blob: 27ae356af30a7f5661114813079392ae037daf8d
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 "frame_thread_encoder.h" |
22 | |
23 | #include "libavutil/fifo.h" |
24 | #include "libavutil/avassert.h" |
25 | #include "libavutil/imgutils.h" |
26 | #include "libavutil/thread.h" |
27 | #include "avcodec.h" |
28 | #include "internal.h" |
29 | #include "thread.h" |
30 | |
31 | #define MAX_THREADS 64 |
32 | #define BUFFER_SIZE (2*MAX_THREADS) |
33 | |
34 | typedef struct{ |
35 | void *indata; |
36 | void *outdata; |
37 | int64_t return_code; |
38 | unsigned index; |
39 | } Task; |
40 | |
41 | typedef struct{ |
42 | AVCodecContext *parent_avctx; |
43 | pthread_mutex_t buffer_mutex; |
44 | |
45 | AVFifoBuffer *task_fifo; |
46 | pthread_mutex_t task_fifo_mutex; |
47 | pthread_cond_t task_fifo_cond; |
48 | |
49 | Task finished_tasks[BUFFER_SIZE]; |
50 | pthread_mutex_t finished_task_mutex; |
51 | pthread_cond_t finished_task_cond; |
52 | |
53 | unsigned task_index; |
54 | unsigned finished_task_index; |
55 | |
56 | pthread_t worker[MAX_THREADS]; |
57 | int exit; |
58 | } ThreadContext; |
59 | |
60 | static void * attribute_align_arg worker(void *v){ |
61 | AVCodecContext *avctx = v; |
62 | ThreadContext *c = avctx->internal->frame_thread_encoder; |
63 | AVPacket *pkt = NULL; |
64 | |
65 | while(!c->exit){ |
66 | int got_packet, ret; |
67 | AVFrame *frame; |
68 | Task task; |
69 | |
70 | if(!pkt) pkt= av_mallocz(sizeof(*pkt)); |
71 | if(!pkt) continue; |
72 | av_init_packet(pkt); |
73 | |
74 | pthread_mutex_lock(&c->task_fifo_mutex); |
75 | while (av_fifo_size(c->task_fifo) <= 0 || c->exit) { |
76 | if(c->exit){ |
77 | pthread_mutex_unlock(&c->task_fifo_mutex); |
78 | goto end; |
79 | } |
80 | pthread_cond_wait(&c->task_fifo_cond, &c->task_fifo_mutex); |
81 | } |
82 | av_fifo_generic_read(c->task_fifo, &task, sizeof(task), NULL); |
83 | pthread_mutex_unlock(&c->task_fifo_mutex); |
84 | frame = task.indata; |
85 | |
86 | ret = avcodec_encode_video2(avctx, pkt, frame, &got_packet); |
87 | pthread_mutex_lock(&c->buffer_mutex); |
88 | av_frame_unref(frame); |
89 | pthread_mutex_unlock(&c->buffer_mutex); |
90 | av_frame_free(&frame); |
91 | if(got_packet) { |
92 | int ret2 = av_dup_packet(pkt); |
93 | if (ret >= 0 && ret2 < 0) |
94 | ret = ret2; |
95 | } else { |
96 | pkt->data = NULL; |
97 | pkt->size = 0; |
98 | } |
99 | pthread_mutex_lock(&c->finished_task_mutex); |
100 | c->finished_tasks[task.index].outdata = pkt; pkt = NULL; |
101 | c->finished_tasks[task.index].return_code = ret; |
102 | pthread_cond_signal(&c->finished_task_cond); |
103 | pthread_mutex_unlock(&c->finished_task_mutex); |
104 | } |
105 | end: |
106 | av_free(pkt); |
107 | pthread_mutex_lock(&c->buffer_mutex); |
108 | avcodec_close(avctx); |
109 | pthread_mutex_unlock(&c->buffer_mutex); |
110 | av_freep(&avctx); |
111 | return NULL; |
112 | } |
113 | |
114 | int ff_frame_thread_encoder_init(AVCodecContext *avctx, AVDictionary *options){ |
115 | int i=0; |
116 | ThreadContext *c; |
117 | |
118 | |
119 | if( !(avctx->thread_type & FF_THREAD_FRAME) |
120 | || !(avctx->codec->capabilities & AV_CODEC_CAP_INTRA_ONLY)) |
121 | return 0; |
122 | |
123 | if( !avctx->thread_count |
