August 2015 - ffmpeg-cvslog - lists.ffmpeg.org

fate: add a test for encoding AAC-Main prediction
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 07:00:54 2015 +0100| [4ff897a5364f85de051ca81058554454f3ad5d62] | committer: Rostislav Pehlivanov fate: add a test for encoding AAC-Main prediction This commit introduces a test for AAC-Main prediction which was just reworked in this series of commits. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=4ff897a5364f85de0… --- tests/fate/aac.mak | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/tests/fate/aac.mak b/tests/fate/aac.mak index a28976e..d04fcc2 100644 --- a/tests/fate/aac.mak +++ b/tests/fate/aac.mak @@ -182,6 +182,14 @@ fate-aac-is-encode: CMP_SHIFT = -4096 fate-aac-is-encode: CMP_TARGET = 613.26 fate-aac-is-encode: SIZE_TOLERANCE = 3560 +FATE_AAC_ENCODE += fate-aac-pred-encode +fate-aac-pred-encode: CMD = enc_dec_pcm adts wav s16le $(TARGET_SAMPLES)/audio-reference/luckynight_2ch_44kHz_s16.wav -strict -2 -profile:a aac_main -c:a aac -aac_pred 1 -b:a 128k +fate-aac-pred-encode: CMP = stddev +fate-aac-pred-encode: REF = $(SAMPLES)/audio-reference/luckynight_2ch_44kHz_s16.wav +fate-aac-pred-encode: CMP_SHIFT = -4096 +fate-aac-pred-encode: CMP_TARGET = 652.60 +fate-aac-pred-encode: SIZE_TOLERANCE = 3560 + FATE_AAC_LATM += fate-aac-latm_000000001180bc60 fate-aac-latm_000000001180bc60: CMD = pcm -i $(TARGET_SAMPLES)/aac/latm_000000001180bc60.mpg fate-aac-latm_000000001180bc60: REF = $(SAMPLES)/aac/latm_000000001180bc60.s16

