diff --git a/libavcodec/opus.h b/libavcodec/opus.h index be042497ea..c3cbaec35d 100644 --- a/libavcodec/opus.h +++ b/libavcodec/opus.h @@ -62,7 +62,9 @@ static const uint8_t opus_default_extradata[30] = { enum OpusMode { OPUS_MODE_SILK, OPUS_MODE_HYBRID, - OPUS_MODE_CELT + OPUS_MODE_CELT, + + OPUS_MODE_NB }; enum OpusBandwidth { @@ -70,12 +72,14 @@ enum OpusBandwidth { OPUS_BANDWIDTH_MEDIUMBAND, OPUS_BANDWIDTH_WIDEBAND, OPUS_BANDWIDTH_SUPERWIDEBAND, - OPUS_BANDWIDTH_FULLBAND + OPUS_BANDWIDTH_FULLBAND, + + OPUS_BANDWITH_NB }; typedef struct SilkContext SilkContext; -typedef struct CeltContext CeltContext; +typedef struct CeltFrame CeltFrame; typedef struct OpusPacket { int packet_size; /**< packet size */ @@ -100,7 +104,7 @@ typedef struct OpusStreamContext { OpusRangeCoder rc; OpusRangeCoder redundancy_rc; SilkContext *silk; - CeltContext *celt; + CeltFrame *celt; AVFloatDSPContext *fdsp; float silk_buf[2][960]; @@ -185,14 +189,4 @@ int ff_silk_decode_superframe(SilkContext *s, OpusRangeCoder *rc, enum OpusBandwidth bandwidth, int coded_channels, int duration_ms); -int ff_celt_init(AVCodecContext *avctx, CeltContext **s, int output_channels); - -void ff_celt_free(CeltContext **s); - -void ff_celt_flush(CeltContext *s); - -int ff_celt_decode_frame(CeltContext *s, OpusRangeCoder *rc, - float **output, int coded_channels, int frame_size, - int startband, int endband); - #endif /* AVCODEC_OPUS_H */ diff --git a/libavcodec/opus_celt.c b/libavcodec/opus_celt.c index 71ef8965e2..af3c100e6e 100644 --- a/libavcodec/opus_celt.c +++ b/libavcodec/opus_celt.c @@ -1,6 +1,7 @@ /* * Copyright (c) 2012 Andrew D'Addesio * Copyright (c) 2013-2014 Mozilla Corporation + * Copyright (c) 2016 Rostislav Pehlivanov * * This file is part of FFmpeg. * @@ -28,7 +29,7 @@ #include "opustab.h" #include "opus_pvq.h" -static void celt_decode_coarse_energy(CeltContext *s, OpusRangeCoder *rc) +static void celt_decode_coarse_energy(CeltFrame *f, OpusRangeCoder *rc) { int i, j; float prev[2] = {0}; @@ -38,29 +39,29 @@ static void celt_decode_coarse_energy(CeltContext *s, OpusRangeCoder *rc) /* use the 2D z-transform to apply prediction in both */ /* the time domain (alpha) and the frequency domain (beta) */ - if (opus_rc_tell(rc)+3 <= s->framebits && ff_opus_rc_dec_log(rc, 3)) { + if (opus_rc_tell(rc)+3 <= f->framebits && ff_opus_rc_dec_log(rc, 3)) { /* intra frame */ alpha = 0; beta = 1.0f - 4915.0f/32768.0f; - model = ff_celt_coarse_energy_dist[s->duration][1]; + model = ff_celt_coarse_energy_dist[f->size][1]; } else { - alpha = ff_celt_alpha_coef[s->duration]; - beta = 1.0f - ff_celt_beta_coef[s->duration]; - model = ff_celt_coarse_energy_dist[s->duration][0]; + alpha = ff_celt_alpha_coef[f->size]; + beta = 1.0f - ff_celt_beta_coef[f->size]; + model = ff_celt_coarse_energy_dist[f->size][0]; } for (i = 0; i < CELT_MAX_BANDS; i++) { - for (j = 0; j < s->coded_channels; j++) { - CeltFrame *frame = &s->frame[j]; + for (j = 0; j < f->channels; j++) { + CeltBlock *block = &f->block[j]; float value; int available; - if (i < s->startband || i >= s->endband) { - frame->energy[i] = 0.0; + if (i < f->start_band || i >= f->end_band) { + block->energy[i] = 0.0; continue; } - available = s->framebits - opus_rc_tell(rc); + available = f->framebits - opus_rc_tell(rc); if (available >= 15) { /* decode using a Laplace distribution */ int k = FFMIN(i, 20) << 1; @@ -72,82 +73,81 @@ static void celt_decode_coarse_energy(CeltContext *s, OpusRangeCoder *rc) value = -(float)ff_opus_rc_dec_log(rc, 1); } else value = -1; - frame->energy[i] = FFMAX(-9.0f, frame->energy[i]) * alpha + prev[j] + value; + block->energy[i] = FFMAX(-9.0f, block->energy[i]) * alpha + prev[j] + value; prev[j] += beta * value; } } } -static void celt_decode_fine_energy(CeltContext *s, OpusRangeCoder *rc) +static void celt_decode_fine_energy(CeltFrame *f, OpusRangeCoder *rc) { int i; - for (i = s->startband; i < s->endband; i++) { + for (i = f->start_band; i < f->end_band; i++) { int j; - if (!s->fine_bits[i]) + if (!f->fine_bits[i]) continue; - for (j = 0; j < s->coded_channels; j++) { - CeltFrame *frame = &s->frame[j]; + for (j = 0; j < f->channels; j++) { + CeltBlock *block = &f->block[j]; int q2; float offset; - q2 = ff_opus_rc_get_raw(rc, s->fine_bits[i]); - offset = (q2 + 0.5f) * (1 << (14 - s->fine_bits[i])) / 16384.0f - 0.5f; - frame->energy[i] += offset; + q2 = ff_opus_rc_get_raw(rc, f->fine_bits[i]); + offset = (q2 + 0.5f) * (1 << (14 - f->fine_bits[i])) / 16384.0f - 0.5f; + block->energy[i] += offset; } } } -static void celt_decode_final_energy(CeltContext *s, OpusRangeCoder *rc, - int bits_left) +static void celt_decode_final_energy(CeltFrame *f, OpusRangeCoder *rc) { int priority, i, j; + int bits_left = f->framebits - opus_rc_tell(rc); for (priority = 0; priority < 2; priority++) { - for (i = s->startband; i < s->endband && bits_left >= s->coded_channels; i++) { - if (s->fine_priority[i] != priority || s->fine_bits[i] >= CELT_MAX_FINE_BITS) + for (i = f->start_band; i < f->end_band && bits_left >= f->channels; i++) { + if (f->fine_priority[i] != priority || f->fine_bits[i] >= CELT_MAX_FINE_BITS) continue; - for (j = 0; j < s->coded_channels; j++) { + for (j = 0; j < f->channels; j++) { int q2; float offset; q2 = ff_opus_rc_get_raw(rc, 1); - offset = (q2 - 0.5f) * (1 << (14 - s->fine_bits[i] - 1)) / 16384.0f; - s->frame[j].energy[i] += offset; + offset = (q2 - 0.5f) * (1 << (14 - f->fine_bits[i] - 1)) / 16384.0f; + f->block[j].energy[i] += offset; bits_left--; } } } } -static void celt_decode_tf_changes(CeltContext *s, OpusRangeCoder *rc, - int transient) +static void celt_decode_tf_changes(CeltFrame *f, OpusRangeCoder *rc) { int i, diff = 0, tf_select = 0, tf_changed = 0, tf_select_bit; - int consumed, bits = transient ? 2 : 4; + int consumed, bits = f->transient ? 2 : 4; consumed = opus_rc_tell(rc); - tf_select_bit = (s->duration != 0 && consumed+bits+1 <= s->framebits); + tf_select_bit = (f->size != 0 && consumed+bits+1 <= f->framebits); - for (i = s->startband; i < s->endband; i++) { - if (consumed+bits+tf_select_bit <= s->framebits) { + for (i = f->start_band; i < f->end_band; i++) { + if (consumed+bits+tf_select_bit <= f->framebits) { diff ^= ff_opus_rc_dec_log(rc, bits); consumed = opus_rc_tell(rc); tf_changed |= diff; } - s->tf_change[i] = diff; - bits = transient ? 4 : 5; + f->tf_change[i] = diff; + bits = f->transient ? 