| 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 | |
| 41 | typedef struct{ |
| 42 | void *indata; |
| 43 | void *outdata; |
| 44 | int64_t return_code; |
| 45 | unsigned index; |
| 46 | } Task; |
| 47 | |
| 48 | typedef 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 | |
| 67 | static 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 | } |
| 110 | end: |
| 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 | |
| 119 | int 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 | int warn = 0; |
| 146 | if (avctx->flags & CODEC_FLAG_PASS1) |
| 147 | warn = 1; |
| 148 | else if(avctx->context_model > 0) { |
| 149 | AVDictionaryEntry *t = av_dict_get(options, "non_deterministic", |
| 150 | NULL, AV_DICT_MATCH_CASE); |
| 151 | warn = !t || !t->value || !atoi(t->value) ? 1 : 0; |
| 152 | } |
| 153 | // huffyuv does not support these with multiple frame threads currently |
| 154 | if (warn) { |
| 155 | av_log(avctx, AV_LOG_WARNING, |
| 156 | "Forcing thread count to 1 for huffyuv encoding with first pass or context 1\n"); |
| 157 | avctx->thread_count = 1; |
| 158 | } |
| 159 | } |
| 160 | |
| 161 | if(!avctx->thread_count) { |
| 162 | avctx->thread_count = av_cpu_count(); |
| 163 | avctx->thread_count = FFMIN(avctx->thread_count, MAX_THREADS); |
| 164 | } |
| 165 | |
| 166 | if(avctx->thread_count <= 1) |
| 167 | return 0; |
| 168 | |
| 169 | if(avctx->thread_count > MAX_THREADS) |
| 170 | return AVERROR(EINVAL); |
| 171 | |
| 172 | av_assert0(!avctx->internal->frame_thread_encoder); |
| 173 | c = avctx->internal->frame_thread_encoder = av_mallocz(sizeof(ThreadContext)); |
| 174 | if(!c) |
| 175 | return AVERROR(ENOMEM); |
| 176 | |
| 177 | c->parent_avctx = avctx; |
| 178 | |
| 179 | c->task_fifo = av_fifo_alloc_array(BUFFER_SIZE, sizeof(Task)); |
| 180 | if(!c->task_fifo) |
| 181 | goto fail; |
| 182 | |
| 183 | pthread_mutex_init(&c->task_fifo_mutex, NULL); |
| 184 | pthread_mutex_init(&c->finished_task_mutex, NULL); |
| 185 | pthread_mutex_init(&c->buffer_mutex, NULL); |
| 186 | pthread_cond_init(&c->task_fifo_cond, NULL); |
| 187 | pthread_cond_init(&c->finished_task_cond, NULL); |
| 188 | |
| 189 | for(i=0; i<avctx->thread_count ; i++){ |
| 190 | AVDictionary *tmp = NULL; |
| 191 | void *tmpv; |
| 192 | AVCodecContext *thread_avctx = avcodec_alloc_context3(avctx->codec); |
| 193 | if(!thread_avctx) |
| 194 | goto fail; |
| 195 | tmpv = thread_avctx->priv_data; |
| 196 | *thread_avctx = *avctx; |
| 197 | thread_avctx->priv_data = tmpv; |
| 198 | thread_avctx->internal = NULL; |
| 199 | memcpy(thread_avctx->priv_data, avctx->priv_data, avctx->codec->priv_data_size); |
| 200 | thread_avctx->thread_count = 1; |
| 201 | thread_avctx->active_thread_type &= ~FF_THREAD_FRAME; |
| 202 | |
| 203 | av_dict_copy(&tmp, options, 0); |
| 204 | av_dict_set(&tmp, "threads", "1", 0); |
| 205 | if(avcodec_open2(thread_avctx, avctx->codec, &tmp) < 0) { |
| 206 | av_dict_free(&tmp); |
| 207 | goto fail; |
| 208 | } |
| 209 | av_dict_free(&tmp); |
| 210 | av_assert0(!thread_avctx->internal->frame_thread_encoder); |
| 211 | thread_avctx->internal->frame_thread_encoder = c; |
| 212 | if(pthread_create(&c->worker[i], NULL, worker, thread_avctx)) { |
| 213 | goto fail; |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | avctx->active_thread_type = FF_THREAD_FRAME; |
| 218 | |
