1 /*
2 * Copyright (c) 2012 Michael Niedermayer <michaelni@gmx.at>
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 #include <stdatomic.h>
22
23 #include "frame_thread_encoder.h"
24
25 #include "libavutil/fifo.h"
26 #include "libavutil/avassert.h"
27 #include "libavutil/imgutils.h"
28 #include "libavutil/opt.h"
29 #include "libavutil/thread.h"
30 #include "avcodec.h"
31 #include "internal.h"
32 #include "thread.h"
33
34 #define MAX_THREADS 64
35 #define BUFFER_SIZE (2*MAX_THREADS)
36
37 typedef struct{
38 void *indata;
39 void *outdata;
40 int64_t return_code;
41 unsigned index;
42 } Task;
43
44 typedef struct{
45 AVCodecContext *parent_avctx;
46 pthread_mutex_t buffer_mutex;
47
48 AVFifoBuffer *task_fifo;
49 pthread_mutex_t task_fifo_mutex;
50 pthread_cond_t task_fifo_cond;
51
52 Task finished_tasks[BUFFER_SIZE];
53 pthread_mutex_t finished_task_mutex;
54 pthread_cond_t finished_task_cond;
55
56 unsigned task_index;
57 unsigned finished_task_index;
58
59 pthread_t worker[MAX_THREADS];
60 atomic_int exit;
61 } ThreadContext;
62
worker(void * v)63 static void * attribute_align_arg worker(void *v){
64 AVCodecContext *avctx = v;
65 ThreadContext *c = avctx->internal->frame_thread_encoder;
66 AVPacket *pkt = NULL;
67
68 while (!atomic_load(&c->exit)) {
69 int got_packet = 0, ret;
70 AVFrame *frame;
71 Task task;
72
73 if(!pkt) pkt = av_packet_alloc();
74 if(!pkt) continue;
75 av_init_packet(pkt);
76
77 pthread_mutex_lock(&c->task_fifo_mutex);
78 while (av_fifo_size(c->task_fifo) <= 0 || atomic_load(&c->exit)) {
79 if (atomic_load(&c->exit)) {
80 pthread_mutex_unlock(&c->task_fifo_mutex);
81 goto end;
82 }
83 pthread_cond_wait(&c->task_fifo_cond, &c->task_fifo_mutex);
84 }
85 av_fifo_generic_read(c->task_fifo, &task, sizeof(task), NULL);
86 pthread_mutex_unlock(&c->task_fifo_mutex);
87 frame = task.indata;
88
89 ret = avctx->codec->encode2(avctx, pkt, frame, &got_packet);
90 if(got_packet) {
91 int ret2 = av_packet_make_refcounted(pkt);
92 if (ret >= 0 && ret2 < 0)
93 ret = ret2;
94 pkt->pts = pkt->dts = frame->pts;
95 } else {
96 pkt->data = NULL;
97 pkt->size = 0;
98 }
99 pthread_mutex_lock(&c->buffer_mutex);
100 av_frame_unref(frame);
101 pthread_mutex_unlock(&c->buffer_mutex);
102 av_frame_free(&frame);
103 pthread_mutex_lock(&c->finished_task_mutex);
104 c->finished_tasks[task.index].outdata = pkt; pkt = NULL;
105 c->finished_tasks[task.index].return_code = ret;
106 pthread_cond_signal(&c->finished_task_cond);
107 pthread_mutex_unlock(&c->finished_task_mutex);
108 }
109 end:
110 av_free(pkt);
111 pthread_mutex_lock(&c->buffer_mutex);
112 avcodec_close(avctx);
113 pthread_mutex_unlock(&c->buffer_mutex);
114 av_freep(&avctx);
115 return NULL;
116 }
117
ff_frame_thread_encoder_init(AVCodecContext * avctx,AVDictionary * options)118 int ff_frame_thread_encoder_init(AVCodecContext *avctx, AVDictionary *options){
119 int i=0;
120 ThreadContext *c;
121 AVCodecContext *thread_avctx = NULL;
122
123 if( !(avctx->thread_type & FF_THREAD_FRAME)
124 || !(avctx->codec->capabilities & AV_CODEC_CAP_FRAME_THREADS))
125 return 0;
126
127 if( !avctx->thread_count
128 && avctx->codec_id == AV_CODEC_ID_MJPEG
129 && !(avctx->flags & AV_CODEC_FLAG_QSCALE)) {
130 av_log(avctx, AV_LOG_DEBUG,
131 "Forcing thread count to 1 for MJPEG encoding, use -thread_type slice "
