Frame-based multithreading framework using pthreads

See doc/multithreading.txt for details on use in codecs.

Alexander Strange committed 15 years ago

Commit b38f0081 parent 8a278ad11 files · +1068 −10Browse files

Unified · Split

11 files

doc/APIchanges +5 −0
doc/multithreading.txt +65 −0
ffplay.c +1 −0
libavcodec/avcodec.h +87 −1
libavcodec/options.c +4 −0
libavcodec/pthread.c +718 −4
libavcodec/thread.h +111 −0
libavcodec/utils.c +54 −5
libavcodec/w32thread.c +6 −0
libavformat/utils.c +6 −0
libavutil/internal.h +11 −0

@@ -13,6 +13,11 @@ libavutil: 2009-03-08 · 12 unmodified lines
13	13
14	14	API changes, most recent first:
15	15
	16	+2011-02-09 - XXXXXXX - lavc 52.111.0 - threading API
	17	+ Add CODEC_CAP_FRAME_THREADS with new restrictions on get_buffer()/
	18	+ release_buffer()/draw_horiz_band() callbacks for appropriate codecs.
	19	+ Add thread_type and active_thread_type fields to AVCodecContext.
	20	+
16	21	2011-02-08 - 3940caa - lavf 52.98.0 - av_probe_input_buffer
17	22	Add av_probe_input_buffer() to avformat.h for probing format from a
18	23	ByteIOContext.

doc/multithreading.txt+65−0

@@ -0,0 +1,65 @@
	1	+FFmpeg multithreading methods
	2	+==============================================
	3	+
	4	+FFmpeg provides two methods for multithreading codecs.
	5	+
	6	+Slice threading decodes multiple parts of a frame at the same time, using
	7	+AVCodecContext execute() and execute2().
	8	+
	9	+Frame threading decodes multiple frames at the same time.
	10	+It accepts N future frames and delays decoded pictures by N-1 frames.
	11	+The later frames are decoded in separate threads while the user is
	12	+displaying the current one.
	13	+
	14	+Restrictions on clients
	15	+==============================================
	16	+
	17	+Slice threading -
	18	+* The client's draw_horiz_band() must be thread-safe according to the comment
	19	+ in avcodec.h.
	20	+
	21	+Frame threading -
	22	+* Restrictions with slice threading also apply.
	23	+* For best performance, the client should set thread_safe_callbacks if it
	24	+ provides a thread-safe get_buffer() callback.
	25	+* There is one frame of delay added for every thread beyond the first one.
	26	+ Clients must be able to handle this; the pkt_dts and pkt_pts fields in
	27	+ AVFrame will work as usual.
	28	+
	29	+Restrictions on codec implementations
	30	+==============================================
	31	+
	32	+Slice threading -
	33	+ None except that there must be something worth executing in parallel.
	34	+
	35	+Frame threading -
	36	+* Codecs can only accept entire pictures per packet.
	37	+* Codecs similar to ffv1, whose streams don't reset across frames,
	38	+ will not work because their bitstreams cannot be decoded in parallel.
	39	+
	40	+* The contents of buffers must not be read before ff_thread_await_progress()
	41	+ has been called on them. reget_buffer() and buffer age optimizations no longer work.
	42	+* The contents of buffers must not be written to after ff_thread_report_progress()
	43	+ has been called on them. This includes draw_edges().
	44	+
	45	+Porting codecs to frame threading
	46	+==============================================
	47	+
	48	+Find all context variables that are needed by the next frame. Move all
	49	+code changing them, as well as code calling get_buffer(), up to before
	50	+the decode process starts. Call ff_thread_finish_setup() afterwards. If
	51	+some code can't be moved, have update_thread_context() run it in the next
	52	+thread.
	53	+
	54	+If the codec allocates writable tables in its init(), add an init_thread_copy()
	55	+which re-allocates them for other threads.
	56	+
	57	+Add CODEC_CAP_FRAME_THREADS to the codec capabilities. There will be very little
	58	+speed gain at this point but it should work.
	59	+
	60	+Call ff_thread_report_progress() after some part of the current picture has decoded.
	61	+A good place to put this is where draw_horiz_band() is called - add this if it isn't
	62	+called anywhere, as it's useful too and the implementation is trivial when you're
	63	+doing this. Note that draw_edges() needs to be called before reporting progress.
	64	+
	65	+Before accessing a reference frame or its MVs, call ff_thread_await_progress().

ffplay.c+1−0

@@ -1691,6 +1691,7 @@ static int input_init(AVFilterContext ctx, const char args, void *opaque) · 1690 unmodified lines
1691	1691	codec->get_buffer = input_get_buffer;
1692	1692	codec->release_buffer = input_release_buffer;
1693	1693	codec->reget_buffer = input_reget_buffer;
	1694	+ codec->thread_safe_callbacks = 1;
1694	1695	}
1695	1696
1696	1697	priv->frame = avcodec_alloc_frame();

