VP8 encoding via libvpx.
[libav.git] / libavcodec / libvpxenc.c
CommitLineData
ee0e9146
JZ
1/*
2 * Copyright (c) 2010, Google, Inc.
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 * VP8 encoder support via libvpx
24 */
25
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>
30
31#include "avcodec.h"
32#include "libavutil/base64.h"
33
34/**
35 * Portion of struct vpx_codec_cx_pkt from vpx_encoder.h.
36 * One encoded frame returned from the library.
37 */
38struct 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;
47};
48
49typedef 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;
55} VP8Context;
56
57/** String mappings for enum vp8e_enc_control_id */
58static 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};
76
77static av_cold void log_encoder_error(AVCodecContext *avctx, const char *desc)
78{
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);
82
83 av_log(avctx, AV_LOG_ERROR, "%s: %s\n", desc, error);
84 if (detail)
85 av_log(avctx, AV_LOG_ERROR, " Additional information: %s\n", detail);
86}
87
88static av_cold void dump_enc_cfg(AVCodecContext *avctx,
89 const struct vpx_codec_enc_cfg *cfg)
90{
91 int width = -30;
92 int level = AV_LOG_DEBUG;
93
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"
118 " %*s%u\n %*s%u\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"
122 " %*s%u\n %*s%u\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");
141}
142
143static void coded_frame_add(void *list, struct FrameListData *cx_frame)
144{
145 struct FrameListData **p = list;
146
147 while (*p != NULL)
148 p = &(*p)->next;
149 *p = cx_frame;
150 cx_frame->next = NULL;
151}
152
153static av_cold void free_coded_frame(struct FrameListData *cx_frame)
154{
155 av_freep(&cx_frame->buf);
156 av_freep(&cx_frame);
157}
158
159static av_cold void free_frame_list(struct FrameListData *list)
160{
161 struct FrameListData *p = list;
162
163 while (p) {
164 list = list->next;
165 free_coded_frame(p);
166 p = list;
167 }
168}
169
170static av_cold int codecctl_int(AVCodecContext *avctx,
171 enum vp8e_enc_control_id id, int val)
172{
173 VP8Context *ctx = avctx->priv_data;
174 char buf[80];
175 int width = -30;
176 int res;
177
178 snprintf(buf, sizeof(buf), "%s:", ctlidstr[id]);
179 av_log(avctx, AV_LOG_DEBUG, " %*s%d\n", width, buf, val);
180
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",
184 ctlidstr[id]);
185 log_encoder_error(avctx, buf);
186 }
187
188 return res == VPX_CODEC_OK ? 0 : AVERROR(EINVAL);
189}
190
191static av_cold int vp8_free(AVCodecContext *avctx)
192{
193 VP8Context *ctx = avctx->priv_data;
194
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);
200 return 0;
201}
202
203static av_cold int vp8_init(AVCodecContext *avctx)
204{
205 VP8Context *ctx = avctx->priv_data;
206 const struct vpx_codec_iface *iface = &vpx_codec_vp8_cx_algo;
207 int cpuused = 3;
208 struct vpx_codec_enc_cfg enccfg;
209 int res;
210
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());
213
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);
218 }
219 dump_enc_cfg(avctx, &enccfg);
220
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;
226
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;
231 else
232 enccfg.g_pass = VPX_RC_ONE_PASS;
233
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,
238 AV_ROUND_NEAR_INF);
239
240 //convert [1,51] -> [0,63]
241 enccfg.rc_min_quantizer = ((avctx->qmin * 5 + 1) >> 2) - 1;
242 enccfg.rc_max_quantizer = ((avctx->qmax * 5 + 1) >> 2) - 1;
243
244 if (avctx->keyint_min == avctx->gop_size)
245 enccfg.kf_mode = VPX_KF_FIXED;
246 //_enc_init() will balk if kf_min_dist is set in this case