4 : 5; } - if (tf_select_bit && ff_celt_tf_select[s->duration][transient][0][tf_changed] != - ff_celt_tf_select[s->duration][transient][1][tf_changed]) + if (tf_select_bit && ff_celt_tf_select[f->size][f->transient][0][tf_changed] != + ff_celt_tf_select[f->size][f->transient][1][tf_changed]) tf_select = ff_opus_rc_dec_log(rc, 1); - for (i = s->startband; i < s->endband; i++) { - s->tf_change[i] = ff_celt_tf_select[s->duration][transient][tf_select][s->tf_change[i]]; + for (i = f->start_band; i < f->end_band; i++) { + f->tf_change[i] = ff_celt_tf_select[f->size][f->transient][tf_select][f->tf_change[i]]; } } -static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) +static void celt_decode_allocation(CeltFrame *f, OpusRangeCoder *rc) { // approx. maximum bit allocation for each band before boost/trim int cap[CELT_MAX_BANDS]; @@ -157,14 +157,14 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) int bits2[CELT_MAX_BANDS]; int trim_offset[CELT_MAX_BANDS]; - int skip_startband = s->startband; + int skip_start_band = f->start_band; int dynalloc = 6; int alloctrim = 5; int extrabits = 0; - int skip_bit = 0; - int intensitystereo_bit = 0; - int dualstereo_bit = 0; + int skip_bit = 0; + int intensity_stereo_bit = 0; + int dual_stereo_bit = 0; int remaining, bandbits; int low, high, total, done; @@ -175,25 +175,25 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) consumed = opus_rc_tell(rc); /* obtain spread flag */ - s->spread = CELT_SPREAD_NORMAL; - if (consumed + 4 <= s->framebits) - s->spread = ff_opus_rc_dec_cdf(rc, ff_celt_model_spread); + f->spread = CELT_SPREAD_NORMAL; + if (consumed + 4 <= f->framebits) + f->spread = ff_opus_rc_dec_cdf(rc, ff_celt_model_spread); /* generate static allocation caps */ for (i = 0; i < CELT_MAX_BANDS; i++) { - cap[i] = (ff_celt_static_caps[s->duration][s->coded_channels - 1][i] + 64) - * ff_celt_freq_range[i] << (s->coded_channels - 1) << s->duration >> 2; + cap[i] = (ff_celt_static_caps[f->size][f->channels - 1][i] + 64) + * ff_celt_freq_range[i] << (f->channels - 1) << f->size >> 2; } /* obtain band boost */ - totalbits = s->framebits << 3; // convert to 1/8 bits + totalbits = f->framebits << 3; // convert to 1/8 bits consumed = opus_rc_tell_frac(rc); - for (i = s->startband; i < s->endband; i++) { + for (i = f->start_band; i < f->end_band; i++) { int quanta, band_dynalloc; boost[i] = 0; - quanta = ff_celt_freq_range[i] << (s->coded_channels - 1) << s->duration; + quanta = ff_celt_freq_range[i] << (f->channels - 1) << f->size; quanta = FFMIN(quanta << 3, FFMAX(6 << 3, quanta)); band_dynalloc = dynalloc; while (consumed + (band_dynalloc<<3) < totalbits && boost[i] < cap[i]) { @@ -216,12 +216,12 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) alloctrim = ff_opus_rc_dec_cdf(rc, ff_celt_model_alloc_trim); /* anti-collapse bit reservation */ - totalbits = (s->framebits << 3) - opus_rc_tell_frac(rc) - 1; - s->anticollapse_bit = 0; - if (s->blocks > 1 && s->duration >= 2 && - totalbits >= ((s->duration + 2) << 3)) - s->anticollapse_bit = 1 << 3; - totalbits -= s->anticollapse_bit; + totalbits = (f->framebits << 3) - opus_rc_tell_frac(rc) - 1; + f->anticollapse_needed = 0; + if (f->blocks > 1 && f->size >= 2 && + totalbits >= ((f->size + 2) << 3)) + f->anticollapse_needed = 1 << 3; + totalbits -= f->anticollapse_needed; /* band skip bit reservation */ if (totalbits >= 1 << 3) @@ -229,33 +229,33 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) totalbits -= skip_bit; /* intensity/dual stereo bit reservation */ - if (s->coded_channels == 2) { - intensitystereo_bit = ff_celt_log2_frac[s->endband - s->startband]; - if (intensitystereo_bit <= totalbits) { - totalbits -= intensitystereo_bit; + if (f->channels == 2) { + intensity_stereo_bit = ff_celt_log2_frac[f->end_band - f->start_band]; + if (intensity_stereo_bit <= totalbits) { + totalbits -= intensity_stereo_bit; if (totalbits >= 1 << 3) { - dualstereo_bit = 1 << 3; + dual_stereo_bit = 1 << 3; totalbits -= 1 << 3; } } else - intensitystereo_bit = 0; + intensity_stereo_bit = 0; } - for (i = s->startband; i < s->endband; i++) { - int trim = alloctrim - 5 - s->duration; - int band = ff_celt_freq_range[i] * (s->endband - i - 1); - int duration = s->duration + 3; - int scale = duration + s->coded_channels - 1; + for (i = f->start_band; i < f->end_band; i++) { + int trim = alloctrim - 5 - f->size; + int band = ff_celt_freq_range[i] * (f->end_band - i - 1); + int duration = f->size + 3; + int scale = duration + f->channels - 1; /* PVQ minimum allocation threshold, below this value the band is * skipped */ threshold[i] = FFMAX(3 * ff_celt_freq_range[i] << duration >> 4, - s->coded_channels << 3); + f->channels << 3); trim_offset[i] = trim * (band << scale) >> 6; - if (ff_celt_freq_range[i] << s->duration == 1) - trim_offset[i] -= s->coded_channels << 3; + if (ff_celt_freq_range[i] << f->size == 1) + trim_offset[i] -= f->channels << 3; } /* bisection */ @@ -265,9 +265,9 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) int center = (low + high) >> 1; done = total = 0; - for (i = s->endband - 1; i >= s->startband; i--) { + for (i = f->end_band - 1; i >= f->start_band; i--) { bandbits = ff_celt_freq_range[i] * ff_celt_static_alloc[center][i] - << (s->coded_channels - 1) << s->duration >> 2; + << (f->channels - 1) << f->size >> 2; if (bandbits) bandbits = FFMAX(0, bandbits + trim_offset[i]); @@ -276,8 +276,8 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) if (bandbits >= threshold[i] || done) { done = 1; total += FFMIN(bandbits, cap[i]); - } else if (bandbits >= s->coded_channels << 3) - total += s->coded_channels << 3; + } else if (bandbits >= f->channels << 3) + total += f->channels << 3; } if (total > totalbits) @@ -287,12 +287,12 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) } high = low--; - for (i = s->startband; i < s->endband; i++) { + for (i = f->start_band; i < f->end_band; i++) { bits1[i] = ff_celt_freq_range[i] * ff_celt_static_alloc[low][i] - << (s->coded_channels - 1) << s->duration >> 2; + << (f->channels - 1) << f->size >> 2; bits2[i] = high >= CELT_VECTORS ? cap[i] : ff_celt_freq_range[i] * ff_celt_static_alloc[high][i] - << (s->coded_channels - 1) << s->duration >> 2; + << (f->channels - 1) << f->size >> 2; if (bits1[i]) bits1[i] = FFMAX(0, bits1[i] + trim_offset[i]); @@ -303,7 +303,7 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) bits2[i] += boost[i]; if (boost[i]) - skip_startband = i; + skip_start_band = i; bits2[i] = FFMAX(0, bits2[i] - bits1[i]); } @@ -314,14 +314,14 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) int center = (low + high) >> 1; done = total = 0; - for (j = s->endband - 1; j >= s->startband; j--) { + for (j = f->end_band - 1; j >= f->start_band; j--) { bandbits = bits1[j] + (center * bits2[j] >> CELT_ALLOC_STEPS); if (bandbits >= threshold[j] || done) { done = 1; total += FFMIN(bandbits, cap[j]); - } else if (bandbits >= s->coded_channels << 3) - total += s->coded_channels << 3; + } else if (bandbits >= f->channels << 3) + total += f->channels << 3; } if (total > totalbits) high = center; @@ -330,26 +330,26 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) } done = total = 0; - for (i = s->endband - 1; i >= s->startband; i--) { + for (i = f->end_band - 1; i >= f->start_band; i--) { bandbits = bits1[i] + (low * bits2[i] >> CELT_ALLOC_STEPS); if (bandbits >= threshold[i] || done) done = 1; else - bandbits = (bandbits >= s->coded_channels << 3) ? - s->coded_channels << 3 : 0; + bandbits = (bandbits >= f->channels << 3) ? + f->channels << 3 : 0; bandbits = FFMIN(bandbits, cap[i]); - s->pulses[i] = bandbits; + f->pulses[i] = bandbits; total += bandbits; } /* band skipping */ - for (s->codedbands = s->endband; ; s->codedbands--) { + for (f->coded_bands = f->end_band; ; f->coded_bands--) { int allocation; - j = s->codedbands - 1; + j = f->coded_bands - 1; - if (j == skip_startband) { + if (j == skip_start_band) { /* all remaining bands are not skipped */ totalbits += skip_bit; break; @@ -357,14 +357,14 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) /* determine the number of bits available for