[k][!!(j & (1 << k))] * offs[dim - 1 - k];
curbits = ff_aac_spectral_bits[cb-1][curidx];
vec = &ff_aac_codebook_vectors[cb-1][curidx*dim];
#else
mincost = INFINITY;
vec = ff_aac_codebook_vectors[cb-1];
for (j = 0; j < ff_aac_spectral_sizes[cb-1]; j++, vec += dim) {
float rd = 0.0f;
int curbits = ff_aac_spectral_bits[cb-1][j];
#endif /* USE_REALLY_FULL_SEARCH */
if (IS_CODEBOOK_UNSIGNED(cb)) {
for (k = 0; k < dim; k++) {
float t = fabsf(in[i+k]);
float di;
if (vec[k] == 64.0f) { //FIXME: slow
//do not code with escape sequence small values
if (t < 39.0f*IQ) {
rd = INFINITY;
break;
}
if (t >= CLIPPED_ESCAPE) {
di = t - CLIPPED_ESCAPE;
curbits += 21;
} else {
int c = av_clip(quant(t, Q), 0, 8191);
di = t - c*cbrtf(c)*IQ;
curbits += av_log2(c)*2 - 4 + 1;
}
} else {
di = t - vec[k]*IQ;
}
if (vec[k] != 0.0f)
curbits++;
rd += di*di;
}
} else {
for (k = 0; k < dim; k++) {
float di = in[i+k] - vec[k]*IQ;
rd += di*di;
}
}
rd = rd * lambda + curbits;
if (rd < mincost) {
mincost = rd;
minidx = j;
minbits = curbits;
}
}
cost += mincost;
resbits += minbits;
if (cost >= uplim)
return uplim;
}
if (bits)
*bits = resbits;
return cost;
}
static void quantize_and_encode_band(struct AACEncContext *s, PutBitContext *pb,
const float *in, int size, int scale_idx,
int cb, const float lambda)
{
const float IQ = ff_aac_pow2sf_tab[200 + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
const float Q = ff_aac_pow2sf_tab[200 - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float CLIPPED_ESCAPE = 165140.0f*IQ;
const int dim = (cb < FIRST_PAIR_BT) ? 4 : 2;
int i, j, k;
#ifndef USE_REALLY_FULL_SEARCH
const float Q34 = sqrtf(Q * sqrtf(Q));
const int range = aac_cb_range[cb];
const int maxval = aac_cb_maxval[cb];
int offs[4];
float *scaled = s->scoefs;
#endif /* USE_REALLY_FULL_SEARCH */
//START_TIMER
if (!cb)
return;
#ifndef USE_REALLY_FULL_SEARCH
offs[0] = 1;
for (i = 1; i < dim; i++)
offs[i] = offs[i-1]*range;
abs_pow34_v(scaled, in, size);
quantize_bands(s->qcoefs, in, scaled, size, Q34, !IS_CODEBOOK_UNSIGNED(cb), maxval);
#endif /* USE_REALLY_FULL_SEARCH */
for (i = 0; i < size; i += dim) {
float mincost;
int minidx = 0;
int minbits = 0;
const float *vec;
#ifndef USE_REALLY_FULL_SEARCH
int (*quants)[2] = &s->qcoefs[i];
mincost = 0.0f;
for (j = 0; j < dim; j++)
mincost += in[i+j]*in[i+j];
minidx = IS_CODEBOOK_UNSIGNED(cb) ? 0 : 40;
minbits = ff_aac_spectral_bits[cb-1][minidx];
mincost = mincost * lambda + minbits;
for (j = 0; j < (1<<dim); j++) {
float rd = 0.0f;
int curbits;
int curidx = IS_CODEBOOK_UNSIGNED(cb) ? 0 : 40;
int same = 0;
for (k = 0; k < dim; k++) {
if ((j & (1 << k)) && quants[k][0] == quants[k][1]) {
same = 1;
break;
}
}
if (same)
continue;
for (k = 0; k < dim; k++)
curidx += quants[k][!!(j & (1 << k))] * offs[dim - 1 - k];
curbits = ff_aac_spectral_bits[cb-1][curidx];
vec = &ff_aac_codebook_vectors[cb-1][curidx*dim];
#else
vec = ff_aac_codebook_vectors[cb-1];
mincost = INFINITY;
for (j = 0; j < ff_aac_spectral_sizes[cb-1]; j++, vec += dim) {
float rd = 0.0f;
int curbits = ff_aac_spectral_bits[cb-1][j];
int curidx = j;
#endif /* USE_REALLY_FULL_SEARCH */
if (IS_CODEBOOK_UNSIGNED(cb)) {
for (k = 0; k < dim; k++) {
float t = fabsf(in[i+k]);
float di;