1 0

aacenc_pred: rework the way prediction is done
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 06:34:08 2015 +0100| [44ddee945a2e8cfc1b7074de6e35595ed41da4e4] | committer: Rostislav Pehlivanov aacenc_pred: rework the way prediction is done This commit completely alters the algorithm of prediction. The original commit which introduced prediction was completely incorrect to even remotely care about what the actual coefficients contain or whether any options were enabled. Not my actual fault. This commit treats prediction the way the decoder does and expects to do: like lossy encryption. Everything related to prediction now happens at the very end but just before quantization and encoding of coefficients. On the decoder side, prediction happens before anything has had a chance to even access the coefficients. Also the original implementation had problems because it actually touched the band_type of special bands which already had their scalefactor indices marked and it's a wonder the asserion wasn't triggered when transmitting those. Overall, this now drastically increases audio quality and you should think about enabling it if you don't plan on playing anything encoded on really old low power ultra-embedded devices since they might not support decoding of prediction or AAC-Main. Though the specifications were written ages ago and as times change so do the FLOPS. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=44ddee945a2e8cfc1… --- libavcodec/aac.h | 2 +- libavcodec/aaccoder.c | 4 - libavcodec/aacenc.c | 41 +++--- libavcodec/aacenc.h | 1 - libavcodec/aacenc_pred.c | 361 ++++++++++++++++++++++------------------------ libavcodec/aacenc_pred.h | 6 +- 6 files changed, 190 insertions(+), 225 deletions(-) diff --git a/libavcodec/aac.h b/libavcodec/aac.h index dc6b439..aa4b53b 100644 --- a/libavcodec/aac.h +++ b/libavcodec/aac.h @@ -247,7 +247,7 @@ typedef struct SingleChannelElement { TemporalNoiseShaping tns; Pulse pulse; enum BandType band_type[128]; ///< band types - enum BandType orig_band_type[128]; ///< band type backups for undoing prediction + enum BandType band_alt[128]; ///< alternative band type (used by encoder) int band_type_run_end[120]; ///< band type run end points INTFLOAT sf[120]; ///< scalefactors int sf_idx[128]; ///< scalefactor indices (used by encoder) diff --git a/libavcodec/aaccoder.c b/libavcodec/aaccoder.c index c273c54..8256f96 100644 --- a/libavcodec/aaccoder.c +++ b/libavcodec/aaccoder.c @@ -964,7 +964,6 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, - ff_aac_update_main_pred, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, @@ -980,7 +979,6 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, - ff_aac_update_main_pred, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, @@ -996,7 +994,6 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, - ff_aac_update_main_pred, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, @@ -1012,7 +1009,6 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, - ff_aac_update_main_pred, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, diff --git a/libavcodec/aacenc.c b/libavcodec/aacenc.c index ece2328..a7c43c7 100644 --- a/libavcodec/aacenc.c +++ b/libavcodec/aacenc.c @@ -354,15 +354,15 @@ static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce) start += sce->ics.swb_sizes[i]; continue; } - for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) + for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) { s->coder->quantize_and_encode_band(s, &s->pb, &sce->coeffs[start + w2*128], - &sce->pqcoeffs[start + w2*128], - sce->ics.swb_sizes[i], + NULL, sce->ics.swb_sizes[i], sce->sf_idx[w*16 + i], sce->band_type[w*16 + i], s->lambda, sce->ics.window_clipping[w]); + } start += sce->ics.swb_sizes[i]; } } @@ -609,12 +609,8 @@ static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, s->coder->search_for_pns(s, avctx, sce); if (s->options.tns && s->coder->search_for_tns) s->coder->search_for_tns(s, sce); - if (s->options.pred && s->coder->search_for_pred) - s->coder->search_for_pred(s, sce); if (sce->tns.present) tns_mode = 1; - if (sce->ics.predictor_present) - pred_mode = 1; } s->cur_channel = start_ch; if (s->options.stereo_mode && cpe->common_window) { @@ -631,15 +627,26 @@ static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, s->coder->search_for_is(s, avctx, cpe); if (cpe->is_mode) is_mode = 1; } - if (s->options.pred && s->coder->adjust_common_prediction) - s->coder->adjust_common_prediction(s, cpe); if (s->coder->set_special_band_scalefactors) for (ch = 0; ch < chans; ch++) s->coder->set_special_band_scalefactors(s, &cpe->ch[ch]); - if (s->options.pred && s->coder->apply_main_pred) - for (ch = 0; ch < chans; ch++) - s->coder->apply_main_pred(s, &cpe->ch[ch]); adjust_frame_information(cpe, chans); + for (ch = 0; ch < chans; ch++) { + sce = &cpe->ch[ch]; + s->cur_channel = start_ch + ch; + if (s->options.pred && s->coder->search_for_pred) + s->coder->search_for_pred(s, sce); + if (cpe->ch[ch].ics.predictor_present) pred_mode = 1; + } + if (s->options.pred && s->coder->adjust_common_prediction) + s->coder->adjust_common_prediction(s, cpe); + for (ch = 0; ch < chans; ch++) { + sce = &cpe->ch[ch]; + s->cur_channel = start_ch + ch; + if (s->options.pred && s->coder->apply_main_pred) + s->coder->apply_main_pred(s, sce); + } + s->cur_channel = start_ch; if (chans == 2) { put_bits(&s->pb, 1, cpe->common_window); if (cpe->common_window) { @@ -676,16 +683,6 @@ static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, } while (1); - // update predictor state - if (s->options.pred && s->coder->update_main_pred) { - for (i = 0; i < s->chan_map[0]; i++) { - cpe = &s->cpe[i]; - for (ch = 0; ch < chans; ch++) - s->coder->update_main_pred(s, &cpe->ch[ch], - (cpe->common_window && !ch) ? cpe : NULL); - } - } - put_bits(&s->pb, 3, TYPE_END); flush_put_bits(&s->pb); avctx->frame_bits = put_bits_count(&s->pb); diff --git a/libavcodec/aacenc.h b/libavcodec/aacenc.h index 9ee854f..69a8c01 100644 --- a/libavcodec/aacenc.h +++ b/libavcodec/aacenc.h @@ -63,7 +63,6 @@ typedef struct AACCoefficientsEncoder { void (*encode_main_pred)(struct AACEncContext *s, SingleChannelElement *sce); void (*adjust_common_prediction)(struct AACEncContext *s, ChannelElement *cpe); void (*apply_main_pred)(struct AACEncContext *s, SingleChannelElement *sce); - void (*update_main_pred)(struct AACEncContext *s, SingleChannelElement *sce, ChannelElement *cpe); void (*set_special_band_scalefactors)(struct AACEncContext *s, SingleChannelElement *sce); void (*search_for_pns)(struct AACEncContext *s, AVCodecContext *avctx, SingleChannelElement *sce); void (*search_for_tns)(struct AACEncContext *s, SingleChannelElement *sce); diff --git a/libavcodec/aacenc_pred.c b/libavcodec/aacenc_pred.c index c638b70..fafe002 100644 --- a/libavcodec/aacenc_pred.c +++ b/libavcodec/aacenc_pred.c @@ -21,15 +21,22 @@ /** * @file - * AAC encoder main prediction + * AAC encoder Intensity Stereo * @author Rostislav Pehlivanov ( atomnuker gmail com ) */ #include "aactab.h" #include "aacenc_pred.h" #include "aacenc_utils.h" +#include "aacenc_is.h" /* <- Needed for common window distortions */ #include "aacenc_quantization.h" +#define RESTORE_PRED(sce, sfb) \ + if (sce->ics.prediction_used[sfb]) {\ + sce->ics.prediction_used[sfb] = 0;\ + sce->band_type[sfb] = sce->band_alt[sfb];\ + } + static inline float flt16_round(float pf) { union av_intfloat32 tmp; @@ -54,73 +61,57 @@ static inline float flt16_trunc(float pf) return pun.f; } -static inline void predict(PredictorState *ps, float *coef, float *rcoef, - int output_enable) +static inline void predict(PredictorState *ps, float *coef, float *rcoef, int set) { - const float a = 0.953125; // 61.0 / 64 float k2; - float r0 = ps->r0, r1 = ps->r1; - float cor0 = ps->cor0, cor1 = ps->cor1; - float var0 = ps->var0, var1 = ps->var1; - - ps->k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0; - k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0; - - ps->x_est = flt16_round(ps->k1*r0 + k2*r1); - - if (output_enable) - *coef -= ps->x_est; - else - *rcoef = *coef - ps->x_est; -} - -static inline void update_predictor(PredictorState *ps, float qcoef) -{ - const float alpha = 0.90625; // 