124 | && avctx->codec_id == AV_CODEC_ID_MJPEG |
125 | && !(avctx->flags & AV_CODEC_FLAG_QSCALE)) { |
126 | av_log(avctx, AV_LOG_DEBUG, |
127 | "Forcing thread count to 1 for MJPEG encoding, use -thread_type slice " |
128 | "or a constant quantizer if you want to use multiple cpu cores\n"); |
129 | avctx->thread_count = 1; |
130 | } |
131 | if( avctx->thread_count > 1 |
132 | && avctx->codec_id == AV_CODEC_ID_MJPEG |
133 | && !(avctx->flags & AV_CODEC_FLAG_QSCALE)) |
134 | av_log(avctx, AV_LOG_WARNING, |
135 | "MJPEG CBR encoding works badly with frame multi-threading, consider " |
136 | "using -threads 1, -thread_type slice or a constant quantizer.\n"); |
137 | |
138 | if (avctx->codec_id == AV_CODEC_ID_HUFFYUV || |
139 | avctx->codec_id == AV_CODEC_ID_FFVHUFF) { |
140 | int warn = 0; |
141 | int context_model = 0; |
142 | AVDictionaryEntry *con = av_dict_get(options, "context", NULL, AV_DICT_MATCH_CASE); |
143 | |
144 | if (con && con->value) |
145 | context_model = atoi(con->value); |
146 | |
147 | if (avctx->flags & AV_CODEC_FLAG_PASS1) |
148 | warn = 1; |
149 | else if(context_model > 0) { |
150 | AVDictionaryEntry *t = av_dict_get(options, "non_deterministic", |
151 | NULL, AV_DICT_MATCH_CASE); |
152 | warn = !t || !t->value || !atoi(t->value) ? 1 : 0; |
153 | } |
154 | // huffyuv does not support these with multiple frame threads currently |
155 | if (warn) { |
156 | av_log(avctx, AV_LOG_WARNING, |
157 | "Forcing thread count to 1 for huffyuv encoding with first pass or context 1\n"); |
158 | avctx->thread_count = 1; |
159 | } |
160 | } |
161 | |
162 | if(!avctx->thread_count) { |
163 | avctx->thread_count = av_cpu_count(); |
164 | avctx->thread_count = FFMIN(avctx->thread_count, MAX_THREADS); |
165 | } |
166 | |
167 | if(avctx->thread_count <= 1) |
168 | return 0; |
169 | |
170 | if(avctx->thread_count > MAX_THREADS) |
171 | return AVERROR(EINVAL); |
172 | |
173 | av_assert0(!avctx->internal->frame_thread_encoder); |
174 | c = avctx->internal->frame_thread_encoder = av_mallocz(sizeof(ThreadContext)); |
175 | if(!c) |
176 | return AVERROR(ENOMEM); |
177 | |
178 | c->parent_avctx = avctx; |
179 | |
180 | c->task_fifo = av_fifo_alloc_array(BUFFER_SIZE, sizeof(Task)); |
181 | if(!c->task_fifo) |
182 | goto fail; |
183 | |
184 | pthread_mutex_init(&c->task_fifo_mutex, NULL); |
185 | pthread_mutex_init(&c->finished_task_mutex, NULL); |
186 | pthread_mutex_init(&c->buffer_mutex, NULL); |
187 | pthread_cond_init(&c->task_fifo_cond, NULL); |
188 | pthread_cond_init(&c->finished_task_cond, NULL); |
189 | |
190 | for(i=0; i<avctx->thread_count ; i++){ |
191 | AVDictionary *tmp = NULL; |
192 | void *tmpv; |
193 | AVCodecContext *thread_avctx = avcodec_alloc_context3(avctx->codec); |
194 | if(!thread_avctx) |
195 | goto fail; |
196 | tmpv = thread_avctx->priv_data; |
197 | *thread_avctx = *avctx; |
198 | thread_avctx->priv_data = tmpv; |
199 | thread_avctx->internal = NULL; |
200 | memcpy(thread_avctx->priv_data, avctx->priv_data, avctx->codec->priv_data_size); |
201 | thread_avctx->thread_count = 1; |
202 | thread_avctx->active_thread_type &= ~FF_THREAD_FRAME; |
203 | |
204 | av_dict_copy(&tmp, options, 0); |
205 | av_dict_set(&tmp, "threads", "1", 0); |
