2 * Copyright (c) 2010, Google, Inc.
4 * This file is part of FFmpeg.
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.
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.
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
23 * VP8 encoder support via libvpx
26 #define VPX_DISABLE_CTRL_TYPECHECKS 1
27 #define VPX_CODEC_DISABLE_COMPAT 1
28 #include <vpx/vpx_encoder.h>
29 #include <vpx/vp8cx.h>
32 #include "libavutil/base64.h"
35 * Portion of struct vpx_codec_cx_pkt from vpx_encoder.h.
36 * One encoded frame returned from the library.
38 struct FrameListData {
39 void *buf; /**< compressed data buffer */
40 size_t sz; /**< length of compressed data */
41 int64_t pts; /**< time stamp to show frame
42 (in timebase units) */
43 unsigned long duration; /**< duration to show frame
44 (in timebase units) */
45 uint32_t flags; /**< flags for this frame */
46 struct FrameListData *next;
49 typedef struct VP8EncoderContext {
50 struct vpx_codec_ctx encoder;
51 struct vpx_image rawimg;
52 struct vpx_fixed_buf twopass_stats;
53 unsigned long deadline; //i.e., RT/GOOD/BEST
54 struct FrameListData *coded_frame_list;
57 /** String mappings for enum vp8e_enc_control_id */
58 static const char *ctlidstr[] = {
59 [VP8E_UPD_ENTROPY] = "VP8E_UPD_ENTROPY",
60 [VP8E_UPD_REFERENCE] = "VP8E_UPD_REFERENCE",
61 [VP8E_USE_REFERENCE] = "VP8E_USE_REFERENCE",
62 [VP8E_SET_ROI_MAP] = "VP8E_SET_ROI_MAP",
63 [VP8E_SET_ACTIVEMAP] = "VP8E_SET_ACTIVEMAP",
64 [VP8E_SET_SCALEMODE] = "VP8E_SET_SCALEMODE",
65 [VP8E_SET_CPUUSED] = "VP8E_SET_CPUUSED",
66 [VP8E_SET_ENABLEAUTOALTREF] = "VP8E_SET_ENABLEAUTOALTREF",
67 [VP8E_SET_NOISE_SENSITIVITY] = "VP8E_SET_NOISE_SENSITIVITY",
68 [VP8E_SET_SHARPNESS] = "VP8E_SET_SHARPNESS",
69 [VP8E_SET_STATIC_THRESHOLD] = "VP8E_SET_STATIC_THRESHOLD",
70 [VP8E_SET_TOKEN_PARTITIONS] = "VP8E_SET_TOKEN_PARTITIONS",
71 [VP8E_GET_LAST_QUANTIZER] = "VP8E_GET_LAST_QUANTIZER",
72 [VP8E_SET_ARNR_MAXFRAMES] = "VP8E_SET_ARNR_MAXFRAMES",
73 [VP8E_SET_ARNR_STRENGTH] = "VP8E_SET_ARNR_STRENGTH",
74 [VP8E_SET_ARNR_TYPE] = "VP8E_SET_ARNR_TYPE",
75 [VP8E_SET_CQ_LEVEL] = "VP8E_SET_CQ_LEVEL",
78 static av_cold void log_encoder_error(AVCodecContext *avctx, const char *desc)
80 VP8Context *ctx = avctx->priv_data;
81 const char *error = vpx_codec_error(&ctx->encoder);
82 const char *detail = vpx_codec_error_detail(&ctx->encoder);
84 av_log(avctx, AV_LOG_ERROR, "%s: %s\n", desc, error);
86 av_log(avctx, AV_LOG_ERROR, " Additional information: %s\n", detail);
89 static av_cold void dump_enc_cfg(AVCodecContext *avctx,
90 const struct vpx_codec_enc_cfg *cfg)
93 int level = AV_LOG_DEBUG;
95 av_log(avctx, level, "vpx_codec_enc_cfg\n");
96 av_log(avctx, level, "generic settings\n"
97 " %*s%u\n %*s%u\n %*s%u\n %*s%u\n %*s%u\n"
98 " %*s{%u/%u}\n %*s%u\n %*s%d\n %*s%u\n",
99 width, "g_usage:", cfg->g_usage,
100 width, "g_threads:", cfg->g_threads,
101 width, "g_profile:", cfg->g_profile,
102 width, "g_w:", cfg->g_w,
