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",
77 static av_cold void log_encoder_error(AVCodecContext *avctx, const char *desc)
79 VP8Context *ctx = avctx->priv_data;
80 const char *error = vpx_codec_error(&ctx->encoder);
81 const char *detail = vpx_codec_error_detail(&ctx->encoder);
83 av_log(avctx, AV_LOG_ERROR, "%s: %s\n", desc, error);
85 av_log(avctx, AV_LOG_ERROR, " Additional information: %s\n", detail);
88 static av_cold void dump_enc_cfg(AVCodecContext *avctx,
89 const struct vpx_codec_enc_cfg *cfg)
92 int level = AV_LOG_DEBUG;
94 av_log(avctx, level, "vpx_codec_enc_cfg\n");
95 av_log(avctx, level, "generic settings\n"
96 " %*s%u\n %*s%u\n %*s%u\n %*s%u\n %*s%u\n"
97 " %*s{%u/%u}\n %*s%u\n %*s%d\n %*s%u\n",
98 width, "g_usage:", cfg->g_usage,
99 width, "g_threads:", cfg->g_threads,
100 width, "g_profile:", cfg->g_profile,
101 width, "g_w:", cfg->g_w,
102 width, "g_h:", cfg->g_h,
103 width, "g_timebase:", cfg->g_timebase.num, cfg->g_timebase.den,
104 width, "g_error_resilient:", cfg->g_error_resilient,
105 width, "g_pass:", cfg->g_pass,
106 width, "g_lag_in_frames:", cfg->g_lag_in_frames);
107 av_log(avctx, level, "rate control settings\n"
108 " %*s%u\n %*s%u\n %*s%u\n %*s%u\n"
109 " %*s%d\n %*s%p(%zu)\n %*s%u\n",
110 width, "rc_dropframe_thresh:", cfg->rc_dropframe_thresh,
111 width, "rc_resize_allowed:", cfg->rc_resize_allowed,
112 width, "rc_resize_up_thresh:", cfg->rc_resize_up_thresh,
113 width, "rc_resize_down_thresh:", cfg->rc_resize_down_thresh,
114 width, "rc_end_usage:", cfg->rc_end_usage,
115 width, "rc_twopass_stats_in:", cfg->rc_twopass_stats_in.buf, cfg->rc_twopass_stats_in.sz,
116 width, "rc_target_bitrate:", cfg->rc_target_bitrate);
117 av_log(avctx, level, "quantizer settings\n"
119 width, "rc_min_quantizer:", cfg->rc_min_quantizer,
120 width, "rc_max_quantizer:", cfg->rc_max_quantizer);
121 av_log(avctx, level, "bitrate tolerance\n"
123 width, "rc_undershoot_pct:", cfg->rc_undershoot_pct,
124 width, "rc_overshoot_pct:", cfg->rc_overshoot_pct);
125 av_log(avctx, level, "decoder buffer model\n"
126 " %*s%u\n %*s%u\n %*s%u\n",
127 width, "rc_buf_sz:", cfg->rc_buf_sz,
128 width, "rc_buf_initial_sz:", cfg->rc_buf_initial_sz,
129 width, "rc_buf_optimal_sz:", cfg->rc_buf_optimal_sz);
130 av_log(avctx, level, "2 pass rate control settings\n"
131 " %*s%u\n %*s%u\n %*s%u\n",
132 width, "rc_2pass_vbr_bias_pct:", cfg->rc_2pass_vbr_bias_pct,
133 width, "rc_2pass_vbr_minsection_pct:", cfg->rc_2pass_vbr_minsection_pct,
134 width, "rc_2pass_vbr_maxsection_pct:", cfg->rc_2pass_vbr_maxsection_pct);
135 av_log(avctx, level, "keyframing settings\n"
136 " %*s%d\n %*s%u\n %*s%u\n",
137 width, "kf_mode:", cfg->kf_mode,
138 width, "kf_min_dist:", cfg->kf_min_dist,
139 width, "kf_max_dist:", cfg->kf_max_dist);
140 av_log(avctx, level, "\n");
143 static void coded_frame_add(void *list, struct FrameListData *cx_frame)
145 struct FrameListData **p = list;
150 cx_frame->next = NULL;
153 static av_cold void free_coded_frame(struct FrameListData *cx_frame)
155 av_freep(&cx_frame->buf);
159 static av_cold void free_frame_list(struct FrameListData *list)
161 struct FrameListData *p = list;
170 static av_cold int codecctl_int(AVCodecContext *avctx,