247 if (enccfg.kf_mode != VPX_KF_AUTO)
248 enccfg.kf_min_dist = avctx->keyint_min;
249 enccfg.kf_max_dist = avctx->gop_size;
250
251 if (enccfg.g_pass == VPX_RC_FIRST_PASS)
252 enccfg.g_lag_in_frames = 0;
253 else if (enccfg.g_pass == VPX_RC_LAST_PASS) {
254 int decode_size;
255
256 if (!avctx->stats_in) {
257 av_log(avctx, AV_LOG_ERROR, "No stats file for second pass\n");
258 return AVERROR_INVALIDDATA;
259 }
260
261 ctx->twopass_stats.sz = strlen(avctx->stats_in) * 3 / 4;
262 ctx->twopass_stats.buf = av_malloc(ctx->twopass_stats.sz);
263 if (!ctx->twopass_stats.buf) {
264 av_log(avctx, AV_LOG_ERROR,
265 "Stat buffer alloc (%zu bytes) failed\n",
266 ctx->twopass_stats.sz);
267 return AVERROR(ENOMEM);
268 }
269 decode_size = av_base64_decode(ctx->twopass_stats.buf, avctx->stats_in,
270 ctx->twopass_stats.sz);
271 if (decode_size < 0) {
272 av_log(avctx, AV_LOG_ERROR, "Stat buffer decode failed\n");
273 return AVERROR_INVALIDDATA;
274 }
275
276 ctx->twopass_stats.sz = decode_size;
277 enccfg.rc_twopass_stats_in = ctx->twopass_stats;
278 }
279
280 ctx->deadline = VPX_DL_GOOD_QUALITY;
281
282 dump_enc_cfg(avctx, &enccfg);
283 /* Construct Encoder Context */
284 res = vpx_codec_enc_init(&ctx->encoder, iface, &enccfg, 0);
285 if (res != VPX_CODEC_OK) {
286 log_encoder_error(avctx, "Failed to initialize encoder");
287 return AVERROR(EINVAL);
288 }
289
290 //codec control failures are currently treated only as warnings
291 av_log(avctx, AV_LOG_DEBUG, "vpx_codec_control\n");
292 codecctl_int(avctx, VP8E_SET_CPUUSED, cpuused);
293 codecctl_int(avctx, VP8E_SET_NOISE_SENSITIVITY, avctx->noise_reduction);
294
295 //provide dummy value to initialize wrapper, values will be updated each _encode()
296 vpx_img_wrap(&ctx->rawimg, VPX_IMG_FMT_I420, avctx->width, avctx->height, 1,
297 (unsigned char*)1);
298
299 avctx->coded_frame = avcodec_alloc_frame();
300 if (!avctx->coded_frame) {
301 av_log(avctx, AV_LOG_ERROR, "Error allocating coded frame\n");
302 vp8_free(avctx);
303 return AVERROR(ENOMEM);
304 }
305 return 0;
306}
307
308static inline void cx_pktcpy(struct FrameListData *dst,
309 const struct vpx_codec_cx_pkt *src)
310{
311 dst->pts = src->data.frame.pts;
312 dst->duration = src->data.frame.duration;
313 dst->flags = src->data.frame.flags;
314 dst->sz = src->data.frame.sz;
315 dst->buf = src->data.frame.buf;
316}
317
318/**
319 * Store coded frame information in format suitable for return from encode().
320 *
321 * Write buffer information from @a cx_frame to @a buf & @a buf_size.
322 * Timing/frame details to @a coded_frame.
323 * @return Frame size written to @a buf on success
324 * @return AVERROR(EINVAL) on error
325 */
326static int storeframe(AVCodecContext *avctx, struct FrameListData *cx_frame,
327 uint8_t *buf, int buf_size, AVFrame *coded_frame)
328{
329 if ((int) cx_frame->sz <= buf_size) {
330 buf_size = cx_frame->sz;
331 memcpy(buf, cx_frame->buf, buf_size);
332 coded_frame->pts = cx_frame->pts;
333 coded_frame->key_frame = !!(cx_frame->flags & VPX_FRAME_IS_KEY);
334
335 if (coded_frame->key_frame)
336 coded_frame->pict_type = FF_I_TYPE;
337 else
338 coded_frame->pict_type = FF_P_TYPE;
339 } else {
340 av_log(avctx, AV_LOG_ERROR,
341 "Compressed frame larger than storage provided! (%zu/%d)\n",
342 cx_frame->sz, buf_size);
343 return AVERROR(EINVAL);
344 }
345 return buf_size;
346}
347
348/**
349 * Queue multiple output frames from the encoder, returning the front-most.
350 * In cases where vpx_codec_get_cx_data() returns more than 1 frame append
351 * the frame queue. Return the head frame if available.
