Turn on AAC rate control.
[ffmpeg.git] / libavcodec / aacenc.c
index 838863b..61e0f52 100644 (file)
  */
 
 /**
- * @file aacenc.c
+ * @file libavcodec/aacenc.c
  * AAC encoder
  */
 
 /***********************************
  *              TODOs:
- * psy model selection with some option
  * add sane pulse detection
+ * add temporal noise shaping
  ***********************************/
 
 #include "avcodec.h"
-#include "bitstream.h"
+#include "put_bits.h"
 #include "dsputil.h"
 #include "mpeg4audio.h"
 
-#include "aacpsy.h"
 #include "aac.h"
 #include "aactab.h"
+#include "aacenc.h"
+
+#include "psymodel.h"
 
 static const uint8_t swb_size_1024_96[] = {
     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8,
@@ -120,24 +122,14 @@ static const uint8_t *swb_size_128[] = {
 
 /** default channel configurations */
 static const uint8_t aac_chan_configs[6][5] = {
- {1, ID_SCE},                         // 1 channel  - single channel element
- {1, ID_CPE},                         // 2 channels - channel pair
- {2, ID_SCE, ID_CPE},                 // 3 channels - center + stereo
- {3, ID_SCE, ID_CPE, ID_SCE},         // 4 channels - front center + stereo + back center
- {3, ID_SCE, ID_CPE, ID_CPE},         // 5 channels - front center + stereo + back stereo
- {4, ID_SCE, ID_CPE, ID_CPE, ID_LFE}, // 6 channels - front center + stereo + back stereo + LFE
+ {1, TYPE_SCE},                               // 1 channel  - single channel element
+ {1, TYPE_CPE},                               // 2 channels - channel pair
+ {2, TYPE_SCE, TYPE_CPE},                     // 3 channels - center + stereo
+ {3, TYPE_SCE, TYPE_CPE, TYPE_SCE},           // 4 channels - front center + stereo + back center
+ {3, TYPE_SCE, TYPE_CPE, TYPE_CPE},           // 5 channels - front center + stereo + back stereo
+ {4, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_LFE}, // 6 channels - front center + stereo + back stereo + LFE
 };
 
-/**
- * AAC encoder context
- */
-typedef struct {
-    PutBitContext pb;
-    MDCTContext mdct1024;                        ///< long (1024 samples) frame transform context
-    MDCTContext mdct128;                         ///< short (128 samples) frame transform context
-    DSPContext  dsp;
-} AACEncContext;
-
 /**
  * Make AAC audio config object.
  * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
@@ -162,83 +154,259 @@ static av_cold int aac_encode_init(AVCodecContext *avctx)
 {
     AACEncContext *s = avctx->priv_data;
     int i;
+    const uint8_t *sizes[2];
+    int lengths[2];
 
     avctx->frame_size = 1024;
 
-    for(i = 0; i < 16; i++)
-        if(avctx->sample_rate == ff_mpeg4audio_sample_rates[i])
+    for (i = 0; i < 16; i++)
+        if (avctx->sample_rate == ff_mpeg4audio_sample_rates[i])
             break;
-    if(i == 16){
+    if (i == 16) {
         av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate %d\n", avctx->sample_rate);
         return -1;
     }
-    if(avctx->channels > 6){
+    if (avctx->channels > 6) {
         av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %d\n", avctx->channels);
         return -1;
     }
     s->samplerate_index = i;
-    s->swb_sizes1024 = swb_size_1024[i];
-    s->swb_num1024   = ff_aac_num_swb_1024[i];
-    s->swb_sizes128  = swb_size_128[i];
-    s->swb_num128    = ff_aac_num_swb_128[i];
 
     dsputil_init(&s->dsp, avctx);
-    ff_mdct_init(&s->mdct1024, 11, 0);
-    ff_mdct_init(&s->mdct128,   8, 0);
+    ff_mdct_init(&s->mdct1024, 11, 0, 1.0);
+    ff_mdct_init(&s->mdct128,   8, 0, 1.0);
     // window init
     ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
     ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
     ff_sine_window_init(ff_sine_1024, 1024);
     ff_sine_window_init(ff_sine_128, 128);
 