171 enum vp8e_enc_control_id id, int val)
173 VP8Context *ctx = avctx->priv_data;
178 snprintf(buf, sizeof(buf), "%s:", ctlidstr[id]);
179 av_log(avctx, AV_LOG_DEBUG, " %*s%d\n", width, buf, val);
181 res = vpx_codec_control(&ctx->encoder, id, val);
182 if (res != VPX_CODEC_OK) {
183 snprintf(buf, sizeof(buf), "Failed to set %s codec control",
185 log_encoder_error(avctx, buf);
188 return res == VPX_CODEC_OK ? 0 : AVERROR(EINVAL);
191 static av_cold int vp8_free(AVCodecContext *avctx)
193 VP8Context *ctx = avctx->priv_data;
195 vpx_codec_destroy(&ctx->encoder);
196 av_freep(&ctx->twopass_stats.buf);
197 av_freep(&avctx->coded_frame);
198 av_freep(&avctx->stats_out);
199 free_frame_list(ctx->coded_frame_list);
203 static av_cold int vp8_init(AVCodecContext *avctx)
205 VP8Context *ctx = avctx->priv_data;
206 const struct vpx_codec_iface *iface = &vpx_codec_vp8_cx_algo;
208 struct vpx_codec_enc_cfg enccfg;
211 av_log(avctx, AV_LOG_INFO, "%s\n", vpx_codec_version_str());
212 av_log(avctx, AV_LOG_VERBOSE, "%s\n", vpx_codec_build_config());
214 if ((res = vpx_codec_enc_config_default(iface, &enccfg, 0)) != VPX_CODEC_OK) {
215 av_log(avctx, AV_LOG_ERROR, "Failed to get config: %s\n",
216 vpx_codec_err_to_string(res));
217 return AVERROR(EINVAL);
219 dump_enc_cfg(avctx, &enccfg);
221 enccfg.g_w = avctx->width;
222 enccfg.g_h = avctx->height;
223 enccfg.g_timebase.num = avctx->time_base.num;
224 enccfg.g_timebase.den = avctx->time_base.den;
225 enccfg.g_threads = avctx->thread_count;
227 if (avctx->flags & CODEC_FLAG_PASS1)
228 enccfg.g_pass = VPX_RC_FIRST_PASS;
229 else if (avctx->flags & CODEC_FLAG_PASS2)
230 enccfg.g_pass = VPX_RC_LAST_PASS;
232 enccfg.g_pass = VPX_RC_ONE_PASS;
234 if (avctx->rc_min_rate == avctx->rc_max_rate &&
235 avctx->rc_min_rate == avctx->bit_rate)
236 enccfg.rc_end_usage = VPX_CBR;
237 enccfg.rc_target_bitrate = av_rescale_rnd(avctx->bit_rate, 1, 1000,
240 enccfg.rc_min_quantizer = avctx->qmin;
241 enccfg.rc_max_quantizer = avctx->qmax;
242 enccfg.rc_dropframe_thresh = avctx->frame_skip_threshold;
244 //0-100 (0 => CBR, 100 => VBR)
245 enccfg.rc_2pass_vbr_bias_pct = round(avctx->qcompress * 100);
246 enccfg.rc_2pass_vbr_minsection_pct =
247 avctx->rc_min_rate * 100LL / avctx->bit_rate;
248 if (avctx->rc_max_rate)
249 enccfg.rc_2pass_vbr_maxsection_pct =
250 avctx->rc_max_rate * 100LL / avctx->bit_rate;
252 if (avctx->rc_buffer_size)
254 avctx->rc_buffer_size * 1000LL / avctx->bit_rate;
255 if (avctx->rc_initial_buffer_occupancy)
256 enccfg.rc_buf_initial_sz =
257 avctx->rc_initial_buffer_occupancy * 1000LL / avctx->bit_rate;
258 enccfg.rc_buf_optimal_sz = enccfg.rc_buf_sz * 5 / 6;
260 //_enc_init() will balk if kf_min_dist differs from max w/VPX_KF_AUTO
261 if (avctx->keyint_min == avctx->gop_size)
262 enccfg.kf_min_dist = avctx->keyint_min;
263 enccfg.kf_max_dist = avctx->gop_size;
265 if (enccfg.g_pass == VPX_RC_FIRST_PASS)
266 enccfg.g_lag_in_frames = 0;
267 else if (enccfg.g_pass == VPX_RC_LAST_PASS) {
270 if (!avctx->stats_in) {
271 av_log(avctx, AV_LOG_ERROR, "No stats file for second pass\n");
272 return AVERROR_INVALIDDATA;
275 ctx->twopass_stats.sz = strlen(avctx->stats_in) * 3 / 4;