352 * @return Stored frame size
353 * @return AVERROR(EINVAL) on output size error
354 * @return AVERROR(ENOMEM) on coded frame queue data allocation error
355 */
356static int queue_frames(AVCodecContext *avctx, uint8_t *buf, int buf_size,
357 AVFrame *coded_frame)
358{
359 VP8Context *ctx = avctx->priv_data;
360 const struct vpx_codec_cx_pkt *pkt;
361 const void *iter = NULL;
362 int size = 0;
363
364 if (ctx->coded_frame_list) {
365 struct FrameListData *cx_frame = ctx->coded_frame_list;
366 /* return the leading frame if we've already begun queueing */
367 size = storeframe(avctx, cx_frame, buf, buf_size, coded_frame);
368 if (size < 0)
369 return AVERROR(EINVAL);
370 ctx->coded_frame_list = cx_frame->next;
371 free_coded_frame(cx_frame);
372 }
373
374 /* consume all available output from the encoder before returning. buffers
375 are only good through the next vpx_codec call */
376 while ((pkt = vpx_codec_get_cx_data(&ctx->encoder, &iter))) {
377 switch (pkt->kind) {
378 case VPX_CODEC_CX_FRAME_PKT:
379 if (!size) {
380 struct FrameListData cx_frame;
381
382 /* avoid storing the frame when the list is empty and we haven't yet
383 provided a frame for output */
384 assert(!ctx->coded_frame_list);
385 cx_pktcpy(&cx_frame, pkt);
386 size = storeframe(avctx, &cx_frame, buf, buf_size, coded_frame);
387 if (size < 0)
388 return AVERROR(EINVAL);
389 } else {
390 struct FrameListData *cx_frame =
391 av_malloc(sizeof(struct FrameListData));
392
393 if (!cx_frame) {
394 av_log(avctx, AV_LOG_ERROR,
395 "Frame queue element alloc failed\n");
396 return AVERROR(ENOMEM);
397 }
398 cx_pktcpy(cx_frame, pkt);
399 cx_frame->buf = av_malloc(cx_frame->sz);
400
401 if (!cx_frame->buf) {
402 av_log(avctx, AV_LOG_ERROR,
403 "Data buffer alloc (%zu bytes) failed\n",
404 cx_frame->sz);
405 return AVERROR(ENOMEM);
406 }
407 memcpy(cx_frame->buf, pkt->data.frame.buf, pkt->data.frame.sz);
408 coded_frame_add(&ctx->coded_frame_list, cx_frame);
409 }
410 break;
411 case VPX_CODEC_STATS_PKT: {
412 struct vpx_fixed_buf *stats = &ctx->twopass_stats;
413 stats->buf = av_realloc(stats->buf,
414 stats->sz + pkt->data.twopass_stats.sz);
415 if (!stats->buf) {
416 av_log(avctx, AV_LOG_ERROR, "Stat buffer realloc failed\n");
417 return AVERROR(ENOMEM);
418 }
419 memcpy((uint8_t*)stats->buf + stats->sz,
420 pkt->data.twopass_stats.buf, pkt->data.twopass_stats.sz);
421 stats->sz += pkt->data.twopass_stats.sz;
422 break;
423 }
424 case VPX_CODEC_PSNR_PKT: //FIXME add support for CODEC_FLAG_PSNR
425 case VPX_CODEC_CUSTOM_PKT:
426 //ignore unsupported/unrecognized packet types
427 break;
428 }
429 }
430
431 return size;
432}
433
434static int vp8_encode(AVCodecContext *avctx, uint8_t *buf, int buf_size,
435 void *data)
436{
437 VP8Context *ctx = avctx->priv_data;
438 AVFrame *frame = data;
439 struct vpx_image *rawimg = NULL;
440 int64_t timestamp = 0;
441 int res, coded_size;
442
443 if (frame) {
444 rawimg = &ctx->rawimg;
445 rawimg->planes[VPX_PLANE_Y] = frame->data[0];
446 rawimg->planes[VPX_PLANE_U] = frame->data[1];
447 rawimg->planes[VPX_PLANE_V] = frame->data[2];
448 rawimg->stride[VPX_PLANE_Y] = frame->linesize[0];
449 rawimg->stride[VPX_PLANE_U] = frame->linesize[1];
450 rawimg->stride[VPX_PLANE_V] = frame->linesize[2];
451 timestamp = frame->pts;
452 }
453
454 res = vpx_codec_encode(&ctx->encoder, rawimg, timestamp,
455 avctx->ticks_per_frame, 0, ctx->deadline);
456 if (res != VPX_CODEC_OK) {
457 log_encoder_error(avctx, "Error encoding frame");
458 return AVERROR_INVALIDDATA;
459 }
460 coded_size = queue_frames(avctx, buf, buf_size, avctx->coded_frame);
461
462 if (!frame && avctx->flags & CODEC_FLAG_PASS1) {
463 unsigned int b64_size = ((ctx->twopass_stats.sz + 2) / 3) * 4 + 1;
464
465 avctx->stats_out = av_malloc(b64_size);
466 if (!avctx->stats_out) {
467 av_log(avctx, AV_LOG_ERROR, "Stat buffer alloc (%d bytes) failed\n",
468 b64_size);
469 return AVERROR(ENOMEM);
470 }
471 av_base64_encode(avctx->stats_out, b64_size, ctx->twopass_stats.buf,
472 ctx->twopass_stats.sz);
473 }
474 return coded_size;
475}
476
477AVCodec libvpx_encoder = {
478 "libvpx",
479 AVMEDIA_TYPE_VIDEO,
480 CODEC_ID_VP8,
481 sizeof(VP8Context),
482 vp8_init,
483 vp8_encode,
484 vp8_free,
485 NULL,
486 CODEC_CAP_DELAY,
487 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
488 .long_name = NULL_IF_CONFIG_SMALL("libvpx VP8"),
489};