libavcodec/avcodec.h+87−1

@@ -32,7 +32,7 @@ · 31 unmodified lines
32	32	#include "libavutil/cpu.h"
33	33
34	34	#define LIBAVCODEC_VERSION_MAJOR 52
35		−#define LIBAVCODEC_VERSION_MINOR 110
	35	+#define LIBAVCODEC_VERSION_MINOR 111
36	36	#define LIBAVCODEC_VERSION_MICRO 0
37	37
38	38	#define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
@@ -728,6 +728,10 @@ typedef struct RcOverride{ · 689 unmodified lines
728	728	* Codec is able to deal with negative linesizes
729	729	*/
730	730	#define CODEC_CAP_NEG_LINESIZES 0x0800
	731	+/**
	732	+ * Codec supports frame-level multithreading.
	733	+ */
	734	+#define CODEC_CAP_FRAME_THREADS 0x1000
731	735
732	736	//The following defines may change, don't expect compatibility if you use them.
733	737	#define MB_TYPE_INTRA4x4 0x0001
@@ -1027,6 +1031,20 @@ typedef struct AVPanScan{ · 293 unmodified lines
1027	1031	* - decoding: Read by user.\
1028	1032	*/\
1029	1033	int64_t pkt_dts;\
	1034	+\
	1035	+ /**\
	1036	+ * the AVCodecContext which ff_thread_get_buffer() was last called on\
	1037	+ * - encoding: Set by libavcodec.\
	1038	+ * - decoding: Set by libavcodec.\
	1039	+ */\
	1040	+ struct AVCodecContext *owner;\
	1041	+\
	1042	+ /**\
	1043	+ * used by multithreading to store frame-specific info\
	1044	+ * - encoding: Set by libavcodec.\
	1045	+ * - decoding: Set by libavcodec.\
	1046	+ */\
	1047	+ void *thread_opaque;\
1030	1048	\
1031	1049	/**\
1032	1050	* frame timestamp estimated using various heuristics, in stream time base\
@@ -1246,6 +1264,10 @@ typedef struct AVCodecContext { · 213 unmodified lines
1246	1264	* decoder to draw a horizontal band. It improves cache usage. Not
1247	1265	* all codecs can do that. You must check the codec capabilities
1248	1266	* beforehand.
	1267	+ * When multithreading is used, it may be called from multiple threads
	1268	+ * at the same time; threads might draw different parts of the same AVFrame,
	1269	+ * or multiple AVFrames, and there is no guarantee that slices will be drawn
	1270	+ * in order.
1249	1271	* The function is also used by hardware acceleration APIs.
1250	1272	* It is called at least once during frame decoding to pass
1251	1273	* the data needed for hardware render.
@@ -1499,6 +1521,9 @@ typedef struct AVCodecContext { · 247 unmodified lines
1499	1521	* if CODEC_CAP_DR1 is not set then get_buffer() must call
1500	1522	* avcodec_default_get_buffer() instead of providing buffers allocated by
1501	1523	* some other means.
	1524	+ * If frame multithreading is used and thread_safe_callbacks is set,
	1525	+ * it may be called from a different thread, but not from more than one at once.
	1526	+ * Does not need to be reentrant.
1502	1527	* - encoding: unused
1503	1528	* - decoding: Set by libavcodec, user can override.
1504	1529	*/
@@ -1508,6 +1533,8 @@ typedef struct AVCodecContext { · 3 unmodified lines
1508	1533	* Called to release buffers which were allocated with get_buffer.
1509	1534	* A released buffer can be reused in get_buffer().
1510	1535	* pic.data[*] must be set to NULL.
	1536	+ * May be called from a different thread if frame multithreading is used,
	1537	+ * but not by more than one thread at once, so does not need to be reentrant.
1511	1538	* - encoding: unused
1512	1539	* - decoding: Set by libavcodec, user can override.
1513	1540	*/
@@ -1811,6 +1838,7 @@ typedef struct AVCodecContext { · 297 unmodified lines
1811	1838	#define FF_DEBUG_VIS_QP 0x00002000
1812	1839	#define FF_DEBUG_VIS_MB_TYPE 0x00004000
1813	1840	#define FF_DEBUG_BUFFERS 0x00008000
	1841	+#define FF_DEBUG_THREADS 0x00010000
1814	1842
1815	1843	/**
1816	1844	* debug
@@ -2835,6 +2863,44 @@ typedef struct AVCodecContext { · 1018 unmodified lines
2835	2863	*/
2836	2864	AVPacket *pkt;
2837	2865
	2866	+ /**
	2867	+ * Whether this is a copy of the context which had init() called on it.
	2868	+ * This is used by multithreading - shared tables and picture pointers
	2869	+ * should be freed from the original context only.
	2870	+ * - encoding: Set by libavcodec.
	2871	+ * - decoding: Set by libavcodec.
	2872	+ */
	2873	+ int is_copy;
	2874	+
	2875	+ /**
	2876	+ * Which multithreading methods to use.
	2877	+ * Use of FF_THREAD_FRAME will increase decoding delay by one frame per thread,
	2878	+ * so clients which cannot provide future frames should not use it.
	2879	+ *
	2880	+ * - encoding: Set by user, otherwise the default is used.
	2881	+ * - decoding: Set by user, otherwise the default is used.
	2882	+ */
	2883	+ int thread_type;
	2884	+#define FF_THREAD_FRAME 1 //< Decode more than one frame at once
	2885	+#define FF_THREAD_SLICE 2 //< Decode more than one part of a single frame at once
	2886	+
	2887	+ /**
	2888	+ * Which multithreading methods are in use by the codec.
	2889	+ * - encoding: Set by libavcodec.
	2890	+ * - decoding: Set by libavcodec.
	2891	+ */
	2892	+ int active_thread_type;
	2893	+
	2894	+ /**
	2895	+ * Set by the client if its custom get_buffer() callback can be called
	2896	+ * from another thread, which allows faster multithreaded decoding.
	2897	+ * draw_horiz_band() will be called from other threads regardless of this setting.
	2898	+ * Ignored if the default get_buffer() is used.
	2899	+ * - encoding: Set by user.
	2900	+ * - decoding: Set by user.
	2901	+ */
	2902	+ int thread_safe_callbacks;
	2903	+
2838	2904	/**
2839	2905	* Current statistics for PTS correction.
2840	2906	* - decoding: maintained and used by libavcodec
@@ -2897,6 +2963,26 @@ typedef struct AVCodec { · 56 unmodified lines
2897	2963	uint8_t max_lowres; ///< maximum value for lowres supported by the decoder
2898	2964	AVClass *priv_class; ///< AVClass for the private context
2899	2965	const AVProfile *profiles; ///< array of recognized profiles, or NULL if unknown, array is terminated by {FF_PROFILE_UNKNOWN}
	2966	+
	2967	+ /**
	2968	+ * @defgroup framethreading Frame-level threading support functions.
	2969	+ * @{
	2970	+ */
	2971	+ /**
	2972	+ * If defined, called on thread contexts when they are created.
	2973	+ * If the codec allocates writable tables in init(), re-allocate them here.
	2974	+ * priv_data will be set to a copy of the original.
	2975	+ */
	2976	+ int (init_thread_copy)(AVCodecContext );
	2977	+ /**
	2978	+ * Copy necessary context variables from a previous thread context to the current one.
	2979	+ * If not defined, the next thread will start automatically; otherwise, the codec
	2980	+ * must call ff_thread_finish_setup().
	2981	+ *
	2982	+ * dst and src will (rarely) point to the same context, in which case memcpy should be skipped.
	2983	+ */
	2984	+ int (update_thread_context)(AVCodecContext dst, const AVCodecContext *src);
	2985	+ /** @} */
2900	2986	} AVCodec;
2901	2987
2902	2988	/**

libavcodec/options.c+4−0

@@ -250,6 +250,7 @@ static const AVOption options[]={ · 249 unmodified lines
250	250	{"vis_qp", "visualize quantization parameter (QP), lower QP are tinted greener", 0, FF_OPT_TYPE_CONST, FF_DEBUG_VIS_QP, INT_MIN, INT_MAX, V\|D, "debug"},
251	251	{"vis_mb_type", "visualize block types", 0, FF_OPT_TYPE_CONST, FF_DEBUG_VIS_MB_TYPE, INT_MIN, INT_MAX, V\|D, "debug"},
252	252	{"buffers", "picture buffer allocations", 0, FF_OPT_TYPE_CONST, FF_DEBUG_BUFFERS, INT_MIN, INT_MAX, V\|D, "debug"},
	253	+{"thread_ops", "threading operations", 0, FF_OPT_TYPE_CONST, FF_DEBUG_THREADS, INT_MIN, INT_MAX, V\|D, "debug"},
253	254	{"vismv", "visualize motion vectors (MVs)", OFFSET(debug_mv), FF_OPT_TYPE_INT, DEFAULT, 0, INT_MAX, V\|D, "debug_mv"},
254	255	{"pf", "forward predicted MVs of P-frames", 0, FF_OPT_TYPE_CONST, FF_DEBUG_VIS_MV_P_FOR, INT_MIN, INT_MAX, V\|D, "debug_mv"},
255	256	{"bf", "forward predicted MVs of B-frames", 0, FF_OPT_TYPE_CONST, FF_DEBUG_VIS_MV_B_FOR, INT_MIN, INT_MAX, V\|D, "debug_mv"},
@@ -431,6 +432,9 @@ static const AVOption options[]={ · 175 unmodified lines
431	432	{"cholesky", NULL, 0, FF_OPT_TYPE_CONST, AV_LPC_TYPE_CHOLESKY, INT_MIN, INT_MAX, A\|E, "lpc_type"},
432	433	{"lpc_passes", "number of passes to use for Cholesky factorization during LPC analysis", OFFSET(lpc_passes), FF_OPT_TYPE_INT, -1, INT_MIN, INT_MAX, A\|E},
433	434	{"slices", "number of slices, used in parallelized decoding", OFFSET(slices), FF_OPT_TYPE_INT, 0, 0, INT_MAX, V\|E},
	435	+{"thread_type", "select multithreading type", OFFSET(thread_type), FF_OPT_TYPE_INT, FF_THREAD_SLICE\|FF_THREAD_FRAME, 0, INT_MAX, V\|E\|D, "thread_type"},
	436	+{"slice", NULL, 0, FF_OPT_TYPE_CONST, FF_THREAD_SLICE, INT_MIN, INT_MAX, V\|E\|D, "thread_type"},
	437	+{"frame", NULL, 0, FF_OPT_TYPE_CONST, FF_THREAD_FRAME, INT_MIN, INT_MAX, V\|E\|D, "thread_type"},
434	438	{NULL},
435	439	};
436	440