if (vec[k] == 64.0f) { //FIXME: slow
//do not code with escape sequence small values
if (t < 39.0f*IQ) {
rd = INFINITY;
break;
}
if (t >= CLIPPED_ESCAPE) {
di = t - CLIPPED_ESCAPE;
curbits += 21;
} else {
int c = av_clip(quant(t, Q), 0, 8191);
di = t - c*cbrtf(c)*IQ;
curbits += av_log2(c)*2 - 4 + 1;
}
} else {
di = t - vec[k]*IQ;
}
if (vec[k] != 0.0f)
curbits++;
rd += di*di;
}
} else {
for (k = 0; k < dim; k++) {
float di = in[i+k] - vec[k]*IQ;
rd += di*di;
}
}
rd = rd * lambda + curbits;
if (rd < mincost) {
mincost = rd;
minidx = curidx;
minbits = curbits;
}
}
put_bits(pb, ff_aac_spectral_bits[cb-1][minidx], ff_aac_spectral_codes[cb-1][minidx]);
if (IS_CODEBOOK_UNSIGNED(cb))
for (j = 0; j < dim; j++)
if (ff_aac_codebook_vectors[cb-1][minidx*dim+j] != 0.0f)
put_bits(pb, 1, in[i+j] < 0.0f);
if (cb == ESC_BT) {
for (j = 0; j < 2; j++) {
if (ff_aac_codebook_vectors[cb-1][minidx*2+j] == 64.0f) {
int coef = av_clip(quant(fabsf(in[i+j]), Q), 0, 8191);
int len = av_log2(coef);
put_bits(pb, len - 4 + 1, (1 << (len - 4 + 1)) - 2);
put_bits(pb, len, coef & ((1 << len) - 1));
}
}
}
}
//STOP_TIMER("quantize_and_encode")
}
/**
* structure used in optimal codebook search
*/
typedef struct BandCodingPath {
int prev_idx; ///< pointer to the previous path point
float cost; ///< path cost
int run;
} BandCodingPath;
/**
* Encode band info for single window group bands.
*/
static void encode_window_bands_info(AACEncContext *s, SingleChannelElement *sce,
int win, int group_len, const float lambda)
{
BandCodingPath path[120][12];
int w, swb, cb, start, start2, size;
int i, j;
const int max_sfb = sce->ics.max_sfb;
const int run_bits = sce->ics.num_windows == 1 ? 5 : 3;
const int run_esc = (1 << run_bits) - 1;
int idx, ppos, count;
int stackrun[120], stackcb[120], stack_len;
float next_minrd = INFINITY;
int next_mincb = 0;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
start = win*128;
for (cb = 0; cb < 12; cb++) {
path[0][cb].cost = 0.0f;
path[0][cb].prev_idx = -1;
path[0][cb].run = 0;
}
for (swb = 0; swb < max_sfb; swb++) {
start2 = start;
size = sce->ics.swb_sizes[swb];
if (sce->zeroes[win*16 + swb]) {
for (cb = 0; cb < 12; cb++) {
path[swb+1][cb].prev_idx = cb;
path[swb+1][cb].cost = path[swb][cb].cost;
path[swb+1][cb].run = path[swb][cb].run + 1;
}
} else {
float minrd = next_minrd;
int mincb = next_mincb;
next_minrd = INFINITY;
next_mincb = 0;
for (cb = 0; cb < 12; cb++) {
float cost_stay_here, cost_get_here;
float rd = 0.0f;
for (w = 0; w < group_len; w++) {
FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(win+w)*16+swb];
rd += quantize_band_cost(s, sce->coeffs + start + w*128,
s->scoefs + start + w*128, size,
sce->sf_idx[(win+w)*16+swb], cb,
lambda / band->threshold, INFINITY, NULL);
}
cost_stay_here = path[swb][cb].cost + rd;
cost_get_here = minrd + rd + run_bits + 4;
if ( run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run]
!= run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run+1])
cost_stay_here += run_bits;
if (cost_get_here < cost_stay_here) {
path[swb+1][cb].prev_idx = mincb;
path[swb+1][cb].cost = cost_get_here;
path[swb+1][cb].run = 1;
} else {
path[swb+1][cb].prev_idx = cb;
path[swb+1][cb].cost = cost_stay_here;