132 "or a constant quantizer if you want to use multiple cpu cores\n");
133 avctx->thread_count = 1;
134 }
135 if( avctx->thread_count > 1
136 && avctx->codec_id == AV_CODEC_ID_MJPEG
137 && !(avctx->flags & AV_CODEC_FLAG_QSCALE))
138 av_log(avctx, AV_LOG_WARNING,
139 "MJPEG CBR encoding works badly with frame multi-threading, consider "
140 "using -threads 1, -thread_type slice or a constant quantizer.\n");
141
142 if (avctx->codec_id == AV_CODEC_ID_HUFFYUV ||
143 avctx->codec_id == AV_CODEC_ID_FFVHUFF) {
144 int warn = 0;
145 int context_model = 0;
146 AVDictionaryEntry *con = av_dict_get(options, "context", NULL, AV_DICT_MATCH_CASE);
147
148 if (con && con->value)
149 context_model = atoi(con->value);
150
151 if (avctx->flags & AV_CODEC_FLAG_PASS1)
152 warn = 1;
153 else if(context_model > 0) {
154 AVDictionaryEntry *t = av_dict_get(options, "non_deterministic",
155 NULL, AV_DICT_MATCH_CASE);
156 warn = !t || !t->value || !atoi(t->value) ? 1 : 0;
157 }
158 // huffyuv does not support these with multiple frame threads currently
159 if (warn) {
160 av_log(avctx, AV_LOG_WARNING,
161 "Forcing thread count to 1 for huffyuv encoding with first pass or context 1\n");
162 avctx->thread_count = 1;
163 }
164 }
165
166 if(!avctx->thread_count) {
167 avctx->thread_count = av_cpu_count();
168 avctx->thread_count = FFMIN(avctx->thread_count, MAX_THREADS);
169 }
170
171 if(avctx->thread_count <= 1)
172 return 0;
173
174 if(avctx->thread_count > MAX_THREADS)
175 return AVERROR(EINVAL);
176
177 av_assert0(!avctx->internal->frame_thread_encoder);
178 c = avctx->internal->frame_thread_encoder = av_mallocz(sizeof(ThreadContext));
179 if(!c)
180 return AVERROR(ENOMEM);
181
182 c->parent_avctx = avctx;
183
184 c->task_fifo = av_fifo_alloc_array(BUFFER_SIZE, sizeof(Task));
185 if(!c->task_fifo)
186 goto fail;
187
188 pthread_mutex_init(&c->task_fifo_mutex, NULL);
189 pthread_mutex_init(&c->finished_task_mutex, NULL);
190 pthread_mutex_init(&c->buffer_mutex, NULL);
191 pthread_cond_init(&c->task_fifo_cond, NULL);
192 pthread_cond_init(&c->finished_task_cond, NULL);
193 atomic_init(&c->exit, 0);
194
195 for(i=0; i<avctx->thread_count ; i++){
196 AVDictionary *tmp = NULL;
197 int ret;
198 void *tmpv;
199 thread_avctx = avcodec_alloc_context3(avctx->codec);
200 if(!thread_avctx)
201 goto fail;
202 tmpv = thread_avctx->priv_data;
203 *thread_avctx = *avctx;
204 thread_avctx->priv_data = tmpv;
205 thread_avctx->internal = NULL;
206 thread_avctx->hw_frames_ctx = NULL;
207 ret = av_opt_copy(thread_avctx, avctx);
208 if (ret < 0)
209 goto fail;
210 if (avctx->codec->priv_class) {
211 int ret = av_opt_copy(thread_avctx->priv_data, avctx->priv_data);
212 if (ret < 0)
213 goto fail;
214 } else if (avctx->codec->priv_data_size) {
215 memcpy(thread_avctx->priv_data, avctx->priv_data, avctx->codec->priv_data_size);
216 }
217 thread_avctx->thread_count = 1;
218 thread_avctx->active_thread_type &= ~FF_THREAD_FRAME;
219
220 av_dict_copy(&tmp, options, 0);
221 av_dict_set(&tmp, "threads", "1", 0);
222 if(avcodec_open2(thread_avctx, avctx->codec, &tmp) < 0) {
223 av_dict_free(&tmp);
224 goto fail;
225 }
226 av_dict_free(&tmp);
227 av_assert0(!thread_avctx->internal->frame_thread_encoder);
228 thread_avctx->internal->frame_thread_encoder = c;
229 if(pthread_create(&c->worker[i], NULL, worker, thread_avctx)) {
