| 1 | /* |
| 2 | * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com> |
| 3 | * |
| 4 | * This file is part of FFmpeg. |
| 5 | * |
| 6 | * FFmpeg is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2.1 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * FFmpeg is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with FFmpeg; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | */ |
| 20 | |
| 21 | /** |
| 22 | * @file |
| 23 | * @brief Theora encoder using libtheora. |
| 24 | * @author Paul Richards <paul.richards@gmail.com> |
| 25 | * |
| 26 | * A lot of this is copy / paste from other output codecs in |
| 27 | * libavcodec or pure guesswork (or both). |
| 28 | * |
| 29 | * I have used t_ prefixes on variables which are libtheora types |
| 30 | * and o_ prefixes on variables which are libogg types. |
| 31 | */ |
| 32 | |
| 33 | /* FFmpeg includes */ |
| 34 | #include "libavutil/common.h" |
| 35 | #include "libavutil/intreadwrite.h" |
| 36 | #include "libavutil/pixdesc.h" |
| 37 | #include "libavutil/log.h" |
| 38 | #include "libavutil/base64.h" |
| 39 | #include "avcodec.h" |
| 40 | #include "internal.h" |
| 41 | |
| 42 | /* libtheora includes */ |
| 43 | #include <theora/theoraenc.h> |
| 44 | |
| 45 | typedef struct TheoraContext { |
| 46 | th_enc_ctx *t_state; |
| 47 | uint8_t *stats; |
| 48 | int stats_size; |
| 49 | int stats_offset; |
| 50 | int uv_hshift; |
| 51 | int uv_vshift; |
| 52 | int keyframe_mask; |
| 53 | } TheoraContext; |
| 54 | |
| 55 | /** Concatenate an ogg_packet into the extradata. */ |
| 56 | static int concatenate_packet(unsigned int* offset, |
| 57 | AVCodecContext* avc_context, |
| 58 | const ogg_packet* packet) |
| 59 | { |
| 60 | const char* message = NULL; |
| 61 | int newsize = avc_context->extradata_size + 2 + packet->bytes; |
| 62 | int err = AVERROR_INVALIDDATA; |
| 63 | |
| 64 | if (packet->bytes < 0) { |
| 65 | message = "ogg_packet has negative size"; |
| 66 | } else if (packet->bytes > 0xffff) { |
| 67 | message = "ogg_packet is larger than 65535 bytes"; |
| 68 | } else if (newsize < avc_context->extradata_size) { |
| 69 | message = "extradata_size would overflow"; |
| 70 | } else { |
| 71 | if ((err = av_reallocp(&avc_context->extradata, newsize)) < 0) { |
| 72 | avc_context->extradata_size = 0; |
| 73 | message = "av_realloc failed"; |
| 74 | } |
| 75 | } |
| 76 | if (message) { |
| 77 | av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message); |
| 78 | return err; |
| 79 | } |
| 80 | |
| 81 | avc_context->extradata_size = newsize; |
| 82 | AV_WB16(avc_context->extradata + (*offset), packet->bytes); |
| 83 | *offset += 2; |
| 84 | memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes); |
| 85 | (*offset) += packet->bytes; |
| 86 | return 0; |
| 87 | } |
| 88 | |
| 89 | static int get_stats(AVCodecContext *avctx, int eos) |
| 90 | { |
| 91 | #ifdef TH_ENCCTL_2PASS_OUT |
| 92 | TheoraContext *h = avctx->priv_data; |
| 93 | uint8_t *buf; |
| 94 | int bytes; |
| 95 | |
| 96 | bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf)); |
| 97 | if (bytes < 0) { |
| 98 | av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n"); |
| 99 | return AVERROR_EXTERNAL; |
| 100 | } |
| 101 | if (!eos) { |
| 102 | h->stats = av_fast_realloc(h->stats, &h->stats_size, |
| 103 | h->stats_offset + bytes); |
| 104 | memcpy(h->stats + h->stats_offset, buf, bytes); |
| 105 | h->stats_offset += bytes; |