-    s->samples = av_malloc(2 * 1024 * avctx->channels * sizeof(s->samples[0]));
-    s->cpe = av_mallocz(sizeof(ChannelElement) * aac_chan_configs[avctx->channels-1][0]);
-    if(ff_aac_psy_init(&s->psy, avctx, AAC_PSY_3GPP, aac_chan_configs[avctx->channels-1][0], 0, s->swb_sizes1024, s->swb_num1024, s->swb_sizes128, s->swb_num128) < 0){
-        av_log(avctx, AV_LOG_ERROR, "Cannot initialize selected model.\n");
-        return -1;
-    }
-    avctx->extradata = av_malloc(2);
+    s->samples            = av_malloc(2 * 1024 * avctx->channels * sizeof(s->samples[0]));
+    s->cpe                = av_mallocz(sizeof(ChannelElement) * aac_chan_configs[avctx->channels-1][0]);
+    avctx->extradata      = av_malloc(2);
     avctx->extradata_size = 2;
     put_audio_specific_config(avctx);
+
+    sizes[0]   = swb_size_1024[i];
+    sizes[1]   = swb_size_128[i];
+    lengths[0] = ff_aac_num_swb_1024[i];
+    lengths[1] = ff_aac_num_swb_128[i];
+    ff_psy_init(&s->psy, avctx, 2, sizes, lengths);
+    s->psypp = ff_psy_preprocess_init(avctx);
+    s->coder = &ff_aac_coders[0];
+
+    s->lambda = avctx->global_quality ? avctx->global_quality : 120;
+#if !CONFIG_HARDCODED_TABLES
+    for (i = 0; i < 428; i++)
+        ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
+#endif /* CONFIG_HARDCODED_TABLES */
+
+    if (avctx->channels > 5)
+        av_log(avctx, AV_LOG_ERROR, "This encoder does not yet enforce the restrictions on LFEs. "
+               "The output will most likely be an illegal bitstream.\n");
+
     return 0;
 }
 
+static void apply_window_and_mdct(AVCodecContext *avctx, AACEncContext *s,
+                                  SingleChannelElement *sce, short *audio, int channel)
+{
+    int i, j, k;
+    const float * lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
+    const float * swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
+    const float * pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
+
+    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
+        memcpy(s->output, sce->saved, sizeof(float)*1024);
+        if (sce->ics.window_sequence[0] == LONG_STOP_SEQUENCE) {
+            memset(s->output, 0, sizeof(s->output[0]) * 448);
+            for (i = 448; i < 576; i++)
+                s->output[i] = sce->saved[i] * pwindow[i - 448];
+            for (i = 576; i < 704; i++)
+                s->output[i] = sce->saved[i];
+        }
+        if (sce->ics.window_sequence[0] != LONG_START_SEQUENCE) {
+            j = channel;
+            for (i = 0; i < 1024; i++, j += avctx->channels) {
+                s->output[i+1024]         = audio[j] * lwindow[1024 - i - 1];
+                sce->saved[i] = audio[j] * lwindow[i];
+            }
+        } else {
+            j = channel;
+            for (i = 0; i < 448; i++, j += avctx->channels)
+                s->output[i+1024]         = audio[j];
+            for (i = 448; i < 576; i++, j += avctx->channels)
+                s->output[i+1024]         = audio[j] * swindow[576 - i - 1];
+            memset(s->output+1024+576, 0, sizeof(s->output[0]) * 448);
+            j = channel;
+            for (i = 0; i < 1024; i++, j += avctx->channels)
+                sce->saved[i] = audio[j];
+        }
+        ff_mdct_calc(&s->mdct1024, sce->coeffs, s->output);
+    } else {
+        j = channel;
+        for (k = 0; k < 1024; k += 128) {
+            for (i = 448 + k; i < 448 + k + 256; i++)
+                s->output[i - 448 - k] = (i < 1024)
+                                         ? sce->saved[i]
+                                         : audio[channel + (i-1024)*avctx->channels];
+            s->dsp.vector_fmul        (s->output,     k ?  swindow : pwindow, 128);
+            s->dsp.vector_fmul_reverse(s->output+128, s->output+128, swindow, 128);
+            ff_mdct_calc(&s->mdct128, sce->coeffs + k, s->output);
+        }
+        j = channel;
+        for (i = 0; i < 1024; i++, j += avctx->channels)
+            sce->saved[i] = audio[j];
+    }
+}
+
 /**
  * Encode ics_info element.
  * @see Table 4.6 (syntax of ics_info)
  */
-static void put_ics_info(AVCodecContext *avctx, IndividualChannelStream *info)
+static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
 {
-    AACEncContext *s = avctx->priv_data;
-    int i;
+    int w;
 