29.0 / 32 const float a = 0.953125; // 61.0 / 64 - float k1 = ps->k1; - float r0 = ps->r0; - float r1 = ps->r1; - float e0 = qcoef + ps->x_est; - float e1 = e0 - k1 * r0; - float cor0 = ps->cor0, cor1 = ps->cor1; - float var0 = ps->var0, var1 = ps->var1; + const float alpha = 0.90625; // 29.0 / 32 + const float k1 = ps->k1; + const float r0 = ps->r0, r1 = ps->r1; + const float cor0 = ps->cor0, cor1 = ps->cor1; + const float var0 = ps->var0, var1 = ps->var1; + const float e0 = *coef - ps->x_est; + const float e1 = e0 - k1 * r0; + + if (set) + *coef = e0; ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1); ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1)); ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0); ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0)); + ps->r1 = flt16_trunc(a * (r0 - k1 * e0)); + ps->r0 = flt16_trunc(a * e0); - ps->r1 = flt16_trunc(a * (r0 - k1 * e0)); - ps->r0 = flt16_trunc(a * e0); + /* Prediction for next frame */ + ps->k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0; + k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0; + *rcoef = ps->x_est = flt16_round(ps->k1*ps->r0 + k2*ps->r1); } static inline void reset_predict_state(PredictorState *ps) { - ps->r0 = 0.0f; - ps->r1 = 0.0f; - ps->cor0 = 0.0f; - ps->cor1 = 0.0f; - ps->var0 = 1.0f; - ps->var1 = 1.0f; - ps->k1 = 0.0f; - ps->x_est= 0.0f; + ps->r0 = 0.0f; + ps->r1 = 0.0f; + ps->k1 = 0.0f; + ps->cor0 = 0.0f; + ps->cor1 = 0.0f; + ps->var0 = 1.0f; + ps->var1 = 1.0f; + ps->x_est = 0.0f; } -static inline void reset_all_predictors(SingleChannelElement *sce) +static inline void reset_all_predictors(PredictorState *ps) { int i; for (i = 0; i < MAX_PREDICTORS; i++) - reset_predict_state(&sce->predictor_state[i]); - for (i = 1; i < 31; i++) - sce->ics.predictor_reset_count[i] = 0; + reset_predict_state(&ps[i]); } static inline void reset_predictor_group(SingleChannelElement *sce, int group_num) { int i; PredictorState *ps = sce->predictor_state; - sce->ics.predictor_reset_count[group_num] = 0; for (i = group_num - 1; i < MAX_PREDICTORS; i += 30) reset_predict_state(&ps[i]); } @@ -128,136 +119,89 @@ static inline void reset_predictor_group(SingleChannelElement *sce, int group_nu void ff_aac_apply_main_pred(AACEncContext *s, SingleChannelElement *sce) { int sfb, k; + const int pmax = FFMIN(sce->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]); if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) { - for (sfb = 0; sfb < ff_aac_pred_sfb_max[s->samplerate_index]; sfb++) { - for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) + for (sfb = 0; sfb < pmax; sfb++) { + for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) { predict(&sce->predictor_state[k], &sce->coeffs[k], &sce->prcoeffs[k], - (sce->ics.predictor_present && sce->ics.prediction_used[sfb])); - } - } -} - -static void decode_joint_stereo(ChannelElement *cpe) -{ - int i, w, w2, g; - SingleChannelElement *sce0 = &cpe->ch[0]; - SingleChannelElement *sce1 = &cpe->ch[1]; - IndividualChannelStream *ics; - - for (i = 0; i < MAX_PREDICTORS; i++) - sce0->prcoeffs[i] = sce0->predictor_state[i].x_est; - - ics = &sce0->ics; - for (w = 0; w < ics->num_windows; w += ics->group_len[w]) { - for (w2 = 0; w2 < ics->group_len[w]; w2++) { - int start = (w+w2) * 128; - for (g = 0; g < ics->num_swb; g++) { - int sfb = w*16 + g; - //apply Intensity stereo coeffs transformation - if (cpe->is_mask[sfb]) { - int p = -1 + 2 * (sce1->band_type[sfb] - 14); - float rscale = ff_aac_pow2sf_tab[-sce1->sf_idx[sfb] + POW_SF2_ZERO]; - p *= 1 - 2 * cpe->ms_mask[sfb]; - for (i = 0; i < ics->swb_sizes[g]; i++) { - sce0->pqcoeffs[start+i] = (sce0->prcoeffs[start+i] + p*sce0->pqcoeffs[start+i]) * rscale; - } - } else if (cpe->ms_mask[sfb] && - sce0->band_type[sfb] < NOISE_BT && - sce1->band_type[sfb] < NOISE_BT) { - for (i = 0; i < ics->swb_sizes[g]; i++) { - float L = sce0->pqcoeffs[start+i] + sce1->pqcoeffs[start+i]; - float R = sce0->pqcoeffs[start+i] - sce1->pqcoeffs[start+i]; - sce0->pqcoeffs[start+i] = L; - sce1->pqcoeffs[start+i] = R; - } - } - start += ics->swb_sizes[g]; + sce->ics.predictor_present && sce->ics.prediction_used[sfb]); } } + if (sce->ics.predictor_reset_group) { + reset_predictor_group(sce, sce->ics.predictor_reset_group); + } + } else { + reset_all_predictors(sce->predictor_state); } } -static inline void prepare_predictors(SingleChannelElement *sce) -{ - int k; - for (k = 0; k < MAX_PREDICTORS; k++) - predict(&sce->predictor_state[k], &sce->coeffs[k], &sce->prcoeffs[k], 0); -} - -void ff_aac_update_main_pred(AACEncContext *s, SingleChannelElement *sce, ChannelElement *cpe) -{ - int k; - - if (sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) - return; - - if (cpe && cpe->common_window) - decode_joint_stereo(cpe); - - for (k = 0; k < MAX_PREDICTORS; k++) - update_predictor(&sce->predictor_state[k], sce->pqcoeffs[k]); - - if (sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) { - reset_all_predictors(sce); - } - - if (sce->ics.predictor_reset_group) - reset_predictor_group(sce, sce->ics.predictor_reset_group); -} - -/* If inc == 0 check if it returns 0 to see if you can reset freely */ +/* If inc = 0 you can check if this returns 0 to see if you can reset freely */ static inline int update_counters(IndividualChannelStream *ics, int inc) { - int i, rg = 0; + int i; for (i = 1; i < 31; i++) { ics->predictor_reset_count[i] += inc; - if (!rg && ics->predictor_reset_count[i] > PRED_RESET_FRAME_MIN) - rg = i; /* Reset this immediately */ + if (ics->predictor_reset_count[i] > PRED_RESET_FRAME_MIN) + return i; /* Reset this immediately */ } - return rg; + return 0; } void ff_aac_adjust_common_prediction(AACEncContext *s, ChannelElement *cpe) { - int start, w, g, count = 0; + int start, w, w2, g, i, count = 0; SingleChannelElement *sce0 = &cpe->ch[0]; SingleChannelElement *sce1 = &cpe->ch[1]; + const int pmax0 = FFMIN(sce0->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]); + const int pmax1 = FFMIN(sce1->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]); + const int pmax = FFMIN(pmax0, pmax1); - if (!cpe->common_window || sce0->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) + if (!cpe->common_window || + sce0->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE || + sce1->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) return; - /* Predict if IS or MS is on and at least one channel is marked or when both are */ 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++) { int sfb = w*16+g; - if (sfb < PRED_SFB_START || sfb > ff_aac_pred_sfb_max[s->samplerate_index]) { - ; - } else if ((cpe->is_mask[sfb] || cpe->ms_mask[sfb]) && - (sce0->ics.prediction_used[sfb] || sce1->ics.prediction_used[sfb])) { - sce0->ics.prediction_used[sfb] = sce1->ics.prediction_used[sfb] = 1; - count++; - } else if (sce0->ics.prediction_used[sfb] && sce1->ics.prediction_used[sfb]) { + int sum = sce0->ics.prediction_used[sfb] + sce1->ics.prediction_used[sfb]; + float ener0 = 0.0f, ener1 = 0.0f, ener01 = 0.0f; + struct AACISError ph_err1, ph_err2, *erf; + if (sfb < PRED_SFB_START || sfb > pmax || sum != 2) { + RESTORE_PRED(sce0, sfb); + RESTORE_PRED(sce1, sfb); + start += sce0->ics.swb_sizes[g]; + continue; + } + for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) { + for (i = 0; i < sce0->ics.swb_sizes[g]; i++) { + float coef0 = sce0->pcoeffs[start+(w+w2)*128+i]; + float coef1 = sce1->pcoeffs[start+(w+w2)*128+i]; + ener0 += coef0*coef0; + ener1 += coef1*coef1; + ener01 += (coef0 + coef1)*(coef0 + coef1); + } + } + ph_err1 = ff_aac_is_encoding_err(s, cpe, start, w, g, + ener0, ener1, ener01, -1); + ph_err2 = ff_aac_is_encoding_err(s, cpe, start, w, g, + ener0, ener1, ener01, +1); + erf = ph_err1.error < ph_err2.error ? &ph_err1 : &ph_err2; + if (erf->pass) { + sce0->ics.prediction_used[sfb] = 1; + sce1->ics.prediction_used[sfb] = 1; count++; } else { - /* Restore band types, if changed - prediction never sets > RESERVED_BT */ - if (sce0->ics.prediction_used[sfb] && sce0->band_type[sfb] < RESERVED_BT) - sce0->band_type[sfb] = sce0->orig_band_type[sfb]; - if (sce1->ics.prediction_used[sfb] && sce1->band_type[sfb] < RESERVED_BT) - sce1->band_type[sfb] = sce1->orig_band_type[sfb]; - sce0->ics.prediction_used[sfb] = sce1->ics.prediction_used[sfb] = 