| 219 | return 0; |
| 220 | fail: |
| 221 | avctx->thread_count = i; |
| 222 | av_log(avctx, AV_LOG_ERROR, "ff_frame_thread_encoder_init failed\n"); |
| 223 | ff_frame_thread_encoder_free(avctx); |
| 224 | return -1; |
| 225 | } |
| 226 | |
| 227 | void ff_frame_thread_encoder_free(AVCodecContext *avctx){ |
| 228 | int i; |
| 229 | ThreadContext *c= avctx->internal->frame_thread_encoder; |
| 230 | |
| 231 | pthread_mutex_lock(&c->task_fifo_mutex); |
| 232 | c->exit = 1; |
| 233 | pthread_cond_broadcast(&c->task_fifo_cond); |
| 234 | pthread_mutex_unlock(&c->task_fifo_mutex); |
| 235 | |
| 236 | for (i=0; i<avctx->thread_count; i++) { |
| 237 | pthread_join(c->worker[i], NULL); |
| 238 | } |
| 239 | |
| 240 | pthread_mutex_destroy(&c->task_fifo_mutex); |
| 241 | pthread_mutex_destroy(&c->finished_task_mutex); |
| 242 | pthread_mutex_destroy(&c->buffer_mutex); |
| 243 | pthread_cond_destroy(&c->task_fifo_cond); |
| 244 | pthread_cond_destroy(&c->finished_task_cond); |
| 245 | av_fifo_freep(&c->task_fifo); |
| 246 | av_freep(&avctx->internal->frame_thread_encoder); |
| 247 | } |
| 248 | |
| 249 | int ff_thread_video_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet_ptr){ |
| 250 | ThreadContext *c = avctx->internal->frame_thread_encoder; |
| 251 | Task task; |
| 252 | int ret; |
| 253 | |
| 254 | av_assert1(!*got_packet_ptr); |
| 255 | |
| 256 | if(frame){ |
| 257 | if(!(avctx->flags & CODEC_FLAG_INPUT_PRESERVED)){ |
| 258 | AVFrame *new = av_frame_alloc(); |
| 259 | if(!new) |
| 260 | return AVERROR(ENOMEM); |
| 261 | pthread_mutex_lock(&c->buffer_mutex); |
| 262 | ret = ff_get_buffer(c->parent_avctx, new, 0); |
| 263 | pthread_mutex_unlock(&c->buffer_mutex); |
| 264 | if(ret<0) |
| 265 | return ret; |
| 266 | new->pts = frame->pts; |
| 267 | new->quality = frame->quality; |
| 268 | new->pict_type = frame->pict_type; |
| 269 | av_image_copy(new->data, new->linesize, (const uint8_t **)frame->data, frame->linesize, |
| 270 | avctx->pix_fmt, avctx->width, avctx->height); |
| 271 | frame = new; |
| 272 | } |
| 273 | |
| 274 | task.index = c->task_index; |
| 275 | task.indata = (void*)frame; |
| 276 | pthread_mutex_lock(&c->task_fifo_mutex); |
| 277 | av_fifo_generic_write(c->task_fifo, &task, sizeof(task), NULL); |
| 278 | pthread_cond_signal(&c->task_fifo_cond); |
| 279 | pthread_mutex_unlock(&c->task_fifo_mutex); |
| 280 | |
| 281 | c->task_index = (c->task_index+1) % BUFFER_SIZE; |
| 282 | |
| 283 | if(!c->finished_tasks[c->finished_task_index].outdata && (c->task_index - c->finished_task_index) % BUFFER_SIZE <= avctx->thread_count) |
| 284 | return 0; |
| 285 | } |
| 286 | |
| 287 | if(c->task_index == c->finished_task_index) |
| 288 | return 0; |
| 289 | |
| 290 | pthread_mutex_lock(&c->finished_task_mutex); |
| 291 | while (!c->finished_tasks[c->finished_task_index].outdata) { |
| 292 | pthread_cond_wait(&c->finished_task_cond, &c->finished_task_mutex); |
| 293 | } |
| 294 | task = c->finished_tasks[c->finished_task_index]; |
| 295 | *pkt = *(AVPacket*)(task.outdata); |
| 296 | if(pkt->data) |
| 297 | *got_packet_ptr = 1; |
| 298 | av_freep(&c->finished_tasks[c->finished_task_index].outdata); |
| 299 | c->finished_task_index = (c->finished_task_index+1) % BUFFER_SIZE; |
| 300 | pthread_mutex_unlock(&c->finished_task_mutex); |
| 301 | |
| 302 | return task.return_code; |
| 303 | } |