coding "do not skip" markers */ remaining = totalbits - total; - bandbits = remaining / (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[s->startband]); - remaining -= bandbits * (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[s->startband]); - allocation = s->pulses[j] + bandbits * ff_celt_freq_range[j] - + FFMAX(0, remaining - (ff_celt_freq_bands[j] - ff_celt_freq_bands[s->startband])); + bandbits = remaining / (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[f->start_band]); + remaining -= bandbits * (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[f->start_band]); + allocation = f->pulses[j] + bandbits * ff_celt_freq_range[j] + + FFMAX(0, remaining - (ff_celt_freq_bands[j] - ff_celt_freq_bands[f->start_band])); /* a "do not skip" marker is only coded if the allocation is above the chosen threshold */ - if (allocation >= FFMAX(threshold[j], (s->coded_channels + 1) <<3 )) { + if (allocation >= FFMAX(threshold[j], (f->channels + 1) <<3 )) { if (ff_opus_rc_dec_log(rc, 1)) break; @@ -373,43 +373,43 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) } /* the band is skipped, so reclaim its bits */ - total -= s->pulses[j]; - if (intensitystereo_bit) { - total -= intensitystereo_bit; - intensitystereo_bit = ff_celt_log2_frac[j - s->startband]; - total += intensitystereo_bit; + total -= f->pulses[j]; + if (intensity_stereo_bit) { + total -= intensity_stereo_bit; + intensity_stereo_bit = ff_celt_log2_frac[j - f->start_band]; + total += intensity_stereo_bit; } - total += s->pulses[j] = (allocation >= s->coded_channels << 3) ? - s->coded_channels << 3 : 0; + total += f->pulses[j] = (allocation >= f->channels << 3) ? + f->channels << 3 : 0; } /* obtain stereo flags */ - s->intensitystereo = 0; - s->dualstereo = 0; - if (intensitystereo_bit) - s->intensitystereo = s->startband + - ff_opus_rc_dec_uint(rc, s->codedbands + 1 - s->startband); - if (s->intensitystereo <= s->startband) - totalbits += dualstereo_bit; /* no intensity stereo means no dual stereo */ - else if (dualstereo_bit) - s->dualstereo = ff_opus_rc_dec_log(rc, 1); + f->intensity_stereo = 0; + f->dual_stereo = 0; + if (intensity_stereo_bit) + f->intensity_stereo = f->start_band + + ff_opus_rc_dec_uint(rc, f->coded_bands + 1 - f->start_band); + if (f->intensity_stereo <= f->start_band) + totalbits += dual_stereo_bit; /* no intensity stereo means no dual stereo */ + else if (dual_stereo_bit) + f->dual_stereo = ff_opus_rc_dec_log(rc, 1); /* supply the remaining bits in this frame to lower bands */ remaining = totalbits - total; - bandbits = remaining / (ff_celt_freq_bands[s->codedbands] - ff_celt_freq_bands[s->startband]); - remaining -= bandbits * (ff_celt_freq_bands[s->codedbands] - ff_celt_freq_bands[s->startband]); - for (i = s->startband; i < s->codedbands; i++) { + bandbits = remaining / (ff_celt_freq_bands[f->coded_bands] - ff_celt_freq_bands[f->start_band]); + remaining -= bandbits * (ff_celt_freq_bands[f->coded_bands] - ff_celt_freq_bands[f->start_band]); + for (i = f->start_band; i < f->coded_bands; i++) { int bits = FFMIN(remaining, ff_celt_freq_range[i]); - s->pulses[i] += bits + bandbits * ff_celt_freq_range[i]; + f->pulses[i] += bits + bandbits * ff_celt_freq_range[i]; remaining -= bits; } - for (i = s->startband; i < s->codedbands; i++) { - int N = ff_celt_freq_range[i] << s->duration; + for (i = f->start_band; i < f->coded_bands; i++) { + int N = ff_celt_freq_range[i] << f->size; int prev_extra = extrabits; - s->pulses[i] += extrabits; + f->pulses[i] += extrabits; if (N > 1) { int dof; // degrees of freedom @@ -419,83 +419,83 @@ static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc) // totalbits/dof int fine_bits, max_bits; - extrabits = FFMAX(0, s->pulses[i] - cap[i]); - s->pulses[i] -= extrabits; + extrabits = FFMAX(0, f->pulses[i] - cap[i]); + f->pulses[i] -= extrabits; /* intensity stereo makes use of an extra degree of freedom */ - dof = N * s->coded_channels - + (s->coded_channels == 2 && N > 2 && !s->dualstereo && i < s->intensitystereo); - temp = dof * (ff_celt_log_freq_range[i] + (s->duration<<3)); + dof = N * f->channels + + (f->channels == 2 && N > 2 && !f->dual_stereo && i < f->intensity_stereo); + temp = dof * (ff_celt_log_freq_range[i] + (f->size<<3)); offset = (temp >> 1) - dof * CELT_FINE_OFFSET; if (N == 2) /* dof=2 is the only case that doesn't fit the model */ offset += dof<<1; /* grant an additional bias for the first and second pulses */ - if (s->pulses[i] + offset < 2 * (dof << 3)) + if (f->pulses[i] + offset < 2 * (dof << 3)) offset += temp >> 2; - else if (s->pulses[i] + offset < 3 * (dof << 3)) + else if (f->pulses[i] + offset < 3 * (dof << 3)) offset += temp >> 3; - fine_bits = (s->pulses[i] + offset + (dof << 2)) / (dof << 3); - max_bits = FFMIN((s->pulses[i]>>3) >> (s->coded_channels - 1), + fine_bits = (f->pulses[i] + offset + (dof << 2)) / (dof << 3); + max_bits = FFMIN((f->pulses[i]>>3) >> (f->channels - 1), CELT_MAX_FINE_BITS); max_bits = FFMAX(max_bits, 0); - s->fine_bits[i] = av_clip(fine_bits, 0, max_bits); + f->fine_bits[i] = av_clip(fine_bits, 0, max_bits); /* if fine_bits was rounded down or capped, give priority for the final fine energy pass */ - s->fine_priority[i] = (s->fine_bits[i] * (dof<<3) >= s->pulses[i] + offset); + f->fine_priority[i] = (f->fine_bits[i] * (dof<<3) >= f->pulses[i] + offset); /* the remaining bits are assigned to PVQ */ - s->pulses[i] -= s->fine_bits[i] << (s->coded_channels - 1) << 3; + f->pulses[i] -= f->fine_bits[i] << (f->channels - 1) << 3; } else { /* all bits go to fine energy except for the sign bit */ - extrabits = FFMAX(0, s->pulses[i] - (s->coded_channels << 3)); - s->pulses[i] -= extrabits; - s->fine_bits[i] = 0; - s->fine_priority[i] = 1; + extrabits = FFMAX(0, f->pulses[i] - (f->channels << 3)); + f->pulses[i] -= extrabits; + f->fine_bits[i] = 0; + f->fine_priority[i] = 1; } /* hand back a limited number of extra fine energy bits to this band */ if (extrabits > 0) { - int fineextra = FFMIN(extrabits >> (s->coded_channels + 2), - CELT_MAX_FINE_BITS - s->fine_bits[i]); - s->fine_bits[i] += fineextra; + int fineextra = FFMIN(extrabits >> (f->channels + 2), + CELT_MAX_FINE_BITS - f->fine_bits[i]); + f->fine_bits[i] += fineextra; - fineextra <<= s->coded_channels + 2; - s->fine_priority[i] = (fineextra >= extrabits - prev_extra); + fineextra <<= f->channels + 2; + f->fine_priority[i] = (fineextra >= extrabits - prev_extra); extrabits -= fineextra; } } - s->remaining = extrabits; + f->remaining = extrabits; /* skipped bands dedicate all of their bits for fine energy */ - for (; i < s->endband; i++) { - s->fine_bits[i] = s->pulses[i] >> (s->coded_channels - 1) >> 3; - s->pulses[i] = 0; - s->fine_priority[i] = s->fine_bits[i] < 1; + for (; i < f->end_band; i++) { + f->fine_bits[i] = f->pulses[i] >> (f->channels - 1) >> 3; + f->pulses[i] = 0; + f->fine_priority[i] = f->fine_bits[i] < 1; } } -static void celt_denormalize(CeltContext *s, CeltFrame *frame, float *data) +static void celt_denormalize(CeltFrame *f, CeltBlock *block, float *data) { int i, j; - for (i = s->startband; i < s->endband; i++) { - float *dst = data + (ff_celt_freq_bands[i] << s->duration); - float norm = exp2(frame->energy[i] + ff_celt_mean_energy[i]); + for (i = f->start_band; i < f->end_band; i++) { + float *dst = data + (ff_celt_freq_bands[i] << f->size); + float norm = exp2(block->energy[i] + ff_celt_mean_energy[i]); - for (j = 0; j < ff_celt_freq_range[i] << s->duration; j++) + for (j = 0; j < ff_celt_freq_range[i] << f->size; j++) dst[j] *= norm; } } -static void celt_postfilter_apply_transition(CeltFrame *frame, float *data) +static void celt_postfilter_apply_transition(CeltBlock *block, float *data) { - const int T0 = frame->pf_period_old; - const int T1 = frame->pf_period; + const int T0 = block->pf_period_old; + const int T1 = block->pf_period; float g00, g01, g02; float g10, g11, g12; @@ -504,16 +504,16 @@ static void celt_postfilter_apply_transition(CeltFrame *frame, float *data) int i; - if (frame->pf_gains[0] == 0.0 && - frame->pf_gains_old[0] == 0.0) + if (block->pf_gains[0] == 0.0 && + block->pf_gains_old[0] == 0.0) return; - g00 = frame->pf_gains_old[0]; - g01 = frame->pf_gains_old[1]; - g02 = frame->pf_gains_old[2]; - g10 = frame->pf_gains[0]; - g11 = frame->pf_gains[1]; - g12 = frame->pf_gains[2]; + g00 = block->pf_gains_old[0]; + g01 = block->pf_gains_old[1]; + g02 = block->pf_gains_old[2]; + g10 = block->pf_gains[0]; + g11 = block->pf_gains[1]; + g12 = block->pf_gains[2]; x1 = data[-T1 + 1]; x2 = data[-T1]; @@ -537,20 +537,19 @@ static void celt_postfilter_apply_transition(CeltFrame *frame, float *data) } } -static void celt_postfilter_apply(CeltFrame *frame, - float *data, int len) +static void celt_postfilter_apply(CeltBlock *block, float *data, int len) { - const int T = frame->pf_period; + const int T = block->pf_period; float g0, g1, g2; float x0, x1, x2, x3, x4; int i; - if (frame->pf_gains[0] == 0.0 || len <= 0) + if (block->pf_gains[0] == 0.0 || len <= 0) return; - g0 = frame->pf_gains[0]; - g1 = frame->pf_gains[1]; - g2 = frame->pf_gains[2]; + g0 = block->pf_gains[0]; + g1 = block->pf_gains[1]; + g2 = block->pf_gains[2]; x4 = data[-T - 2]; x3 = data[-T - 1]; @@ -569,31 +568,31 @@ static void celt_postfilter_apply(CeltFrame *frame, } } -static void celt_postfilter(CeltContext *s, CeltFrame *frame) +static void celt_postfilter(CeltFrame *f, CeltBlock *block) { - int len = s->blocksize * s->blocks; + int len = f->blocksize * f->blocks; - celt_postfilter_apply_transition(frame, frame->buf + 1024); + celt_postfilter_apply_transition(block, block->buf + 1024); - frame->pf_period_old = frame->pf_period; - memcpy(frame->pf_gains_old, frame->pf_gains, sizeof(frame->pf_gains)); + block->pf_period_old = block->pf_period; + memcpy(block->pf_gains_old, block->pf_gains, sizeof(block->pf_gains)); - frame->pf_period = frame->pf_period_new; - memcpy(frame->pf_gains, frame->pf_gains_new, sizeof(frame->pf_gains)); + block->pf_period = block->pf_period_new; + memcpy(block->pf_gains, block->pf_gains_new, sizeof(block->pf_gains)); if (len > CELT_OVERLAP) { - celt_postfilter_apply_transition(frame, frame->buf + 1024 + CELT_OVERLAP); - celt_postfilter_apply(frame, frame->buf + 1024 + 2 * CELT_OVERLAP, + celt_postfilter_apply_transition(block, block->buf + 1024 + CELT_OVERLAP); + celt_postfilter_apply(block, block->buf + 1024 + 2 * CELT_OVERLAP, len - 2 * CELT_OVERLAP); - frame->pf_period_old = frame->pf_period; - memcpy(frame->pf_gains_old, frame->pf_gains, sizeof(frame->pf_gains)); + block->pf_period_old = block->pf_period; + memcpy(block->pf_gains_old, block->pf_gains, sizeof(block->pf_gains)); } - memmove(frame->buf, frame->buf + len, (1024 + CELT_OVERLAP / 2) * sizeof(float)); + memmove(block->buf, block->buf + len, (1024 + CELT_OVERLAP / 2) * sizeof(float)); } -static int parse_postfilter(CeltContext *s, OpusRangeCoder *rc, int consumed) +static int parse_postfilter(CeltFrame *f, OpusRangeCoder *rc, int consumed) { static const float postfilter_taps[3][3] = { { 0.3066406250f, 0.2170410156f, 0.1296386719f }, @@ -602,10 +601,10 @@ static int parse_postfilter(CeltContext *s, OpusRangeCoder *rc, int consumed) }; int i; - memset(s->frame[0].pf_gains_new, 0, sizeof(s->frame[0].pf_gains_new)); - memset(s->frame[1].pf_gains_new, 0, sizeof(s->frame[1].pf_gains_new)); + memset(f->block[0].pf_gains_new, 0, sizeof(f->block[0].pf_gains_new)); + memset(f->block[1].pf_gains_new, 0, sizeof(f->block[1].pf_gains_new)); - if (s->startband == 0 && consumed + 16 <= s->framebits) { + if (f->start_band == 0 && consumed + 16 <= f->framebits) { int has_postfilter = ff_opus_rc_dec_log(rc, 1); if (has_postfilter) { float gain; @@ -614,16 +613,16 @@ static int parse_postfilter(CeltContext *s, OpusRangeCoder *rc, int consumed) octave = ff_opus_rc_dec_uint(rc, 6); period = (16 << octave) + ff_opus_rc_get_raw(rc, 4 + octave) - 1; gain = 0.09375f * (ff_opus_rc_get_raw(rc, 3) + 1); - tapset = (opus_rc_tell(rc) + 2 <= s->framebits) ? + tapset = (opus_rc_tell(rc) + 2 <= f->framebits) ? ff_opus_rc_dec_cdf(rc, ff_celt_model_tapset) : 0; for (i = 0; i < 2; i++) { - CeltFrame *frame = &s->frame[i]; + CeltBlock *block = &f->block[i]; - frame->pf_period_new = FFMAX(period, CELT_POSTFILTER_MINPERIOD); - frame->pf_gains_new[0] = gain * postfilter_taps[tapset][0]; - frame->pf_gains_new[1] = gain * postfilter_taps[tapset][1]; - frame->pf_gains_new[2] = gain * postfilter_taps[tapset][2]; + block->pf_period_new = FFMAX(period, CELT_POSTFILTER_MINPERIOD); + block->pf_gains_new[0] = gain * postfilter_taps[tapset][0]; + block->pf_gains_new[1] = gain * postfilter_taps[tapset][1]; + block->pf_gains_new[2] = gain * postfilter_taps[tapset][2]; } } @@ -633,11 +632,11 @@ static int parse_postfilter(CeltContext *s, OpusRangeCoder *rc, int consumed) return consumed; } -static void process_anticollapse(CeltContext *s, CeltFrame *frame, float *X) +static void process_anticollapse(CeltFrame *f, CeltBlock *block, float *X) { int i, j, k; - for (i = s->startband; i < s->endband; i++) { + for (i = f->start_band; i < f->end_band; i++) { int renormalize = 0; float *xptr; float prev[2]; @@ -646,64 +645,65 @@ static void process_anticollapse(CeltContext *s, CeltFrame *frame, float *X) int depth; /* depth in 1/8 bits */ - depth = (1 + s->pulses[i]) / (ff_celt_freq_range[i] << s->duration); + depth = (1 + f->pulses[i]) / (ff_celt_freq_range[i] << f->size); thresh = exp2f(-1.0 - 0.125f * depth); - sqrt_1 = 1.0f / sqrtf(ff_celt_freq_range[i] << s->duration); + sqrt_1 = 1.0f / sqrtf(ff_celt_freq_range[i] << f->size); - xptr = X + (ff_celt_freq_bands[i] << s->duration); + xptr = X + (ff_celt_freq_bands[i] << f->size); - prev[0] = frame->prev_energy[0][i]; - prev[1] = frame->prev_energy[1][i]; - if (s->coded_channels == 1) { - CeltFrame *frame1 = &s->frame[1]; + prev[0] = block->prev_energy[0][i]; + prev[1] = block->prev_energy[1][i]; + if (f->channels == 1) { + CeltBlock *block1 = &f->block[1]; - prev[0] = FFMAX(prev[0], frame1->prev_energy[0][i]); - prev[1] = FFMAX(prev[1], frame1->prev_energy[1][i]); + prev[0] = FFMAX(prev[0], block1->prev_energy[0][i]); + prev[1] = FFMAX(prev[1], block1->prev_energy[1][i]); } - Ediff = frame->energy[i] - FFMIN(prev[0], prev[1]); + Ediff = block->energy[i] - FFMIN(prev[0], prev[1]); Ediff = FFMAX(0, Ediff); /* r needs to be multiplied by 2 or 2*sqrt(2) depending on LM because short blocks don't have the same energy as long */ r = exp2(1 - Ediff); - if (s->duration == 3) + if (f->size == 3) r *= M_SQRT2; r = FFMIN(thresh, r) * sqrt_1; - for (k = 0; k < 1 << s->duration; k++) { + for (k = 0; k < 1 << f->size; k++) { /* Detect collapse */ - if (!