0; + RESTORE_PRED(sce0, sfb); + RESTORE_PRED(sce1, sfb); } start += sce0->ics.swb_sizes[g]; } } sce1->ics.predictor_present = sce0->ics.predictor_present = !!count; - - if (!count) - return; - - sce1->ics.predictor_reset_group = sce0->ics.predictor_reset_group; } static void update_pred_resets(SingleChannelElement *sce) @@ -266,14 +210,12 @@ static void update_pred_resets(SingleChannelElement *sce) float avg_frame = 0.0f; IndividualChannelStream *ics = &sce->ics; - /* Some other code probably chose the reset group */ - if (ics->predictor_reset_group) - return; - + /* Update the counters and immediately update any frame behind schedule */ if ((ics->predictor_reset_group = update_counters(&sce->ics, 1))) return; for (i = 1; i < 31; i++) { + /* Count-based */ if (ics->predictor_reset_count[i] > max_frame) { max_group_id_c = i; max_frame = ics->predictor_reset_count[i]; @@ -281,8 +223,7 @@ static void update_pred_resets(SingleChannelElement *sce) avg_frame = (ics->predictor_reset_count[i] + avg_frame)/2; } - if (avg_frame*2 > max_frame && max_frame > PRED_RESET_MIN || - max_frame > (2*PRED_RESET_MIN)/3) { + if (max_frame > PRED_RESET_MIN) { ics->predictor_reset_group = max_group_id_c; } else { ics->predictor_reset_group = 0; @@ -291,56 +232,91 @@ static void update_pred_resets(SingleChannelElement *sce) void ff_aac_search_for_pred(AACEncContext *s, SingleChannelElement *sce) { - int sfb, i, count = 0; - float *O34 = &s->scoefs[256*0], *P34 = &s->scoefs[256*1]; - int cost_coeffs = PRICE_OFFSET; - int cost_pred = 1+(sce->ics.predictor_reset_group ? 5 : 0) + - FFMIN(sce->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]); + int sfb, i, count = 0, cost_coeffs = 0, cost_pred = 0; + const int pmax = FFMIN(sce->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]); + float *O34 = &s->scoefs[128*0], *P34 = &s->scoefs[128*1]; + float *SENT = &s->scoefs[128*2], *S34 = &s->scoefs[128*3]; + float *QERR = &s->scoefs[128*4]; - memcpy(sce->orig_band_type, sce->band_type, 128*sizeof(enum BandType)); + if (sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) { + sce->ics.predictor_present = 0; + return; + } - if (!sce->ics.predictor_initialized || - sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) { - reset_all_predictors(sce); + if (!sce->ics.predictor_initialized) { + reset_all_predictors(sce->predictor_state); + sce->ics.predictor_initialized = 1; + memcpy(sce->prcoeffs, sce->coeffs, 1024*sizeof(float)); for (i = 1; i < 31; i++) sce->ics.predictor_reset_count[i] = i; - sce->ics.predictor_initialized = 1; } update_pred_resets(sce); - prepare_predictors(sce); - sce->ics.predictor_reset_group = 0; - - for (sfb = PRED_SFB_START; sfb < ff_aac_pred_sfb_max[s->samplerate_index]; sfb++) { - float dist1 = 0.0f, dist2 = 0.0f; - int swb_start = sce->ics.swb_offset[sfb]; - int swb_len = sce->ics.swb_offset[sfb + 1] - swb_start; - int cb1 = sce->band_type[sfb], cb2, bits1 = 0, bits2 = 0; - FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[sfb]; - abs_pow34_v(O34, &sce->coeffs[swb_start], swb_len); - abs_pow34_v(P34, &sce->prcoeffs[swb_start], swb_len); - cb2 = find_min_book(find_max_val(1, swb_len, P34), sce->sf_idx[sfb]); - if (cb2 <= cb1) { - dist1 += quantize_band_cost(s, &sce->coeffs[swb_start], O34, swb_len, - sce->sf_idx[sfb], cb1, s->lambda / band->threshold, - INFINITY, &bits1, 0); - dist2 += quantize_band_cost(s, &sce->prcoeffs[swb_start], P34, swb_len, - sce->sf_idx[sfb], cb2, s->lambda / band->threshold, - INFINITY, &bits2, 0); - if (dist2 <= dist1) { - sce->ics.prediction_used[sfb] = 1; - sce->band_type[sfb] = cb2; - count++; - } - cost_coeffs += bits1; - cost_pred += bits2; + memcpy(sce->band_alt, sce->band_type, sizeof(sce->band_type)); + + for (sfb = PRED_SFB_START; sfb < pmax; sfb++) { + int cost1, cost2, cb_p; + float dist1, dist2, dist_spec_err = 0.0f; + const int cb_n = sce->band_type[sfb]; + const int start_coef = sce->ics.swb_offset[sfb]; + const int num_coeffs = sce->ics.swb_offset[sfb + 1] - start_coef; + const FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[sfb]; + + if (start_coef + num_coeffs > MAX_PREDICTORS) + continue; + + /* Normal coefficients */ + abs_pow34_v(O34, &sce->coeffs[start_coef], num_coeffs); + dist1 = quantize_and_encode_band_cost(s, NULL, &sce->coeffs[start_coef], NULL, + O34, num_coeffs, sce->sf_idx[sfb], + cb_n, s->lambda / band->threshold, INFINITY, &cost1, 0); + cost_coeffs += cost1; + + /* Encoded coefficients - needed for #bits, band type and quant. error */ + for (i = 0; i < num_coeffs; i++) + SENT[i] = sce->coeffs[start_coef + i] - sce->prcoeffs[start_coef + i]; + abs_pow34_v(S34, SENT, num_coeffs); + if (cb_n < RESERVED_BT) + cb_p = find_min_book(find_max_val(1, num_coeffs, S34), sce->sf_idx[sfb]); + else + cb_p = cb_n; + quantize_and_encode_band_cost(s, NULL, SENT, QERR, S34, num_coeffs, + sce->sf_idx[sfb], cb_p, s->lambda / band->threshold, INFINITY, + &cost2, 0); + + /* Reconstructed coefficients - needed for distortion measurements */ + for (i = 0; i < num_coeffs; i++) + sce->prcoeffs[start_coef + i] += QERR[i] != 0.0f ? (sce->prcoeffs[start_coef + i] - QERR[i]) : 0.0f; + abs_pow34_v(P34, &sce->prcoeffs[start_coef], num_coeffs); + if (cb_n < RESERVED_BT) + cb_p = find_min_book(find_max_val(1, num_coeffs, P34), sce->sf_idx[sfb]); + else + cb_p = cb_n; + dist2 = quantize_and_encode_band_cost(s, NULL, &sce->prcoeffs[start_coef], NULL, + P34, num_coeffs, sce->sf_idx[sfb], + cb_p, s->lambda / band->threshold, INFINITY, NULL, 0); + for (i = 0; i < num_coeffs; i++) + dist_spec_err += (O34[i] - P34[i])*(O34[i] - P34[i]); + dist_spec_err *= s->lambda / band->threshold; + dist2 += dist_spec_err; + + if (dist2 <= dist1 && cb_p <= cb_n) { + cost_pred += cost2; + sce->ics.prediction_used[sfb] = 1; + sce->band_alt[sfb] = cb_n; + sce->band_type[sfb] = cb_p; + count++; + } else { + cost_pred += cost1; + sce->band_alt[sfb] = cb_p; } } - if (count && cost_pred > cost_coeffs) { - memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used)); - memcpy(sce->band_type, sce->orig_band_type, sizeof(sce->band_type)); + if (count && cost_coeffs < cost_pred) { count = 0; + for (sfb = PRED_SFB_START; sfb < pmax; sfb++) + RESTORE_PRED(sce, sfb); + memset(&sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used)); } sce->ics.predictor_present = !!count; @@ -352,14 +328,15 @@ void ff_aac_search_for_pred(AACEncContext *s, SingleChannelElement *sce) void ff_aac_encode_main_pred(AACEncContext *s, SingleChannelElement *sce) { int sfb; + IndividualChannelStream *ics = &sce->ics; + const int pmax = FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]); - if (!sce->ics.predictor_present || - sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) + if (!ics->predictor_present) return; - put_bits(&s->pb, 1, !!sce->ics.predictor_reset_group); - if (sce->ics.predictor_reset_group) - put_bits(&s->pb, 5, sce->ics.predictor_reset_group); - for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]); sfb++) - put_bits(&s->pb, 1, sce->ics.prediction_used[sfb]); + put_bits(&s->pb, 1, !!ics->predictor_reset_group); + if (ics->predictor_reset_group) + put_bits(&s->pb, 5, ics->predictor_reset_group); + for (sfb = 0; sfb < pmax; sfb++) + put_bits(&s->pb, 1, ics->prediction_used[sfb]); } diff --git a/libavcodec/aacenc_pred.h b/libavcodec/aacenc_pred.h index 0694c8a..999af86 100644 --- a/libavcodec/aacenc_pred.h +++ b/libavcodec/aacenc_pred.h @@ -34,16 +34,12 @@ #define PRED_RESET_FRAME_MIN 240 /* Any frame with less than this amount of frames since last reset is ok */ -#define PRED_RESET_MIN 128 +#define PRED_RESET_MIN 64 /* Raise to filter any low frequency artifacts due to prediction */ #define PRED_SFB_START 10 -/* Offset for the number of bits to encode normal coefficients */ -#define PRICE_OFFSET 440 - void ff_aac_apply_main_pred(AACEncContext *s, SingleChannelElement *sce); -void ff_aac_update_main_pred(AACEncContext *s, SingleChannelElement *sce, ChannelElement *cpe); void ff_aac_adjust_common_prediction(AACEncContext *s, ChannelElement *cpe); void ff_aac_search_for_pred(AACEncContext *s, SingleChannelElement *sce); void ff_aac_encode_main_pred(AACEncContext *s, SingleChannelElement *sce);

