PeerStreamer

 * Copyright (c) 2008 Alexander Strange (astrange@ithinksw.com)

/**

 * @file

 * Multithreading support functions

 * @see doc/multithreading.txt

 */

#include "thread.h"

/// Max number of frame buffers that can be allocated when using frame threads.

#define MAX_BUFFERS 32

/**

 * Context used by codec threads and stored in their AVCodecContext thread_opaque.

 */

typedef struct PerThreadContext {

    struct FrameThreadContext *parent;

    pthread_t      thread;

    pthread_cond_t input_cond;      ///< Used to wait for a new packet from the main thread.

    pthread_cond_t progress_cond;   ///< Used by child threads to wait for progress to change.

    pthread_cond_t output_cond;     ///< Used by the main thread to wait for frames to finish.

    pthread_mutex_t mutex;          ///< Mutex used to protect the contents of the PerThreadContext.

    pthread_mutex_t progress_mutex; ///< Mutex used to protect frame progress values and progress_cond.

    AVCodecContext *avctx;          ///< Context used to decode packets passed to this thread.

    AVPacket       avpkt;           ///< Input packet (for decoding) or output (for encoding).

    int            allocated_buf_size; ///< Size allocated for avpkt.data

    AVFrame frame;                  ///< Output frame (for decoding) or input (for encoding).

    int     got_frame;              ///< The output of got_picture_ptr from the last avcodec_decode_video() call.

    int     result;                 ///< The result of the last codec decode/encode() call.

    enum {

        STATE_INPUT_READY,          ///< Set when the thread is awaiting a packet.

        STATE_SETTING_UP,           ///< Set before the codec has called ff_thread_finish_setup().

        STATE_GET_BUFFER,           /**<

                                     * Set when the codec calls get_buffer().

                                     * State is returned to STATE_SETTING_UP afterwards.

                                     */

        STATE_SETUP_FINISHED        ///< Set after the codec has called ff_thread_finish_setup().

    } state;

    /**

     * Array of frames passed to ff_thread_release_buffer().

     * Frames are released after all threads referencing them are finished.

     */

    AVFrame released_buffers[MAX_BUFFERS];

    int     num_released_buffers;

    /**

     * Array of progress values used by ff_thread_get_buffer().

     */

    int     progress[MAX_BUFFERS][2];

    uint8_t progress_used[MAX_BUFFERS];

    AVFrame *requested_frame;       ///< AVFrame the codec passed to get_buffer()

} PerThreadContext;

/**

 * Context stored in the client AVCodecContext thread_opaque.

 */

typedef struct FrameThreadContext {

    PerThreadContext *threads;     ///< The contexts for each thread.

    PerThreadContext *prev_thread; ///< The last thread submit_packet() was called on.

    pthread_mutex_t buffer_mutex;  ///< Mutex used to protect get/release_buffer().

    int next_decoding;             ///< The next context to submit a packet to.

    int next_finished;             ///< The next context to return output from.

    int delaying;                  /**<

                                    * Set for the first N packets, where N is the number of threads.

                                    * While it is set, ff_thread_en/decode_frame won't return any results.

                                    */

    int die;                       ///< Set when threads should exit.

} FrameThreadContext;

static void thread_free(AVCodecContext *avctx)

    if (!(avctx->active_thread_type&FF_THREAD_SLICE) || avctx->thread_count <= 1)

        return avcodec_default_execute(avctx, func, arg, ret, job_count, job_size);

static int thread_init(AVCodecContext *avctx)

    int thread_count = avctx->thread_count;

/**

 * Codec worker thread.

 *

 * Automatically calls ff_thread_finish_setup() if the codec does

 * not provide an update_thread_context method, or if the codec returns

 * before calling it.