libavcodec/pthread.c+718−4

@@ -1,5 +1,6 @@
1	1	/*
2	2	* Copyright (c) 2004 Roman Shaposhnik
	3	+ * Copyright (c) 2008 Alexander Strange (astrange@ithinksw.com)
3	4	*
4	5	* Many thanks to Steven M. Schultz for providing clever ideas and
5	6	* to Michael Niedermayer <michaelni@gmx.at> for writing initial
@@ -21,9 +22,17 @@ · 15 unmodified lines
21	22	* License along with FFmpeg; if not, write to the Free Software
22	23	* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23	24	*/
	25	+
	26	+/**
	27	+ * @file
	28	+ * Multithreading support functions
	29	+ * @see doc/multithreading.txt
	30	+ */
	31	+
24	32	#include <pthread.h>
25	33
26	34	#include "avcodec.h"
	35	+#include "thread.h"
27	36
28	37	typedef int (action_func)(AVCodecContext c, void arg);
29	38	typedef int (action_func2)(AVCodecContext c, void arg, int jobnr, int threadnr);
@@ -45,6 +54,78 @@ typedef struct ThreadContext { · 15 unmodified lines
45	54	int done;
46	55	} ThreadContext;
47	56
	57	+/// Max number of frame buffers that can be allocated when using frame threads.
	58	+#define MAX_BUFFERS 32
	59	+
	60	+/**
	61	+ * Context used by codec threads and stored in their AVCodecContext thread_opaque.
	62	+ */
	63	+typedef struct PerThreadContext {
	64	+ struct FrameThreadContext *parent;
	65	+
	66	+ pthread_t thread;
	67	+ pthread_cond_t input_cond; ///< Used to wait for a new packet from the main thread.
	68	+ pthread_cond_t progress_cond; ///< Used by child threads to wait for progress to change.
	69	+ pthread_cond_t output_cond; ///< Used by the main thread to wait for frames to finish.
	70	+
	71	+ pthread_mutex_t mutex; ///< Mutex used to protect the contents of the PerThreadContext.
	72	+ pthread_mutex_t progress_mutex; ///< Mutex used to protect frame progress values and progress_cond.
	73	+
	74	+ AVCodecContext *avctx; ///< Context used to decode packets passed to this thread.
	75	+
	76	+ AVPacket avpkt; ///< Input packet (for decoding) or output (for encoding).
	77	+ int allocated_buf_size; ///< Size allocated for avpkt.data
	78	+
	79	+ AVFrame frame; ///< Output frame (for decoding) or input (for encoding).
	80	+ int got_frame; ///< The output of got_picture_ptr from the last avcodec_decode_video() call.
	81	+ int result; ///< The result of the last codec decode/encode() call.
	82	+
	83	+ enum {
	84	+ STATE_INPUT_READY, ///< Set when the thread is awaiting a packet.
	85	+ STATE_SETTING_UP, ///< Set before the codec has called ff_thread_finish_setup().
	86	+ STATE_GET_BUFFER, /**<
	87	+ * Set when the codec calls get_buffer().
	88	+ * State is returned to STATE_SETTING_UP afterwards.
	89	+ */
	90	+ STATE_SETUP_FINISHED ///< Set after the codec has called ff_thread_finish_setup().
	91	+ } state;
	92	+
	93	+ /**
	94	+ * Array of frames passed to ff_thread_release_buffer().
	95	+ * Frames are released after all threads referencing them are finished.
	96	+ */
	97	+ AVFrame released_buffers[MAX_BUFFERS];
	98	+ int num_released_buffers;
	99	+
	100	+ /**
	101	+ * Array of progress values used by ff_thread_get_buffer().
	102	+ */
	103	+ int progress[MAX_BUFFERS][2];
	104	+ uint8_t progress_used[MAX_BUFFERS];
	105	+
	106	+ AVFrame *requested_frame; ///< AVFrame the codec passed to get_buffer()
	107	+} PerThreadContext;
	108	+
	109	+/**
	110	+ * Context stored in the client AVCodecContext thread_opaque.
	111	+ */
	112	+typedef struct FrameThreadContext {
	113	+ PerThreadContext *threads; ///< The contexts for each thread.
	114	+ PerThreadContext *prev_thread; ///< The last thread submit_packet() was called on.
	115	+
	116	+ pthread_mutex_t buffer_mutex; ///< Mutex used to protect get/release_buffer().
	117	+
	118	+ int next_decoding; ///< The next context to submit a packet to.
	119	+ int next_finished; ///< The next context to return output from.
	120	+
	121	+ int delaying; /**<
	122	+ * Set for the first N packets, where N is the number of threads.
	123	+ * While it is set, ff_thread_en/decode_frame won't return any results.
	124	+ */
	125	+
	126	+ int die; ///< Set when threads should exit.
	127	+} FrameThreadContext;
	128	+
48	129	static void* attribute_align_arg worker(void *v)
49	130	{
50	131	AVCodecContext *avctx = v;
@@ -84,7 +165,7 @@ static av_always_inline void avcodec_thread_park_workers(ThreadContext *c, int t · 33 unmodified lines
84	165	pthread_mutex_unlock(&c->current_job_lock);
85	166	}
86	167
87		−void avcodec_thread_free(AVCodecContext *avctx)
	168	+static void thread_free(AVCodecContext *avctx)
88	169	{
89	170	ThreadContext *c = avctx->thread_opaque;
90	171	int i;
@@ -109,6 +190,9 @@ static int avcodec_thread_execute(AVCodecContext avctx, action_func func, void · 18 unmodified lines
109	190	ThreadContext *c= avctx->thread_opaque;
110	191	int dummy_ret;
111	192
	193	+ if (!(avctx->active_thread_type&FF_THREAD_SLICE) \|\| avctx->thread_count <= 1)
	194	+ return avcodec_default_execute(avctx, func, arg, ret, job_count, job_size);
	195	+
112	196	if (job_count <= 0)
113	197	return 0;
114	198
@@ -140,12 +224,11 @@ static int avcodec_thread_execute2(AVCodecContext avctx, action_func2 func2, v · 25 unmodified lines
140	224	return avcodec_thread_execute(avctx, NULL, arg, ret, job_count, 0);
141	225	}
142	226
143		−int avcodec_thread_init(AVCodecContext *avctx, int thread_count)
	227	+static int thread_init(AVCodecContext *avctx)
144	228	{
145	229	int i;
146	230	ThreadContext *c;
147		−
148		− avctx->thread_count = thread_count;
	231	+ int thread_count = avctx->thread_count;
149	232
150	233	if (thread_count <= 1)
151	234	return 0;
@@ -184,3 +267,634 @@ int avcodec_thread_init(AVCodecContext *avctx, int thread_count) · 32 unmodified lines
184	267	avctx->execute2 = avcodec_thread_execute2;
185	268	return 0;
186	269	}
	270	+
	271	+/**
	272	+ * Codec worker thread.
	273	+ *
	274	+ * Automatically calls ff_thread_finish_setup() if the codec does
	275	+ * not provide an update_thread_context method, or if the codec returns
	276	+ * before calling it.
	277	+ */
	278	+static attribute_align_arg void frame_worker_thread(void arg)
	279	+{
	280	+ PerThreadContext *p = arg;
	281	+ FrameThreadContext *fctx = p->parent;
	282	+ AVCodecContext *avctx = p->avctx;
	283	+ AVCodec *codec = avctx->codec;
	284	+
	285	+ while (1) {
	286	+ if (p->state == STATE_INPUT_READY && !fctx->die) {
	287	+ pthread_mutex_lock(&p->mutex);
	288	+ while (p->state == STATE_INPUT_READY && !fctx->die)
	289	+ pthread_cond_wait(&p->input_cond, &p->mutex);
	290	+ pthread_mutex_unlock(&p->mutex);