276 ctx->twopass_stats.buf = av_malloc(ctx->twopass_stats.sz);
277 if (!ctx->twopass_stats.buf) {
278 av_log(avctx, AV_LOG_ERROR,
279 "Stat buffer alloc (%zu bytes) failed\n",
280 ctx->twopass_stats.sz);
281 return AVERROR(ENOMEM);
283 decode_size = av_base64_decode(ctx->twopass_stats.buf, avctx->stats_in,
284 ctx->twopass_stats.sz);
285 if (decode_size < 0) {
286 av_log(avctx, AV_LOG_ERROR, "Stat buffer decode failed\n");
287 return AVERROR_INVALIDDATA;
290 ctx->twopass_stats.sz = decode_size;
291 enccfg.rc_twopass_stats_in = ctx->twopass_stats;
294 ctx->deadline = VPX_DL_GOOD_QUALITY;
295 /* 0-3: For non-zero values the encoder increasingly optimizes for reduced
296 complexity playback on low powered devices at the expense of encode
298 if (avctx->profile != FF_PROFILE_UNKNOWN)
299 enccfg.g_profile = avctx->profile;
301 dump_enc_cfg(avctx, &enccfg);
302 /* Construct Encoder Context */
303 res = vpx_codec_enc_init(&ctx->encoder, iface, &enccfg, 0);
304 if (res != VPX_CODEC_OK) {
305 log_encoder_error(avctx, "Failed to initialize encoder");
306 return AVERROR(EINVAL);
309 //codec control failures are currently treated only as warnings
310 av_log(avctx, AV_LOG_DEBUG, "vpx_codec_control\n");
311 codecctl_int(avctx, VP8E_SET_CPUUSED, cpuused);
312 codecctl_int(avctx, VP8E_SET_NOISE_SENSITIVITY, avctx->noise_reduction);
313 codecctl_int(avctx, VP8E_SET_TOKEN_PARTITIONS, av_log2(avctx->slices));
315 //provide dummy value to initialize wrapper, values will be updated each _encode()
316 vpx_img_wrap(&ctx->rawimg, VPX_IMG_FMT_I420, avctx->width, avctx->height, 1,
319 avctx->coded_frame = avcodec_alloc_frame();
320 if (!avctx->coded_frame) {
321 av_log(avctx, AV_LOG_ERROR, "Error allocating coded frame\n");
323 return AVERROR(ENOMEM);
328 static inline void cx_pktcpy(struct FrameListData *dst,
329 const struct vpx_codec_cx_pkt *src)
331 dst->pts = src->data.frame.pts;
332 dst->duration = src->data.frame.duration;
333 dst->flags = src->data.frame.flags;
334 dst->sz = src->data.frame.sz;
335 dst->buf = src->data.frame.buf;
339 * Store coded frame information in format suitable for return from encode().
341 * Write buffer information from @a cx_frame to @a buf & @a buf_size.
342 * Timing/frame details to @a coded_frame.
343 * @return Frame size written to @a buf on success
344 * @return AVERROR(EINVAL) on error
346 static int storeframe(AVCodecContext *avctx, struct FrameListData *cx_frame,
347 uint8_t *buf, int buf_size, AVFrame *coded_frame)
349 if ((int) cx_frame->sz <= buf_size) {
350 buf_size = cx_frame->sz;
351 memcpy(buf, cx_frame->buf, buf_size);
352 coded_frame->pts = cx_frame->pts;
353 coded_frame->key_frame = !!(cx_frame->flags & VPX_FRAME_IS_KEY);
355 if (coded_frame->key_frame)
356 coded_frame->pict_type = FF_I_TYPE;
358 coded_frame->pict_type = FF_P_TYPE;
360 av_log(avctx, AV_LOG_ERROR,
361 "Compressed frame larger than storage provided! (%zu/%d)\n",
362 cx_frame->sz, buf_size);
363 return AVERROR(EINVAL);
369 * Queue multiple output frames from the encoder, returning the front-most.
370 * In cases where vpx_codec_get_cx_data() returns more than 1 frame append
371 * the frame queue. Return the head frame if available.