1 0

fate: reenable TNS test
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 07:00:17 2015 +0100| [fe12ba6f30127a78effdc7cdc8ec92f0830d0cf0] | committer: Rostislav Pehlivanov fate: reenable TNS test Hopefully without errors like last time, but I'm prepared. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=fe12ba6f30127a78e… --- tests/fate/aac.mak | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/tests/fate/aac.mak b/tests/fate/aac.mak index 61b4a57..a28976e 100644 --- a/tests/fate/aac.mak +++ b/tests/fate/aac.mak @@ -166,13 +166,13 @@ fate-aac-pns-encode: CMP_SHIFT = -4096 fate-aac-pns-encode: CMP_TARGET = 646.89 fate-aac-pns-encode: SIZE_TOLERANCE = 3560 -#FATE_AAC_ENCODE += fate-aac-tns-encode -#fate-aac-tns-encode: CMD = enc_dec_pcm adts wav s16le $(TARGET_SAMPLES)/audio-reference/luckynight_2ch_44kHz_s16.wav -strict -2 -c:a aac -aac_tns 1 -b:a 128k -#fate-aac-tns-encode: CMP = stddev -#fate-aac-tns-encode: REF = $(SAMPLES)/audio-reference/luckynight_2ch_44kHz_s16.wav -#fate-aac-tns-encode: CMP_SHIFT = -4096 -#fate-aac-tns-encode: CMP_TARGET = 648.02 -#fate-aac-tns-encode: SIZE_TOLERANCE = 3560 +FATE_AAC_ENCODE += fate-aac-tns-encode +fate-aac-tns-encode: CMD = enc_dec_pcm adts wav s16le $(TARGET_SAMPLES)/audio-reference/luckynight_2ch_44kHz_s16.wav -strict -2 -c:a aac -aac_tns 1 -b:a 128k +fate-aac-tns-encode: CMP = stddev +fate-aac-tns-encode: REF = $(SAMPLES)/audio-reference/luckynight_2ch_44kHz_s16.wav +fate-aac-tns-encode: CMP_SHIFT = -4096 +fate-aac-tns-encode: CMP_TARGET = 646.19 +fate-aac-tns-encode: SIZE_TOLERANCE = 3560 FATE_AAC_ENCODE += fate-aac-is-encode fate-aac-is-encode: CMD = enc_dec_pcm adts wav s16le $(TARGET_SAMPLES)/audio-reference/luckynight_2ch_44kHz_s16.wav -strict -2 -c:a aac -aac_is 1 -b:a 128k

1 0

aacenc: remove TNS from the todo list
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 06:57:08 2015 +0100| [f04d86c16a5dd884114653441101c42d98fb28f7] | committer: Rostislav Pehlivanov aacenc: remove TNS from the todo list Pulses are already on the way so expect to see the list gone in the close future. TNS is already of sufficiently high quality to be enabled by default (but isn't yet, so you too can help by testing!). Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=f04d86c16a5dd8841… --- libavcodec/aacenc.c | 1 - 1 file changed, 1 deletion(-) diff --git a/libavcodec/aacenc.c b/libavcodec/aacenc.c index 9f59b92..09dd0dd 100644 --- a/libavcodec/aacenc.c +++ b/libavcodec/aacenc.c @@ -27,7 +27,6 @@ /*********************************** * TODOs: * add sane pulse detection - * add temporal noise shaping ***********************************/ #include "libavutil/float_dsp.h"