	291	+ }
	292	+
	293	+ if (fctx->die) break;
	294	+
	295	+ if (!codec->update_thread_context) ff_thread_finish_setup(avctx);
	296	+
	297	+ pthread_mutex_lock(&p->mutex);
	298	+ avcodec_get_frame_defaults(&p->frame);
	299	+ p->got_frame = 0;
	300	+ p->result = codec->decode(avctx, &p->frame, &p->got_frame, &p->avpkt);
	301	+
	302	+ if (p->state == STATE_SETTING_UP) ff_thread_finish_setup(avctx);
	303	+
	304	+ p->state = STATE_INPUT_READY;
	305	+
	306	+ pthread_mutex_lock(&p->progress_mutex);
	307	+ pthread_cond_signal(&p->output_cond);
	308	+ pthread_mutex_unlock(&p->progress_mutex);
	309	+
	310	+ pthread_mutex_unlock(&p->mutex);
	311	+ }
	312	+
	313	+ return NULL;
	314	+}
	315	+
	316	+/**
	317	+ * Updates the next thread's AVCodecContext with values from the reference thread's context.
	318	+ *
	319	+ * @param dst The destination context.
	320	+ * @param src The source context.
	321	+ * @param for_user 0 if the destination is a codec thread, 1 if the destination is the user's thread
	322	+ */
	323	+static int update_context_from_thread(AVCodecContext dst, AVCodecContext src, int for_user)
	324	+{
	325	+ int err = 0;
	326	+
	327	+ if (dst != src) {
	328	+ dst->sub_id = src->sub_id;
	329	+ dst->time_base = src->time_base;
	330	+ dst->width = src->width;
	331	+ dst->height = src->height;
	332	+ dst->pix_fmt = src->pix_fmt;
	333	+
	334	+ dst->has_b_frames = src->has_b_frames;
	335	+ dst->idct_algo = src->idct_algo;
	336	+ dst->slice_count = src->slice_count;
	337	+
	338	+ dst->bits_per_coded_sample = src->bits_per_coded_sample;
	339	+ dst->sample_aspect_ratio = src->sample_aspect_ratio;
	340	+ dst->dtg_active_format = src->dtg_active_format;
	341	+
	342	+ dst->profile = src->profile;
	343	+ dst->level = src->level;
	344	+
	345	+ dst->bits_per_raw_sample = src->bits_per_raw_sample;
	346	+ dst->ticks_per_frame = src->ticks_per_frame;
	347	+ dst->color_primaries = src->color_primaries;
	348	+
	349	+ dst->color_trc = src->color_trc;
	350	+ dst->colorspace = src->colorspace;
	351	+ dst->color_range = src->color_range;
	352	+ dst->chroma_sample_location = src->chroma_sample_location;
	353	+ }
	354	+
	355	+ if (for_user) {
	356	+ dst->coded_frame = src->coded_frame;
	357	+ dst->has_b_frames += src->thread_count - 1;
	358	+ } else {
	359	+ if (dst->codec->update_thread_context)
	360	+ err = dst->codec->update_thread_context(dst, src);
	361	+ }
	362	+
	363	+ return err;
	364	+}
	365	+
	366	+/**
	367	+ * Update the next thread's AVCodecContext with values set by the user.
	368	+ *
	369	+ * @param dst The destination context.
	370	+ * @param src The source context.
	371	+ */
	372	+static void update_context_from_user(AVCodecContext dst, AVCodecContext src)
	373	+{
	374	+#define copy_fields(s, e) memcpy(&dst->s, &src->s, (char)&dst->e - (char)&dst->s);
	375	+ dst->flags = src->flags;
	376	+
	377	+ dst->draw_horiz_band= src->draw_horiz_band;
	378	+ dst->get_buffer = src->get_buffer;
	379	+ dst->release_buffer = src->release_buffer;
	380	+
	381	+ dst->opaque = src->opaque;
	382	+ dst->hurry_up = src->hurry_up;
	383	+ dst->dsp_mask = src->dsp_mask;
	384	+ dst->debug = src->debug;
	385	+ dst->debug_mv = src->debug_mv;
	386	+
	387	+ dst->slice_flags = src->slice_flags;
	388	+ dst->flags2 = src->flags2;
	389	+
	390	+ copy_fields(skip_loop_filter, bidir_refine);
	391	+
	392	+ dst->frame_number = src->frame_number;
	393	+ dst->reordered_opaque = src->reordered_opaque;
	394	+#undef copy_fields
	395	+}
	396	+
	397	+static void free_progress(AVFrame *f)
	398	+{
	399	+ PerThreadContext *p = f->owner->thread_opaque;
	400	+ int *progress = f->thread_opaque;
	401	+
	402	+ p->progress_used[(progress - p->progress[0]) / 2] = 0;
	403	+}
	404	+
	405	+/// Releases the buffers that this decoding thread was the last user of.
	406	+static void release_delayed_buffers(PerThreadContext *p)
	407	+{
	408	+ FrameThreadContext *fctx = p->parent;
	409	+
	410	+ while (p->num_released_buffers > 0) {
	411	+ AVFrame *f = &p->released_buffers[--p->num_released_buffers];
	412	+
	413	+ pthread_mutex_lock(&fctx->buffer_mutex);
	414	+ free_progress(f);
	415	+ f->thread_opaque = NULL;
	416	+
	417	+ f->owner->release_buffer(f->owner, f);
	418	+ pthread_mutex_unlock(&fctx->buffer_mutex);
	419	+ }
	420	+}
	421	+
	422	+static int submit_packet(PerThreadContext p, AVPacket avpkt)
	423	+{
	424	+ FrameThreadContext *fctx = p->parent;
	425	+ PerThreadContext *prev_thread = fctx->prev_thread;
	426	+ AVCodec *codec = p->avctx->codec;
	427	+ uint8_t *buf = p->avpkt.data;
	428	+
	429	+ if (!avpkt->size && !(codec->capabilities & CODEC_CAP_DELAY)) return 0;
	430	+
	431	+ pthread_mutex_lock(&p->mutex);
	432	+
	433	+ release_delayed_buffers(p);
	434	+
	435	+ if (prev_thread) {
	436	+ int err;
	437	+ if (prev_thread->state == STATE_SETTING_UP) {
	438	+ pthread_mutex_lock(&prev_thread->progress_mutex);
	439	+ while (prev_thread->state == STATE_SETTING_UP)
	440	+ pthread_cond_wait(&prev_thread->progress_cond, &prev_thread->progress_mutex);
	441	+ pthread_mutex_unlock(&prev_thread->progress_mutex);
	442	+ }
	443	+
	444	+ err = update_context_from_thread(p->avctx, prev_thread->avctx, 0);
	445	+ if (err) {
	446	+ pthread_mutex_unlock(&p->mutex);
	447	+ return err;
	448	+ }
	449	+ }
	450	+
	451	+ av_fast_malloc(&buf, &p->allocated_buf_size, avpkt->size + FF_INPUT_BUFFER_PADDING_SIZE);
	452	+ p->avpkt = *avpkt;
	453	+ p->avpkt.data = buf;
	454	+ memcpy(buf, avpkt->data, avpkt->size);
	455	+ memset(buf + avpkt->size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
	456	+
	457	+ p->state = STATE_SETTING_UP;
	458	+ pthread_cond_signal(&p->input_cond);
	459	+ pthread_mutex_unlock(&p->mutex);
	460	+
	461	+ /*
	462	+ * If the client doesn't have a thread-safe get_buffer(),
	463	+ * then decoding threads call back to the main thread,
	464	+ * and it calls back to the client here.
	465	+ */
	466	+
	467	+ if (!p->avctx->thread_safe_callbacks &&
	468	+ p->avctx->get_buffer != avcodec_default_get_buffer) {
	469	+ while (p->state != STATE_SETUP_FINISHED && p->state != STATE_INPUT_READY) {
	470	+ pthread_mutex_lock(&p->progress_mutex);
	471	+ while (p->state == STATE_SETTING_UP)
	472	+ pthread_cond_wait(&p->progress_cond, &p->progress_mutex);
	473	+
	474	+ if (p->state == STATE_GET_BUFFER) {
	475	+ p->result = p->avctx->get_buffer(p->avctx, p->requested_frame);
	476	+ p->state = STATE_SETTING_UP;
	477	+ pthread_cond_signal(&p->progress_cond);
	478	+ }
	479	+ pthread_mutex_unlock(&p->progress_mutex);