     put_bits(&s->pb, 1, 0);                // ics_reserved bit
     put_bits(&s->pb, 2, info->window_sequence[0]);
     put_bits(&s->pb, 1, info->use_kb_window[0]);
-    if(info->window_sequence[0] != EIGHT_SHORT_SEQUENCE){
+    if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
         put_bits(&s->pb, 6, info->max_sfb);
         put_bits(&s->pb, 1, 0);            // no prediction
-    }else{
+    } else {
         put_bits(&s->pb, 4, info->max_sfb);
-        for(i = 1; i < info->num_windows; i++)
-            put_bits(&s->pb, 1, info->group_len[i]);
+        for (w = 1; w < 8; w++)
+            put_bits(&s->pb, 1, !info->group_len[w]);
+    }
+}
+
+/**
+ * Encode MS data.
+ * @see 4.6.8.1 "Joint Coding - M/S Stereo"
+ */
+static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
+{
+    int i, w;
+
+    put_bits(pb, 2, cpe->ms_mode);
+    if (cpe->ms_mode == 1)
+        for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
+            for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
+                put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
+}
+
+/**
+ * Produce integer coefficients from scalefactors provided by the model.
+ */
+static void adjust_frame_information(AACEncContext *apc, ChannelElement *cpe, int chans)
+{
+    int i, w, w2, g, ch;
+    int start, sum, maxsfb, cmaxsfb;
+
+    for (ch = 0; ch < chans; ch++) {
+        IndividualChannelStream *ics = &cpe->ch[ch].ics;
+        start = 0;
+        maxsfb = 0;
+        cpe->ch[ch].pulse.num_pulse = 0;
+        for (w = 0; w < ics->num_windows*16; w += 16) {
+            for (g = 0; g < ics->num_swb; g++) {
+                sum = 0;
+                //apply M/S
+                if (!ch && cpe->ms_mask[w + g]) {
+                    for (i = 0; i < ics->swb_sizes[g]; i++) {
+                        cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0;
+                        cpe->ch[1].coeffs[start+i] =  cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i];
+                    }
+                }
+                start += ics->swb_sizes[g];
+            }
+            for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w+cmaxsfb-1]; cmaxsfb--)
+                ;
+            maxsfb = FFMAX(maxsfb, cmaxsfb);
+        }
+        ics->max_sfb = maxsfb;
+
+        //adjust zero bands for window groups
+        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
+            for (g = 0; g < ics->max_sfb; g++) {
+                i = 1;
+                for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
+                    if (!cpe->ch[ch].zeroes[w2*16 + g]) {
+                        i = 0;
+                        break;
+                    }
+                }
+                cpe->ch[ch].zeroes[w*16 + g] = i;
+            }
+        }
+    }
+
+    if (chans > 1 && cpe->common_window) {
+        IndividualChannelStream *ics0 = &cpe->ch[0].ics;
+        IndividualChannelStream *ics1 = &cpe->ch[1].ics;
+        int msc = 0;
+        ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
+        ics1->max_sfb = ics0->max_sfb;
+        for (w = 0; w < ics0->num_windows*16; w += 16)
+            for (i = 0; i < ics0->max_sfb; i++)
+                if (cpe->ms_mask[w+i])
+                    msc++;
+        if (msc == 0 || ics0->max_sfb == 0)
+            cpe->ms_mode = 0;
+        else
+            cpe->ms_mode = msc < ics0->max_sfb ? 1 : 2;
+    }
+}
+
+/**
+ * Encode scalefactor band coding type.
+ */
+static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
+{
+    int w;
+
+    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
+        s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
+}
+
+/**
+ * Encode scalefactors.
+ */
+static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
+                                 SingleChannelElement *sce)
+{
+    int off = sce->sf_idx[0], diff;
+    int i, w;
+
+    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
+        for (i = 0; i < sce->ics.max_sfb; i++) {
+            if (!sce->zeroes[w*16 + i]) {
+                diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO;
+                if (diff < 0 || diff > 120)
+                    av_log(avctx, AV_LOG_ERROR, "Scalefactor difference is too big to be coded\n");
+                off = sce->sf_idx[w*16 + i];
+                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
+            }
+        }
     }
 }
 