230 goto fail;
231 }
232 }
233
234 avctx->active_thread_type = FF_THREAD_FRAME;
235
236 return 0;
237 fail:
238 avcodec_close(thread_avctx);
239 av_freep(&thread_avctx);
240 avctx->thread_count = i;
241 av_log(avctx, AV_LOG_ERROR, "ff_frame_thread_encoder_init failed\n");
242 ff_frame_thread_encoder_free(avctx);
243 return -1;
244 }
245
ff_frame_thread_encoder_free(AVCodecContext * avctx)246 void ff_frame_thread_encoder_free(AVCodecContext *avctx){
247 int i;
248 ThreadContext *c= avctx->internal->frame_thread_encoder;
249
250 pthread_mutex_lock(&c->task_fifo_mutex);
251 atomic_store(&c->exit, 1);
252 pthread_cond_broadcast(&c->task_fifo_cond);
253 pthread_mutex_unlock(&c->task_fifo_mutex);
254
255 for (i=0; i<avctx->thread_count; i++) {
256 pthread_join(c->worker[i], NULL);
257 }
258
259 while (av_fifo_size(c->task_fifo) > 0) {
260 Task task;
261 AVFrame *frame;
262 av_fifo_generic_read(c->task_fifo, &task, sizeof(task), NULL);
263 frame = task.indata;
264 av_frame_free(&frame);
265 task.indata = NULL;
266 }
267
268 for (i=0; i<BUFFER_SIZE; i++) {
269 if (c->finished_tasks[i].outdata != NULL) {
270 AVPacket *pkt = c->finished_tasks[i].outdata;
271 av_packet_free(&pkt);
272 c->finished_tasks[i].outdata = NULL;
273 }
274 }
275
276 pthread_mutex_destroy(&c->task_fifo_mutex);
277 pthread_mutex_destroy(&c->finished_task_mutex);
278 pthread_mutex_destroy(&c->buffer_mutex);
279 pthread_cond_destroy(&c->task_fifo_cond);
280 pthread_cond_destroy(&c->finished_task_cond);
281 av_fifo_freep(&c->task_fifo);
282 av_freep(&avctx->internal->frame_thread_encoder);
283 }
284
ff_thread_video_encode_frame(AVCodecContext * avctx,AVPacket * pkt,const AVFrame * frame,int * got_packet_ptr)285 int ff_thread_video_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet_ptr){
286 ThreadContext *c = avctx->internal->frame_thread_encoder;
287 Task task;
288 int ret;
289
290 av_assert1(!*got_packet_ptr);
291
292 if(frame){
293 AVFrame *new = av_frame_alloc();
294 if(!new)
295 return AVERROR(ENOMEM);
296 ret = av_frame_ref(new, frame);
297 if(ret < 0) {
298 av_frame_free(&new);
299 return ret;
300 }
301
302 task.index = c->task_index;
303 task.indata = (void*)new;
304 pthread_mutex_lock(&c->task_fifo_mutex);
305 av_fifo_generic_write(c->task_fifo, &task, sizeof(task), NULL);
306 pthread_cond_signal(&c->task_fifo_cond);
307 pthread_mutex_unlock(&c->task_fifo_mutex);
308
309 c->task_index = (c->task_index+1) % BUFFER_SIZE;
310 }
311
312 pthread_mutex_lock(&c->finished_task_mutex);
313 if (c->task_index == c->finished_task_index ||
314 (frame && !c->finished_tasks[c->finished_task_index].outdata &&
315 (c->task_index - c->finished_task_index) % BUFFER_SIZE <= avctx->thread_count)) {
316 pthread_mutex_unlock(&c->finished_task_mutex);
317 return 0;
318 }
319
320 while (!c->finished_tasks[c->finished_task_index].outdata) {
321 pthread_cond_wait(&c->finished_task_cond, &c->finished_task_mutex);
322 }
323 task = c->finished_tasks[c->finished_task_index];
324 *pkt = *(AVPacket*)(task.outdata);
325 if(pkt->data)
326 *got_packet_ptr = 1;
327 av_freep(&c->finished_tasks[c->finished_task_index].outdata);
328 c->finished_task_index = (c->finished_task_index+1) % BUFFER_SIZE;
329 pthread_mutex_unlock(&c->finished_task_mutex);
330
331 return task.return_code;
332 }
333