	480	+ }
	481	+ }
	482	+
	483	+ fctx->prev_thread = p;
	484	+
	485	+ return 0;
	486	+}
	487	+
	488	+int ff_thread_decode_frame(AVCodecContext *avctx,
	489	+ AVFrame picture, int got_picture_ptr,
	490	+ AVPacket *avpkt)
	491	+{
	492	+ FrameThreadContext *fctx = avctx->thread_opaque;
	493	+ int finished = fctx->next_finished;
	494	+ PerThreadContext *p;
	495	+ int err;
	496	+
	497	+ /*
	498	+ * Submit a packet to the next decoding thread.
	499	+ */
	500	+
	501	+ p = &fctx->threads[fctx->next_decoding];
	502	+ update_context_from_user(p->avctx, avctx);
	503	+ err = submit_packet(p, avpkt);
	504	+ if (err) return err;
	505	+
	506	+ fctx->next_decoding++;
	507	+
	508	+ /*
	509	+ * If we're still receiving the initial packets, don't return a frame.
	510	+ */
	511	+
	512	+ if (fctx->delaying && avpkt->size) {
	513	+ if (fctx->next_decoding >= (avctx->thread_count-1)) fctx->delaying = 0;
	514	+
	515	+ *got_picture_ptr=0;
	516	+ return 0;
	517	+ }
	518	+
	519	+ /*
	520	+ * Return the next available frame from the oldest thread.
	521	+ * If we're at the end of the stream, then we have to skip threads that
	522	+ * didn't output a frame, because we don't want to accidentally signal
	523	+ * EOF (avpkt->size == 0 && *got_picture_ptr == 0).
	524	+ */
	525	+
	526	+ do {
	527	+ p = &fctx->threads[finished++];
	528	+
	529	+ if (p->state != STATE_INPUT_READY) {
	530	+ pthread_mutex_lock(&p->progress_mutex);
	531	+ while (p->state != STATE_INPUT_READY)
	532	+ pthread_cond_wait(&p->output_cond, &p->progress_mutex);
	533	+ pthread_mutex_unlock(&p->progress_mutex);
	534	+ }
	535	+
	536	+ *picture = p->frame;
	537	+ *got_picture_ptr = p->got_frame;
	538	+ picture->pkt_dts = p->avpkt.dts;
	539	+
	540	+ /*
	541	+ * A later call with avkpt->size == 0 may loop over all threads,
	542	+ * including this one, searching for a frame to return before being
	543	+ * stopped by the "finished != fctx->next_finished" condition.
	544	+ * Make sure we don't mistakenly return the same frame again.
	545	+ */
	546	+ p->got_frame = 0;
	547	+
	548	+ if (finished >= avctx->thread_count) finished = 0;
	549	+ } while (!avpkt->size && !*got_picture_ptr && finished != fctx->next_finished);
	550	+
	551	+ update_context_from_thread(avctx, p->avctx, 1);
	552	+
	553	+ if (fctx->next_decoding >= avctx->thread_count) fctx->next_decoding = 0;
	554	+
	555	+ fctx->next_finished = finished;
	556	+
	557	+ return p->result;
	558	+}
	559	+
	560	+void ff_thread_report_progress(AVFrame *f, int n, int field)
	561	+{
	562	+ PerThreadContext *p;
	563	+ int *progress = f->thread_opaque;
	564	+
	565	+ if (!progress \|\| progress[field] >= n) return;
	566	+
	567	+ p = f->owner->thread_opaque;
	568	+
	569	+ if (f->owner->debug&FF_DEBUG_THREADS)
	570	+ av_log(f->owner, AV_LOG_DEBUG, "%p finished %d field %d\n", progress, n, field);
	571	+
	572	+ pthread_mutex_lock(&p->progress_mutex);
	573	+ progress[field] = n;
	574	+ pthread_cond_broadcast(&p->progress_cond);
	575	+ pthread_mutex_unlock(&p->progress_mutex);
	576	+}
	577	+
	578	+void ff_thread_await_progress(AVFrame *f, int n, int field)
	579	+{
	580	+ PerThreadContext *p;
	581	+ int *progress = f->thread_opaque;
	582	+
	583	+ if (!progress \|\| progress[field] >= n) return;
	584	+
	585	+ p = f->owner->thread_opaque;
	586	+
	587	+ if (f->owner->debug&FF_DEBUG_THREADS)
	588	+ av_log(f->owner, AV_LOG_DEBUG, "thread awaiting %d field %d from %p\n", n, field, progress);
	589	+
	590	+ pthread_mutex_lock(&p->progress_mutex);
	591	+ while (progress[field] < n)
	592	+ pthread_cond_wait(&p->progress_cond, &p->progress_mutex);
	593	+ pthread_mutex_unlock(&p->progress_mutex);
	594	+}
	595	+
	596	+void ff_thread_finish_setup(AVCodecContext *avctx) {
	597	+ PerThreadContext *p = avctx->thread_opaque;
	598	+
	599	+ if (!(avctx->active_thread_type&FF_THREAD_FRAME)) return;
	600	+
	601	+ pthread_mutex_lock(&p->progress_mutex);
	602	+ p->state = STATE_SETUP_FINISHED;
	603	+ pthread_cond_broadcast(&p->progress_cond);
	604	+ pthread_mutex_unlock(&p->progress_mutex);
	605	+}
	606	+
	607	+/// Waits for all threads to finish.
	608	+static void park_frame_worker_threads(FrameThreadContext *fctx, int thread_count)
	609	+{
	610	+ int i;
	611	+
	612	+ for (i = 0; i < thread_count; i++) {
	613	+ PerThreadContext *p = &fctx->threads[i];
	614	+
	615	+ if (p->state != STATE_INPUT_READY) {
	616	+ pthread_mutex_lock(&p->progress_mutex);
	617	+ while (p->state != STATE_INPUT_READY)
	618	+ pthread_cond_wait(&p->output_cond, &p->progress_mutex);
	619	+ pthread_mutex_unlock(&p->progress_mutex);
	620	+ }
	621	+ }
	622	+}
	623	+
	624	+static void frame_thread_free(AVCodecContext *avctx, int thread_count)
	625	+{
	626	+ FrameThreadContext *fctx = avctx->thread_opaque;
	627	+ AVCodec *codec = avctx->codec;
	628	+ int i;
	629	+
	630	+ park_frame_worker_threads(fctx, thread_count);
	631	+
	632	+ if (fctx->prev_thread)
	633	+ update_context_from_thread(fctx->threads->avctx, fctx->prev_thread->avctx, 0);
	634	+
	635	+ fctx->die = 1;
	636	+
	637	+ for (i = 0; i < thread_count; i++) {
	638	+ PerThreadContext *p = &fctx->threads[i];
	639	+
	640	+ pthread_mutex_lock(&p->mutex);
	641	+ pthread_cond_signal(&p->input_cond);
	642	+ pthread_mutex_unlock(&p->mutex);
	643	+
	644	+ pthread_join(p->thread, NULL);
	645	+
	646	+ if (codec->close)
	647	+ codec->close(p->avctx);
	648	+
	649	+ avctx->codec = NULL;
	650	+
	651	+ release_delayed_buffers(p);
	652	+ }
	653	+
	654	+ for (i = 0; i < thread_count; i++) {
	655	+ PerThreadContext *p = &fctx->threads[i];
	656	+
	657	+ avcodec_default_free_buffers(p->avctx);
	658	+
	659	+ pthread_mutex_destroy(&p->mutex);
	660	+ pthread_mutex_destroy(&p->progress_mutex);
	661	+ pthread_cond_destroy(&p->input_cond);
	662	+ pthread_cond_destroy(&p->progress_cond);
	663	+ pthread_cond_destroy(&p->output_cond);
	664	+ av_freep(&p->avpkt.data);
	665	+
	666	+ if (i)
	667	+ av_freep(&p->avctx->priv_data);
	668	+
	669	+ av_freep(&p->avctx);
	670	+ }
	671	+
	672	+ av_freep(&fctx->threads);
	673	+ pthread_mutex_destroy(&fctx->buffer_mutex);
	674	+ av_freep(&avctx->thread_opaque);
	675	+}
	676	+
	677	+static int frame_thread_init(AVCodecContext *avctx)
	678	+{
	679	+ int thread_count = avctx->thread_count;
	680	+ AVCodec *codec = avctx->codec;
	681	+ AVCodecContext *src = avctx;
	682	+ FrameThreadContext *fctx;
	683	+ int i, err = 0;
	684	+
	685	+ avctx->thread_opaque = fctx = av_mallocz(sizeof(FrameThreadContext));
	686	+