 */

static attribute_align_arg void *frame_worker_thread(void *arg)

{

    PerThreadContext *p = arg;

    FrameThreadContext *fctx = p->parent;

    AVCodecContext *avctx = p->avctx;

    AVCodec *codec = avctx->codec;

    while (1) {

        if (p->state == STATE_INPUT_READY && !fctx->die) {

            pthread_mutex_lock(&p->mutex);

            while (p->state == STATE_INPUT_READY && !fctx->die)

                pthread_cond_wait(&p->input_cond, &p->mutex);

            pthread_mutex_unlock(&p->mutex);

        }

        if (fctx->die) break;

        if (!codec->update_thread_context) ff_thread_finish_setup(avctx);

        pthread_mutex_lock(&p->mutex);

        avcodec_get_frame_defaults(&p->frame);

        p->got_frame = 0;

        p->result = codec->decode(avctx, &p->frame, &p->got_frame, &p->avpkt);

        if (p->state == STATE_SETTING_UP) ff_thread_finish_setup(avctx);

        p->state = STATE_INPUT_READY;

        pthread_mutex_lock(&p->progress_mutex);

        pthread_cond_signal(&p->output_cond);

        pthread_mutex_unlock(&p->progress_mutex);

        pthread_mutex_unlock(&p->mutex);

    }

    return NULL;

}

/**

 * Updates the next thread's AVCodecContext with values from the reference thread's context.

 *

 * @param dst The destination context.

 * @param src The source context.

 * @param for_user 0 if the destination is a codec thread, 1 if the destination is the user's thread

 */

static int update_context_from_thread(AVCodecContext *dst, AVCodecContext *src, int for_user)

{

    int err = 0;

    if (dst != src) {

        dst->sub_id    = src->sub_id;

        dst->time_base = src->time_base;

        dst->width     = src->width;

        dst->height    = src->height;

        dst->pix_fmt   = src->pix_fmt;

        dst->has_b_frames = src->has_b_frames;

        dst->idct_algo    = src->idct_algo;

        dst->slice_count  = src->slice_count;

        dst->bits_per_coded_sample = src->bits_per_coded_sample;

        dst->sample_aspect_ratio   = src->sample_aspect_ratio;

        dst->dtg_active_format     = src->dtg_active_format;

        dst->profile = src->profile;

        dst->level   = src->level;

        dst->bits_per_raw_sample = src->bits_per_raw_sample;

        dst->ticks_per_frame     = src->ticks_per_frame;

        dst->color_primaries     = src->color_primaries;

        dst->color_trc   = src->color_trc;

        dst->colorspace  = src->colorspace;

        dst->color_range = src->color_range;

        dst->chroma_sample_location = src->chroma_sample_location;

    }

    if (for_user) {

        dst->coded_frame   = src->coded_frame;

        dst->has_b_frames += src->thread_count - 1;

    } else {

        if (dst->codec->update_thread_context)

            err = dst->codec->update_thread_context(dst, src);

    }

    return err;

}

/**

 * Update the next thread's AVCodecContext with values set by the user.

 *

 * @param dst The destination context.

 * @param src The source context.

 */

static void update_context_from_user(AVCodecContext *dst, AVCodecContext *src)

{

#define copy_fields(s, e) memcpy(&dst->s, &src->s, (char*)&dst->e - (char*)&dst->s);

    dst->flags          = src->flags;

    dst->draw_horiz_band= src->draw_horiz_band;

    dst->get_buffer     = src->get_buffer;

    dst->release_buffer = src->release_buffer;

    dst->opaque   = src->opaque;

    dst->hurry_up = src->hurry_up;

    dst->dsp_mask = src->dsp_mask;

    dst->debug    = src->debug;

    dst->debug_mv = src->debug_mv;

    dst->slice_flags = src->slice_flags;

    dst->flags2      = src->flags2;

    copy_fields(skip_loop_filter, bidir_refine);

    dst->frame_number     = src->frame_number;

    dst->reordered_opaque = src->reordered_opaque;

#undef copy_fields

}

static void free_progress(AVFrame *f)

{

    PerThreadContext *p = f->owner->thread_opaque;

    int *progress = f->thread_opaque;

    p->progress_used[(progress - p->progress[0]) / 2] = 0;

}

/// Releases the buffers that this decoding thread was the last user of.

static void release_delayed_buffers(PerThreadContext *p)