| 106 | } else { |
| 107 | int b64_size = AV_BASE64_SIZE(h->stats_offset); |
| 108 | // libtheora generates a summary header at the end |
| 109 | memcpy(h->stats, buf, bytes); |
| 110 | avctx->stats_out = av_malloc(b64_size); |
| 111 | av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset); |
| 112 | } |
| 113 | return 0; |
| 114 | #else |
| 115 | av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n"); |
| 116 | return AVERROR(ENOSUP); |
| 117 | #endif |
| 118 | } |
| 119 | |
| 120 | // libtheora won't read the entire buffer we give it at once, so we have to |
| 121 | // repeatedly submit it... |
| 122 | static int submit_stats(AVCodecContext *avctx) |
| 123 | { |
| 124 | #ifdef TH_ENCCTL_2PASS_IN |
| 125 | TheoraContext *h = avctx->priv_data; |
| 126 | int bytes; |
| 127 | if (!h->stats) { |
| 128 | if (!avctx->stats_in) { |
| 129 | av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n"); |
| 130 | return AVERROR(EINVAL); |
| 131 | } |
| 132 | h->stats_size = strlen(avctx->stats_in) * 3/4; |
| 133 | h->stats = av_malloc(h->stats_size); |
| 134 | h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size); |
| 135 | } |
| 136 | while (h->stats_size - h->stats_offset > 0) { |
| 137 | bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN, |
| 138 | h->stats + h->stats_offset, |
| 139 | h->stats_size - h->stats_offset); |
| 140 | if (bytes < 0) { |
| 141 | av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n"); |
| 142 | return AVERROR_EXTERNAL; |
| 143 | } |
| 144 | if (!bytes) |
| 145 | return 0; |
| 146 | h->stats_offset += bytes; |
| 147 | } |
| 148 | return 0; |
| 149 | #else |
| 150 | av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n"); |
| 151 | return AVERROR(ENOSUP); |
| 152 | #endif |
| 153 | } |
| 154 | |
| 155 | static av_cold int encode_init(AVCodecContext* avc_context) |
| 156 | { |
| 157 | th_info t_info; |
| 158 | th_comment t_comment; |
| 159 | ogg_packet o_packet; |
| 160 | unsigned int offset; |
| 161 | TheoraContext *h = avc_context->priv_data; |
| 162 | uint32_t gop_size = avc_context->gop_size; |
| 163 | int ret; |
| 164 | |
| 165 | /* Set up the theora_info struct */ |
| 166 | th_info_init(&t_info); |
| 167 | t_info.frame_width = FFALIGN(avc_context->width, 16); |
| 168 | t_info.frame_height = FFALIGN(avc_context->height, 16); |
| 169 | t_info.pic_width = avc_context->width; |
| 170 | t_info.pic_height = avc_context->height; |
| 171 | t_info.pic_x = 0; |
| 172 | t_info.pic_y = 0; |
| 173 | /* Swap numerator and denominator as time_base in AVCodecContext gives the |
| 174 | * time period between frames, but theora_info needs the framerate. */ |
| 175 | t_info.fps_numerator = avc_context->time_base.den; |
| 176 | t_info.fps_denominator = avc_context->time_base.num; |
| 177 | if (avc_context->sample_aspect_ratio.num) { |
| 178 | t_info.aspect_numerator = avc_context->sample_aspect_ratio.num; |
| 179 | t_info.aspect_denominator = avc_context->sample_aspect_ratio.den; |
| 180 | } else { |
| 181 | t_info.aspect_numerator = 1; |
| 182 | t_info.aspect_denominator = 1; |
| 183 | } |
| 184 | |
| 185 | if (avc_context->color_primaries == AVCOL_PRI_BT470M) |
| 186 | t_info.colorspace = TH_CS_ITU_REC_470M; |
| 187 | else if (avc_context->color_primaries == AVCOL_PRI_BT470BG) |
| 188 | t_info.colorspace = TH_CS_ITU_REC_470BG; |
| 189 | else |
| 190 | t_info.colorspace = TH_CS_UNSPECIFIED; |
| 191 | |
| 192 | if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P) |