103 width, "g_h:", cfg->g_h,
104 width, "g_timebase:", cfg->g_timebase.num, cfg->g_timebase.den,
105 width, "g_error_resilient:", cfg->g_error_resilient,
106 width, "g_pass:", cfg->g_pass,
107 width, "g_lag_in_frames:", cfg->g_lag_in_frames);
108 av_log(avctx, level, "rate control settings\n"
109 " %*s%u\n %*s%u\n %*s%u\n %*s%u\n"
110 " %*s%d\n %*s%p(%zu)\n %*s%u\n",
111 width, "rc_dropframe_thresh:", cfg->rc_dropframe_thresh,
112 width, "rc_resize_allowed:", cfg->rc_resize_allowed,
113 width, "rc_resize_up_thresh:", cfg->rc_resize_up_thresh,
114 width, "rc_resize_down_thresh:", cfg->rc_resize_down_thresh,
115 width, "rc_end_usage:", cfg->rc_end_usage,
116 width, "rc_twopass_stats_in:", cfg->rc_twopass_stats_in.buf, cfg->rc_twopass_stats_in.sz,
117 width, "rc_target_bitrate:", cfg->rc_target_bitrate);
118 av_log(avctx, level, "quantizer settings\n"
120 width, "rc_min_quantizer:", cfg->rc_min_quantizer,
121 width, "rc_max_quantizer:", cfg->rc_max_quantizer);
122 av_log(avctx, level, "bitrate tolerance\n"
124 width, "rc_undershoot_pct:", cfg->rc_undershoot_pct,
125 width, "rc_overshoot_pct:", cfg->rc_overshoot_pct);
126 av_log(avctx, level, "decoder buffer model\n"
127 " %*s%u\n %*s%u\n %*s%u\n",
128 width, "rc_buf_sz:", cfg->rc_buf_sz,
129 width, "rc_buf_initial_sz:", cfg->rc_buf_initial_sz,
130 width, "rc_buf_optimal_sz:", cfg->rc_buf_optimal_sz);
131 av_log(avctx, level, "2 pass rate control settings\n"
132 " %*s%u\n %*s%u\n %*s%u\n",
133 width, "rc_2pass_vbr_bias_pct:", cfg->rc_2pass_vbr_bias_pct,
134 width, "rc_2pass_vbr_minsection_pct:", cfg->rc_2pass_vbr_minsection_pct,
135 width, "rc_2pass_vbr_maxsection_pct:", cfg->rc_2pass_vbr_maxsection_pct);
136 av_log(avctx, level, "keyframing settings\n"
137 " %*s%d\n %*s%u\n %*s%u\n",
138 width, "kf_mode:", cfg->kf_mode,
139 width, "kf_min_dist:", cfg->kf_min_dist,
140 width, "kf_max_dist:", cfg->kf_max_dist);
141 av_log(avctx, level, "\n");
144 static void coded_frame_add(void *list, struct FrameListData *cx_frame)
146 struct FrameListData **p = list;
151 cx_frame->next = NULL;
154 static av_cold void free_coded_frame(struct FrameListData *cx_frame)
156 av_freep(&cx_frame->buf);
160 static av_cold void free_frame_list(struct FrameListData *list)
162 struct FrameListData *p = list;
171 static av_cold int codecctl_int(AVCodecContext *avctx,
172 enum vp8e_enc_control_id id, int val)
174 VP8Context *ctx = avctx->priv_data;
179 snprintf(buf, sizeof(buf), "%s:", ctlidstr[id]);
180 av_log(avctx, AV_LOG_DEBUG, " %*s%d\n", width, buf, val);
182 res = vpx_codec_control(&ctx->encoder, id, val);
183 if (res != VPX_CODEC_OK) {
184 snprintf(buf, sizeof(buf), "Failed to set %s codec control",
186 log_encoder_error(avctx, buf);
189 return res == VPX_CODEC_OK ? 0 : AVERROR(EINVAL);
192 static av_cold int vp8_free(AVCodecContext *avctx)
194 VP8Context *ctx = avctx->priv_data;
196 vpx_codec_destroy(&ctx->encoder);
197 av_freep(&ctx->twopass_stats.buf);