 /**
  * Encode pulse data.
  */
-static void encode_pulses(AVCodecContext *avctx, AACEncContext *s, Pulse *pulse, int channel)
+static void encode_pulses(AACEncContext *s, Pulse *pulse)
 {
     int i;
 
     put_bits(&s->pb, 1, !!pulse->num_pulse);
-    if(!pulse->num_pulse) return;
+    if (!pulse->num_pulse)
+        return;
 
     put_bits(&s->pb, 2, pulse->num_pulse - 1);
     put_bits(&s->pb, 6, pulse->start);
-    for(i = 0; i < pulse->num_pulse; i++){
-        put_bits(&s->pb, 5, pulse->offset[i]);
+    for (i = 0; i < pulse->num_pulse; i++) {
+        put_bits(&s->pb, 5, pulse->pos[i]);
         put_bits(&s->pb, 4, pulse->amp[i]);
     }
 }
@@ -246,54 +414,197 @@ static void encode_pulses(AVCodecContext *avctx, AACEncContext *s, Pulse *pulse,
 /**
  * Encode spectral coefficients processed by psychoacoustic model.
  */
-static void encode_spectral_coeffs(AVCodecContext *avctx, AACEncContext *s, ChannelElement *cpe, int channel)
+static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
 {
-    int start, i, w, w2, wg;
+    int start, i, w, w2;
 
-    w = 0;
-    for(wg = 0; wg < cpe->ch[channel].ics.num_window_groups; wg++){
+    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
         start = 0;
-        for(i = 0; i < cpe->ch[channel].ics.max_sfb; i++){
-            if(cpe->ch[channel].zeroes[w][i]){
-                start += cpe->ch[channel].ics.swb_sizes[i];
+        for (i = 0; i < sce->ics.max_sfb; i++) {
+            if (sce->zeroes[w*16 + i]) {
+                start += sce->ics.swb_sizes[i];
                 continue;
             }
-            for(w2 = w; w2 < w + cpe->ch[channel].ics.group_len[wg]; w2++){
-                encode_band_coeffs(s, cpe, channel, start + w2*128, cpe->ch[channel].ics.swb_sizes[i], cpe->ch[channel].band_type[w][i]);
-            }
-            start += cpe->ch[channel].ics.swb_sizes[i];
+            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->ics.swb_sizes[i],
+                                                   sce->sf_idx[w*16 + i],
+                                                   sce->band_type[w*16 + i],
+                                                   s->lambda);
+            start += sce->ics.swb_sizes[i];
         }
-        w += cpe->ch[channel].ics.group_len[wg];
     }
 }
 
