Switching the vorbis encoder to use a buffer queue for input frames allows saving lookahead samples more easily and safely for psychoacoustic systems, requiring less pointer arithmetic in the case of transient windows.
Unified · Split
| @@ -1044,20 +1044,117 @@ static int apply_window_and_mdct(vorbis_enc_context *venc, · 1043 unmodified lines | |||
| 1044 | return 1; | 1044 | return 1; |
| 1045 | } | 1045 | } |
| 1046 | 1046 | ||
| 1047 | /* Used for padding the last encoded packet */ | ||
| 1048 | static AVFrame *spawn_empty_frame(AVCodecContext *avctx, int channels) | ||
| 1049 | { | ||
| 1050 | AVFrame *f = av_frame_alloc(); | ||
| 1051 | |||
| 1052 | if (!f) | ||
| 1053 | return NULL; | ||
| 1054 | |||
| 1055 | f->format = avctx->sample_fmt; | ||
| 1056 | f->nb_samples = avctx->frame_size; | ||
| 1057 | f->channel_layout = avctx->channel_layout; | ||
| 1058 | |||
| 1059 | if (av_frame_get_buffer(f, 4)) { | ||
| 1060 | av_frame_free(&f); | ||
| 1061 | return NULL; | ||
| 1062 | } | ||
| 1063 | |||
| 1064 | for (int ch = 0; ch < channels; ch++) { | ||
| 1065 | size_t bps = av_get_bytes_per_sample(f->format); | ||
| 1066 | memset(f->extended_data[ch], 0, bps * f->nb_samples); | ||
| 1067 | } | ||
| 1068 | return f; | ||
| 1069 | } | ||
| 1070 | |||
| 1071 | static float **alloc_audio_arrays(int channels, int frame_size) | ||
| 1072 | { | ||
| 1073 | float **audio = av_mallocz_array(channels, sizeof(float *)); | ||
| 1074 | if (!audio) | ||
| 1075 | return NULL; | ||
| 1076 | |||
| 1077 | for (int ch = 0; ch < channels; ch++) { | ||
| 1078 | audio[ch] = av_mallocz_array(frame_size, sizeof(float)); | ||
| 1079 | if (!audio[ch]) { | ||
| 1080 | // alloc has failed, free everything allocated thus far | ||
| 1081 | for (ch--; ch >= 0; ch--) | ||
| 1082 | av_free(audio[ch]); | ||
| 1083 | av_free(audio); | ||
| 1084 | return NULL; | ||
| 1085 | } | ||
| 1086 | } | ||
| 1087 | |||
| 1088 | return audio; | ||
| 1089 | } | ||
| 1090 | |||
| 1091 | /* Concatenate audio frames into an appropriately sized array of samples */ | ||
| 1092 | static void move_audio(vorbis_enc_context *venc, float **audio, int *samples, int sf_size) | ||
| 1093 | { | ||
| 1094 | AVFrame *cur = NULL; | ||
| 1095 | int frame_size = 1 << (venc->log2_blocksize[1] - 1); | ||
| 1096 | int subframes = frame_size / sf_size; | ||
| 1097 | |||
| 1098 | for (int sf = 0; sf < subframes; sf++) { | ||
| 1099 | cur = ff_bufqueue_get(&venc->bufqueue); | ||
| 1100 | *samples += cur->nb_samples; | ||
| 1101 | |||
| 1102 | for (int ch = 0; ch < venc->channels; ch++) { | ||
| 1103 | const float *input = (float *) cur->extended_data[ch]; | ||
| 1104 | const size_t len = cur->nb_samples * sizeof(float); | ||
| 1105 | memcpy(&audio[ch][sf*sf_size], input, len); | ||
| 1106 | } | ||
| 1107 | av_frame_free(&cur); | ||
| 1108 | } | ||
| 1109 | } | ||
| 1110 | |||
| 1047 | static int vorbis_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, | 1111 | static int vorbis_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
| 1048 | const AVFrame *frame, int *got_packet_ptr) | 1112 | const AVFrame *frame, int *got_packet_ptr) |
| 1049 | { | 1113 | { |
| 1050 | vorbis_enc_context *venc = avctx->priv_data; | 1114 | vorbis_enc_context *venc = avctx->priv_data; |
| 1051 | float **audio = frame ? (float **)frame->extended_data : NULL; | 1115 | float **audio = NULL; |
| 1052 | int samples = frame ? frame->nb_samples : 0; | 1116 | int i, ret, need_more; |
| 1117 | int samples = 0, frame_size = 1 << (venc->log2_blocksize[1] - 1); | ||