path[swb+1][cb].run = path[swb][cb].run + 1;
}
if (path[swb+1][cb].cost < next_minrd) {
next_minrd = path[swb+1][cb].cost;
next_mincb = cb;
}
}
}
start += sce->ics.swb_sizes[swb];
}
//convert resulting path from backward-linked list
stack_len = 0;
idx = 0;
for (cb = 1; cb < 12; cb++)
if (path[max_sfb][cb].cost < path[max_sfb][idx].cost)
idx = cb;
ppos = max_sfb;
while (ppos > 0) {
cb = idx;
stackrun[stack_len] = path[ppos][cb].run;
stackcb [stack_len] = cb;
idx = path[ppos-path[ppos][cb].run+1][cb].prev_idx;
ppos -= path[ppos][cb].run;
stack_len++;
}
//perform actual band info encoding
start = 0;
for (i = stack_len - 1; i >= 0; i--) {
put_bits(&s->pb, 4, stackcb[i]);
count = stackrun[i];
memset(sce->zeroes + win*16 + start, !stackcb[i], count);
//XXX: memset when band_type is also uint8_t
for (j = 0; j < count; j++) {
sce->band_type[win*16 + start] = stackcb[i];
start++;
}
while (count >= run_esc) {
put_bits(&s->pb, run_bits, run_esc);
count -= run_esc;
}
put_bits(&s->pb, run_bits, count);
}
}
typedef struct TrellisPath {
float cost;
int prev;
int min_val;
int max_val;
} TrellisPath;
#define TRELLIS_STAGES 121
#define TRELLIS_STATES 256
static void search_for_quantizers_anmr(AVCodecContext *avctx, AACEncContext *s,
SingleChannelElement *sce,
const float lambda)
{
int q, w, w2, g, start = 0;
int i, j;
int idx;
TrellisPath paths[TRELLIS_STAGES][TRELLIS_STATES];
int bandaddr[TRELLIS_STAGES];
int minq;
float mincost;
for (i = 0; i < TRELLIS_STATES; i++) {
paths[0][i].cost = 0.0f;
paths[0][i].prev = -1;
paths[0][i].min_val = i;
paths[0][i].max_val = i;
}
for (j = 1; j < TRELLIS_STAGES; j++) {
for (i = 0; i < TRELLIS_STATES; i++) {
paths[j][i].cost = INFINITY;
paths[j][i].prev = -2;
paths[j][i].min_val = INT_MAX;
paths[j][i].max_val = 0;
}
}
idx = 1;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *coefs = sce->coeffs + start;
float qmin, qmax;
int nz = 0;
bandaddr[idx] = w * 16 + g;
qmin = INT_MAX;
qmax = 0.0f;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g];
if (band->energy <= band->threshold || band->threshold == 0.0f) {
sce->zeroes[(w+w2)*16+g] = 1;
continue;
}
sce->zeroes[(w+w2)*16+g] = 0;
nz = 1;
for (i = 0; i < sce->ics.swb_sizes[g]; i++) {
float t = fabsf(coefs[w2*128+i]);
if (t > 0.0f)
qmin = FFMIN(qmin, t);
qmax = FFMAX(qmax, t);
}
}
if (nz) {
int minscale, maxscale;
float minrd = INFINITY;
//minimum scalefactor index is when minimum nonzero coefficient after quantizing is not clipped
minscale = av_clip_uint8(log2(qmin)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512);
//maximum scalefactor index is when maximum coefficient after quantizing is still not zero
maxscale = av_clip_uint8(log2(qmax)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512);
for (q = minscale; q < maxscale; q++) {
float dists[12], dist;
memset(dists, 0, sizeof(dists));
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g];
int cb;
for (cb = 0; cb <= ESC_BT; cb++)
dists[cb] += quantize_band_cost(s, coefs + w2*128, s->scoefs + start + w2*128, sce->ics.swb_sizes[g],
q, cb, lambda / band->threshold, INFINITY, NULL);
}
dist = dists[0];
for (i = 1; i <= ESC_BT; i++)
dist = FFMIN(dist, dists[i]);
minrd = FFMIN(minrd, dist);