1 0

aacenc_tns: rework the way coefficients are calculated
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 06:47:31 2015 +0100| [f20b67173ca6a05b8c3dee02dad3b7243b96292b] | committer: Rostislav Pehlivanov aacenc_tns: rework the way coefficients are calculated This commit abandons the way the specifications state to quantize the coefficients, makes use of the new LPC float functions and is much better. The original way of converting non-normalized float samples to int32_t which out LPC system expects was wrong and it was wrong to assume the coefficients that are generated are also valid. It was essentially a full garbage-in, garbage-out system and it definitely shows when looking at spectrals and listening. The high frequencies were very overattenuated. The new LPC function performs the analysis directly. The specifications state to quantize the coefficients into four bit index values using an asin() function which of course had to have ugly ternary operators because the function turns negative if the coefficients are negative which when encoding causes invalid bitstream to get generated. This deviates from this by using the direct TNS tables, which are fairly small since you only have 4 bits at most for index values. The LPC values are directly quantized against the tables and are then used to perform filtering after the requantization, which simply fetches the array values. The end result is that TNS works much better now and doesn't attenuate anything but the actual signal, e.g. TNS removes quantization errors and does it's job correctly now. It might be enabled by default soon since it doesn't hurt and helps reduce nastyness at low bitrates. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=f20b67173ca6a05b8… --- libavcodec/aaccoder.c | 4 + libavcodec/aacenc.c | 5 +- libavcodec/aacenc.h | 1 + libavcodec/aacenc_tns.c | 211 ++++++++++++++++++++++------------------------- libavcodec/aacenc_tns.h | 11 +-- 5 files changed, 110 insertions(+), 122 deletions(-) diff --git a/libavcodec/aaccoder.c b/libavcodec/aaccoder.c index 8256f96..86d598f 100644 --- a/libavcodec/aaccoder.c +++ b/libavcodec/aaccoder.c @@ -964,6 +964,7 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, + ff_aac_apply_tns, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, @@ -979,6 +980,7 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, + ff_aac_apply_tns, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, @@ -994,6 +996,7 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, + ff_aac_apply_tns, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, @@ -1009,6 +1012,7 @@ AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = { ff_aac_encode_main_pred, ff_aac_adjust_common_prediction, ff_aac_apply_main_pred, + ff_aac_apply_tns, set_special_band_scalefactors, search_for_pns, ff_aac_search_for_tns, diff --git a/libavcodec/aacenc.c b/libavcodec/aacenc.c index a7c43c7..9f59b92 100644 --- a/libavcodec/aacenc.c +++ b/libavcodec/aacenc.c @@ -404,10 +404,9 @@ static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s, encode_band_info(s, sce); encode_scale_factors(avctx, s, sce); encode_pulses(s, &sce->pulse); + put_bits(&s->pb, 1, !!sce->tns.present); if (s->coder->encode_tns_info) s->coder->encode_tns_info(s, sce); - else - put_bits(&s->pb, 1, 0); put_bits(&s->pb, 1, 0); //ssr encode_spectral_coeffs(s, sce); return 0; @@ -609,6 +608,8 @@ static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, s->coder->search_for_pns(s, avctx, sce); if (s->options.tns && s->coder->search_for_tns) s->coder->search_for_tns(s, sce); + if (s->options.tns && s->coder->apply_tns_filt) + s->coder->apply_tns_filt(sce); if (sce->tns.present) tns_mode = 1; } diff --git a/libavcodec/aacenc.h b/libavcodec/aacenc.h index 69a8c01..51dce8a 100644 --- a/libavcodec/aacenc.h +++ b/libavcodec/aacenc.h @@ -63,6 +63,7 @@ typedef struct AACCoefficientsEncoder { void (*encode_main_pred)(struct AACEncContext *s, SingleChannelElement *sce); void (*adjust_common_prediction)(struct AACEncContext *s, ChannelElement *cpe); void (*apply_main_pred)(struct AACEncContext *s, SingleChannelElement *sce); + void (*apply_tns_filt)(SingleChannelElement *sce); void (*set_special_band_scalefactors)(struct AACEncContext *s, SingleChannelElement *sce); void (*search_for_pns)(struct AACEncContext *s, AVCodecContext *avctx, SingleChannelElement *sce); void (*search_for_tns)(struct AACEncContext *s, SingleChannelElement *sce); diff --git a/libavcodec/aacenc_tns.c b/libavcodec/aacenc_tns.c index 8c21255..912229e 100644 --- a/libavcodec/aacenc_tns.c +++ b/libavcodec/aacenc_tns.c @@ -31,62 +31,30 @@ #include "aacenc_utils.h" #include "aacenc_quantization.h" -static inline void conv_to_int32(int32_t *loc, float *samples, int num, float norm) -{ - int i; - for (i = 0; i < num; i++) - loc[i] = ceilf((samples[i]/norm)*INT32_MAX); -} - -static inline void conv_to_float(float *arr, int32_t *cof, int num) -{ - int i; - for (i = 0; i < num; i++) - arr[i] = (float)cof[i]/INT32_MAX; -} - -/* Input: quantized 4 bit coef, output: 1 if first (MSB) 2 bits are the same */ -static inline int coef_test_compression(int coef) -{ - int tmp = coef >> 2; - int res = ff_ctz(tmp); - if (res > 1) - return 1; /* ...00 -> compressable */ - else if (res == 1) - return 0; /* ...10 -> uncompressable */ - else if (ff_ctz(tmp >> 1) > 0) - return 0; /* ...0 1 -> uncompressable */ - else - return 1; /* ...1 1 -> compressable */ -} - static inline int compress_coef(int *coefs, int num) { - int i, res = 0; + int i, c = 0; for (i = 0; i < num; i++) - res += coef_test_compression(coefs[i]); - return res == num ? 1 : 0; + c += coefs[i] < 4 || coefs[i] > 11; + return c == num; } /** * Encode TNS data. - * Coefficient compression saves a single bit. + * Coefficient compression saves a single bit per coefficient. */ void ff_aac_encode_tns_info(AACEncContext *s, SingleChannelElement *sce) { int i, w, filt, coef_len, coef_compress; - const int coef_res = MAX_LPC_PRECISION == 4 ? 1 : 0; const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE; - put_bits(&s->pb, 1, !!sce->tns.present); - if (!sce->tns.present) return; for (i = 0; i < sce->ics.num_windows; i++) { put_bits(&s->pb, 2 - is8, sce->tns.n_filt[i]); if (sce->tns.n_filt[i]) { - put_bits(&s->pb, 1, !!coef_res); + put_bits(&s->pb, 1, 1); for (filt = 0; filt < sce->tns.n_filt[i]; filt++) { put_bits(&s->pb, 6 - 2 * is8, sce->tns.length[i][filt]); put_bits(&s->pb, 5 - 2 * is8, sce->tns.order[i][filt]); @@ -95,7 +63,7 @@ void ff_aac_encode_tns_info(AACEncContext *s, SingleChannelElement *sce) sce->tns.order[i][filt]); put_bits(&s->pb, 1, !!sce->tns.direction[i][filt]); put_bits(&s->pb, 1, !!coef_compress); - coef_len = coef_res + 3 - coef_compress; + coef_len = 4 - coef_compress; for (w = 0; w < sce->tns.order[i][filt]; w++) put_bits(&s->pb, coef_len, sce->tns.coef_idx[i][filt][w]); } @@ -104,24 +72,25 @@ void ff_aac_encode_tns_info(AACEncContext *s, SingleChannelElement *sce) } } -static int process_tns_coeffs(TemporalNoiseShaping *tns, float *tns_coefs_raw, - int order, int w, int filt) +static void process_tns_coeffs(TemporalNoiseShaping *tns, double *coef_raw, + int *order_p, int w, int filt) { - int i, j; + int i, j, order = *order_p; int *idx = tns->coef_idx[w][filt]; float *lpc = tns->coef[w][filt]; - const int iqfac_p = ((1 << (MAX_LPC_PRECISION-1)) - 0.5)/(M_PI/2.0); - const int iqfac_m = ((1 << (MAX_LPC_PRECISION-1)) + 0.5)/(M_PI/2.0); float temp[TNS_MAX_ORDER] = {0.0f}, out[TNS_MAX_ORDER] = {0.0f}; - /* Quantization */ + if (!order) + return; + + /* Not what the specs say, but it's better */ for (i = 0; i < order; i++) { - idx[i] = ceilf(asin(tns_coefs_raw[i])*((tns_coefs_raw[i] >= 0) ? iqfac_p : iqfac_m)); - lpc[i] = 2*sin(idx[i]/((idx[i] >= 0) ? iqfac_p : iqfac_m)); + idx[i] = quant_array_idx(coef_raw[i], tns_tmp2_map_0_4, 16); + lpc[i] = tns_tmp2_map_0_4[idx[i]]; } /* Trim any coeff less than 0.1f from the end */ - for (i = order; i > -1; i--) { + for (i = order-1; i > -1; i--) { lpc[i] = (fabs(lpc[i]) > 0.1f) ? lpc[i] : 0.0f; if (lpc[i] != 0.0 ) { order = i; @@ -129,9 +98,6 @@ static int process_tns_coeffs(TemporalNoiseShaping *tns, float *tns_coefs_raw, } } - if (!order) - return 0; - /* Step up procedure, convert to LPC coeffs */ out[0] = 1.0f; for (i = 1; i <= order; i++) { @@ -143,35 +109,59 @@ static int process_tns_coeffs(TemporalNoiseShaping *tns, float *tns_coefs_raw, } out[i] = lpc[i-1]; } + *order_p = order; memcpy(lpc, out, TNS_MAX_ORDER*sizeof(float)); - - return order; } -static void apply_tns_filter(float *out, float *in, int order, int direction, - float *tns_coefs, int ltp_used, int w, int filt, - int start_i, int len) +/* Apply TNS filter */ +void ff_aac_apply_tns(SingleChannelElement *sce) { - int i, j, inc, start = start_i; - float tmp[TNS_MAX_ORDER+1]; - if (direction) { - inc = -1; - start = (start + len) - 1; - } else { - inc = 1; - } - if (!ltp_used) { /* AR filter */ - for (i = 0; i < len; i++, start += inc) - out[i] = in[start]; - for (j = 1; j <= FFMIN(i, order); j++) - out[i] += tns_coefs[j]*in[start - j*inc]; - } else { /* MA filter */ - for (i = 0; i < len; i++, start += inc) { - tmp[0] = out[i] = in[start]; - for (j = 1; j <= FFMIN(i, order); j++) - out[i] += tmp[j]*tns_coefs[j]; - for (j = order; j > 0; j--) - tmp[j] = tmp[j - 1]; + const int mmm = FFMIN(sce->ics.tns_max_bands, sce->ics.max_sfb); + float *coef = sce->pcoeffs; + TemporalNoiseShaping *tns = &sce->tns; + int w, filt, m, i; + int bottom, top, order, start, end, size, inc; + float *lpc, tmp[TNS_MAX_ORDER+1]; + + return; + + for (w = 0; w < sce->ics.num_windows; w++) { + bottom = sce->ics.num_swb; + for (filt = 0; filt < tns->n_filt[w]; filt++) { + top = bottom; + bottom = FFMAX(0, top - tns->length[w][filt]); + order = tns->order[w][filt]; + lpc = tns->coef[w][filt]; + if (!order) + continue; + + start = sce->ics.swb_offset[FFMIN(bottom, mmm)]; + end = sce->ics.swb_offset[FFMIN( top, mmm)]; + if ((size = end - start) <= 0) + continue; + if (tns->direction[w][filt]) { + inc = -1; + start = end - 1; + } else { + inc = 1; + } + start += w * 128; + + if (!sce->ics.ltp.present) { + // ar filter + for (m = 0; m < size; m++, start += inc) + for (i = 1; i <= FFMIN(m, order); i++) + coef[start] += coef[start - i * inc]*lpc[i - 1]; + } else { + // ma filter + for (m = 0; m < size; m++, start += inc) { + tmp[0] = coef[start]; + for (i = 1; i <= FFMIN(m, order); i++) + coef[start] += tmp[i]*lpc[i - 1]; + for (i = order; i > 0; i--) + tmp[i] = tmp[i - 1]; + } + } } } } @@ -179,57 +169,54 @@ static void apply_tns_filter(float *out, float *in, int order, int direction, void ff_aac_search_for_tns(AACEncContext *s, SingleChannelElement *sce) { TemporalNoiseShaping *tns = &sce->tns; - int w, g, order, sfb_start, sfb_len, coef_start, shift[MAX_LPC_ORDER], count = 0; + int w, g, w2, prev_end_sfb = 0, count = 0; const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE; const int tns_max_order = is8 ? 7 : s->profile == FF_PROFILE_AAC_LOW ? 12 : TNS_MAX_ORDER; - const float freq_mult = mpeg4audio_sample_rates[s->samplerate_index]/(1024.0f/sce->ics.num_windows)/2.0f; - float max_coef = 0.0f; - - sce->tns.present = 0; - return; - - for (coef_start = 0; coef_start < 1024; coef_start++) - max_coef = FFMAX(max_coef, sce->pcoeffs[coef_start]); - for (w = 0; w < sce->ics.num_windows; w++) { - int filters = 1, start = 0, coef_len = 0; - int32_t conv_coeff[1024] = {0}; - int32_t coefs_t[MAX_LPC_ORDER][MAX_LPC_ORDER] = {{0}}; - - /* Determine start sfb + coef - excludes anything below threshold */ + for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { + int order = 0, filters = 1; + int sfb_start = 0, sfb_len = 0; + int coef_start = 0, coef_len = 0; + float energy = 0.0f, threshold = 0.0f; + double coefs[MAX_LPC_ORDER][MAX_LPC_ORDER] = {{0}}; for (g = 0; g < sce->ics.num_swb; g++) { - if (start*freq_mult > TNS_LOW_LIMIT) { + if (!sfb_start && w*16+g > TNS_LOW_LIMIT && w*16+g > prev_end_sfb) { sfb_start = w*16+g; - sfb_len = (w+1)*16 + g - sfb_start; - coef_start = sce->ics.swb_offset[sfb_start]; - coef_len = sce->ics.swb_offset[sfb_start + sfb_len] - coef_start; - break; + coef_start = sce->ics.swb_offset[sfb_start]; + } + if (sfb_start) { + 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 (!sfb_len && band->energy < band->threshold*1.3f) { + sfb_len = (w+w2)*16+g - sfb_start; + prev_end_sfb = sfb_start + sfb_len; + coef_len = sce->ics.swb_offset[sfb_start + sfb_len] - coef_start; + break; + } + energy += band->energy; + threshold += band->threshold; + } + if (!sfb_len) { + sfb_len = (w+sce->ics.group_len[w])*16+g - sfb_start; + coef_len = sce->ics.swb_offset[sfb_start + sfb_len] - coef_start; + } } - start += sce->ics.swb_sizes[g]; } - if (coef_len <= 0) + if (sfb_len <= 0 || coef_len <= 0) continue; - - conv_to_int32(conv_coeff, &sce->pcoeffs[coef_start], coef_len, max_coef); + if (coef_start + coef_len > 1024) + coef_len = 1024 - coef_start; /* LPC */ - order = ff_lpc_calc_coefs(&s->lpc, conv_coeff, coef_len, - TNS_MIN_PRED_ORDER, tns_max_order, - 32, coefs_t, shift, - FF_LPC_TYPE_LEVINSON, 10, - ORDER_METHOD_EST, MAX_LPC_SHIFT, 0) - 1; - - /* Works surprisingly well, remember to tweak MAX_LPC_SHIFT if you want to play around with this */ - if (shift[order] > 3) { + order = ff_lpc_calc_levinsion(&s->lpc, &sce->coeffs[coef_start], coef_len, + coefs, 0, tns_max_order, ORDER_METHOD_LOG); + + if (energy > threshold) { int direction = 0; - float tns_coefs_raw[TNS_MAX_ORDER]; tns->n_filt[w] = filters++; - conv_to_float(tns_coefs_raw, coefs_t[order], order); for (g = 0; g < tns->n_filt[w]; g++) { - process_tns_coeffs(tns, tns_coefs_raw, order, w, g); - apply_tns_filter(&sce->coeffs[coef_start], sce->pcoeffs, order, direction, tns->coef[w][g], - sce->ics.ltp.present, w, g, coef_start, coef_len); + process_tns_coeffs(tns, coefs[order], &order, w, g); tns->order[w][g] = order; tns->length[w][g] = sfb_len; tns->direction[w][g] = direction; diff --git a/libavcodec/aacenc_tns.h b/libavcodec/aacenc_tns.h index 789e9a6..72c9123 100644 --- a/libavcodec/aacenc_tns.h +++ b/libavcodec/aacenc_tns.h @@ -30,16 +30,11 @@ #include "aacenc.h" -/** Frequency in Hz for lower limit of TNS **/ -#define TNS_LOW_LIMIT 2150 - -/** LPC settings */ -#define TNS_MIN_PRED_ORDER 0 -#define MAX_LPC_PRECISION 4 /* 4 bits ltp coeff precision */ -#define TNS_LPC_PASSES 2 -#define MAX_LPC_SHIFT 4 +/** Lower limit of TNS in SFBs **/ +#define TNS_LOW_LIMIT 24 void ff_aac_encode_tns_info(AACEncContext *s, SingleChannelElement *sce); +void ff_aac_apply_tns(SingleChannelElement *sce); void ff_aac_search_for_tns(AACEncContext *s, SingleChannelElement *sce); #endif /* AVCODEC_AACENC_TNS_H */