| 193 | t_info.pixel_fmt = TH_PF_420; |
| 194 | else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P) |
| 195 | t_info.pixel_fmt = TH_PF_422; |
| 196 | else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P) |
| 197 | t_info.pixel_fmt = TH_PF_444; |
| 198 | else { |
| 199 | av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n"); |
| 200 | return AVERROR(EINVAL); |
| 201 | } |
| 202 | avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift); |
| 203 | |
| 204 | if (avc_context->flags & CODEC_FLAG_QSCALE) { |
| 205 | /* Clip global_quality in QP units to the [0 - 10] range |
| 206 | to be consistent with the libvorbis implementation. |
| 207 | Theora accepts a quality parameter which is an int value in |
| 208 | the [0 - 63] range. |
| 209 | */ |
| 210 | t_info.quality = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3; |
| 211 | t_info.target_bitrate = 0; |
| 212 | } else { |
| 213 | t_info.target_bitrate = avc_context->bit_rate; |
| 214 | t_info.quality = 0; |
| 215 | } |
| 216 | |
| 217 | /* Now initialise libtheora */ |
| 218 | h->t_state = th_encode_alloc(&t_info); |
| 219 | if (!h->t_state) { |
| 220 | av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n"); |
| 221 | return AVERROR_EXTERNAL; |
| 222 | } |
| 223 | |
| 224 | h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1; |
| 225 | /* Clear up theora_info struct */ |
| 226 | th_info_clear(&t_info); |
| 227 | |
| 228 | if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE, |
| 229 | &gop_size, sizeof(gop_size))) { |
| 230 | av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n"); |
| 231 | return AVERROR_EXTERNAL; |
| 232 | } |
| 233 | |
| 234 | // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers |
| 235 | if (avc_context->flags & CODEC_FLAG_PASS1) { |
| 236 | if ((ret = get_stats(avc_context, 0)) < 0) |
| 237 | return ret; |
| 238 | } else if (avc_context->flags & CODEC_FLAG_PASS2) { |
| 239 | if ((ret = submit_stats(avc_context)) < 0) |
| 240 | return ret; |
| 241 | } |
| 242 | |
| 243 | /* |
| 244 | Output first header packet consisting of theora |
| 245 | header, comment, and tables. |
| 246 | |
| 247 | Each one is prefixed with a 16bit size, then they |
| 248 | are concatenated together into libavcodec's extradata. |
| 249 | */ |
| 250 | offset = 0; |
| 251 | |
| 252 | /* Headers */ |
| 253 | th_comment_init(&t_comment); |
| 254 | |
| 255 | while (th_encode_flushheader(h->t_state, &t_comment, &o_packet)) |
| 256 | if ((ret = concatenate_packet(&offset, avc_context, &o_packet)) < 0) |
| 257 | return ret; |
| 258 | |
| 259 | th_comment_clear(&t_comment); |
| 260 | |
| 261 | /* Set up the output AVFrame */ |
| 262 | avc_context->coded_frame = av_frame_alloc(); |
| 263 | |
| 264 | return 0; |
| 265 | } |
| 266 | |
| 267 | static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt, |
| 268 | const AVFrame *frame, int *got_packet) |
| 269 | { |
| 270 | th_ycbcr_buffer t_yuv_buffer; |
| 271 | TheoraContext *h = avc_context->priv_data; |
| 272 | ogg_packet o_packet; |
| 273 | int result, i, ret; |
| 274 | |
| 275 | // EOS, finish and get 1st pass stats if applicable |
| 276 | if (!frame) { |
| 277 | th_encode_packetout(h->t_state, 1, &o_packet); |
| 278 | if (avc_context->flags & CODEC_FLAG_PASS1) |
| 279 | if ((ret = get_stats(avc_context, 1)) < 0) |
| 280 | return ret; |
| 281 | return 0; |
| 282 | } |
| 283 | |