for (i = FFMAX(q - SCALE_MAX_DIFF, 0); i < FFMIN(q + SCALE_MAX_DIFF, TRELLIS_STATES); i++) {
float cost;
int minv, maxv;
if (isinf(paths[idx - 1][i].cost))
continue;
cost = paths[idx - 1][i].cost + dist
+ ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO];
minv = FFMIN(paths[idx - 1][i].min_val, q);
maxv = FFMAX(paths[idx - 1][i].max_val, q);
if (cost < paths[idx][q].cost && maxv-minv < SCALE_MAX_DIFF) {
paths[idx][q].cost = cost;
paths[idx][q].prev = i;
paths[idx][q].min_val = minv;
paths[idx][q].max_val = maxv;
}
}
}
} else {
for (q = 0; q < TRELLIS_STATES; q++) {
if (!isinf(paths[idx - 1][q].cost)) {
paths[idx][q].cost = paths[idx - 1][q].cost + 1;
paths[idx][q].prev = q;
paths[idx][q].min_val = FFMIN(paths[idx - 1][q].min_val, q);
paths[idx][q].max_val = FFMAX(paths[idx - 1][q].max_val, q);
continue;
}
for (i = FFMAX(q - SCALE_MAX_DIFF, 0); i < FFMIN(q + SCALE_MAX_DIFF, TRELLIS_STATES); i++) {
float cost;
int minv, maxv;
if (isinf(paths[idx - 1][i].cost))
continue;
cost = paths[idx - 1][i].cost + ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO];
minv = FFMIN(paths[idx - 1][i].min_val, q);
maxv = FFMAX(paths[idx - 1][i].max_val, q);
if (cost < paths[idx][q].cost && maxv-minv < SCALE_MAX_DIFF) {
paths[idx][q].cost = cost;
paths[idx][q].prev = i;
paths[idx][q].min_val = minv;
paths[idx][q].max_val = maxv;
}
}
}
}
sce->zeroes[w*16+g] = !nz;
start += sce->ics.swb_sizes[g];
idx++;
}
}
idx--;
mincost = paths[idx][0].cost;
minq = 0;
for (i = 1; i < TRELLIS_STATES; i++) {
if (paths[idx][i].cost < mincost) {
mincost = paths[idx][i].cost;
minq = i;
}
}
while (idx) {
sce->sf_idx[bandaddr[idx]] = minq;
minq = paths[idx][minq].prev;
idx--;
}
//set the same quantizers inside window groups
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
for (g = 0; g < sce->ics.num_swb; g++)
for (w2 = 1; w2 < sce->ics.group_len[w]; w2++)
sce->sf_idx[(w+w2)*16+g] = sce->sf_idx[w*16+g];
}
static void search_for_ms(AACEncContext *s, ChannelElement *cpe)
{
int start = 0, i, w, w2, g, sid_sf_boost;
float M[128], S[128];
float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3;
const float lambda = s->lambda;
const float mslambda = FFMIN(1.0f, lambda / 120.f);
SingleChannelElement *sce0 = &cpe->ch[0];
SingleChannelElement *sce1 = &cpe->ch[1];
if (!cpe->common_window)
return;
for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
int min_sf_idx_mid = SCALE_MAX_POS;
int min_sf_idx_side = SCALE_MAX_POS;
for (g = 0; g < sce0->ics.num_swb; g++) {
if (!sce0->zeroes[w*16+g] && sce0->band_type[w*16+g] < RESERVED_BT)
min_sf_idx_mid = FFMIN(min_sf_idx_mid, sce0->sf_idx[w*16+g]);
if (!sce1->zeroes[w*16+g] && sce1->band_type[w*16+g] < RESERVED_BT)
min_sf_idx_side = FFMIN(min_sf_idx_side, sce1->sf_idx[w*16+g]);
}
start = 0;
for (g = 0; g < sce0->ics.num_swb; g++) {
float bmax = bval2bmax(g * 17.0f / sce0->ics.num_swb) / 0.0045f;
cpe->ms_mask[w*16+g] = 0;
if (!cpe->ch[0].zeroes[w*16+g] && !cpe->ch[1].zeroes[w*16+g]) {
float Mmax = 0.0f, Smax = 0.0f;
/* Must compute mid/side SF and book for the whole window group */
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
M[i] = (sce0->coeffs[start+(w+w2)*128+i]
+ sce1->coeffs[start+(w+w2)*128+i]) * 0.5;
S[i] = M[i]
- sce1->coeffs[start+(w+w2)*128+i];