| 1053 | vorbis_enc_mode *mode; | 1118 | vorbis_enc_mode *mode; |
| 1054 | vorbis_enc_mapping *mapping; | 1119 | vorbis_enc_mapping *mapping; |
| 1055 | PutBitContext pb; | 1120 | PutBitContext pb; |
| 1056 | int i, ret; | 1121 | |
| 1122 | if (frame) { | ||
| 1123 | if ((ret = ff_af_queue_add(&venc->afq, frame)) < 0) | ||
| 1124 | return ret; | ||
| 1125 | ff_bufqueue_add(avctx, &venc->bufqueue, av_frame_clone(frame)); | ||
| 1126 | } else | ||
| 1127 | if (!venc->afq.remaining_samples) | ||
| 1128 | return 0; | ||
| 1129 | |||
| 1130 | need_more = venc->bufqueue.available * avctx->frame_size < frame_size; | ||
| 1131 | need_more = frame && need_more; | ||
| 1132 | if (need_more) | ||
| 1133 | return 0; | ||
| 1134 | |||
| 1135 | audio = alloc_audio_arrays(venc->channels, frame_size); | ||
| 1136 | if (!audio) | ||
| 1137 | return AVERROR(ENOMEM); | ||
| 1138 | |||
| 1139 | /* Pad the bufqueue with empty frames for encoding the last packet. */ | ||
| 1140 | if (!frame) { | ||
| 1141 | if (venc->bufqueue.available * avctx->frame_size < frame_size) { | ||
| 1142 | int frames_needed = (frame_size/avctx->frame_size) - venc->bufqueue.available; | ||
| 1143 | |||
| 1144 | for (int i = 0; i < frames_needed; i++) { | ||
| 1145 | AVFrame *empty = spawn_empty_frame(avctx, venc->channels); | ||
| 1146 | if (!empty) | ||
| 1147 | return AVERROR(ENOMEM); | ||
| 1148 | |||
| 1149 | ff_bufqueue_add(avctx, &venc->bufqueue, empty); | ||
| 1150 | } | ||
| 1151 | } | ||
| 1152 | } | ||
| 1153 | |||
| 1154 | move_audio(venc, audio, &samples, avctx->frame_size); | ||
| 1057 | 1155 | ||
| 1058 | if (!apply_window_and_mdct(venc, audio, samples)) | 1156 | if (!apply_window_and_mdct(venc, audio, samples)) |
| 1059 | return 0; | 1157 | return 0; |
| 1060 | samples = 1 << (venc->log2_blocksize[0] - 1); | ||
| 1061 | 1158 | ||
| 1062 | if ((ret = ff_alloc_packet2(avctx, avpkt, 8192, 0)) < 0) | 1159 | if ((ret = ff_alloc_packet2(avctx, avpkt, 8192, 0)) < 0) |
| 1063 | return ret; | 1160 | return ret; |
| @@ -1116,16 +1213,11 @@ static int vorbis_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, · 52 unmodified lines | |||
| 1116 | flush_put_bits(&pb); | 1213 | flush_put_bits(&pb); |
| 1117 | avpkt->size = put_bits_count(&pb) >> 3; | 1214 | avpkt->size = put_bits_count(&pb) >> 3; |
| 1118 | 1215 | ||
| 1119 | avpkt->duration = ff_samples_to_time_base(avctx, avctx->frame_size); | 1216 | for (int ch = 0; ch < venc->channels; ch++) |
| 1120 | if (frame) { | 1217 | av_free(audio[ch]); |
| 1121 | if (frame->pts != AV_NOPTS_VALUE) | 1218 | av_free(audio); |
| 1122 | avpkt->pts = ff_samples_to_time_base(avctx, frame->pts); | ||
| 1123 | } else { | ||
| 1124 | avpkt->pts = venc->next_pts; | ||
| 1125 | } | ||
| 1126 | if (avpkt->pts != AV_NOPTS_VALUE) | ||
| 1127 | venc->next_pts = avpkt->pts + avpkt->duration; | ||
| 1128 | 1219 | ||
| 1220 | ff_af_queue_remove(&venc->afq, frame_size, &avpkt->pts, &avpkt->duration); | ||
| 1129 | *got_packet_ptr = 1; | 1221 | *got_packet_ptr = 1; |
| 1130 | return 0; | 1222 | return 0; |
| 1131 | } | 1223 | } |
| @@ -1217,7 +1309,7 @@ static av_cold int vorbis_encode_init(AVCodecContext *avctx) · 85 unmodified lines | |||
| 1217 | goto error; | 1309 | goto error; |
| 1218 | avctx->extradata_size = ret; | 1310 | avctx->extradata_size = ret; |
| 1219 | 1311 | ||
| 1220 | avctx->frame_size = 1 << (venc->log2_blocksize[0] - 1); | 1312 | avctx->frame_size = 64; |
| 1221 | 1313 | ||
| 1222 | ff_af_queue_init(avctx, &venc->afq); | 1314 | ff_af_queue_init(avctx, &venc->afq); |
| 1223 | 1315 | ||