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 "avcodec.h"
27#include "internal.h"
28#include "thread.h"
29
30#if HAVE_PTHREADS
31#include <pthread.h>
32#elif HAVE_W32THREADS
33#include "compat/w32pthreads.h"
34#elif HAVE_OS2THREADS
35#include "compat/os2threads.h"
36#endif
37
38#define MAX_THREADS 64
39#define BUFFER_SIZE (2*MAX_THREADS)
40
41typedef struct{
42    void *indata;
43    void *outdata;
44    int64_t return_code;
45    unsigned index;
46} Task;
47
48typedef struct{
49    AVCodecContext *parent_avctx;
50    pthread_mutex_t buffer_mutex;
51
52    AVFifoBuffer *task_fifo;
53    pthread_mutex_t task_fifo_mutex;
54    pthread_cond_t task_fifo_cond;
55
56    Task finished_tasks[BUFFER_SIZE];
57    pthread_mutex_t finished_task_mutex;
58    pthread_cond_t finished_task_cond;
59
60    unsigned task_index;
61    unsigned finished_task_index;
62
63    pthread_t worker[MAX_THREADS];
64    int exit;
65} ThreadContext;
66
67static void * attribute_align_arg worker(void *v){
68    AVCodecContext *avctx = v;
69    ThreadContext *c = avctx->internal->frame_thread_encoder;
70    AVPacket *pkt = NULL;
71
72    while(!c->exit){
73        int got_packet, ret;
74        AVFrame *frame;
75        Task task;
76
77        if(!pkt) pkt= av_mallocz(sizeof(*pkt));
78        if(!pkt) continue;
79        av_init_packet(pkt);
80
81        pthread_mutex_lock(&c->task_fifo_mutex);
82        while (av_fifo_size(c->task_fifo) <= 0 || c->exit) {
83            if(c->exit){
84                pthread_mutex_unlock(&c->task_fifo_mutex);
85                goto end;
86            }
87            pthread_cond_wait(&c->task_fifo_cond, &c->task_fifo_mutex);
88        }
89        av_fifo_generic_read(c->task_fifo, &task, sizeof(task), NULL);
90        pthread_mutex_unlock(&c->task_fifo_mutex);
91        frame = task.indata;
92
93        ret = avcodec_encode_video2(avctx, pkt, frame, &got_packet);
94        pthread_mutex_lock(&c->buffer_mutex);
95        av_frame_unref(frame);
96        pthread_mutex_unlock(&c->buffer_mutex);
97        av_frame_free(&frame);
98        if(got_packet) {
99            av_dup_packet(pkt);
100        } else {
101            pkt->data = NULL;
102            pkt->size = 0;
103        }
104        pthread_mutex_lock(&c->finished_task_mutex);
105        c->finished_tasks[task.index].outdata = pkt; pkt = NULL;
106        c->finished_tasks[task.index].return_code = ret;
107        pthread_cond_signal(&c->finished_task_cond);
108        pthread_mutex_unlock(&c->finished_task_mutex);
109    }
110end:
111    av_free(pkt);
112    pthread_mutex_lock(&c->buffer_mutex);
113    avcodec_close(avctx);
114    pthread_mutex_unlock(&c->buffer_mutex);
115    av_freep(&avctx);
116    return NULL;
117}
118
119int ff_frame_thread_encoder_init(AVCodecContext *avctx, AVDictionary *options){
120    int i=0;
121    ThreadContext *c;
122
123
124    if(   !(avctx->thread_type & FF_THREAD_FRAME)
125       || !(avctx->codec->capabilities & CODEC_CAP_INTRA_ONLY))
126        return 0;
127
128    if(   !avctx->thread_count
129       && avctx->codec_id == AV_CODEC_ID_MJPEG
130       && !(avctx->flags & CODEC_FLAG_QSCALE)) {
131        av_log(avctx, AV_LOG_DEBUG,
132               "Forcing thread count to 1 for MJPEG encoding, use -thread_type slice "
133               "or a constant quantizer if you want to use multiple cpu cores\n");
134        avctx->thread_count = 1;
135    }
136    if(   avctx->thread_count > 1
137       && avctx->codec_id == AV_CODEC_ID_MJPEG
138       && !(avctx->flags & CODEC_FLAG_QSCALE))
139        av_log(avctx, AV_LOG_WARNING,
140               "MJPEG CBR encoding works badly with frame multi-threading, consider "
141               "using -threads 1, -thread_type slice or a constant quantizer.\n");
142
143    if (avctx->codec_id == AV_CODEC_ID_HUFFYUV ||
144        avctx->codec_id == AV_CODEC_ID_FFVHUFF) {
145        // huffyuv does not support these with multiple frame threads currently
146        if (avctx->context_model > 0 || (avctx->flags & CODEC_FLAG_PASS1)) {
147            av_log(avctx, AV_LOG_WARNING,
148               "Forcing thread count to 1 for huffyuv encoding with first pass or context 1\n");
149            avctx->thread_count = 1;
150        }
151    }
152
153    if(!avctx->thread_count) {
154        avctx->thread_count = av_cpu_count();
155        avctx->thread_count = FFMIN(avctx->thread_count, MAX_THREADS);
156    }
157
158    if(avctx->thread_count <= 1)
159        return 0;
160
161    if(avctx->thread_count > MAX_THREADS)
162        return AVERROR(EINVAL);
163
164    av_assert0(!avctx->internal->frame_thread_encoder);
165    c = avctx->internal->frame_thread_encoder = av_mallocz(sizeof(ThreadContext));
166    if(!c)
167        return AVERROR(ENOMEM);
168
169    c->parent_avctx = avctx;
170