}
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
for (i = 0; i < sce0->ics.swb_sizes[g]; i++ ) {
Mmax = FFMAX(Mmax, M34[i]);
Smax = FFMAX(Smax, S34[i]);
}
}
for (sid_sf_boost = 0; sid_sf_boost < 4; sid_sf_boost++) {
float dist1 = 0.0f, dist2 = 0.0f;
int B0 = 0, B1 = 0;
int minidx;
int mididx, sididx;
int midcb, sidcb;
minidx = FFMIN(sce0->sf_idx[w*16+g], sce1->sf_idx[w*16+g]);
mididx = av_clip(minidx, min_sf_idx_mid, min_sf_idx_mid + SCALE_MAX_DIFF);
sididx = av_clip(minidx - sid_sf_boost * 3, min_sf_idx_side, min_sf_idx_side + SCALE_MAX_DIFF);
midcb = find_min_book(Mmax, mididx);
sidcb = find_min_book(Smax, sididx);
if ((mididx > minidx) || (sididx > minidx)) {
/* scalefactor range violation, bad stuff, will decrease quality unacceptably */
continue;
}
/* No CB can be zero */
midcb = FFMAX(1,midcb);
sidcb = FFMAX(1,sidcb);
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];
float minthr = FFMIN(band0->threshold, band1->threshold);
int b1,b2,b3,b4;
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
M[i] = (sce0->coeffs[start+(w+w2)*128+i]
+ sce1->coeffs[start+(w+w2)*128+i]) * 0.5;
S[i] = M[i]
- sce1->coeffs[start+(w+w2)*128+i];
}
abs_pow34_v(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128],
L34,
sce0->ics.swb_sizes[g],
sce0->sf_idx[(w+w2)*16+g],
sce0->band_type[(w+w2)*16+g],
lambda / band0->threshold, INFINITY, &b1, NULL, 0);
dist1 += quantize_band_cost(s, &sce1->coeffs[start + (w+w2)*128],
R34,
sce1->ics.swb_sizes[g],
sce1->sf_idx[(w+w2)*16+g],
sce1->band_type[(w+w2)*16+g],
lambda / band1->threshold, INFINITY, &b2, NULL, 0);
dist2 += quantize_band_cost(s, M,
M34,
sce0->ics.swb_sizes[g],
sce0->sf_idx[(w+w2)*16+g],
sce0->band_type[(w+w2)*16+g],
lambda / minthr, INFINITY, &b3, NULL, 0);
dist2 += quantize_band_cost(s, S,
S34,
sce1->ics.swb_sizes[g],
sce1->sf_idx[(w+w2)*16+g],
sce1->band_type[(w+w2)*16+g],
mslambda / (minthr * bmax), INFINITY, &b4, NULL, 0);
B0 += b1+b2;
B1 += b3+b4;
dist1 -= B0;
dist2 -= B1;
}
cpe->ms_mask[w*16+g] = dist2 <= dist1 && B1 < B0;
if (cpe->ms_mask[w*16+g]) {
/* Setting the M/S mask is useful with I/S or PNS, but only the flag */
if (!cpe->is_mask[w*16+g] && sce0->band_type[w*16+g] != NOISE_BT && sce1->band_type[w*16+g] != NOISE_BT) {
sce0->sf_idx[w*16+g] = mididx;
sce1->sf_idx[w*16+g] = sididx;
sce0->band_type[w*16+g] = midcb;
sce1->band_type[w*16+g] = sidcb;
}
break;
} else if (B1 > B0) {
/* More boost won't fix this */
break;
}
}
}
start += sce0->ics.swb_sizes[g];
}
}
}
/**
* two-loop quantizers search taken from ISO 13818-7 Appendix C
*/
static void search_for_quantizers_twoloop(AVCodecContext *avctx,
AACEncContext *s,
SingleChannelElement *sce,
const float lambda)
{
int start = 0, i, w, w2, g;
int destbits = avctx->bit_rate * 1024.0 / avctx->sample_rate / avctx->channels * (lambda / 120.f);
const float freq_mult = avctx->sample_rate/(1024.0f/sce->ics.num_windows)/2.0f;
float dists[128] = { 0 }, uplims[128] = { 0 };
float maxvals[128];
int noise_sf[128] = { 0 };
int fflag, minscaler, minscaler_n;
int its = 0;
int allz = 0;
float minthr = INFINITY;
// for values above this the decoder might end up in an endless loop
// due to always having more bits than what can be encoded.