| 284 | /* Copy planes to the theora yuv_buffer */ |
| 285 | for (i = 0; i < 3; i++) { |
| 286 | t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift); |
| 287 | t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift); |
| 288 | t_yuv_buffer[i].stride = frame->linesize[i]; |
| 289 | t_yuv_buffer[i].data = frame->data[i]; |
| 290 | } |
| 291 | |
| 292 | if (avc_context->flags & CODEC_FLAG_PASS2) |
| 293 | if ((ret = submit_stats(avc_context)) < 0) |
| 294 | return ret; |
| 295 | |
| 296 | /* Now call into theora_encode_YUVin */ |
| 297 | result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer); |
| 298 | if (result) { |
| 299 | const char* message; |
| 300 | switch (result) { |
| 301 | case -1: |
| 302 | message = "differing frame sizes"; |
| 303 | break; |
| 304 | case TH_EINVAL: |
| 305 | message = "encoder is not ready or is finished"; |
| 306 | break; |
| 307 | default: |
| 308 | message = "unknown reason"; |
| 309 | break; |
| 310 | } |
| 311 | av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result); |
| 312 | return AVERROR_EXTERNAL; |
| 313 | } |
| 314 | |
| 315 | if (avc_context->flags & CODEC_FLAG_PASS1) |
| 316 | if ((ret = get_stats(avc_context, 0)) < 0) |
| 317 | return ret; |
| 318 | |
| 319 | /* Pick up returned ogg_packet */ |
| 320 | result = th_encode_packetout(h->t_state, 0, &o_packet); |
| 321 | switch (result) { |
| 322 | case 0: |
| 323 | /* No packet is ready */ |
| 324 | return 0; |
| 325 | case 1: |
| 326 | /* Success, we have a packet */ |
| 327 | break; |
| 328 | default: |
| 329 | av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result); |
| 330 | return AVERROR_EXTERNAL; |
| 331 | } |
| 332 | |
| 333 | /* Copy ogg_packet content out to buffer */ |
| 334 | if ((ret = ff_alloc_packet2(avc_context, pkt, o_packet.bytes)) < 0) |
| 335 | return ret; |
| 336 | memcpy(pkt->data, o_packet.packet, o_packet.bytes); |
| 337 | |
| 338 | // HACK: assumes no encoder delay, this is true until libtheora becomes |
| 339 | // multithreaded (which will be disabled unless explicitly requested) |
| 340 | pkt->pts = pkt->dts = frame->pts; |
| 341 | avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask); |
| 342 | if (avc_context->coded_frame->key_frame) |
| 343 | pkt->flags |= AV_PKT_FLAG_KEY; |
| 344 | *got_packet = 1; |
| 345 | |
| 346 | return 0; |
| 347 | } |
| 348 | |
| 349 | static av_cold int encode_close(AVCodecContext* avc_context) |
| 350 | { |
| 351 | TheoraContext *h = avc_context->priv_data; |
| 352 | |
| 353 | th_encode_free(h->t_state); |
| 354 | av_freep(&h->stats); |
| 355 | av_freep(&avc_context->coded_frame); |
| 356 | av_freep(&avc_context->stats_out); |
| 357 | av_freep(&avc_context->extradata); |
| 358 | avc_context->extradata_size = 0; |
| 359 | |
| 360 | return 0; |
| 361 | } |
| 362 | |
| 363 | /** AVCodec struct exposed to libavcodec */ |
| 364 | AVCodec ff_libtheora_encoder = { |
| 365 | .name = "libtheora", |
| 366 | .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"), |
| 367 | .type = AVMEDIA_TYPE_VIDEO, |
| 368 | .id = AV_CODEC_ID_THEORA, |
| 369 | .priv_data_size = sizeof(TheoraContext), |
| 370 | .init = encode_init, |
| 371 | .close = encode_close, |
| 372 | .encode2 = encode_frame, |
| 373 | .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary |
| 374 | .pix_fmts = (const enum AVPixelFormat[]){ |
| 375 | AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_NONE |
| 376 | }, |
| 377 | }; |