{

    FrameThreadContext *fctx = p->parent;

    while (p->num_released_buffers > 0) {

        AVFrame *f = &p->released_buffers[--p->num_released_buffers];

        pthread_mutex_lock(&fctx->buffer_mutex);

        free_progress(f);

        f->thread_opaque = NULL;

        f->owner->release_buffer(f->owner, f);

        pthread_mutex_unlock(&fctx->buffer_mutex);

    }

}

static int submit_packet(PerThreadContext *p, AVPacket *avpkt)

{

    FrameThreadContext *fctx = p->parent;

    PerThreadContext *prev_thread = fctx->prev_thread;

    AVCodec *codec = p->avctx->codec;

    uint8_t *buf = p->avpkt.data;

    if (!avpkt->size && !(codec->capabilities & CODEC_CAP_DELAY)) return 0;

    pthread_mutex_lock(&p->mutex);

    release_delayed_buffers(p);

    if (prev_thread) {

        int err;

        if (prev_thread->state == STATE_SETTING_UP) {

            pthread_mutex_lock(&prev_thread->progress_mutex);

            while (prev_thread->state == STATE_SETTING_UP)

                pthread_cond_wait(&prev_thread->progress_cond, &prev_thread->progress_mutex);

            pthread_mutex_unlock(&prev_thread->progress_mutex);

        }

        err = update_context_from_thread(p->avctx, prev_thread->avctx, 0);

        if (err) {

            pthread_mutex_unlock(&p->mutex);

            return err;

        }

    }

    av_fast_malloc(&buf, &p->allocated_buf_size, avpkt->size + FF_INPUT_BUFFER_PADDING_SIZE);

    p->avpkt = *avpkt;

    p->avpkt.data = buf;

    memcpy(buf, avpkt->data, avpkt->size);

    memset(buf + avpkt->size, 0, FF_INPUT_BUFFER_PADDING_SIZE);

    p->state = STATE_SETTING_UP;

    pthread_cond_signal(&p->input_cond);

    pthread_mutex_unlock(&p->mutex);

    /*

     * If the client doesn't have a thread-safe get_buffer(),

     * then decoding threads call back to the main thread,

     * and it calls back to the client here.

     */

    if (!p->avctx->thread_safe_callbacks &&

         p->avctx->get_buffer != avcodec_default_get_buffer) {

        while (p->state != STATE_SETUP_FINISHED && p->state != STATE_INPUT_READY) {

            pthread_mutex_lock(&p->progress_mutex);

            while (p->state == STATE_SETTING_UP)

                pthread_cond_wait(&p->progress_cond, &p->progress_mutex);

            if (p->state == STATE_GET_BUFFER) {

                p->result = p->avctx->get_buffer(p->avctx, p->requested_frame);

                p->state  = STATE_SETTING_UP;

                pthread_cond_signal(&p->progress_cond);

            }

            pthread_mutex_unlock(&p->progress_mutex);

        }

    }

    fctx->prev_thread = p;

    return 0;

}

int ff_thread_decode_frame(AVCodecContext *avctx,

                           AVFrame *picture, int *got_picture_ptr,

                           AVPacket *avpkt)

{

    FrameThreadContext *fctx = avctx->thread_opaque;

    int finished = fctx->next_finished;

    PerThreadContext *p;

    int err;

    /*

     * Submit a packet to the next decoding thread.

     */

    p = &fctx->threads[fctx->next_decoding];

    update_context_from_user(p->avctx, avctx);

    err = submit_packet(p, avpkt);

    if (err) return err;

    fctx->next_decoding++;

    /*

     * If we're still receiving the initial packets, don't return a frame.

     */

    if (fctx->delaying && avpkt->size) {

        if (fctx->next_decoding >= (avctx->thread_count-1)) fctx->delaying = 0;

        *got_picture_ptr=0;

        return 0;

    }

    /*

     * Return the next available frame from the oldest thread.

     * If we're at the end of the stream, then we have to skip threads that

     * didn't output a frame, because we don't want to accidentally signal

     * EOF (avpkt->size == 0 && *got_picture_ptr == 0).