destbits = FFMIN(destbits, 5800);
//XXX: some heuristic to determine initial quantizers will reduce search time
//determine zero bands and upper limits
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = 0;
for (g = 0; g < sce->ics.num_swb; g++) {
int nz = 0;
float uplim = 0.0f, energy = 0.0f;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
uplim += band->threshold;
energy += band->energy;
if (band->energy <= band->threshold || band->threshold == 0.0f) {
sce->zeroes[(w+w2)*16+g] = 1;
continue;
}
nz = 1;
}
uplims[w*16+g] = uplim *512;
if (s->options.pns && start*freq_mult > NOISE_LOW_LIMIT && energy < uplim * 1.2f) {
noise_sf[w*16+g] = av_clip(4+FFMIN(log2f(energy)*2,255), -100, 155);
sce->band_type[w*16+g] = NOISE_BT;
nz= 1;
} else { /** Band type will be determined by the twoloop algorithm */
sce->band_type[w*16+g] = 0;
}
sce->zeroes[w*16+g] = !nz;
if (nz)
minthr = FFMIN(minthr, uplim);
allz |= nz;
start += sce->ics.swb_sizes[g];
}
}
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
for (g = 0; g < sce->ics.num_swb; g++) {
if (sce->zeroes[w*16+g]) {
sce->sf_idx[w*16+g] = SCALE_ONE_POS;
continue;
}
sce->sf_idx[w*16+g] = SCALE_ONE_POS + FFMIN(log2f(uplims[w*16+g]/minthr)*4,59);
}
}
if (!allz)
return;
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *scaled = s->scoefs + start;
maxvals[w*16+g] = find_max_val(sce->ics.group_len[w], sce->ics.swb_sizes[g], scaled);
start += sce->ics.swb_sizes[g];
}
}
//perform two-loop search
//outer loop - improve quality
do {
int tbits, qstep;
minscaler = sce->sf_idx[0];
minscaler_n = sce->sf_idx[0];
//inner loop - quantize spectrum to fit into given number of bits
qstep = its ? 1 : 32;
do {
int prev = -1;
tbits = 0;
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = w*128;
for (g = 0; g < sce->ics.num_swb; g++) {
const float *coefs = sce->coeffs + start;
const float *scaled = s->scoefs + start;
int bits = 0;
int cb;
float dist = 0.0f;
if (sce->band_type[w*16+g] == NOISE_BT) {
minscaler_n = FFMIN(minscaler_n, noise_sf[w*16+g]);
start += sce->ics.swb_sizes[g];
continue;
} else if (sce->zeroes[w*16+g] || sce->sf_idx[w*16+g] >= 218) {
start += sce->ics.swb_sizes[g];
continue;
}
minscaler = FFMIN(minscaler, sce->sf_idx[w*16+g]);
cb = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
int b;
dist += quantize_band_cost(s, coefs + w2*128,
scaled + w2*128,
sce->ics.swb_sizes[g],
sce->sf_idx[w*16+g],
cb,
1.0f,
INFINITY,
&b);
bits += b;
}
dists[w*16+g] = dist - bits;
if (prev != -1) {
bits += ff_aac_scalefactor_bits[sce->sf_idx[w*16+g] - prev + SCALE_DIFF_ZERO];
}
tbits += bits;
start += sce->ics.swb_sizes[g];
prev = sce->sf_idx[w*16+g];
}
}
if (tbits > destbits) {
for (i = 0; i < 128; i++)
if (sce->sf_idx[i] < 218 - qstep)
sce->sf_idx[i] += qstep;
} else {
for (i = 0; i < 128; i++)
if (sce->sf_idx[i] > 60 - qstep)
sce->sf_idx[i] -= qstep;
}
qstep >>= 1;
if (!qstep && tbits > destbits*1.02 && sce->sf_idx[0] < 217)
qstep = 1;
} while (qstep);
fflag = 0;
minscaler = av_clip(minscaler, 60, 255 - SCALE_MAX_DIFF);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
for (g = 0; g < sce->ics.num_swb; g++)
if (sce->band_type[w*16+g] == NOISE_BT)