206 | if(avcodec_open2(thread_avctx, avctx->codec, &tmp) < 0) { |
207 | av_dict_free(&tmp); |
208 | goto fail; |
209 | } |
210 | av_dict_free(&tmp); |
211 | av_assert0(!thread_avctx->internal->frame_thread_encoder); |
212 | thread_avctx->internal->frame_thread_encoder = c; |
213 | if(pthread_create(&c->worker[i], NULL, worker, thread_avctx)) { |
214 | goto fail; |
215 | } |
216 | } |
217 | |
218 | avctx->active_thread_type = FF_THREAD_FRAME; |
219 | |
220 | return 0; |
221 | fail: |
222 | avctx->thread_count = i; |
223 | av_log(avctx, AV_LOG_ERROR, "ff_frame_thread_encoder_init failed\n"); |
224 | ff_frame_thread_encoder_free(avctx); |
225 | return -1; |
226 | } |
227 | |
228 | void ff_frame_thread_encoder_free(AVCodecContext *avctx){ |
229 | int i; |
230 | ThreadContext *c= avctx->internal->frame_thread_encoder; |
231 | |
232 | pthread_mutex_lock(&c->task_fifo_mutex); |
233 | c->exit = 1; |
234 | pthread_cond_broadcast(&c->task_fifo_cond); |
235 | pthread_mutex_unlock(&c->task_fifo_mutex); |
236 | |
237 | for (i=0; i<avctx->thread_count; i++) { |
238 | pthread_join(c->worker[i], NULL); |
239 | } |
240 | |
241 | pthread_mutex_destroy(&c->task_fifo_mutex); |
242 | pthread_mutex_destroy(&c->finished_task_mutex); |
243 | pthread_mutex_destroy(&c->buffer_mutex); |
244 | pthread_cond_destroy(&c->task_fifo_cond); |
245 | pthread_cond_destroy(&c->finished_task_cond); |
246 | av_fifo_freep(&c->task_fifo); |
247 | av_freep(&avctx->internal->frame_thread_encoder); |
248 | } |
249 | |
250 | int ff_thread_video_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet_ptr){ |
251 | ThreadContext *c = avctx->internal->frame_thread_encoder; |
252 | Task task; |
253 | int ret; |
254 | |
255 | av_assert1(!*got_packet_ptr); |
256 | |
257 | if(frame){ |
258 | AVFrame *new = av_frame_alloc(); |
259 | if(!new) |
260 | return AVERROR(ENOMEM); |
261 | ret = av_frame_ref(new, frame); |
262 | if(ret < 0) { |
263 | av_frame_free(&new); |
264 | return ret; |
265 | } |
266 | |
267 | task.index = c->task_index; |
268 | task.indata = (void*)new; |
269 | pthread_mutex_lock(&c->task_fifo_mutex); |
270 | av_fifo_generic_write(c->task_fifo, &task, sizeof(task), NULL); |
271 | pthread_cond_signal(&c->task_fifo_cond); |
272 | pthread_mutex_unlock(&c->task_fifo_mutex); |
273 | |
274 | c->task_index = (c->task_index+1) % BUFFER_SIZE; |
275 | |
276 | if(!c->finished_tasks[c->finished_task_index].outdata && (c->task_index - c->finished_task_index) % BUFFER_SIZE <= avctx->thread_count) |
277 | return 0; |
278 | } |
279 | |
280 | if(c->task_index == c->finished_task_index) |
281 | return 0; |
282 | |
283 | pthread_mutex_lock(&c->finished_task_mutex); |
284 | while (!c->finished_tasks[c->finished_task_index].outdata) { |
285 | pthread_cond_wait(&c->finished_task_cond, &c->finished_task_mutex); |
286 | } |
287 | task = c->finished_tasks[c->finished_task_index]; |
288 | *pkt = *(AVPacket*)(task.outdata); |
289 | if(pkt->data) |
290 | *got_packet_ptr = 1; |
291 | av_freep(&c->finished_tasks[c->finished_task_index].outdata); |
292 | c->finished_task_index = (c->finished_task_index+1) % BUFFER_SIZE; |
293 | pthread_mutex_unlock(&c->finished_task_mutex); |
294 | |
295 | return task.return_code; |
296 | } |
297 |