     */

    do {

        p = &fctx->threads[finished++];

        if (p->state != STATE_INPUT_READY) {

            pthread_mutex_lock(&p->progress_mutex);

            while (p->state != STATE_INPUT_READY)

                pthread_cond_wait(&p->output_cond, &p->progress_mutex);

            pthread_mutex_unlock(&p->progress_mutex);

        }

        *picture = p->frame;

        *got_picture_ptr = p->got_frame;

        picture->pkt_dts = p->avpkt.dts;

        /*

         * A later call with avkpt->size == 0 may loop over all threads,

         * including this one, searching for a frame to return before being

         * stopped by the "finished != fctx->next_finished" condition.

         * Make sure we don't mistakenly return the same frame again.

         */

        p->got_frame = 0;

        if (finished >= avctx->thread_count) finished = 0;

    } while (!avpkt->size && !*got_picture_ptr && finished != fctx->next_finished);

    update_context_from_thread(avctx, p->avctx, 1);

    if (fctx->next_decoding >= avctx->thread_count) fctx->next_decoding = 0;

    fctx->next_finished = finished;

    return p->result;

}

void ff_thread_report_progress(AVFrame *f, int n, int field)

{

    PerThreadContext *p;

    int *progress = f->thread_opaque;

    if (!progress || progress[field] >= n) return;

    p = f->owner->thread_opaque;

    if (f->owner->debug&FF_DEBUG_THREADS)

        av_log(f->owner, AV_LOG_DEBUG, "%p finished %d field %d\n", progress, n, field);

    pthread_mutex_lock(&p->progress_mutex);

    progress[field] = n;

    pthread_cond_broadcast(&p->progress_cond);

    pthread_mutex_unlock(&p->progress_mutex);

}

void ff_thread_await_progress(AVFrame *f, int n, int field)

{

    PerThreadContext *p;

    int *progress = f->thread_opaque;

    if (!progress || progress[field] >= n) return;

    p = f->owner->thread_opaque;

    if (f->owner->debug&FF_DEBUG_THREADS)

        av_log(f->owner, AV_LOG_DEBUG, "thread awaiting %d field %d from %p\n", n, field, progress);

    pthread_mutex_lock(&p->progress_mutex);

    while (progress[field] < n)

        pthread_cond_wait(&p->progress_cond, &p->progress_mutex);

    pthread_mutex_unlock(&p->progress_mutex);

}

void ff_thread_finish_setup(AVCodecContext *avctx) {

    PerThreadContext *p = avctx->thread_opaque;

    if (!(avctx->active_thread_type&FF_THREAD_FRAME)) return;

    pthread_mutex_lock(&p->progress_mutex);

    p->state = STATE_SETUP_FINISHED;

    pthread_cond_broadcast(&p->progress_cond);

    pthread_mutex_unlock(&p->progress_mutex);

}

/// Waits for all threads to finish.

static void park_frame_worker_threads(FrameThreadContext *fctx, int thread_count)

{

    int i;

    for (i = 0; i < thread_count; i++) {

        PerThreadContext *p = &fctx->threads[i];

        if (p->state != STATE_INPUT_READY) {

            pthread_mutex_lock(&p->progress_mutex);

            while (p->state != STATE_INPUT_READY)

                pthread_cond_wait(&p->output_cond, &p->progress_mutex);

            pthread_mutex_unlock(&p->progress_mutex);

        }

    }

}

static void frame_thread_free(AVCodecContext *avctx, int thread_count)

{

    FrameThreadContext *fctx = avctx->thread_opaque;

    AVCodec *codec = avctx->codec;

    int i;

    park_frame_worker_threads(fctx, thread_count);

    if (fctx->prev_thread)

        update_context_from_thread(fctx->threads->avctx, fctx->prev_thread->avctx, 0);

    fctx->die = 1;

    for (i = 0; i < thread_count; i++) {

        PerThreadContext *p = &fctx->threads[i];

        pthread_mutex_lock(&p->mutex);

        pthread_cond_signal(&p->input_cond);

        pthread_mutex_unlock(&p->mutex);

        pthread_join(p->thread, NULL);

        if (codec->close)

            codec->close(p->avctx);

        avctx->codec = NULL;

        release_delayed_buffers(p);