	687	+ fctx->threads = av_mallocz(sizeof(PerThreadContext) * thread_count);
	688	+ pthread_mutex_init(&fctx->buffer_mutex, NULL);
	689	+ fctx->delaying = 1;
	690	+
	691	+ for (i = 0; i < thread_count; i++) {
	692	+ AVCodecContext *copy = av_malloc(sizeof(AVCodecContext));
	693	+ PerThreadContext *p = &fctx->threads[i];
	694	+
	695	+ pthread_mutex_init(&p->mutex, NULL);
	696	+ pthread_mutex_init(&p->progress_mutex, NULL);
	697	+ pthread_cond_init(&p->input_cond, NULL);
	698	+ pthread_cond_init(&p->progress_cond, NULL);
	699	+ pthread_cond_init(&p->output_cond, NULL);
	700	+
	701	+ p->parent = fctx;
	702	+ p->avctx = copy;
	703	+
	704	+ copy = src;
	705	+ copy->thread_opaque = p;
	706	+ copy->pkt = &p->avpkt;
	707	+
	708	+ if (!i) {
	709	+ src = copy;
	710	+
	711	+ if (codec->init)
	712	+ err = codec->init(copy);
	713	+
	714	+ update_context_from_thread(avctx, copy, 1);
	715	+ } else {
	716	+ copy->is_copy = 1;
	717	+ copy->priv_data = av_malloc(codec->priv_data_size);
	718	+ memcpy(copy->priv_data, src->priv_data, codec->priv_data_size);
	719	+
	720	+ if (codec->init_thread_copy)
	721	+ err = codec->init_thread_copy(copy);
	722	+ }
	723	+
	724	+ if (err) goto error;
	725	+
	726	+ pthread_create(&p->thread, NULL, frame_worker_thread, p);
	727	+ }
	728	+
	729	+ return 0;
	730	+
	731	+error:
	732	+ frame_thread_free(avctx, i+1);
	733	+
	734	+ return err;
	735	+}
	736	+
	737	+void ff_thread_flush(AVCodecContext *avctx)
	738	+{
	739	+ FrameThreadContext *fctx = avctx->thread_opaque;
	740	+
	741	+ if (!avctx->thread_opaque) return;
	742	+
	743	+ park_frame_worker_threads(fctx, avctx->thread_count);
	744	+
	745	+ if (fctx->prev_thread)
	746	+ update_context_from_thread(fctx->threads->avctx, fctx->prev_thread->avctx, 0);
	747	+
	748	+ fctx->next_decoding = fctx->next_finished = 0;
	749	+ fctx->delaying = 1;
	750	+ fctx->prev_thread = NULL;
	751	+}
	752	+
	753	+static int allocate_progress(PerThreadContext p)
	754	+{
	755	+ int i;
	756	+
	757	+ for (i = 0; i < MAX_BUFFERS; i++)
	758	+ if (!p->progress_used[i]) break;
	759	+
	760	+ if (i == MAX_BUFFERS) {
	761	+ av_log(p->avctx, AV_LOG_ERROR, "allocate_progress() overflow\n");
	762	+ return NULL;
	763	+ }
	764	+
	765	+ p->progress_used[i] = 1;
	766	+
	767	+ return p->progress[i];
	768	+}
	769	+
	770	+int ff_thread_get_buffer(AVCodecContext avctx, AVFrame f)
	771	+{
	772	+ PerThreadContext *p = avctx->thread_opaque;
	773	+ int *progress, err;
	774	+
	775	+ f->owner = avctx;
	776	+
	777	+ if (!(avctx->active_thread_type&FF_THREAD_FRAME)) {
	778	+ f->thread_opaque = NULL;
	779	+ return avctx->get_buffer(avctx, f);
	780	+ }
	781	+
	782	+ if (p->state != STATE_SETTING_UP) {
	783	+ av_log(avctx, AV_LOG_ERROR, "get_buffer() cannot be called after ff_thread_finish_setup()\n");
	784	+ return -1;
	785	+ }
	786	+
	787	+ pthread_mutex_lock(&p->parent->buffer_mutex);
	788	+ f->thread_opaque = progress = allocate_progress(p);
	789	+
	790	+ if (!progress) {
	791	+ pthread_mutex_unlock(&p->parent->buffer_mutex);
	792	+ return -1;
	793	+ }
	794	+
	795	+ progress[0] =
	796	+ progress[1] = -1;
	797	+
	798	+ if (avctx->thread_safe_callbacks \|\|
	799	+ avctx->get_buffer == avcodec_default_get_buffer) {
	800	+ err = avctx->get_buffer(avctx, f);
	801	+ } else {
	802	+ p->requested_frame = f;
	803	+ p->state = STATE_GET_BUFFER;
	804	+ pthread_mutex_lock(&p->progress_mutex);
	805	+ pthread_cond_signal(&p->progress_cond);
	806	+
	807	+ while (p->state != STATE_SETTING_UP)
	808	+ pthread_cond_wait(&p->progress_cond, &p->progress_mutex);
	809	+
	810	+ err = p->result;
	811	+
	812	+ pthread_mutex_unlock(&p->progress_mutex);
	813	+ }
	814	+
	815	+ pthread_mutex_unlock(&p->parent->buffer_mutex);
	816	+
	817	+ /*
	818	+ * Buffer age is difficult to keep track of between
	819	+ * multiple threads, and the optimizations it allows
	820	+ * are not worth the effort. It is disabled for now.
	821	+ */
	822	+ f->age = INT_MAX;
	823	+
	824	+ return err;
	825	+}
	826	+
	827	+void ff_thread_release_buffer(AVCodecContext avctx, AVFrame f)
	828	+{
	829	+ PerThreadContext *p = avctx->thread_opaque;
	830	+
	831	+ if (!(avctx->active_thread_type&FF_THREAD_FRAME)) {
	832	+ avctx->release_buffer(avctx, f);
	833	+ return;
	834	+ }
	835	+
	836	+ if (p->num_released_buffers >= MAX_BUFFERS) {
	837	+ av_log(p->avctx, AV_LOG_ERROR, "too many thread_release_buffer calls!\n");
	838	+ return;
	839	+ }
	840	+
	841	+ if(avctx->debug & FF_DEBUG_BUFFERS)
	842	+ av_log(avctx, AV_LOG_DEBUG, "thread_release_buffer called on pic %p, %d buffers used\n",
	843	+ f, f->owner->internal_buffer_count);
	844	+
	845	+ p->released_buffers[p->num_released_buffers++] = *f;
	846	+ memset(f->data, 0, sizeof(f->data));
	847	+}
	848	+
	849	+/**
	850	+ * Set the threading algorithms used.
	851	+ *
	852	+ * Threading requires more than one thread.
	853	+ * Frame threading requires entire frames to be passed to the codec,
	854	+ * and introduces extra decoding delay, so is incompatible with low_delay.
	855	+ *
	856	+ * @param avctx The context.
	857	+ */
	858	+static void validate_thread_parameters(AVCodecContext *avctx)
	859	+{
	860	+ int frame_threading_supported = (avctx->codec->capabilities & CODEC_CAP_FRAME_THREADS)
	861	+ && !(avctx->flags & CODEC_FLAG_TRUNCATED)
	862	+ && !(avctx->flags & CODEC_FLAG_LOW_DELAY)
	863	+ && !(avctx->flags2 & CODEC_FLAG2_CHUNKS);
	864	+ if (avctx->thread_count == 1) {
	865	+ avctx->active_thread_type = 0;
	866	+ } else if (frame_threading_supported && (avctx->thread_type & FF_THREAD_FRAME)) {
	867	+ avctx->active_thread_type = FF_THREAD_FRAME;
	868	+ } else {
	869	+ avctx->active_thread_type = FF_THREAD_SLICE;
	870	+ }
	871	+}
	872	+
	873	+int avcodec_thread_init(AVCodecContext *avctx, int thread_count)
	874	+{
	875	+ if (avctx->thread_opaque) {
	876	+ av_log(avctx, AV_LOG_ERROR, "avcodec_thread_init is ignored after avcodec_open\n");
	877	+ return -1;
	878	+ }
	879	+
	880	+ avctx->thread_count = FFMAX(1, thread_count);
	881	+
	882	+ if (avctx->codec) {
	883	+ validate_thread_parameters(avctx);
	884	+
	885	+ if (avctx->active_thread_type&FF_THREAD_SLICE)
	886	+ return thread_init(avctx);
	887	+ else if (avctx->active_thread_type&FF_THREAD_FRAME)
	888	+ return frame_thread_init(avctx);
	889	+ }
	890	+
	891	+ return 0;
	892	+}
	893	+
	894	+void avcodec_thread_free(AVCodecContext *avctx)
	895	+{
	896	+ if (avctx->active_thread_type&FF_THREAD_FRAME)
	897	+ frame_thread_free(avctx, avctx->thread_count);
	898	+ else
	899	+ thread_free(avctx);
	900	+}