171    c->task_fifo = av_fifo_alloc_array(BUFFER_SIZE, sizeof(Task));
172    if(!c->task_fifo)
173        goto fail;
174
175    pthread_mutex_init(&c->task_fifo_mutex, NULL);
176    pthread_mutex_init(&c->finished_task_mutex, NULL);
177    pthread_mutex_init(&c->buffer_mutex, NULL);
178    pthread_cond_init(&c->task_fifo_cond, NULL);
179    pthread_cond_init(&c->finished_task_cond, NULL);
180
181    for(i=0; i<avctx->thread_count ; i++){
182        AVDictionary *tmp = NULL;
183        void *tmpv;
184        AVCodecContext *thread_avctx = avcodec_alloc_context3(avctx->codec);
185        if(!thread_avctx)
186            goto fail;
187        tmpv = thread_avctx->priv_data;
188        *thread_avctx = *avctx;
189        thread_avctx->priv_data = tmpv;
190        thread_avctx->internal = NULL;
191        memcpy(thread_avctx->priv_data, avctx->priv_data, avctx->codec->priv_data_size);
192        thread_avctx->thread_count = 1;
193        thread_avctx->active_thread_type &= ~FF_THREAD_FRAME;
194
195        av_dict_copy(&tmp, options, 0);
196        av_dict_set(&tmp, "threads", "1", 0);
197        if(avcodec_open2(thread_avctx, avctx->codec, &tmp) < 0) {
198            av_dict_free(&tmp);
199            goto fail;
200        }
201        av_dict_free(&tmp);
202        av_assert0(!thread_avctx->internal->frame_thread_encoder);
203        thread_avctx->internal->frame_thread_encoder = c;
204        if(pthread_create(&c->worker[i], NULL, worker, thread_avctx)) {
205            goto fail;
206        }
207    }
208
209    avctx->active_thread_type = FF_THREAD_FRAME;
210
211    return 0;
212fail:
213    avctx->thread_count = i;
214    av_log(avctx, AV_LOG_ERROR, "ff_frame_thread_encoder_init failed\n");
215    ff_frame_thread_encoder_free(avctx);
216    return -1;
217}
218
219void ff_frame_thread_encoder_free(AVCodecContext *avctx){
220    int i;
221    ThreadContext *c= avctx->internal->frame_thread_encoder;
222
223    pthread_mutex_lock(&c->task_fifo_mutex);
224    c->exit = 1;
225    pthread_cond_broadcast(&c->task_fifo_cond);
226    pthread_mutex_unlock(&c->task_fifo_mutex);
227
228    for (i=0; i<avctx->thread_count; i++) {
229         pthread_join(c->worker[i], NULL);
230    }
231
232    pthread_mutex_destroy(&c->task_fifo_mutex);
233    pthread_mutex_destroy(&c->finished_task_mutex);
234    pthread_mutex_destroy(&c->buffer_mutex);
235    pthread_cond_destroy(&c->task_fifo_cond);
236    pthread_cond_destroy(&c->finished_task_cond);
237    av_fifo_freep(&c->task_fifo);
238    av_freep(&avctx->internal->frame_thread_encoder);
239}
240
241int ff_thread_video_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet_ptr){
242    ThreadContext *c = avctx->internal->frame_thread_encoder;
243    Task task;
244    int ret;
245
246    av_assert1(!*got_packet_ptr);
247
248    if(frame){
249        if(!(avctx->flags & CODEC_FLAG_INPUT_PRESERVED)){
250            AVFrame *new = av_frame_alloc();
251            if(!new)
252                return AVERROR(ENOMEM);
253            pthread_mutex_lock(&c->buffer_mutex);
254            ret = ff_get_buffer(c->parent_avctx, new, 0);
255            pthread_mutex_unlock(&c->buffer_mutex);
256            if(ret<0)
257                return ret;
258            new->pts = frame->pts;
259            new->quality = frame->quality;
260            new->pict_type = frame->pict_type;
261            av_image_copy(new->data, new->linesize, (const uint8_t **)frame->data, frame->linesize,
262                          avctx->pix_fmt, avctx->width, avctx->height);
263            frame = new;
264        }
265
266        task.index = c->task_index;
267        task.indata = (void*)frame;
268        pthread_mutex_lock(&c->task_fifo_mutex);
269        av_fifo_generic_write(c->task_fifo, &task, sizeof(task), NULL);
270        pthread_cond_signal(&c->task_fifo_cond);
271        pthread_mutex_unlock(&c->task_fifo_mutex);
272
273        c->task_index = (c->task_index+1) % BUFFER_SIZE;
274
275        if(!c->finished_tasks[c->finished_task_index].outdata && (c->task_index - c->finished_task_index) % BUFFER_SIZE <= avctx->thread_count)
276            return 0;
277    }
278
279    if(c->task_index == c->finished_task_index)
280        return 0;
281
282    pthread_mutex_lock(&c->finished_task_mutex);
283    while (!c->finished_tasks[c->finished_task_index].outdata) {
284        pthread_cond_wait(&c->finished_task_cond, &c->finished_task_mutex);
285    }
286    task = c->finished_tasks[c->finished_task_index];
287    *pkt = *(AVPacket*)(task.outdata);
288    if(pkt->data)
289        *got_packet_ptr = 1;
290    av_freep(&c->finished_tasks[c->finished_task_index].outdata);
291    c->finished_task_index = (c->finished_task_index+1) % BUFFER_SIZE;
292    pthread_mutex_unlock(&c->finished_task_mutex);
293
294    return task.return_code;
295}
296