198 av_freep(&avctx->coded_frame);
199 av_freep(&avctx->stats_out);
200 free_frame_list(ctx->coded_frame_list);
204 static av_cold int vp8_init(AVCodecContext *avctx)
206 VP8Context *ctx = avctx->priv_data;
207 const struct vpx_codec_iface *iface = &vpx_codec_vp8_cx_algo;
209 struct vpx_codec_enc_cfg enccfg;
212 av_log(avctx, AV_LOG_INFO, "%s\n", vpx_codec_version_str());
213 av_log(avctx, AV_LOG_VERBOSE, "%s\n", vpx_codec_build_config());
215 if ((res = vpx_codec_enc_config_default(iface, &enccfg, 0)) != VPX_CODEC_OK) {
216 av_log(avctx, AV_LOG_ERROR, "Failed to get config: %s\n",
217 vpx_codec_err_to_string(res));
218 return AVERROR(EINVAL);
220 dump_enc_cfg(avctx, &enccfg);
222 enccfg.g_w = avctx->width;
223 enccfg.g_h = avctx->height;
224 enccfg.g_timebase.num = avctx->time_base.num;
225 enccfg.g_timebase.den = avctx->time_base.den;
226 enccfg.g_threads = avctx->thread_count;
228 if (avctx->flags & CODEC_FLAG_PASS1)
229 enccfg.g_pass = VPX_RC_FIRST_PASS;
230 else if (avctx->flags & CODEC_FLAG_PASS2)
231 enccfg.g_pass = VPX_RC_LAST_PASS;
233 enccfg.g_pass = VPX_RC_ONE_PASS;
235 if (avctx->rc_min_rate == avctx->rc_max_rate &&
236 avctx->rc_min_rate == avctx->bit_rate)
237 enccfg.rc_end_usage = VPX_CBR;
239 enccfg.rc_end_usage = VPX_CQ;
240 enccfg.rc_target_bitrate = av_rescale_rnd(avctx->bit_rate, 1, 1000,
243 enccfg.rc_min_quantizer = avctx->qmin;
244 enccfg.rc_max_quantizer = avctx->qmax;
245 enccfg.rc_dropframe_thresh = avctx->frame_skip_threshold;
247 //0-100 (0 => CBR, 100 => VBR)
248 enccfg.rc_2pass_vbr_bias_pct = round(avctx->qcompress * 100);
249 enccfg.rc_2pass_vbr_minsection_pct =
250 avctx->rc_min_rate * 100LL / avctx->bit_rate;
251 if (avctx->rc_max_rate)
252 enccfg.rc_2pass_vbr_maxsection_pct =
253 avctx->rc_max_rate * 100LL / avctx->bit_rate;
255 if (avctx->rc_buffer_size)
257 avctx->rc_buffer_size * 1000LL / avctx->bit_rate;
258 if (avctx->rc_initial_buffer_occupancy)
259 enccfg.rc_buf_initial_sz =
260 avctx->rc_initial_buffer_occupancy * 1000LL / avctx->bit_rate;
261 enccfg.rc_buf_optimal_sz = enccfg.rc_buf_sz * 5 / 6;
263 //_enc_init() will balk if kf_min_dist differs from max w/VPX_KF_AUTO
264 if (avctx->keyint_min == avctx->gop_size)
265 enccfg.kf_min_dist = avctx->keyint_min;
266 enccfg.kf_max_dist = avctx->gop_size;
268 if (enccfg.g_pass == VPX_RC_FIRST_PASS)
269 enccfg.g_lag_in_frames = 0;
270 else if (enccfg.g_pass == VPX_RC_LAST_PASS) {
273 if (!avctx->stats_in) {
274 av_log(avctx, AV_LOG_ERROR, "No stats file for second pass\n");
275 return AVERROR_INVALIDDATA;
278 ctx->twopass_stats.sz = strlen(avctx->stats_in) * 3 / 4;
279 ctx->twopass_stats.buf = av_malloc(ctx->twopass_stats.sz);
280 if (!ctx->twopass_stats.buf) {
281 av_log(avctx, AV_LOG_ERROR,
282 "Stat buffer alloc (%zu bytes) failed\n",
283 ctx->twopass_stats.sz);
284 return AVERROR(ENOMEM);
286 decode_size = av_base64_decode(ctx->twopass_stats.buf, avctx->stats_in,
287 ctx->twopass_stats.sz);
288 if (decode_size < 0) {