libavcodec/thread.h+111−0

@@ -0,0 +1,111 @@
	1	+/*
	2	+ * Copyright (c) 2008 Alexander Strange <astrange@ithinksw.com>
	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	+ * Multithreading support functions
	24	+ * @author Alexander Strange <astrange@ithinksw.com>
	25	+ */
	26	+
	27	+#ifndef AVCODEC_THREAD_H
	28	+#define AVCODEC_THREAD_H
	29	+
	30	+#include "config.h"
	31	+#include "avcodec.h"
	32	+
	33	+/**
	34	+ * Waits for decoding threads to finish and resets internal
	35	+ * state. Called by avcodec_flush_buffers().
	36	+ *
	37	+ * @param avctx The context.
	38	+ */
	39	+void ff_thread_flush(AVCodecContext *avctx);
	40	+
	41	+/**
	42	+ * Submits a new frame to a decoding thread.
	43	+ * Returns the next available frame in picture. *got_picture_ptr
	44	+ * will be 0 if none is available.
	45	+ *
	46	+ * Parameters are the same as avcodec_decode_video2().
	47	+ */
	48	+int ff_thread_decode_frame(AVCodecContext avctx, AVFrame picture,
	49	+ int got_picture_ptr, AVPacket avpkt);
	50	+
	51	+/**
	52	+ * If the codec defines update_thread_context(), call this
	53	+ * when they are ready for the next thread to start decoding
	54	+ * the next frame. After calling it, do not change any variables
	55	+ * read by the update_thread_context() method, or call ff_thread_get_buffer().
	56	+ *
	57	+ * @param avctx The context.
	58	+ */
	59	+void ff_thread_finish_setup(AVCodecContext *avctx);
	60	+
	61	+/**
	62	+ * Notifies later decoding threads when part of their reference picture
	63	+ * is ready.
	64	+ * Call this when some part of the picture is finished decoding.
	65	+ * Later calls with lower values of progress have no effect.
	66	+ *
	67	+ * @param f The picture being decoded.
	68	+ * @param progress Value, in arbitrary units, of how much of the picture has decoded.
	69	+ * @param field The field being decoded, for field-picture codecs.
	70	+ * 0 for top field or frame pictures, 1 for bottom field.
	71	+ */
	72	+void ff_thread_report_progress(AVFrame *f, int progress, int field);
	73	+
	74	+/**
	75	+ * Waits for earlier decoding threads to finish reference pictures
	76	+ * Call this before accessing some part of a picture, with a given
	77	+ * value for progress, and it will return after the responsible decoding
	78	+ * thread calls ff_thread_report_progress() with the same or
	79	+ * higher value for progress.
	80	+ *
	81	+ * @param f The picture being referenced.
	82	+ * @param progress Value, in arbitrary units, to wait for.
	83	+ * @param field The field being referenced, for field-picture codecs.
	84	+ * 0 for top field or frame pictures, 1 for bottom field.
	85	+ */
	86	+void ff_thread_await_progress(AVFrame *f, int progress, int field);
	87	+
	88	+/**
	89	+ * Wrapper around get_buffer() for frame-multithreaded codecs.
	90	+ * Call this function instead of avctx->get_buffer(f).
	91	+ * Cannot be called after the codec has called ff_thread_finish_setup().
	92	+ *
	93	+ * @param avctx The current context.
	94	+ * @param f The frame to write into.
	95	+ */
	96	+int ff_thread_get_buffer(AVCodecContext avctx, AVFrame f);
	97	+
	98	+/**
	99	+ * Wrapper around release_buffer() frame-for multithreaded codecs.
	100	+ * Call this function instead of avctx->release_buffer(f).
	101	+ * The AVFrame will be copied and the actual release_buffer() call
	102	+ * will be performed later. The contents of data pointed to by the
	103	+ * AVFrame should not be changed until ff_thread_get_buffer() is called
	104	+ * on it.
	105	+ *
	106	+ * @param avctx The current context.
	107	+ * @param f The picture being released.
	108	+ */
	109	+void ff_thread_release_buffer(AVCodecContext avctx, AVFrame f);
	110	+
	111	+#endif /* AVCODEC_THREAD_H */

