X264 - lookahead

1.   x264_lookahead_t

typedef struct x264_lookahead_t

{
    volatile uint8_t              b_exit_thread;   // notify thread exit
    uint8_t                       b_thread_active; // thread in active status
    uint8_t                       b_analyse_keyframe; // analyse key frame
    int                           i_last_keyframe;     // last keyframe
    int                           i_slicetype_length; 
    x264_frame_t                  *last_nonb;         // last non-bframe
    x264_pthread_t                thread_handle;      // thread handle
    x264_sync_frame_list_t        ifbuf;               // buffer list
    x264_sync_frame_list_t        next;
    x264_sync_frame_list_t        ofbuf;

} x264_lookahead_t;

2.  x264_lookahead_init

int x264_lookahead_init( x264_t *h, int i_slicetype_length )

{
    x264_lookahead_t *look;

    // malloc lookahead and zero it
    CHECKED_MALLOCZERO( look, sizeof(x264_lookahead_t) );
    for( int i = 0; i < h->param.i_threads; i++ ) // set x264_t's lookahead for each thread
        h->thread[i]->lookahead = look;


    look->i_last_keyframe = - h->param.i_keyint_max;
    look->b_analyse_keyframe = (h->param.rc.b_mb_tree || (h->param.rc.i_vbv_buffer_size && h->param.rc.i_lookahead))
                               && !h->param.rc.b_stat_read;
    look->i_slicetype_length = i_slicetype_length;


    /* init frame lists */
    if( x264_sync_frame_list_init( &look->ifbuf, h->param.i_sync_lookahead+3 ) ||
        x264_sync_frame_list_init( &look->next, h->frames.i_delay+3 ) ||
        x264_sync_frame_list_init( &look->ofbuf, h->frames.i_delay+3 ) )
        goto fail;


    if( !h->param.i_sync_lookahead )
        return 0;


    x264_t *look_h = h->thread[h->param.i_threads];
    *look_h = *h;
    if( x264_macroblock_cache_allocate( look_h ) )
        goto fail;


    if( x264_macroblock_thread_allocate( look_h, 1 ) < 0 )
        goto fail;

    // create lookahead_thread
    if( x264_pthread_create( &look->thread_handle, NULL, (void*)x264_lookahead_thread, look_h ) )
        goto fail;
    look->b_thread_active = 1;


    return 0;
fail:
    x264_free( look );
    return -1;
}

3.   x264_lookahead_shift

// shift frame from src to dst

static void x264_lookahead_shift( x264_sync_frame_list_t *dst, x264_sync_frame_list_t *src, int count )
{
    int i = count;
    while( i-- )
    {
        assert( dst->i_size < dst->i_max_size );
        assert( src->i_size );
        dst->list[ dst->i_size++ ] = x264_frame_shift( src->list ); // get first from src->list, and move others forward
        src->i_size--;
    }
    if( count )
    {
        x264_pthread_cond_broadcast( &dst->cv_fill );
        x264_pthread_cond_broadcast( &src->cv_empty );
    }
}

4.  x264_lookahead_update_last_nonb

// subtract 1 from h->lookahead->last_nonb->i_reference_count

// and if it == 0, insert h->lookahead->last_nonb into h->frames.unused[frame->b_fdec]

static void x264_lookahead_update_last_nonb( x264_t *h, x264_frame_t *new_nonb )
{
    if( h->lookahead->last_nonb )
        x264_frame_push_unused( h, h->lookahead->last_nonb );
    h->lookahead->last_nonb = new_nonb;
    new_nonb->i_reference_count++;
}

5. x264_frame_pop_unused

// pop unused from  h->frames.unused[b_fdec] trail, and update its trail.

x264_frame_t *x264_frame_pop_unused( x264_t *h, int b_fdec )
{
    x264_frame_t *frame;
    if( h->frames.unused[b_fdec][0] )
        frame = x264_frame_pop( h->frames.unused[b_fdec] );
    else
        frame = x264_frame_new( h, b_fdec );
    if( !frame )
        return NULL;
    frame->b_last_minigop_bframe = 0;
    frame->i_reference_count = 1;
    frame->b_intra_calculated = 0;
    frame->b_scenecut = 1;
    frame->b_keyframe = 0;
    frame->b_corrupt = 0;
    frame->i_slice_count = h->param.b_sliced_threads ? h->param.i_threads : 1;


    memset( frame->weight, 0, sizeof(frame->weight) );
    memset( frame->f_weighted_cost_delta, 0, sizeof(frame->f_weighted_cost_delta) );


    return frame;
}

6. x264_lookahead_slicetype_decide and x264_lookahead_thread

#if HAVE_THREAD
static void x264_lookahead_slicetype_decide( x264_t *h )
{
    x264_stack_align( x264_slicetype_decide, h );


    x264_lookahead_update_last_nonb( h, h->lookahead->next.list[0] );
    int shift_frames = h->lookahead->next.list[0]->i_bframes + 1;


    x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
    while( h->lookahead->ofbuf.i_size == h->lookahead->ofbuf.i_max_size )
        x264_pthread_cond_wait( &h->lookahead->ofbuf.cv_empty, &h->lookahead->ofbuf.mutex );


    x264_pthread_mutex_lock( &h->lookahead->next.mutex );

    // shift from next to ofbuf
    x264_lookahead_shift( &h->lookahead->ofbuf, &h->lookahead->next, shift_frames );
    x264_pthread_mutex_unlock( &h->lookahead->next.mutex );