+/**
+ * Encode one channel of audio data.
+ */
+static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
+                                     SingleChannelElement *sce,
+                                     int common_window)
+{
+    put_bits(&s->pb, 8, sce->sf_idx[0]);
+    if (!common_window)
+        put_ics_info(s, &sce->ics);
+    encode_band_info(s, sce);
+    encode_scale_factors(avctx, s, sce);
+    encode_pulses(s, &sce->pulse);
+    put_bits(&s->pb, 1, 0); //tns
+    put_bits(&s->pb, 1, 0); //ssr
+    encode_spectral_coeffs(s, sce);
+    return 0;
+}
+
 /**
  * Write some auxiliary information about the created AAC file.
  */
-static void put_bitstream_info(AVCodecContext *avctx, AACEncContext *s, const char *name)
+static void put_bitstream_info(AVCodecContext *avctx, AACEncContext *s,
+                               const char *name)
 {
     int i, namelen, padbits;
 
     namelen = strlen(name) + 2;
-    put_bits(&s->pb, 3, ID_FIL);
+    put_bits(&s->pb, 3, TYPE_FIL);
     put_bits(&s->pb, 4, FFMIN(namelen, 15));
-    if(namelen >= 15)
+    if (namelen >= 15)
         put_bits(&s->pb, 8, namelen - 16);
     put_bits(&s->pb, 4, 0); //extension type - filler
     padbits = 8 - (put_bits_count(&s->pb) & 7);
     align_put_bits(&s->pb);
-    for(i = 0; i < namelen - 2; i++)
+    for (i = 0; i < namelen - 2; i++)
         put_bits(&s->pb, 8, name[i]);
     put_bits(&s->pb, 12 - padbits, 0);
 }
 