(frame->collapse_masks[i] & 1 << k)) { + if (!(block->collapse_masks[i] & 1 << k)) { /* Fill with noise */ for (j = 0; j < ff_celt_freq_range[i]; j++) - xptr[(j << s->duration) + k] = (celt_rng(s) & 0x8000) ? r : -r; + xptr[(j << f->size) + k] = (celt_rng(f) & 0x8000) ? r : -r; renormalize = 1; } } /* We just added some energy, so we need to renormalize */ if (renormalize) - celt_renormalize_vector(xptr, ff_celt_freq_range[i] << s->duration, 1.0f); + celt_renormalize_vector(xptr, ff_celt_freq_range[i] << f->size, 1.0f); } } -static void celt_decode_bands(CeltContext *s, OpusRangeCoder *rc) +static void celt_decode_bands(CeltFrame *f, OpusRangeCoder *rc) { float lowband_scratch[8 * 22]; float norm[2 * 8 * 100]; - int totalbits = (s->framebits << 3) - s->anticollapse_bit; + int totalbits = (f->framebits << 3) - f->anticollapse_needed; int update_lowband = 1; int lowband_offset = 0; int i, j; - memset(s->coeffs, 0, sizeof(s->coeffs)); + memset(f->block[0].coeffs, 0, sizeof(f->block[0].coeffs)); + memset(f->block[1].coeffs, 0, sizeof(f->block[0].coeffs)); - for (i = s->startband; i < s->endband; i++) { - int band_offset = ff_celt_freq_bands[i] << s->duration; - int band_size = ff_celt_freq_range[i] << s->duration; - float *X = s->coeffs[0] + band_offset; - float *Y = (s->coded_channels == 2) ? s->coeffs[1] + band_offset : NULL; + for (i = f->start_band; i < f->end_band; i++) { + int band_offset = ff_celt_freq_bands[i] << f->size; + int band_size = ff_celt_freq_range[i] << f->size; + float *X = f->block[0].coeffs + band_offset; + float *Y = (f->channels == 2) ? f->block[1].coeffs + band_offset : NULL; int consumed = opus_rc_tell_frac(rc); float *norm2 = norm + 8 * 100; @@ -712,27 +712,27 @@ static void celt_decode_bands(CeltContext *s, OpusRangeCoder *rc) int b; /* Compute how many bits we want to allocate to this band */ - if (i != s->startband) - s->remaining -= consumed; - s->remaining2 = totalbits - consumed - 1; - if (i <= s->codedbands - 1) { - int curr_balance = s->remaining / FFMIN(3, s->codedbands-i); - b = av_clip_uintp2(FFMIN(s->remaining2 + 1, s->pulses[i] + curr_balance), 14); + if (i != f->start_band) + f->remaining -= consumed; + f->remaining2 = totalbits - consumed - 1; + if (i <= f->coded_bands - 1) { + int curr_balance = f->remaining / FFMIN(3, f->coded_bands-i); + b = av_clip_uintp2(FFMIN(f->remaining2 + 1, f->pulses[i] + curr_balance), 14); } else b = 0; - if (ff_celt_freq_bands[i] - ff_celt_freq_range[i] >= ff_celt_freq_bands[s->startband] && + if (ff_celt_freq_bands[i] - ff_celt_freq_range[i] >= ff_celt_freq_bands[f->start_band] && (update_lowband || lowband_offset == 0)) lowband_offset = i; /* Get a conservative estimate of the collapse_mask's for the bands we're going to be folding from. */ - if (lowband_offset != 0 && (s->spread != CELT_SPREAD_AGGRESSIVE || - s->blocks > 1 || s->tf_change[i] < 0)) { + if (lowband_offset != 0 && (f->spread != CELT_SPREAD_AGGRESSIVE || + f->blocks > 1 || f->tf_change[i] < 0)) { int foldstart, foldend; /* This ensures we never repeat spectral content within one band */ - effective_lowband = FFMAX(ff_celt_freq_bands[s->startband], + effective_lowband = FFMAX(ff_celt_freq_bands[f->start_band], ff_celt_freq_bands[lowband_offset] - ff_celt_freq_range[i]); foldstart = lowband_offset; while (ff_celt_freq_bands[--foldstart] > effective_lowband); @@ -741,260 +741,260 @@ static void celt_decode_bands(CeltContext *s, OpusRangeCoder *rc) cm[0] = cm[1] = 0; for (j = foldstart; j < foldend; j++) { - cm[0] |= s->frame[0].collapse_masks[j]; - cm[1] |= s->frame[s->coded_channels - 1].collapse_masks[j]; + cm[0] |= f->block[0].collapse_masks[j]; + cm[1] |= f->block[f->channels - 1].collapse_masks[j]; } } else /* Otherwise, we'll be using the LCG to fold, so all blocks will (almost always) be non-zero.*/ - cm[0] = cm[1] = (1 << s->blocks) - 1; + cm[0] = cm[1] = (1 << f->blocks) - 1; - if (s->dualstereo && i == s->intensitystereo) { + if (f->dual_stereo && i == f->intensity_stereo) { /* Switch off dual stereo to do intensity */ - s->dualstereo = 0; - for (j = ff_celt_freq_bands[s->startband] << s->duration; j < band_offset; j++) + f->dual_stereo = 0; + for (j = ff_celt_freq_bands[f->start_band] << f->size; j < band_offset; j++) norm[j] = (norm[j] + norm2[j]) / 2; } - if (s->dualstereo) { - cm[0] = ff_celt_decode_band(s, rc, i, X, NULL, band_size, b / 2, s->blocks, - effective_lowband != -1 ? norm + (effective_lowband << s->duration) : NULL, s->duration, + if (f->dual_stereo) { + cm[0] = ff_celt_decode_band(f, rc, i, X, NULL, band_size, b / 2, f->blocks, + effective_lowband != -1 ? norm + (effective_lowband << f->size) : NULL, f->size, norm + band_offset, 0, 1.0f, lowband_scratch, cm[0]); - cm[1] = ff_celt_decode_band(s, rc, i, Y, NULL, band_size, b/2, s->blocks, - effective_lowband != -1 ? norm2 + (effective_lowband << s->duration) : NULL, s->duration, + cm[1] = ff_celt_decode_band(f, rc, i, Y, NULL, band_size, b/2, f->blocks, + effective_lowband != -1 ? norm2 + (effective_lowband << f->size) : NULL, f->size, norm2 + band_offset, 0, 1.0f, lowband_scratch, cm[1]); } else { - cm[0] = ff_celt_decode_band(s, rc, i, X, Y, band_size, b, s->blocks, - effective_lowband != -1 ? norm + (effective_lowband << s->duration) : NULL, s->duration, + cm[0] = ff_celt_decode_band(f, rc, i, X, Y, band_size, b, f->blocks, + effective_lowband != -1 ? norm + (effective_lowband << f->size) : NULL, f->size, norm + band_offset, 0, 1.0f, lowband_scratch, cm[0]|cm[1]); cm[1] = cm[0]; } - s->frame[0].collapse_masks[i] = (uint8_t)cm[0]; - s->frame[s->coded_channels - 1].collapse_masks[i] = (uint8_t)cm[1]; - s->remaining += s->pulses[i] + consumed; + f->block[0].collapse_masks[i] = (uint8_t)cm[0]; + f->block[f->channels - 1].collapse_masks[i] = (uint8_t)cm[1]; + f->remaining += f->pulses[i] + consumed; /* Update the folding position only as long as we have 1 bit/sample depth */ update_lowband = (b > band_size << 3); } } -int ff_celt_decode_frame(CeltContext *s, OpusRangeCoder *rc, - float **output, int coded_channels, int frame_size, - int startband, int endband) +int ff_celt_decode_frame(CeltFrame *f, OpusRangeCoder *rc, + float **output, int channels, int frame_size, + int start_band, int end_band) { int i, j; - int consumed; // bits of entropy consumed thus far for this frame - int silence = 0; - int transient = 0; - int anticollapse = 0; MDCT15Context *imdct; float imdct_scale = 1.0; - if (coded_channels != 1 && coded_channels != 2) { - av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coded channels: %d\n", - coded_channels); + if (channels != 1 && channels != 2) { + av_log(f->avctx, AV_LOG_ERROR, "Invalid number of coded channels: %d\n", + channels); return AVERROR_INVALIDDATA; } - if (startband < 0 || startband > endband || endband > CELT_MAX_BANDS) { - av_log(s->avctx, AV_LOG_ERROR, "Invalid start/end band: %d %d\n", - startband, endband); + if (start_band < 0 || start_band > end_band || end_band > CELT_MAX_BANDS) { + av_log(f->avctx, AV_LOG_ERROR, "Invalid start/end band: %d %d\n", + start_band, end_band); return AVERROR_INVALIDDATA; } - s->flushed = 