372 * @return Stored frame size
373 * @return AVERROR(EINVAL) on output size error
374 * @return AVERROR(ENOMEM) on coded frame queue data allocation error
376 static int queue_frames(AVCodecContext *avctx, uint8_t *buf, int buf_size,
377 AVFrame *coded_frame)
379 VP8Context *ctx = avctx->priv_data;
380 const struct vpx_codec_cx_pkt *pkt;
381 const void *iter = NULL;
384 if (ctx->coded_frame_list) {
385 struct FrameListData *cx_frame = ctx->coded_frame_list;
386 /* return the leading frame if we've already begun queueing */
387 size = storeframe(avctx, cx_frame, buf, buf_size, coded_frame);
389 return AVERROR(EINVAL);
390 ctx->coded_frame_list = cx_frame->next;
391 free_coded_frame(cx_frame);
394 /* consume all available output from the encoder before returning. buffers
395 are only good through the next vpx_codec call */
396 while ((pkt = vpx_codec_get_cx_data(&ctx->encoder, &iter))) {
398 case VPX_CODEC_CX_FRAME_PKT:
400 struct FrameListData cx_frame;
402 /* avoid storing the frame when the list is empty and we haven't yet
403 provided a frame for output */
404 assert(!ctx->coded_frame_list);
405 cx_pktcpy(&cx_frame, pkt);
406 size = storeframe(avctx, &cx_frame, buf, buf_size, coded_frame);
408 return AVERROR(EINVAL);
410 struct FrameListData *cx_frame =
411 av_malloc(sizeof(struct FrameListData));
414 av_log(avctx, AV_LOG_ERROR,
415 "Frame queue element alloc failed\n");
416 return AVERROR(ENOMEM);
418 cx_pktcpy(cx_frame, pkt);
419 cx_frame->buf = av_malloc(cx_frame->sz);
421 if (!cx_frame->buf) {
422 av_log(avctx, AV_LOG_ERROR,
423 "Data buffer alloc (%zu bytes) failed\n",
425 return AVERROR(ENOMEM);
427 memcpy(cx_frame->buf, pkt->data.frame.buf, pkt->data.frame.sz);
428 coded_frame_add(&ctx->coded_frame_list, cx_frame);
431 case VPX_CODEC_STATS_PKT: {
432 struct vpx_fixed_buf *stats = &ctx->twopass_stats;
433 stats->buf = av_realloc(stats->buf,
434 stats->sz + pkt->data.twopass_stats.sz);
436 av_log(avctx, AV_LOG_ERROR, "Stat buffer realloc failed\n");
437 return AVERROR(ENOMEM);
439 memcpy((uint8_t*)stats->buf + stats->sz,
440 pkt->data.twopass_stats.buf, pkt->data.twopass_stats.sz);
441 stats->sz += pkt->data.twopass_stats.sz;
444 case VPX_CODEC_PSNR_PKT: //FIXME add support for CODEC_FLAG_PSNR
445 case VPX_CODEC_CUSTOM_PKT:
446 //ignore unsupported/unrecognized packet types
454 static int vp8_encode(AVCodecContext *avctx, uint8_t *buf, int buf_size,
457 VP8Context *ctx = avctx->priv_data;
458 AVFrame *frame = data;
459 struct vpx_image *rawimg = NULL;
460 int64_t timestamp = 0;
464 rawimg = &ctx->rawimg;
465 rawimg->planes[VPX_PLANE_Y] = frame->data[0];
466 rawimg->planes[VPX_PLANE_U] = frame->data[1];
467 rawimg->planes[VPX_PLANE_V] = frame->data[2];
468 rawimg->stride[VPX_PLANE_Y] = frame->linesize[0];
469 rawimg->stride[VPX_PLANE_U] = frame->linesize[1];
470 rawimg->stride[VPX_PLANE_V] = frame->linesize[2];
471 timestamp = frame->pts;
474 res = vpx_codec_encode(&ctx->encoder, rawimg, timestamp,
475 avctx->ticks_per_frame, 0, ctx->deadline);
476 if (res != VPX_CODEC_OK) {
477 log_encoder_error(avctx, "Error encoding frame");
478 return AVERROR_INVALIDDATA;
480 coded_size = queue_frames(avctx, buf, buf_size, avctx->coded_frame);
482 if (!frame && avctx->flags & CODEC_FLAG_PASS1) {
483 unsigned int b64_size = AV_BASE64_SIZE(ctx->twopass_stats.sz);
485 avctx->stats_out = av_malloc(b64_size);
486 if (!avctx->stats_out) {
487 av_log(avctx, AV_LOG_ERROR, "Stat buffer alloc (%d bytes) failed\n",
489 return AVERROR(ENOMEM);
491 av_base64_encode(avctx->stats_out, b64_size, ctx->twopass_stats.buf,
492 ctx->twopass_stats.sz);
497 AVCodec libvpx_encoder = {
507 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
508 .long_name = NULL_IF_CONFIG_SMALL("libvpx VP8"),