289 av_log(avctx, AV_LOG_ERROR, "Stat buffer decode failed\n");
290 return AVERROR_INVALIDDATA;
293 ctx->twopass_stats.sz = decode_size;
294 enccfg.rc_twopass_stats_in = ctx->twopass_stats;
297 ctx->deadline = VPX_DL_GOOD_QUALITY;
298 /* 0-3: For non-zero values the encoder increasingly optimizes for reduced
299 complexity playback on low powered devices at the expense of encode
301 if (avctx->profile != FF_PROFILE_UNKNOWN)
302 enccfg.g_profile = avctx->profile;
304 dump_enc_cfg(avctx, &enccfg);
305 /* Construct Encoder Context */
306 res = vpx_codec_enc_init(&ctx->encoder, iface, &enccfg, 0);
307 if (res != VPX_CODEC_OK) {
308 log_encoder_error(avctx, "Failed to initialize encoder");
309 return AVERROR(EINVAL);
312 //codec control failures are currently treated only as warnings
313 av_log(avctx, AV_LOG_DEBUG, "vpx_codec_control\n");
314 codecctl_int(avctx, VP8E_SET_CPUUSED, cpuused);
315 codecctl_int(avctx, VP8E_SET_NOISE_SENSITIVITY, avctx->noise_reduction);
316 codecctl_int(avctx, VP8E_SET_TOKEN_PARTITIONS, av_log2(avctx->slices));
317 codecctl_int(avctx, VP8E_SET_STATIC_THRESHOLD, avctx->mb_threshold);
318 codecctl_int(avctx, VP8E_SET_CQ_LEVEL, (int)avctx->crf);
320 //provide dummy value to initialize wrapper, values will be updated each _encode()
321 vpx_img_wrap(&ctx->rawimg, VPX_IMG_FMT_I420, avctx->width, avctx->height, 1,
324 avctx->coded_frame = avcodec_alloc_frame();
325 if (!avctx->coded_frame) {
326 av_log(avctx, AV_LOG_ERROR, "Error allocating coded frame\n");
328 return AVERROR(ENOMEM);
333 static inline void cx_pktcpy(struct FrameListData *dst,
334 const struct vpx_codec_cx_pkt *src)
336 dst->pts = src->data.frame.pts;
337 dst->duration = src->data.frame.duration;
338 dst->flags = src->data.frame.flags;
339 dst->sz = src->data.frame.sz;
340 dst->buf = src->data.frame.buf;
344 * Store coded frame information in format suitable for return from encode().
346 * Write buffer information from @a cx_frame to @a buf & @a buf_size.
347 * Timing/frame details to @a coded_frame.
348 * @return Frame size written to @a buf on success
349 * @return AVERROR(EINVAL) on error
351 static int storeframe(AVCodecContext *avctx, struct FrameListData *cx_frame,
352 uint8_t *buf, int buf_size, AVFrame *coded_frame)
354 if ((int) cx_frame->sz <= buf_size) {
355 buf_size = cx_frame->sz;
356 memcpy(buf, cx_frame->buf, buf_size);
357 coded_frame->pts = cx_frame->pts;
358 coded_frame->key_frame = !!(cx_frame->flags & VPX_FRAME_IS_KEY);
360 if (coded_frame->key_frame)
361 coded_frame->pict_type = FF_I_TYPE;
363 coded_frame->pict_type = FF_P_TYPE;
365 av_log(avctx, AV_LOG_ERROR,
366 "Compressed frame larger than storage provided! (%zu/%d)\n",
367 cx_frame->sz, buf_size);
368 return AVERROR(EINVAL);
374 * Queue multiple output frames from the encoder, returning the front-most.
375 * In cases where vpx_codec_get_cx_data() returns more than 1 frame append
376 * the frame queue. Return the head frame if available.