libavcodec/utils.c+54−5

@@ -37,6 +37,7 @@ · 36 unmodified lines
37	37	#include "dsputil.h"
38	38	#include "libavutil/opt.h"
39	39	#include "imgconvert.h"
	40	+#include "thread.h"
40	41	#include "audioconvert.h"
41	42	#include "internal.h"
42	43	#include <stdlib.h>
@@ -261,6 +262,11 @@ int avcodec_default_get_buffer(AVCodecContext s, AVFrame pic){ · 218 unmodified lines
261	262	(*picture_number)++;
262	263
263	264	if(buf->base[0] && (buf->width != w \|\| buf->height != h \|\| buf->pix_fmt != s->pix_fmt)){
	265	+ if(s->active_thread_type&FF_THREAD_FRAME) {
	266	+ av_log_missing_feature(s, "Width/height changing with frame threads is", 0);
	267	+ return -1;
	268	+ }
	269	+
264	270	for(i=0; i<4; i++){
265	271	av_freep(&buf->base[i]);
266	272	buf->data[i]= NULL;
@@ -532,6 +538,14 @@ int attribute_align_arg avcodec_open(AVCodecContext avctx, AVCodec codec) · 265 unmodified lines
532	538	goto free_and_end;
533	539	}
534	540	avctx->frame_number = 0;
	541	+
	542	+ if (HAVE_THREADS && !avctx->thread_opaque) {
	543	+ ret = avcodec_thread_init(avctx, avctx->thread_count);
	544	+ if (ret < 0) {
	545	+ goto free_and_end;
	546	+ }
	547	+ }
	548	+
535	549	if (avctx->codec->max_lowres < avctx->lowres) {
536	550	av_log(avctx, AV_LOG_ERROR, "The maximum value for lowres supported by the decoder is %d\n",
537	551	avctx->codec->max_lowres);
@@ -543,7 +557,7 @@ int attribute_align_arg avcodec_open(AVCodecContext avctx, AVCodec codec) · 5 unmodified lines
543	557	avctx->pts_correction_last_pts =
544	558	avctx->pts_correction_last_dts = INT64_MIN;
545	559
546		− if(avctx->codec->init){
	560	+ if(avctx->codec->init && !(avctx->active_thread_type&FF_THREAD_FRAME)){
547	561	ret = avctx->codec->init(avctx);
548	562	if (ret < 0) {
549	563	goto free_and_end;
@@ -674,13 +688,18 @@ int attribute_align_arg avcodec_decode_video2(AVCodecContext avctx, AVFrame pi · 124 unmodified lines
674	688
675	689	avctx->pkt = avpkt;
676	690
677		− if((avctx->codec->capabilities & CODEC_CAP_DELAY) \|\| avpkt->size){
678		− ret = avctx->codec->decode(avctx, picture, got_picture_ptr,
679		− avpkt);
	691	+ if((avctx->codec->capabilities & CODEC_CAP_DELAY) \|\| avpkt->size \|\| (avctx->active_thread_type&FF_THREAD_FRAME)){
	692	+ if (HAVE_PTHREADS && avctx->active_thread_type&FF_THREAD_FRAME)
	693	+ ret = ff_thread_decode_frame(avctx, picture, got_picture_ptr,
	694	+ avpkt);
	695	+ else {
	696	+ ret = avctx->codec->decode(avctx, picture, got_picture_ptr,
	697	+ avpkt);
	698	+ picture->pkt_dts= avpkt->dts;
	699	+ }
680	700
681	701	emms_c(); //needed to avoid an emms_c() call before every return;
682	702
683		− picture->pkt_dts= avpkt->dts;
684	703	picture->best_effort_timestamp = guess_correct_pts(avctx,
685	704	picture->pkt_pts,
686	705	picture->pkt_dts);
@@ -809,6 +828,7 @@ av_cold int avcodec_close(AVCodecContext *avctx) · 122 unmodified lines
809	828	if(avctx->codec && avctx->codec->encode)
810	829	av_freep(&avctx->extradata);
811	830	avctx->codec = NULL;
	831	+ avctx->active_thread_type = 0;
812	832	entangled_thread_counter--;
813	833
814	834	/* Release any user-supplied mutex. */
@@ -1070,6 +1090,8 @@ void avcodec_init(void) · 255 unmodified lines
1070	1090
1071	1091	void avcodec_flush_buffers(AVCodecContext *avctx)
1072	1092	{
	1093	+ if(HAVE_PTHREADS && avctx->active_thread_type&FF_THREAD_FRAME)
	1094	+ ff_thread_flush(avctx);
1073	1095	if(avctx->codec->flush)
1074	1096	avctx->codec->flush(avctx);
1075	1097	}
@@ -1270,3 +1292,30 @@ unsigned int ff_toupper4(unsigned int x) · 194 unmodified lines
1270	1292	+ (toupper((x>>16)&0xFF)<<16)
1271	1293	+ (toupper((x>>24)&0xFF)<<24);
1272	1294	}
	1295	+
	1296	+#if !HAVE_PTHREADS
	1297	+
	1298	+int ff_thread_get_buffer(AVCodecContext avctx, AVFrame f)
	1299	+{
	1300	+ f->owner = avctx;
	1301	+ return avctx->get_buffer(avctx, f);
	1302	+}
	1303	+
	1304	+void ff_thread_release_buffer(AVCodecContext avctx, AVFrame f)
	1305	+{
	1306	+ f->owner->release_buffer(f->owner, f);
	1307	+}
	1308	+
	1309	+void ff_thread_finish_setup(AVCodecContext *avctx)
	1310	+{
	1311	+}
	1312	+
	1313	+void ff_thread_report_progress(AVFrame *f, int progress, int field)
	1314	+{
	1315	+}
	1316	+
	1317	+void ff_thread_await_progress(AVFrame *f, int progress, int field)
	1318	+{
	1319	+}
	1320	+
	1321	+#endif

libavcodec/w32thread.c+6−0

@@ -129,7 +129,13 @@ int avcodec_thread_init(AVCodecContext *s, int thread_count){ · 128 unmodified lines
129	129	ThreadContext *c;
130	130	uint32_t threadid;
131	131
	132	+ if(!(s->thread_type & FF_THREAD_SLICE)){
	133	+ av_log(s, AV_LOG_WARNING, "The requested thread algorithm is not supported with this thread library.\n");
	134	+ return 0;
	135	+ }
	136	+
132	137	s->thread_count= thread_count;
	138	+ s->active_thread_type= FF_THREAD_SLICE;
133	139
134	140	if (thread_count <= 1)
135	141	return 0;

libavformat/utils.c+6−0

@@ -930,6 +930,12 @@ static void compute_pkt_fields(AVFormatContext s, AVStream st, · 929 unmodified lines
930	930	/* do we have a video B-frame ? */
931	931	delay= st->codec->has_b_frames;
932	932	presentation_delayed = 0;
	933	+
	934	+ // ignore delay caused by frame threading so that the mpeg2-without-dts
	935	+ // warning will not trigger
	936	+ if (delay && st->codec->active_thread_type&FF_THREAD_FRAME)
	937	+ delay -= st->codec->thread_count-1;
	938	+
933	939	/* XXX: need has_b_frame, but cannot get it if the codec is
934	940	not initialized */
935	941	if (delay &&

libavutil/internal.h+11−0

@@ -210,4 +210,15 @@ · 209 unmodified lines
210	210	type ff_##name args
211	211	#endif
212	212
	213	+/**
	214	+ * Returns NULL if a threading library has not been enabled.
	215	+ * Used to disable threading functions in AVCodec definitions
	216	+ * when not needed.
	217	+ */
	218	+#if HAVE_THREADS
	219	+# define ONLY_IF_THREADS_ENABLED(x) x
	220	+#else
	221	+# define ONLY_IF_THREADS_ENABLED(x) NULL
	222	+#endif
	223	+
213	224	#endif /* AVUTIL_INTERNAL_H */

Frame-based multithreading framework using pthreads

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code