sce->sf_idx[w*16+g] = av_clip(noise_sf[w*16+g], minscaler_n, minscaler_n + SCALE_MAX_DIFF);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
for (g = 0; g < sce->ics.num_swb; g++) {
int prevsc = sce->sf_idx[w*16+g];
if (sce->band_type[w*16+g] == NOISE_BT)
continue;
if (dists[w*16+g] > uplims[w*16+g] && sce->sf_idx[w*16+g] > 60) {
if (find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]-1))
sce->sf_idx[w*16+g]--;
else //Try to make sure there is some energy in every band
sce->sf_idx[w*16+g]-=2;
}
sce->sf_idx[w*16+g] = av_clip(sce->sf_idx[w*16+g], minscaler, minscaler + SCALE_MAX_DIFF);
sce->sf_idx[w*16+g] = FFMIN(sce->sf_idx[w*16+g], 219);
if (sce->sf_idx[w*16+g] != prevsc)
fflag = 1;
sce->band_type[w*16+g] = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);
}
}
its++;
} while (fflag && its < 10);
}
static void search_for_pns(AACEncContext *s, AVCodecContext *avctx, SingleChannelElement *sce)
{
FFPsyBand *band;
int w, g, w2, i, start, count = 0;
float *PNS = &s->scoefs[0*128], *PNS34 = &s->scoefs[1*128];
float *NOR34 = &s->scoefs[3*128];
const float lambda = s->lambda;
const float freq_mult = avctx->sample_rate/(1024.0f/sce->ics.num_windows)/2.0f;
const float thr_mult = NOISE_LAMBDA_REPLACE*(100.0f/lambda);
const float spread_threshold = NOISE_SPREAD_THRESHOLD*(lambda/100.f);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = 0;
for (g = 0; g < sce->ics.num_swb; g++) {
int noise_sfi, try_pns = 0;
float dist1 = 0.0f, dist2 = 0.0f, noise_amp;
float energy = 0.0f, threshold = 0.0f, spread = 0.0f;
if (start*freq_mult < NOISE_LOW_LIMIT) {
start += sce->ics.swb_sizes[g];
continue;
}
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
energy += band->energy;
spread += band->spread;
threshold += band->threshold;
}
sce->pns_ener[w*16+g] = energy;
if (sce->zeroes[w*16+g]) {
try_pns = 1;
} else if (energy < threshold) {
try_pns = 1;
} else if (spread > spread_threshold) {
try_pns = 0;
} else if (energy < threshold*thr_mult) {
try_pns = 1;
}
if (!try_pns || !energy) {
start += sce->ics.swb_sizes[g];
continue;
}
noise_sfi = av_clip(roundf(log2f(energy)*2), -100, 155); /* Quantize */
noise_amp = -ff_aac_pow2sf_tab[noise_sfi + POW_SF2_ZERO]; /* Dequantize */
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
float band_energy, scale;
band = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
for (i = 0; i < sce->ics.swb_sizes[g]; i++)
PNS[i] = s->random_state = lcg_random(s->random_state);
band_energy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]);
scale = noise_amp/sqrtf(band_energy);
s->fdsp->vector_fmul_scalar(PNS, PNS, scale, sce->ics.swb_sizes[g]);
abs_pow34_v(NOR34, &sce->coeffs[start+(w+w2)*128], sce->ics.swb_sizes[g]);
abs_pow34_v(PNS34, PNS, sce->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, &sce->coeffs[start + (w+w2)*128],
NOR34,
sce->ics.swb_sizes[g],
sce->sf_idx[(w+w2)*16+g],
sce->band_alt[(w+w2)*16+g],
lambda/band->threshold, INFINITY, NULL, 0);
dist2 += quantize_band_cost(s, PNS,
PNS34,
sce->ics.swb_sizes[g],
noise_sfi,
NOISE_BT,
lambda/band->threshold, INFINITY, NULL, 0);
}
if (dist2 < dist1) {
sce->band_type[w*16+g] = NOISE_BT;
sce->zeroes[w*16+g] = 0;
if (sce->band_type[w*16+g-1] != NOISE_BT && /* Prevent holes */