377 * @return Stored frame size
378 * @return AVERROR(EINVAL) on output size error
379 * @return AVERROR(ENOMEM) on coded frame queue data allocation error
381 static int queue_frames(AVCodecContext *avctx, uint8_t *buf, int buf_size,
382 AVFrame *coded_frame)
384 VP8Context *ctx = avctx->priv_data;
385 const struct vpx_codec_cx_pkt *pkt;
386 const void *iter = NULL;
389 if (ctx->coded_frame_list) {
390 struct FrameListData *cx_frame = ctx->coded_frame_list;
391 /* return the leading frame if we've already begun queueing */
392 size = storeframe(avctx, cx_frame, buf, buf_size, coded_frame);
394 return AVERROR(EINVAL);
395 ctx->coded_frame_list = cx_frame->next;
396 free_coded_frame(cx_frame);
399 /* consume all available output from the encoder before returning. buffers
400 are only good through the next vpx_codec call */
401 while ((pkt = vpx_codec_get_cx_data(&ctx->encoder, &iter))) {
403 case VPX_CODEC_CX_FRAME_PKT:
405 struct FrameListData cx_frame;
407 /* avoid storing the frame when the list is empty and we haven't yet
408 provided a frame for output */
409 assert(!ctx->coded_frame_list);
410 cx_pktcpy(&cx_frame, pkt);
411 size = storeframe(avctx, &cx_frame, buf, buf_size, coded_frame);
413 return AVERROR(EINVAL);
415 struct FrameListData *cx_frame =
416 av_malloc(sizeof(struct FrameListData));
419 av_log(avctx, AV_LOG_ERROR,
420 "Frame queue element alloc failed\n");
421 return AVERROR(ENOMEM);
423 cx_pktcpy(cx_frame, pkt);
424 cx_frame->buf = av_malloc(cx_frame->sz);
426 if (!cx_frame->buf) {
427 av_log(avctx, AV_LOG_ERROR,
428 "Data buffer alloc (%zu bytes) failed\n",
430 return AVERROR(ENOMEM);
432 memcpy(cx_frame->buf, pkt->data.frame.buf, pkt->data.frame.sz);
433 coded_frame_add(&ctx->coded_frame_list, cx_frame);
436 case VPX_CODEC_STATS_PKT: {
437 struct vpx_fixed_buf *stats = &ctx->twopass_stats;
438 stats->buf = av_realloc(stats->buf,
439 stats->sz + pkt->data.twopass_stats.sz);
441 av_log(avctx, AV_LOG_ERROR, "Stat buffer realloc failed\n");
442 return AVERROR(ENOMEM);
444 memcpy((uint8_t*)stats->buf + stats->sz,
445 pkt->data.twopass_stats.buf, pkt->data.twopass_stats.sz);
446 stats->sz += pkt->data.twopass_stats.sz;
449 case VPX_CODEC_PSNR_PKT: //FIXME add support for CODEC_FLAG_PSNR
450 case VPX_CODEC_CUSTOM_PKT:
451 //ignore unsupported/unrecognized packet types
459 static int vp8_encode(AVCodecContext *avctx, uint8_t *buf, int buf_size,
462 VP8Context *ctx = avctx->priv_data;
463 AVFrame *frame = data;
464 struct vpx_image *rawimg = NULL;
465 int64_t timestamp = 0;
469 rawimg = &ctx->rawimg;
470 rawimg->planes[VPX_PLANE_Y] = frame->data[0];
471 rawimg->planes[VPX_PLANE_U] = frame->data[1];
472 rawimg->planes[VPX_PLANE_V] = frame->data[2];
473 rawimg->stride[VPX_PLANE_Y] = frame->linesize[0];
474 rawimg->stride[VPX_PLANE_U] = frame->linesize[1];
475 rawimg->stride[VPX_PLANE_V] = frame->linesize[2];
476 timestamp = frame->pts;
479 res = vpx_codec_encode(&ctx->encoder, rawimg, timestamp,
480 avctx->ticks_per_frame, 0, ctx->deadline);
481 if (res != VPX_CODEC_OK) {
482 log_encoder_error(avctx, "Error encoding frame");
483 return AVERROR_INVALIDDATA;
485 coded_size = queue_frames(avctx, buf, buf_size, avctx->coded_frame);
487 if (!frame && avctx->flags & CODEC_FLAG_PASS1) {
488 unsigned int b64_size = AV_BASE64_SIZE(ctx->twopass_stats.sz);
490 avctx->stats_out = av_malloc(b64_size);
491 if (!avctx->stats_out) {
492 av_log(avctx, AV_LOG_ERROR, "Stat buffer alloc (%d bytes) failed\n",
494 return AVERROR(ENOMEM);
496 av_base64_encode(avctx->stats_out, b64_size, ctx->twopass_stats.buf,
497 ctx->twopass_stats.sz);
502 AVCodec ff_libvpx_encoder = {
512 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
513 .long_name = NULL_IF_CONFIG_SMALL("libvpx VP8"),