0; - s->coded_channels = coded_channels; - s->startband = startband; - s->endband = endband; - s->framebits = rc->rb.bytes * 8; + f->silence = 0; + f->transient = 0; + f->anticollapse = 0; + f->flushed = 0; + f->channels = channels; + f->start_band = start_band; + f->end_band = end_band; + f->framebits = rc->rb.bytes * 8; - s->duration = av_log2(frame_size / CELT_SHORT_BLOCKSIZE); - if (s->duration > CELT_MAX_LOG_BLOCKS || - frame_size != CELT_SHORT_BLOCKSIZE * (1 << s->duration)) { - av_log(s->avctx, AV_LOG_ERROR, "Invalid CELT frame size: %d\n", + f->size = av_log2(frame_size / CELT_SHORT_BLOCKSIZE); + if (f->size > CELT_MAX_LOG_BLOCKS || + frame_size != CELT_SHORT_BLOCKSIZE * (1 << f->size)) { + av_log(f->avctx, AV_LOG_ERROR, "Invalid CELT frame size: %d\n", frame_size); return AVERROR_INVALIDDATA; } - if (!s->output_channels) - s->output_channels = coded_channels; + if (!f->output_channels) + f->output_channels = channels; - memset(s->frame[0].collapse_masks, 0, sizeof(s->frame[0].collapse_masks)); - memset(s->frame[1].collapse_masks, 0, sizeof(s->frame[1].collapse_masks)); + memset(f->block[0].collapse_masks, 0, sizeof(f->block[0].collapse_masks)); + memset(f->block[1].collapse_masks, 0, sizeof(f->block[1].collapse_masks)); consumed = opus_rc_tell(rc); /* obtain silence flag */ - if (consumed >= s->framebits) - silence = 1; + if (consumed >= f->framebits) + f->silence = 1; else if (consumed == 1) - silence = ff_opus_rc_dec_log(rc, 15); + f->silence = ff_opus_rc_dec_log(rc, 15); - if (silence) { - consumed = s->framebits; - rc->total_bits += s->framebits - opus_rc_tell(rc); + if (f->silence) { + consumed = f->framebits; + rc->total_bits += f->framebits - opus_rc_tell(rc); } /* obtain post-filter options */ - consumed = parse_postfilter(s, rc, consumed); + consumed = parse_postfilter(f, rc, consumed); /* obtain transient flag */ - if (s->duration != 0 && consumed+3 <= s->framebits) - transient = ff_opus_rc_dec_log(rc, 3); + if (f->size != 0 && consumed+3 <= f->framebits) + f->transient = ff_opus_rc_dec_log(rc, 3); - s->blocks = transient ? 1 << s->duration : 1; - s->blocksize = frame_size / s->blocks; + f->blocks = f->transient ? 1 << f->size : 1; + f->blocksize = frame_size / f->blocks; - imdct = s->imdct[transient ? 0 : s->duration]; + imdct = f->imdct[f->transient ? 0 : f->size]; - if (coded_channels == 1) { + if (channels == 1) { for (i = 0; i < CELT_MAX_BANDS; i++) - s->frame[0].energy[i] = FFMAX(s->frame[0].energy[i], s->frame[1].energy[i]); + f->block[0].energy[i] = FFMAX(f->block[0].energy[i], f->block[1].energy[i]); } - celt_decode_coarse_energy(s, rc); - celt_decode_tf_changes (s, rc, transient); - celt_decode_allocation (s, rc); - celt_decode_fine_energy (s, rc); - celt_decode_bands (s, rc); + celt_decode_coarse_energy(f, rc); + celt_decode_tf_changes (f, rc); + celt_decode_allocation (f, rc); + celt_decode_fine_energy (f, rc); + celt_decode_bands (f, rc); - if (s->anticollapse_bit) - anticollapse = ff_opus_rc_get_raw(rc, 1); + if (f->anticollapse_needed) + f->anticollapse = ff_opus_rc_get_raw(rc, 1); - celt_decode_final_energy(s, rc, s->framebits - opus_rc_tell(rc)); + celt_decode_final_energy(f, rc); /* apply anti-collapse processing and denormalization to * each coded channel */ - for (i = 0; i < s->coded_channels; i++) { - CeltFrame *frame = &s->frame[i]; + for (i = 0; i < f->channels; i++) { + CeltBlock *block = &f->block[i]; - if (anticollapse) - process_anticollapse(s, frame, s->coeffs[i]); + if (f->anticollapse) + process_anticollapse(f, block, f->block[i].coeffs); - celt_denormalize(s, frame, s->coeffs[i]); + celt_denormalize(f, block, f->block[i].coeffs); } /* stereo -> mono downmix */ - if (s->output_channels < s->coded_channels) { - s->dsp->vector_fmac_scalar(s->coeffs[0], s->coeffs[1], 1.0, FFALIGN(frame_size, 16)); + if (f->output_channels < f->channels) { + f->dsp->vector_fmac_scalar(f->block[0].coeffs, f->block[1].coeffs, 1.0, FFALIGN(frame_size, 16)); imdct_scale = 0.5; - } else if (s->output_channels > s->coded_channels) - memcpy(s->coeffs[1], s->coeffs[0], frame_size * sizeof(float)); + } else if (f->output_channels > f->channels) + memcpy(f->block[1].coeffs, f->block[0].coeffs, frame_size * sizeof(float)); - if (silence) { + if (f->silence) { for (i = 0; i < 2; i++) { - CeltFrame *frame = &s->frame[i]; + CeltBlock *block = &f->block[i]; - for (j = 0; j < FF_ARRAY_ELEMS(frame->energy); j++) - frame->energy[j] = CELT_ENERGY_SILENCE; + for (j = 0; j < FF_ARRAY_ELEMS(block->energy); j++) + block->energy[j] = CELT_ENERGY_SILENCE; } - memset(s->coeffs, 0, sizeof(s->coeffs)); + memset(f->block[0].coeffs, 0, sizeof(f->block[0].coeffs)); + memset(f->block[1].coeffs, 0, sizeof(f->block[1].coeffs)); } /* transform and output for each output channel */ - for (i = 0; i < s->output_channels; i++) { - CeltFrame *frame = &s->frame[i]; - float m = frame->deemph_coeff; + for (i = 0; i < f->output_channels; i++) { + CeltBlock *block = &f->block[i]; + float m = block->emph_coeff; /* iMDCT and overlap-add */ - for (j = 0; j < s->blocks; j++) { - float *dst = frame->buf + 1024 + j * s->blocksize; + for (j = 0; j < f->blocks; j++) { + float *dst = block->buf + 1024 + j * f->blocksize; - imdct->imdct_half(imdct, dst + CELT_OVERLAP / 2, s->coeffs[i] + j, - s->blocks, imdct_scale); - s->dsp->vector_fmul_window(dst, dst, dst + CELT_OVERLAP / 2, + imdct->imdct_half(imdct, dst + CELT_OVERLAP / 2, f->block[i].coeffs + j, + f->blocks, imdct_scale); + f->dsp->vector_fmul_window(dst, dst, dst + CELT_OVERLAP / 2, ff_celt_window, CELT_OVERLAP / 2); } /* postfilter */ - celt_postfilter(s, frame); + celt_postfilter(f, block); /* deemphasis and output scaling */ for (j = 0; j < frame_size; j++) { - float tmp = frame->buf[1024 - frame_size + j] + m; - m = tmp * CELT_DEEMPH_COEFF; + float tmp = block->buf[1024 - frame_size + j] + m; + m = tmp * CELT_EMPH_COEFF; output[i][j] = tmp / 32768.; } - frame->deemph_coeff = m; + block->emph_coeff = m; } - if (coded_channels == 1) - memcpy(s->frame[1].energy, s->frame[0].energy, sizeof(s->frame[0].energy)); + if (channels == 1) + memcpy(f->block[1].energy, f->block[0].energy, sizeof(f->block[0].energy)); for (i = 0; i < 2; i++ ) { - CeltFrame *frame = &s->frame[i]; + CeltBlock *block = &f->block[i]; - if (!transient) { - memcpy(frame->prev_energy[1], frame->prev_energy[0], sizeof(frame->prev_energy[0])); - memcpy(frame->prev_energy[0], frame->energy, sizeof(frame->prev_energy[0])); + if (!f->transient) { + memcpy(block->prev_energy[1], block->prev_energy[0], sizeof(block->prev_energy[0])); + memcpy(block->prev_energy[0], block->energy, sizeof(block->prev_energy[0])); } else { for (j = 0; j < CELT_MAX_BANDS; j++) - frame->prev_energy[0][j] = FFMIN(frame->prev_energy[0][j], frame->energy[j]); + block->prev_energy[0][j] = FFMIN(block->prev_energy[0][j], block->energy[j]); } - for (j = 0; j < s->startband; j++) { - frame->prev_energy[0][j] = CELT_ENERGY_SILENCE; - frame->energy[j] = 0.0; + for (j = 0; j < f->start_band; j++) { + block->prev_energy[0][j] = CELT_ENERGY_SILENCE; + block->energy[j] = 0.0; } - for (j = s->endband; j < CELT_MAX_BANDS; j++) { - frame->prev_energy[0][j] = CELT_ENERGY_SILENCE; - frame->energy[j] = 0.0; + for (j = f->end_band; j < CELT_MAX_BANDS; j++) { + block->prev_energy[0][j] = CELT_ENERGY_SILENCE; + block->energy[j] = 0.0; } } - s->seed = rc->range; + f->seed = rc->range; return 0; } -void ff_celt_flush(CeltContext *s) +void ff_celt_flush(CeltFrame *f) { int i, j; - if (s->flushed) + if (f->flushed) return; for (i = 0; i < 2; i++) { - CeltFrame *frame = &s->frame[i]; + CeltBlock *block = &f->block[i]; for (j = 0; j < CELT_MAX_BANDS; j++) - frame->prev_energy[0][j] = frame->prev_energy[1][j] = CELT_ENERGY_SILENCE; + block->prev_energy[0][j] = block->prev_energy[1][j] = CELT_ENERGY_SILENCE; - memset(frame->energy, 0, sizeof(frame->energy)); - memset(frame->buf, 0, sizeof(frame->buf)); + memset(block->energy, 0, sizeof(block->energy)); + memset(block->buf, 0, sizeof(block->buf)); - memset(frame->pf_gains, 0, sizeof(frame->pf_gains)); - memset(frame->pf_gains_old, 0, sizeof(frame->pf_gains_old)); - memset(frame->pf_gains_new, 0, sizeof(frame->pf_gains_new)); + memset(block->pf_gains, 0, sizeof(block->pf_gains)); + memset(block->pf_gains_old, 0, sizeof(block->pf_gains_old)); + memset(block->pf_gains_new, 0, sizeof(block->pf_gains_new)); - frame->deemph_coeff = 0.0; + block->emph_coeff = 0.0; } - s->seed = 0; + f->seed = 0; - s->flushed = 1; + f->flushed = 1; } -void ff_celt_free(CeltContext **ps) +void ff_celt_free(CeltFrame **f) { - CeltContext *s = *ps; + CeltFrame *frm = *f; int i; - if (!s) + if (!frm) return; - for (i = 0; i < FF_ARRAY_ELEMS(s->imdct); i++) - ff_mdct15_uninit(&s->imdct[i]); + for (i = 0; i < FF_ARRAY_ELEMS(frm->imdct); i++) + ff_mdct15_uninit(&frm->imdct[i]); - av_freep(&s->dsp); - av_freep(ps); + av_freep(&frm->dsp); + av_freep(f); } -int ff_celt_init(AVCodecContext *avctx, CeltContext **ps, int output_channels) +int ff_celt_init(AVCodecContext *avctx, CeltFrame **f, int output_channels) { - CeltContext *s; + CeltFrame *frm; int i, ret; if (output_channels != 1 && output_channels != 2) { @@ -1003,31 +1003,31 @@ int ff_celt_init(AVCodecContext *avctx, CeltContext **ps, int output_channels) return AVERROR(EINVAL); } - s = av_mallocz(sizeof(*s)); - if (!s) + frm = av_mallocz(sizeof(*frm)); + if (!frm) return AVERROR(ENOMEM); - s->avctx = avctx; - s->output_channels = output_channels; + frm->avctx = avctx; + frm->output_channels = output_channels; - for (i = 0; i < FF_ARRAY_ELEMS(s->imdct); i++) { - ret = ff_mdct15_init(&s->imdct[i], 1, i + 3, -1.0f); + for (i = 0; i < FF_ARRAY_ELEMS(frm->imdct); i++) { + ret = ff_mdct15_init(&frm->imdct[i], 1, i + 3, -1.0f); if (ret < 0) goto fail; } - s->dsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT); - if (!s->dsp) { + frm->dsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT); + if (!frm->dsp) { ret = AVERROR(ENOMEM); goto fail; } - ff_celt_flush(s); + ff_celt_flush(frm); - *ps = s; + *f = frm; return 0; fail: - ff_celt_free(&s); + ff_celt_free(&frm); return ret; } diff --git a/libavcodec/opus_celt.h b/libavcodec/opus_celt.h index e9b5946642..7b46a5116b 100644 --- a/libavcodec/opus_celt.h +++ b/libavcodec/opus_celt.h @@ -24,6 +24,8 @@ #ifndef AVCODEC_OPUS_CELT_H #define AVCODEC_OPUS_CELT_H +#include + #include "opus.h" #include "mdct15.h" @@ -37,7 +39,7 @@ #define CELT_NORM_SCALE 16384 #define CELT_QTHETA_OFFSET 4 #define CELT_QTHETA_OFFSET_TWOPHASE 16 -#define CELT_DEEMPH_COEFF 0.85000610f +#define CELT_EMPH_COEFF 0.85000610f #define CELT_POSTFILTER_MINPERIOD 15 #define CELT_ENERGY_SILENCE (-28.0f) @@ -48,7 +50,16 @@ enum CeltSpread { CELT_SPREAD_AGGRESSIVE }; -typedef struct CeltFrame { +enum CeltBlockSize { + CELT_BLOCK_120, + CELT_BLOCK_240, + CELT_BLOCK_480, + CELT_BLOCK_960, + + CELT_BLOCK_NB +}; + +typedef struct CeltBlock { float energy[CELT_MAX_BANDS]; float prev_energy[2][CELT_MAX_BANDS]; @@ -56,50 +67,46 @@ typedef struct CeltFrame { /* buffer for mdct output + postfilter */ DECLARE_ALIGNED(32, float, buf)[2048]; + DECLARE_ALIGNED(32, float, coeffs)[CELT_MAX_FRAME_SIZE]; /* postfilter parameters */ - int pf_period_new; + int pf_period_new; float pf_gains_new[3]; - int pf_period; + int pf_period; float pf_gains[3]; - int pf_period_old; + int pf_period_old; float pf_gains_old[3]; - float deemph_coeff; -} CeltFrame; + float emph_coeff; +} CeltBlock; -struct CeltContext { +struct CeltFrame { // constant values that do not change during context lifetime - AVCodecContext *avctx; - MDCT15Context *imdct[4]; - AVFloatDSPContext *dsp; + AVCodecContext *avctx; + MDCT15Context *imdct[4]; + AVFloatDSPContext *dsp; + CeltBlock block[2]; + int channels; int output_channels; - // values that have inter-frame effect and must be reset on flush - CeltFrame frame[2]; - uint32_t seed; + enum CeltBlockSize size; + int start_band; + int end_band; + int coded_bands; + int transient; + int blocks; /* number of iMDCT blocks in the frame, depends on transient */ + int blocksize; /* size of each block */ + int silence; /* Frame is filled with silence */ + int anticollapse_needed; /* Whether to expect an anticollapse bit */ + int anticollapse; /* Encoded anticollapse bit */ + int intensity_stereo; + int dual_stereo; int flushed; - - // values that only affect a single frame - int coded_channels; - int framebits; - int duration; - - /* number of iMDCT blocks in the frame */ - int blocks; - /* size of each block */ - int blocksize; - - int startband; - int endband; - int codedbands; - - int anticollapse_bit; - - int intensitystereo; - int dualstereo; + uint32_t seed; enum CeltSpread spread; + /* Bit allocation */ + int framebits; int remaining; int remaining2; int fine_bits [CELT_MAX_BANDS]; @@ -107,15 +114,14 @@ struct CeltContext { int pulses [CELT_MAX_BANDS]; int tf_change [CELT_MAX_BANDS]; - DECLARE_ALIGNED(32, float, coeffs)[2][CELT_MAX_FRAME_SIZE]; DECLARE_ALIGNED(32, float, scratch)[22 * 8]; // MAX(ff_celt_freq_range) * 1<seed = 1664525 * s->seed + 1013904223; - return s->seed; + f->seed = 1664525 * f->seed + 1013904223; + return f->seed; } static av_always_inline void celt_renormalize_vector(float *X, int N, float gain) @@ -130,4 +136,13 @@ static av_always_inline void celt_renormalize_vector(float *X, int N, float gain X[i] *= g; } +int ff_celt_init(AVCodecContext *avctx, CeltFrame **f, int output_channels); + +void ff_celt_free(CeltFrame **f); + +void ff_celt_flush(CeltFrame *f); + +int ff_celt_decode_frame(CeltFrame *f, OpusRangeCoder *rc, float **output, + int coded_channels, int frame_size, int startband, int endband); + #endif /* AVCODEC_OPUS_CELT_H */ diff --git a/libavcodec/opus_pvq.c b/libavcodec/opus_pvq.c index b4e23c86b8..ddc5fc2895 100644 --- a/libavcodec/opus_pvq.c +++ b/libavcodec/opus_pvq.c @@ -375,7 +375,7 @@ static uint32_t celt_alg_unquant(OpusRangeCoder *rc, float *X, uint32_t N, uint3 return celt_extract_collapse_mask(y, N, blocks); } -uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, +uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, float *X, float *Y, int N, int b, uint32_t blocks, float *lowband, int duration, float *lowband_out, int level, float gain, float *lowband_scratch, int fill) @@ -403,9 +403,9 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, float *x = X; for (i = 0; i <= dualstereo; i++) { int sign = 0; - if (s->remaining2 >= 1<<3) { + if (f->remaining2 >= 1<<3) { sign = ff_opus_rc_get_raw(rc, 1); - s->remaining2 -= 1 << 3; + f->remaining2 -= 1 << 3; b -= 1 << 3; } x[0] = sign ? -1.0f : 1.0f; @@ -417,7 +417,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, } if (!dualstereo && level == 0) { - int tf_change = s->tf_change[band]; + int tf_change = f->tf_change[band]; int k; if (tf_change > 0) recombine = tf_change; @@ -454,7 +454,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, /* Reorganize the samples in time order instead of frequency order */ if (B0 > 1 && lowband) - celt_deinterleave_hadamard(s->scratch, lowband, N_B >> recombine, + celt_deinterleave_hadamard(f->scratch, lowband, N_B >> recombine, B0 << recombine, longblocks); } @@ -485,7 +485,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, pulse_cap = ff_celt_log_freq_range[band] + duration * 8; offset = (pulse_cap >> 1) - (dualstereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE : CELT_QTHETA_OFFSET); - qn = (dualstereo && band >= s->intensitystereo) ? 