1 0

aac: remove now-unused redundant array
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 06:44:20 2015 +0100| [1cd5daee20602e1fa271bd38b3aa47c9bc30ac67] | committer: Rostislav Pehlivanov aac: remove now-unused redundant array This commit removes the array which was made redundant with the last commit. The current prediction system gets the quantization error directly (and without the single-frame delay) in the search_for_pred function. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=1cd5daee20602e1fa… --- libavcodec/aac.h | 1 - 1 file changed, 1 deletion(-) diff --git a/libavcodec/aac.h b/libavcodec/aac.h index aa4b53b..17af49c 100644 --- a/libavcodec/aac.h +++ b/libavcodec/aac.h @@ -259,7 +259,6 @@ typedef struct SingleChannelElement { DECLARE_ALIGNED(32, INTFLOAT, saved)[1536]; ///< overlap DECLARE_ALIGNED(32, INTFLOAT, ret_buf)[2048]; ///< PCM output buffer DECLARE_ALIGNED(16, INTFLOAT, ltp_state)[3072]; ///< time signal for LTP - DECLARE_ALIGNED(32, AAC_FLOAT, pqcoeffs)[1024]; ///< quantization error of coefs (used by encoder) DECLARE_ALIGNED(32, AAC_FLOAT, prcoeffs)[1024]; ///< Main prediction coefs (used by encoder) PredictorState predictor_state[MAX_PREDICTORS]; INTFLOAT *ret; ///< PCM output

1 0

aacenc: change FF_PROFILE_UNKNOWN to AAC-Main if prediction is enabled
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 06:28:21 2015 +0100| [949a4892fa1dd6da028229cb1674340542fc9bc4] | committer: Rostislav Pehlivanov aacenc: change FF_PROFILE_UNKNOWN to AAC-Main if prediction is enabled This was missed when the original commits were done. FF_PROFILE_UNKNOWN is what's in avctx->profile when no audio profile is specified. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=949a4892fa1dd6da0… --- libavcodec/aacenc.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/libavcodec/aacenc.c b/libavcodec/aacenc.c index 71071eb..ece2328 100644 --- a/libavcodec/aacenc.c +++ b/libavcodec/aacenc.c @@ -786,11 +786,12 @@ static av_cold int aac_encode_init(AVCodecContext *avctx) "Too many bits per frame requested, clamping to max\n"); if (avctx->profile == FF_PROFILE_AAC_MAIN) { s->options.pred = 1; - } else if (avctx->profile == FF_PROFILE_AAC_LOW && s->options.pred) { + } else if ((avctx->profile == FF_PROFILE_AAC_LOW || + avctx->profile == FF_PROFILE_UNKNOWN) && s->options.pred) { s->profile = 0; /* Main */ WARN_IF(1, "Prediction requested, changing profile to AAC-Main\n"); } else if (avctx->profile == FF_PROFILE_AAC_LOW || - avctx->profile == FF_PROFILE_UNKNOWN) { + avctx->profile == FF_PROFILE_UNKNOWN) { s->profile = 1; /* Low */ } else { ERROR_IF(1, "Unsupported profile %d\n", avctx->profile);

1 0

lpc: create a simplified Levinson-Durbin LPC handling float samples
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 06:14:13 2015 +0100| [20962b567b4d1fe791ca4b537121f13f6e0f5fa0] | committer: Rostislav Pehlivanov lpc: create a simplified Levinson-Durbin LPC handling float samples This commit simply duplicates the functionality of ff_lpc_calc_coefs() for the case of a Levinson-Durbin LPC with the only difference being that floating point samples are accepted and the resulting coefficients are raw and unquantized. The motivation behind doing this is the fact that the AAC encoder requires LPC in TNS and LTP and converting non-normalized floating point coefficients to int32_t using SWR and again back for the LPC coefficients was very impractical. The current LPC interfaces were designed for int32_t in mind possibly because FLAC and ALAC use this type for most internal operations. The mathematics in case of floats remains of course identical. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=20962b567b4d1fe79… --- libavcodec/lpc.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ libavcodec/lpc.h | 4 ++++ 2 files changed, 60 insertions(+) diff --git a/libavcodec/lpc.c b/libavcodec/lpc.c index 07fc292..6699eb8 100644 --- a/libavcodec/lpc.c +++ b/libavcodec/lpc.c @@ -271,6 +271,62 @@ int ff_lpc_calc_coefs(LPCContext *s, return opt_order; } +/** + * Simplified Levinson LPC accepting float samples + * + * @param lpc_type LPC method for determining coefficients, + * see #FFLPCType for details + */ +int ff_lpc_calc_levinsion(LPCContext *s, const float *samples, int len, + double lpc[][MAX_LPC_ORDER], int min_order, + int max_order, int omethod) +{ + double ref[MAX_LPC_ORDER] = { 0 }; + double autoc[MAX_LPC_ORDER+1]; + double *w_data = s->windowed_samples; + int i, n2 = (len >> 1); + double w, c = 2.0 / (len - 1.0); + + av_assert2(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER); + + /* reinit LPC context if parameters have changed */ + if (len > s->blocksize || max_order > s->max_order) { + ff_lpc_end(s); + ff_lpc_init(s, len, max_order, FF_LPC_TYPE_LEVINSON); + } + + /* Apply welch window */ + if (len & 1) { + for(i=0; i<n2; i++) { + w = c - i - 1.0; + w = 1.0 - (w * w); + w_data[i] = samples[i] * w; + w_data[len-1-i] = samples[len-1-i] * w; + } + } else { + w_data+=n2; + samples+=n2; + for(i=0; i<n2; i++) { + w = c - n2 + i; + w = 1.0 - (w * w); + w_data[-i-1] = samples[-i-1] * w; + w_data[+i ] = samples[+i ] * w; + } + } + + s->lpc_compute_autocorr(w_data, len, max_order, autoc); + + compute_lpc_coefs(autoc, max_order, &lpc[0][0], max_order, 0, 1); + + if(omethod == ORDER_METHOD_EST) { + for(i=0; i<max_order; i++) + ref[i] = fabs(lpc[i][i]); + return estimate_best_order(ref, min_order, max_order); + } + + return max_order; +} + av_cold int ff_lpc_init(LPCContext *s, int blocksize, int max_order, enum FFLPCType lpc_type) { diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h index 9b76e2f..ec59cea 100644 --- a/libavcodec/lpc.h +++ b/libavcodec/lpc.h @@ -100,6 +100,10 @@ int ff_lpc_calc_coefs(LPCContext *s, int ff_lpc_calc_ref_coefs(LPCContext *s, const int32_t *samples, int order, double *ref); +int ff_lpc_calc_levinsion(LPCContext *s, const float *samples, int len, + double lpc[][MAX_LPC_ORDER], int min_order, + int max_order, int omethod); + /** * Initialize LPCContext. */