sce->band_type[w*16+g-2] == NOISE_BT) {
sce->band_type[w*16+g-1] = NOISE_BT;
sce->zeroes[w*16+g-1] = 0;
}
count++;
}
start += sce->ics.swb_sizes[g];
}
}
}
static void search_for_pns(AACEncContext *s, AVCodecContext *avctx, SingleChannelElement *sce)
{
FFPsyBand *band;
int w, g, w2, i;
float *PNS = &s->scoefs[0*128], *PNS34 = &s->scoefs[1*128];
float *NOR34 = &s->scoefs[3*128];
const float lambda = s->lambda;
const float freq_mult = avctx->sample_rate/(1024.0f/sce->ics.num_windows)/2.0f;
const float thr_mult = NOISE_LAMBDA_REPLACE*(100.0f/lambda);
const float spread_threshold = NOISE_SPREAD_THRESHOLD*(lambda/100.f);
if (sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE)
return;
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
for (g = 0; g < sce->ics.num_swb; g++) {
int noise_sfi;
float dist1 = 0.0f, dist2 = 0.0f, noise_amp;
float pns_energy = 0.0f, energy_ratio, dist_thresh;
float sfb_energy = 0.0f, threshold = 0.0f, spread = 0.0f;
const int start = sce->ics.swb_offset[w*16+g];
const float freq = start*freq_mult;
const float freq_boost = FFMAX(0.88f*freq/NOISE_LOW_LIMIT, 1.0f);
if (freq < NOISE_LOW_LIMIT || avctx->cutoff && freq >= avctx->cutoff)
continue;
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
sfb_energy += band->energy;
spread += band->spread;
threshold += band->threshold;
}
/* Ramps down at ~8000Hz and loosens the dist threshold */
dist_thresh = FFMIN(2.5f*NOISE_LOW_LIMIT/freq, 1.27f);
if (sce->zeroes[w*16+g] || spread < spread_threshold ||
sfb_energy > threshold*thr_mult*freq_boost) {
sce->pns_ener[w*16+g] = sfb_energy;
continue;
}
noise_sfi = av_clip(roundf(log2f(sfb_energy)*2), -100, 155); /* Quantize */
noise_amp = -ff_aac_pow2sf_tab[noise_sfi + POW_SF2_ZERO]; /* Dequantize */
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
float band_energy, scale;
const int start_c = sce->ics.swb_offset[(w+w2)*16+g];
band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
for (i = 0; i < sce->ics.swb_sizes[g]; i++)
PNS[i] = s->random_state = lcg_random(s->random_state);
band_energy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]);
scale = noise_amp/sqrtf(band_energy);
s->fdsp->vector_fmul_scalar(PNS, PNS, scale, sce->ics.swb_sizes[g]);
pns_energy += s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]);
abs_pow34_v(NOR34, &sce->coeffs[start_c], sce->ics.swb_sizes[g]);
abs_pow34_v(PNS34, PNS, sce->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, &sce->coeffs[start_c],
NOR34,
sce->ics.swb_sizes[g],
sce->sf_idx[(w+w2)*16+g],
sce->band_alt[(w+w2)*16+g],
lambda/band->threshold, INFINITY, NULL, 0);
dist2 += quantize_band_cost(s, PNS,
PNS34,
sce->ics.swb_sizes[g],
noise_sfi,
NOISE_BT,
lambda/band->threshold, INFINITY, NULL, 0);
}
energy_ratio = sfb_energy/pns_energy; /* Compensates for quantization error */
sce->pns_ener[w*16+g] = energy_ratio*sfb_energy;
if (energy_ratio > 0.85f && energy_ratio < 1.25f && dist1/dist2 > dist_thresh) {
sce->band_type[w*16+g] = NOISE_BT;
sce->zeroes[w*16+g] = 0;
if (sce->band_type[w*16+g-1] != NOISE_BT && /* Prevent holes */
sce->band_type[w*16+g-2] == NOISE_BT) {
sce->band_type[w*16+g-1] = NOISE_BT;
sce->zeroes[w*16+g-1] = 0;
}
}
}
}
}