    }

    for (i = 0; i < thread_count; i++) {

        PerThreadContext *p = &fctx->threads[i];

        avcodec_default_free_buffers(p->avctx);

        pthread_mutex_destroy(&p->mutex);

        pthread_mutex_destroy(&p->progress_mutex);

        pthread_cond_destroy(&p->input_cond);

        pthread_cond_destroy(&p->progress_cond);

        pthread_cond_destroy(&p->output_cond);

        av_freep(&p->avpkt.data);

        if (i)

            av_freep(&p->avctx->priv_data);

        av_freep(&p->avctx);

    }

    av_freep(&fctx->threads);

    pthread_mutex_destroy(&fctx->buffer_mutex);

    av_freep(&avctx->thread_opaque);

}

static int frame_thread_init(AVCodecContext *avctx)

{

    int thread_count = avctx->thread_count;

    AVCodec *codec = avctx->codec;

    AVCodecContext *src = avctx;

    FrameThreadContext *fctx;

    int i, err = 0;

    avctx->thread_opaque = fctx = av_mallocz(sizeof(FrameThreadContext));

    fctx->threads = av_mallocz(sizeof(PerThreadContext) * thread_count);

    pthread_mutex_init(&fctx->buffer_mutex, NULL);

    fctx->delaying = 1;

    for (i = 0; i < thread_count; i++) {

        AVCodecContext *copy = av_malloc(sizeof(AVCodecContext));

        PerThreadContext *p  = &fctx->threads[i];

        pthread_mutex_init(&p->mutex, NULL);

        pthread_mutex_init(&p->progress_mutex, NULL);

        pthread_cond_init(&p->input_cond, NULL);

        pthread_cond_init(&p->progress_cond, NULL);

        pthread_cond_init(&p->output_cond, NULL);

        p->parent = fctx;

        p->avctx  = copy;

        *copy = *src;

        copy->thread_opaque = p;

        copy->pkt = &p->avpkt;

        if (!i) {

            src = copy;

            if (codec->init)

                err = codec->init(copy);

            update_context_from_thread(avctx, copy, 1);

        } else {

            copy->is_copy   = 1;

            copy->priv_data = av_malloc(codec->priv_data_size);

            memcpy(copy->priv_data, src->priv_data, codec->priv_data_size);

            if (codec->init_thread_copy)

                err = codec->init_thread_copy(copy);

        }

        if (err) goto error;

        pthread_create(&p->thread, NULL, frame_worker_thread, p);

    }

    return 0;

error:

    frame_thread_free(avctx, i+1);

    return err;

}

void ff_thread_flush(AVCodecContext *avctx)

{

    FrameThreadContext *fctx = avctx->thread_opaque;

    if (!avctx->thread_opaque) return;

    park_frame_worker_threads(fctx, avctx->thread_count);

    if (fctx->prev_thread)

        update_context_from_thread(fctx->threads->avctx, fctx->prev_thread->avctx, 0);

    fctx->next_decoding = fctx->next_finished = 0;

    fctx->delaying = 1;

    fctx->prev_thread = NULL;

}

static int *allocate_progress(PerThreadContext *p)

{

    int i;

    for (i = 0; i < MAX_BUFFERS; i++)

        if (!p->progress_used[i]) break;

    if (i == MAX_BUFFERS) {

        av_log(p->avctx, AV_LOG_ERROR, "allocate_progress() overflow\n");

        return NULL;

    }

    p->progress_used[i] = 1;

    return p->progress[i];

}

int ff_thread_get_buffer(AVCodecContext *avctx, AVFrame *f)

{

    PerThreadContext *p = avctx->thread_opaque;

    int *progress, err;

    f->owner = avctx;

    if (!(avctx->active_thread_type&FF_THREAD_FRAME)) {

        f->thread_opaque = NULL;

        return avctx->get_buffer(avctx, f);