    /* For MB-tree and VBV lookahead, we have to perform propagation analysis on I-frames too. */
    if( h->lookahead->b_analyse_keyframe && IS_X264_TYPE_I( h->lookahead->last_nonb->i_type ) )
        x264_stack_align( x264_slicetype_analyse, h, shift_frames );


    x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
}


static void *x264_lookahead_thread( x264_t *h )
{
    int shift;
#if HAVE_MMX
    if( h->param.cpu&X264_CPU_SSE_MISALIGN )
        x264_cpu_mask_misalign_sse();
#endif
    while( !h->lookahead->b_exit_thread )
    {
        x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
        x264_pthread_mutex_lock( &h->lookahead->next.mutex );

        // calc shift number from ifbuf to next
        shift = X264_MIN( h->lookahead->next.i_max_size - h->lookahead->next.i_size, h->lookahead->ifbuf.i_size );

        // shift items from ifbuf to next
        x264_lookahead_shift( &h->lookahead->next, &h->lookahead->ifbuf, shift );
        x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
        if( h->lookahead->next.i_size <= h->lookahead->i_slicetype_length + h->param.b_vfr_input )
        {
            while( !h->lookahead->ifbuf.i_size && !h->lookahead->b_exit_thread ) 
                x264_pthread_cond_wait( &h->lookahead->ifbuf.cv_fill, &h->lookahead->ifbuf.mutex );
            x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
        }
        else
        {
            x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
            x264_lookahead_slicetype_decide( h );  // lookahead_slicetype_decide
        }
    }   /* end of input frames */
    x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
    x264_pthread_mutex_lock( &h->lookahead->next.mutex );

    // shift
    x264_lookahead_shift( &h->lookahead->next, &h->lookahead->ifbuf, h->lookahead->ifbuf.i_size );
    x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
    x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
    while( h->lookahead->next.i_size )
        x264_lookahead_slicetype_decide( h );  // decide
    x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
    h->lookahead->b_thread_active = 0;
    x264_pthread_cond_broadcast( &h->lookahead->ofbuf.cv_fill );
    x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
    return NULL;
}
#endif

7. x264_lookahead_delete etc

void x264_lookahead_delete( x264_t *h )
{
    if( h->param.i_sync_lookahead )
    {
        x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
        h->lookahead->b_exit_thread = 1;
        x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
        x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
        x264_pthread_join( h->lookahead->thread_handle, NULL );
        x264_macroblock_cache_free( h->thread[h->param.i_threads] );
        x264_macroblock_thread_free( h->thread[h->param.i_threads], 1 );
        x264_free( h->thread[h->param.i_threads] );
    }
    x264_sync_frame_list_delete( &h->lookahead->ifbuf );
    x264_sync_frame_list_delete( &h->lookahead->next );
    if( h->lookahead->last_nonb )
        x264_frame_push_unused( h, h->lookahead->last_nonb );
    x264_sync_frame_list_delete( &h->lookahead->ofbuf );
    x264_free( h->lookahead );
}


void x264_lookahead_put_frame( x264_t *h, x264_frame_t *frame )
{
    if( h->param.i_sync_lookahead )
        x264_sync_frame_list_push( &h->lookahead->ifbuf, frame );
    else
        x264_sync_frame_list_push( &h->lookahead->next, frame );
}


int x264_lookahead_is_empty( x264_t *h )
{
    x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
    x264_pthread_mutex_lock( &h->lookahead->next.mutex );
    int b_empty = !h->lookahead->next.i_size && !h->lookahead->ofbuf.i_size;
    x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
    x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
    return b_empty;
}


static void x264_lookahead_encoder_shift( x264_t *h )
{
    if( !h->lookahead->ofbuf.i_size )
        return;
    int i_frames = h->lookahead->ofbuf.list[0]->i_bframes + 1;
    while( i_frames-- )
    {
        x264_frame_push( h->frames.current, x264_frame_shift( h->lookahead->ofbuf.list ) );
        h->lookahead->ofbuf.i_size--;
    }
    x264_pthread_cond_broadcast( &h->lookahead->ofbuf.cv_empty );
}


void x264_lookahead_get_frames( x264_t *h )
{
    if( h->param.i_sync_lookahead )
    {   /* We have a lookahead thread, so get frames from there */
        x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
        while( !h->lookahead->ofbuf.i_size && h->lookahead->b_thread_active )
            x264_pthread_cond_wait( &h->lookahead->ofbuf.cv_fill, &h->lookahead->ofbuf.mutex );
        x264_lookahead_encoder_shift( h );
        x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
    }
    else
    {   /* We are not running a lookahead thread, so perform all the slicetype decide on the fly */


        if( h->frames.current[0] || !h->lookahead->next.i_size )
            return;


        x264_stack_align( x264_slicetype_decide, h );
        x264_lookahead_update_last_nonb( h, h->lookahead->next.list[0] );
        int shift_frames = h->lookahead->next.list[0]->i_bframes + 1;
        x264_lookahead_shift( &h->lookahead->ofbuf, &h->lookahead->next, shift_frames );


        /* For MB-tree and VBV lookahead, we have to perform propagation analysis on I-frames too. */
        if( h->lookahead->b_analyse_keyframe && IS_X264_TYPE_I( h->lookahead->last_nonb->i_type ) )
            x264_stack_align( x264_slicetype_analyse, h, shift_frames );


        x264_lookahead_encoder_shift( h );
    }
}