+static int aac_encode_frame(AVCodecContext *avctx,
+                            uint8_t *frame, int buf_size, void *data)
+{
+    AACEncContext *s = avctx->priv_data;
+    int16_t *samples = s->samples, *samples2, *la;
+    ChannelElement *cpe;
+    int i, j, chans, tag, start_ch;
+    const uint8_t *chan_map = aac_chan_configs[avctx->channels-1];
+    int chan_el_counter[4];
+
+    if (s->last_frame)
+        return 0;
+    if (data) {
+        if (!s->psypp) {
+            memcpy(s->samples + 1024 * avctx->channels, data,
+                   1024 * avctx->channels * sizeof(s->samples[0]));
+        } else {
+            start_ch = 0;
+            samples2 = s->samples + 1024 * avctx->channels;
+            for (i = 0; i < chan_map[0]; i++) {
+                tag = chan_map[i+1];
+                chans = tag == TYPE_CPE ? 2 : 1;
+                ff_psy_preprocess(s->psypp, (uint16_t*)data + start_ch,
+                                  samples2 + start_ch, start_ch, chans);
+                start_ch += chans;
+            }
+        }
+    }
+    if (!avctx->frame_number) {
+        memcpy(s->samples, s->samples + 1024 * avctx->channels,
+               1024 * avctx->channels * sizeof(s->samples[0]));
+        return 0;
+    }
+
+    init_put_bits(&s->pb, frame, buf_size*8);
+    if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & CODEC_FLAG_BITEXACT))
+        put_bitstream_info(avctx, s, LIBAVCODEC_IDENT);
+    start_ch = 0;
+    memset(chan_el_counter, 0, sizeof(chan_el_counter));
+    for (i = 0; i < chan_map[0]; i++) {
+        FFPsyWindowInfo wi[2];
+        tag      = chan_map[i+1];
+        chans    = tag == TYPE_CPE ? 2 : 1;
+        cpe      = &s->cpe[i];
+        samples2 = samples + start_ch;
+        la       = samples2 + 1024 * avctx->channels + start_ch;
+        if (!data)
+            la = NULL;
+        for (j = 0; j < chans; j++) {
+            IndividualChannelStream *ics = &cpe->ch[j].ics;
+            int k;
+            wi[j] = ff_psy_suggest_window(&s->psy, samples2, la, start_ch + j, ics->window_sequence[0]);
+            ics->window_sequence[1] = ics->window_sequence[0];
+            ics->window_sequence[0] = wi[j].window_type[0];
+            ics->use_kb_window[1]   = ics->use_kb_window[0];
+            ics->use_kb_window[0]   = wi[j].window_shape;
+            ics->num_windows        = wi[j].num_windows;
+            ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];
+            ics->num_swb            = s->psy.num_bands[ics->num_windows == 8];
+            for (k = 0; k < ics->num_windows; k++)
+                ics->group_len[k] = wi[j].grouping[k];
+
+            s->cur_channel = start_ch + j;
+            apply_window_and_mdct(avctx, s, &cpe->ch[j], samples2, j);
+            s->coder->search_for_quantizers(avctx, s, &cpe->ch[j], s->lambda);
+        }
+        cpe->common_window = 0;
+        if (chans > 1
+            && wi[0].window_type[0] == wi[1].window_type[0]
+            && wi[0].window_shape   == wi[1].window_shape) {
+
+            cpe->common_window = 1;
+            for (j = 0; j < wi[0].num_windows; j++) {
+                if (wi[0].grouping[j] != wi[1].grouping[j]) {
+                    cpe->common_window = 0;
+                    break;
+                }
+            }
+        }
+        if (cpe->common_window && s->coder->search_for_ms)
+            s->coder->search_for_ms(s, cpe, s->lambda);
+        adjust_frame_information(s, cpe, chans);
+        put_bits(&s->pb, 3, tag);
+        put_bits(&s->pb, 4, chan_el_counter[tag]++);
+        if (chans == 2) {
+            put_bits(&s->pb, 1, cpe->common_window);
+            if (cpe->common_window) {
+                put_ics_info(s, &cpe->ch[0].ics);
+                encode_ms_info(&s->pb, cpe);
+            }
+        }
+        for (j = 0; j < chans; j++) {
+            s->cur_channel = start_ch + j;
+            ff_psy_set_band_info(&s->psy, s->cur_channel, cpe->ch[j].coeffs, &wi[j]);
+            encode_individual_channel(avctx, s, &cpe->ch[j], cpe->common_window);
+        }
+        start_ch += chans;
+    }
+
+    put_bits(&s->pb, 3, TYPE_END);
+    flush_put_bits(&s->pb);
+    avctx->frame_bits = put_bits_count(&s->pb);
+
+    // rate control stuff
+    if (!(avctx->flags & CODEC_FLAG_QSCALE)) {
+        float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
+        s->lambda *= ratio;
+        s->lambda = fminf(s->lambda, 65536.f);
+    }
+
+    if (avctx->frame_bits > 6144*avctx->channels)
+        av_log(avctx, AV_LOG_ERROR, "input buffer violation %d > %d.\n",
+               avctx->frame_bits, 6144*avctx->channels);
+
+    if (!data)
+        s->last_frame = 1;
+    memcpy(s->samples, s->samples + 1024 * avctx->channels,
+           1024 * avctx->channels * sizeof(s->samples[0]));
+    return put_bits_count(&s->pb)>>3;
+}
+
 static av_cold int aac_encode_end(AVCodecContext *avctx)
 {
     AACEncContext *s = avctx->priv_data;
 
     ff_mdct_end(&s->mdct1024);
     ff_mdct_end(&s->mdct128);
-    ff_aac_psy_end(&s->psy);
+    ff_psy_end(&s->psy);
+    ff_psy_preprocess_end(s->psypp);
     av_freep(&s->samples);
     av_freep(&s->cpe);
     return 0;