    }

    if (p->state != STATE_SETTING_UP) {

        av_log(avctx, AV_LOG_ERROR, "get_buffer() cannot be called after ff_thread_finish_setup()\n");

        return -1;

    }

    pthread_mutex_lock(&p->parent->buffer_mutex);

    f->thread_opaque = progress = allocate_progress(p);

    if (!progress) {

        pthread_mutex_unlock(&p->parent->buffer_mutex);

        return -1;

    }

    progress[0] =

    progress[1] = -1;

    if (avctx->thread_safe_callbacks ||

        avctx->get_buffer == avcodec_default_get_buffer) {

        err = avctx->get_buffer(avctx, f);

    } else {

        p->requested_frame = f;

        p->state = STATE_GET_BUFFER;

        pthread_mutex_lock(&p->progress_mutex);

        pthread_cond_signal(&p->progress_cond);

        while (p->state != STATE_SETTING_UP)

            pthread_cond_wait(&p->progress_cond, &p->progress_mutex);

        err = p->result;

        pthread_mutex_unlock(&p->progress_mutex);

    }

    pthread_mutex_unlock(&p->parent->buffer_mutex);

    /*

     * Buffer age is difficult to keep track of between

     * multiple threads, and the optimizations it allows

     * are not worth the effort. It is disabled for now.

     */

    f->age = INT_MAX;

    return err;

}

void ff_thread_release_buffer(AVCodecContext *avctx, AVFrame *f)

{

    PerThreadContext *p = avctx->thread_opaque;

    if (!(avctx->active_thread_type&FF_THREAD_FRAME)) {

        avctx->release_buffer(avctx, f);

        return;

    }

    if (p->num_released_buffers >= MAX_BUFFERS) {

        av_log(p->avctx, AV_LOG_ERROR, "too many thread_release_buffer calls!\n");

        return;

    }

    if(avctx->debug & FF_DEBUG_BUFFERS)

        av_log(avctx, AV_LOG_DEBUG, "thread_release_buffer called on pic %p, %d buffers used\n",

                                    f, f->owner->internal_buffer_count);

    p->released_buffers[p->num_released_buffers++] = *f;

    memset(f->data, 0, sizeof(f->data));

}

/**

 * Set the threading algorithms used.

 *

 * Threading requires more than one thread.

 * Frame threading requires entire frames to be passed to the codec,

 * and introduces extra decoding delay, so is incompatible with low_delay.

 *

 * @param avctx The context.

 */

static void validate_thread_parameters(AVCodecContext *avctx)

{

    int frame_threading_supported = (avctx->codec->capabilities & CODEC_CAP_FRAME_THREADS)

                                && !(avctx->flags & CODEC_FLAG_TRUNCATED)

                                && !(avctx->flags & CODEC_FLAG_LOW_DELAY)

                                && !(avctx->flags2 & CODEC_FLAG2_CHUNKS);

    if (avctx->thread_count == 1) {

        avctx->active_thread_type = 0;

    } else if (frame_threading_supported && (avctx->thread_type & FF_THREAD_FRAME)) {

        avctx->active_thread_type = FF_THREAD_FRAME;

    } else {

        avctx->active_thread_type = FF_THREAD_SLICE;

    }

}

int avcodec_thread_init(AVCodecContext *avctx, int thread_count)

{

    if (avctx->thread_opaque) {

        av_log(avctx, AV_LOG_ERROR, "avcodec_thread_init is ignored after avcodec_open\n");

        return -1;

    }

    avctx->thread_count = FFMAX(1, thread_count);

    if (avctx->codec) {

        validate_thread_parameters(avctx);

        if (avctx->active_thread_type&FF_THREAD_SLICE)

            return thread_init(avctx);

        else if (avctx->active_thread_type&FF_THREAD_FRAME)

            return frame_thread_init(avctx);

    }

    return 0;

}

void avcodec_thread_free(AVCodecContext *avctx)

{

    if (avctx->active_thread_type&FF_THREAD_FRAME)

        frame_thread_free(avctx, avctx->thread_count);

    else

        thread_free(avctx);

}