1 0

aac: move the TNS tables from aacdectab to the shared aactab
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Sat Aug 29 06:07:12 2015 +0100| [f55cc579115e4bcfe26eb05ecaa1c823343e9e1e] | committer: Rostislav Pehlivanov aac: move the TNS tables from aacdectab to the shared aactab This commit simply moves the TNS tables to a more appropriate aactab.h since then they can be accessed by both the decoder and encoder. The encoder _shouldn't_ normally need the tables since the specs describe a specific quantization process, but the exact reason for this can be seen in the TNS commit following. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=f55cc579115e4bcfe… --- libavcodec/aacdectab.h | 35 ----------------------------------- libavcodec/aactab.h | 35 +++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 35 deletions(-) diff --git a/libavcodec/aacdectab.h b/libavcodec/aacdectab.h index 62c7f87..5c89a93 100644 --- a/libavcodec/aacdectab.h +++ b/libavcodec/aacdectab.h @@ -43,41 +43,6 @@ static const INTFLOAT ltp_coef[8] = { Q30(0.984900f), Q30(1.067894f), Q30(1.194601f), Q30(1.369533f), }; -/* @name tns_tmp2_map - * Tables of the tmp2[] arrays of LPC coefficients used for TNS. - * The suffix _M_N[] indicate the values of coef_compress and coef_res - * respectively. - * @{ - */ -static const INTFLOAT tns_tmp2_map_1_3[4] = { - Q31(0.00000000f), Q31(-0.43388373f), Q31(0.64278758f), Q31(0.34202015f), -}; - -static const INTFLOAT tns_tmp2_map_0_3[8] = { - Q31(0.00000000f), Q31(-0.43388373f), Q31(-0.78183150f), Q31(-0.97492790f), - Q31(0.98480773f), Q31( 0.86602539f), Q31( 0.64278758f), Q31( 0.34202015f), -}; - -static const INTFLOAT tns_tmp2_map_1_4[8] = { - Q31(0.00000000f), Q31(-0.20791170f), Q31(-0.40673664f), Q31(-0.58778524f), - Q31(0.67369562f), Q31( 0.52643216f), Q31( 0.36124167f), Q31( 0.18374951f), -}; - -static const INTFLOAT tns_tmp2_map_0_4[16] = { - Q31( 0.00000000f), Q31(-0.20791170f), Q31(-0.40673664f), Q31(-0.58778524f), - Q31(-0.74314481f), Q31(-0.86602539f), Q31(-0.95105654f), Q31(-0.99452192f), - Q31( 0.99573416f), Q31( 0.96182561f), Q31( 0.89516330f), Q31( 0.79801720f), - Q31( 0.67369562f), Q31( 0.52643216f), Q31( 0.36124167f), Q31( 0.18374951f), -}; - -static const INTFLOAT * const tns_tmp2_map[4] = { - tns_tmp2_map_0_3, - tns_tmp2_map_0_4, - tns_tmp2_map_1_3, - tns_tmp2_map_1_4 -}; -// @} - static const int8_t tags_per_config[16] = { 0, 1, 1, 2, 3, 3, 4, 5, 0, 0, 0, 4, 5, 0, 5, 0 }; static const uint8_t aac_channel_layout_map[16][5][3] = { diff --git a/libavcodec/aactab.h b/libavcodec/aactab.h index c115b05..9f33380 100644 --- a/libavcodec/aactab.h +++ b/libavcodec/aactab.h @@ -41,6 +41,41 @@ * encoder. */ +/* @name tns_tmp2_map + * Tables of the tmp2[] arrays of LPC coefficients used for TNS. + * The suffix _M_N[] indicate the values of coef_compress and coef_res + * respectively. + * @{ + */ +static const INTFLOAT tns_tmp2_map_1_3[4] = { + Q31(0.00000000f), Q31(-0.43388373f), Q31(0.64278758f), Q31(0.34202015f), +}; + +static const INTFLOAT tns_tmp2_map_0_3[8] = { + Q31(0.00000000f), Q31(-0.43388373f), Q31(-0.78183150f), Q31(-0.97492790f), + Q31(0.98480773f), Q31( 0.86602539f), Q31( 0.64278758f), Q31( 0.34202015f), +}; + +static const INTFLOAT tns_tmp2_map_1_4[8] = { + Q31(0.00000000f), Q31(-0.20791170f), Q31(-0.40673664f), Q31(-0.58778524f), + Q31(0.67369562f), Q31( 0.52643216f), Q31( 0.36124167f), Q31( 0.18374951f), +}; + +static const INTFLOAT tns_tmp2_map_0_4[16] = { + Q31( 0.00000000f), Q31(-0.20791170f), Q31(-0.40673664f), Q31(-0.58778524f), + Q31(-0.74314481f), Q31(-0.86602539f), Q31(-0.95105654f), Q31(-0.99452192f), + Q31( 0.99573416f), Q31( 0.96182561f), Q31( 0.89516330f), Q31( 0.79801720f), + Q31( 0.67369562f), Q31( 0.52643216f), Q31( 0.36124167f), Q31( 0.18374951f), +}; + +static const INTFLOAT * const tns_tmp2_map[4] = { + tns_tmp2_map_0_3, + tns_tmp2_map_0_4, + tns_tmp2_map_1_3, + tns_tmp2_map_1_4 +}; +// @} + /* @name window coefficients * @{ */

1 0

avcodec/vc1dec: Re-order init to avoid initting hwaccel too early
by git＠videolan.org 29 Aug '15

29 Aug '15

ffmpeg | branch: master | Philip Langdale <philipl(a)overt.org> | Sat Aug 8 14:59:09 2015 -0700| [91f1115a0e027074c90bac6e57c2d0f4fe9efe8c] | committer: Philip Langdale avcodec/vc1dec: Re-order init to avoid initting hwaccel too early At least for vdpau, the hwaccel init code tries to check the video profile and ensure that there is a matching vdpau profile available. If it can't find a match, it will fail to initialise. In the case of wmv3/vc1, I observed initialisation to fail all the time. It turns out that this is due to the hwaccel being initialised very early in the codec init, before the profile has been extracted and set. Conceptually, it's a simple fix to reorder the init code, but it gets messy really fast because ff_get_format(), which is what implicitly trigger hwaccel init, is called multiple times through various shared init calls from h263, etc. It's incredibly hard to prove to my own satisfaction that it's safe to move the vc1 specific init code ahead of this generic code, but all the vc1 fate tests pass, and I've visually inspected a couple of samples and things seem correct. Signed-off-by: Philip Langdale <philipl(a)overt.org> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=91f1115a0e027074c… --- libavcodec/vc1dec.c | 52 +++++++++++++++++++++++++++------------------------ 1 file changed, 28 insertions(+), 24 deletions(-) diff --git a/libavcodec/vc1dec.c b/libavcodec/vc1dec.c index 04e22cc..acd29bc 100644 --- a/libavcodec/vc1dec.c +++ b/libavcodec/vc1dec.c @@ -428,30 +428,10 @@ static av_cold int vc1_decode_init(AVCodecContext *avctx) if (!avctx->extradata_size || !avctx->extradata) return -1; - if (!CONFIG_GRAY || !(avctx->flags & AV_CODEC_FLAG_GRAY)) - avctx->pix_fmt = ff_get_format(avctx, avctx->codec->pix_fmts); - else { - avctx->pix_fmt = AV_PIX_FMT_GRAY8; - if (avctx->color_range == AVCOL_RANGE_UNSPECIFIED) - avctx->color_range = AVCOL_RANGE_MPEG; - } v->s.avctx = avctx; if ((ret = ff_vc1_init_common(v)) < 0) return ret; - // ensure static VLC tables are initialized - if ((ret = ff_msmpeg4_decode_init(avctx)) < 0) - return ret; - if ((ret = ff_vc1_decode_init_alloc_tables(v)) < 0) - return ret; - // Hack to ensure the above functions will be called - // again once we know all necessary settings. - // That this is necessary might indicate a bug. - ff_vc1_decode_end(avctx); - - ff_blockdsp_init(&s->bdsp, avctx); - ff_h264chroma_init(&v->h264chroma, 8); - ff_qpeldsp_init(&s->qdsp); if (avctx->codec_id == AV_CODEC_ID_WMV3 || avctx->codec_id == AV_CODEC_ID_WMV3IMAGE) { int count = 0; @@ -524,14 +504,38 @@ static av_cold int vc1_decode_init(AVCodecContext *avctx) v->res_sprite = (avctx->codec_id == AV_CODEC_ID_VC1IMAGE); } - v->sprite_output_frame = av_frame_alloc(); - if (!v->sprite_output_frame) - return AVERROR(ENOMEM); - avctx->profile = v->profile; if (v->profile == PROFILE_ADVANCED) avctx->level = v->level; + if (!CONFIG_GRAY || !(avctx->flags & AV_CODEC_FLAG_GRAY)) + avctx->pix_fmt = ff_get_format(avctx, avctx->codec->pix_fmts); + else { + avctx->pix_fmt = AV_PIX_FMT_GRAY8; + if (avctx->color_range == AVCOL_RANGE_UNSPECIFIED) + avctx->color_range = AVCOL_RANGE_MPEG; + } + + // ensure static VLC tables are initialized + if ((ret = ff_msmpeg4_decode_init(avctx)) < 0) + return ret; + if ((ret = ff_vc1_decode_init_alloc_tables(v)) < 0) + return ret; + // Hack to ensure the above functions will be called + // again once we know all necessary settings. + // That this is necessary might indicate a bug. + ff_vc1_decode_end(avctx); + + ff_blockdsp_init(&s->bdsp, avctx); + ff_h264chroma_init(&v->h264chroma, 8); + ff_qpeldsp_init(&s->qdsp); + + // Must happen after calling ff_vc1_decode_end + // to avoid de-allocating the sprite_output_frame + v->sprite_output_frame = av_frame_alloc(); + if (!v->sprite_output_frame) + return AVERROR(ENOMEM); + avctx->has_b_frames = !!avctx->max_b_frames; if (v->color_prim == 1 || v->color_prim == 5 || v->color_prim == 6)

1 0