1 : + qn = (dualstereo && band >= f->intensity_stereo) ? 1 : celt_compute_qn(N, b, offset, pulse_cap, dualstereo); tell = opus_rc_tell_frac(rc); if (qn != 1) { @@ -501,7 +501,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. Let's do that at higher complexity */ } else if (dualstereo) { - inv = (b > 2 << 3 && s->remaining2 > 2 << 3) ? ff_opus_rc_dec_log(rc, 2) : 0; + inv = (b > 2 << 3 && f->remaining2 > 2 << 3) ? ff_opus_rc_dec_log(rc, 2) : 0; itheta = 0; } qalloc = opus_rc_tell_frac(rc) - tell; @@ -542,7 +542,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, sbits = (itheta != 0 && itheta != 16384) ? 1 << 3 : 0; mbits -= sbits; c = (itheta > 8192); - s->remaining2 -= qalloc+sbits; + f->remaining2 -= qalloc+sbits; x2 = c ? Y : X; y2 = c ? X : Y; @@ -551,7 +551,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, sign = 1 - 2 * sign; /* We use orig_fill here because we want to fold the side, but if itheta==16384, we'll have cleared the low bits of fill. */ - cm = ff_celt_decode_band(s, rc, band, x2, NULL, N, mbits, blocks, + cm = ff_celt_decode_band(f, rc, band, x2, NULL, N, mbits, blocks, lowband, duration, lowband_out, level, gain, lowband_scratch, orig_fill); /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), @@ -588,7 +588,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, } mbits = av_clip((b - delta) / 2, 0, b); sbits = b - mbits; - s->remaining2 -= qalloc; + f->remaining2 -= qalloc; if (lowband && !dualstereo) next_lowband2 = lowband + N; /* >32-bit split case */ @@ -600,40 +600,40 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, else next_level = level + 1; - rebalance = s->remaining2; + rebalance = f->remaining2; if (mbits >= sbits) { /* In stereo mode, we do not apply a scaling to the mid * because we need the normalized mid for folding later */ - cm = ff_celt_decode_band(s, rc, band, X, NULL, N, mbits, blocks, + cm = ff_celt_decode_band(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, next_level, dualstereo ? 1.0f : (gain * mid), lowband_scratch, fill); - rebalance = mbits - (rebalance - s->remaining2); + rebalance = mbits - (rebalance - f->remaining2); if (rebalance > 3 << 3 && itheta != 0) sbits += rebalance - (3 << 3); /* For a stereo split, the high bits of fill are always zero, * so no folding will be done to the side. */ - cm |= ff_celt_decode_band(s, rc, band, Y, NULL, N, sbits, blocks, + cm |= ff_celt_decode_band(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, fill >> blocks) << ((B0 >> 1) & (dualstereo - 1)); } else { /* For a stereo split, the high bits of fill are always zero, * so no folding will be done to the side. */ - cm = ff_celt_decode_band(s, rc, band, Y, NULL, N, sbits, blocks, + cm = ff_celt_decode_band(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, fill >> blocks) << ((B0 >> 1) & (dualstereo - 1)); - rebalance = sbits - (rebalance - s->remaining2); + rebalance = sbits - (rebalance - f->remaining2); if (rebalance > 3 << 3 && itheta != 16384) mbits += rebalance - (3 << 3); /* In stereo mode, we do not apply a scaling to the mid because * we need the normalized mid for folding later */ - cm |= ff_celt_decode_band(s, rc, band, X, NULL, N, mbits, blocks, + cm |= ff_celt_decode_band(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, next_level, dualstereo ? 1.0f : (gain * mid), lowband_scratch, fill); @@ -643,19 +643,19 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, /* This is the basic no-split case */ uint32_t q = celt_bits2pulses(cache, b); uint32_t curr_bits = celt_pulses2bits(cache, q); - s->remaining2 -= curr_bits; + f->remaining2 -= curr_bits; /* Ensures we can never bust the budget */ - while (s->remaining2 < 0 && q > 0) { - s->remaining2 += curr_bits; + while (f->remaining2 < 0 && q > 0) { + f->remaining2 += curr_bits; curr_bits = celt_pulses2bits(cache, --q); - s->remaining2 -= curr_bits; + f->remaining2 -= curr_bits; } if (q != 0) { /* Finally do the actual quantization */ cm = celt_alg_unquant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1), - s->spread, blocks, gain); + f->spread, blocks, gain); } else { /* If there's no pulse, fill the band anyway */ int j; @@ -668,13 +668,13 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, if (!lowband) { /* Noise */ for (j = 0; j < N; j++) - X[j] = (((int32_t)celt_rng(s)) >> 20); + X[j] = (((int32_t)celt_rng(f)) >> 20); cm = cm_mask; } else { /* Folded spectrum */ for (j = 0; j < N; j++) { /* About 48 dB below the "normal" folding level */ - X[j] = lowband[j] + (((celt_rng(s)) & 0x8000) ? 1.0f / 256 : -1.0f / 256); + X[j] = lowband[j] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256); } cm = fill; } @@ -697,7 +697,7 @@ uint32_t ff_celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, /* Undo the sample reorganization going from time order to frequency order */ if (B0 > 1) - celt_interleave_hadamard(s->scratch, X, N_B>>recombine, + celt_interleave_hadamard(f->scratch, X, N_B>>recombine, B0<celt, &s->redundancy_rc, s->redundancy_output, s->packet.stereo + 1, 240, - 0, celt_band_end[s->packet.bandwidth]); + 0, ff_celt_band_end[s->packet.bandwidth]); if (ret < 0) goto fail; @@ -279,7 +278,7 @@ static int opus_decode_frame(OpusStreamContext *s, const uint8_t *data, int size s->packet.stereo + 1, s->packet.frame_duration, (s->packet.mode == OPUS_MODE_HYBRID) ? 17 : 0, - celt_band_end[s->packet.bandwidth]); + ff_celt_band_end[s->packet.bandwidth]); if (ret < 0) return ret; diff --git a/libavcodec/opustab.c b/libavcodec/opustab.c index cfe56ba94f..02d548e71a 100644 --- a/libavcodec/opustab.c +++ b/libavcodec/opustab.c @@ -22,6 +22,10 @@ #include "opustab.h" +const uint8_t ff_opus_default_coupled_streams[] = { 0, 1, 1, 2, 2, 2, 2, 3 }; + +const uint8_t ff_celt_band_end[] = { 13, 17, 17, 19, 21 }; + const uint16_t ff_silk_model_stereo_s1[] = { 256, 7, 9, 10, 11, 12, 22, 46, 54, 55, 56, 59, 82, 174, 197, 200, 201, 202, 210, 234, 244, 245, 246, 247, 249, 256 diff --git a/libavcodec/opustab.h b/libavcodec/opustab.h index e6527c0270..b6be073659 100644 --- a/libavcodec/opustab.h +++ b/libavcodec/opustab.h @@ -27,6 +27,10 @@ #include +extern const uint8_t ff_celt_band_end[]; + +extern const uint8_t ff_opus_default_coupled_streams[]; + extern const uint16_t ff_silk_model_stereo_s1[]; extern const uint16_t ff_silk_model_stereo_s2[]; extern const uint16_t ff_silk_model_stereo_s3[];