comparison h264.c @ 2392:299f2d85d27d libavcodec

add multi slice support for main profile H.264 streams. I have tested this on all H264 conformance streams and comparing the result with the current CVS version, there is 8 streams which decode correctly more frames and 2 streams which are now completely correct with my patch. This patch also correct some typo in comments. patch by (Loic Le Loarer <lll+ffmpeg >at< m4x org>)

2391:336a239ad9a4

#endif
}

/**
 * fill a rectangle.
 * @param h height of the recatangle, should be a constant
 * @param w width of the recatangle, should be a constant
 * @param size the size of val (1 or 4), should be a constant
 */
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
    uint8_t *p= (uint8_t*)vp;
    assert(size==1 || size==4);

static inline int check_intra_pred_mode(H264Context *h, int mode){
    MpegEncContext * const s = &h->s;
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
    
    if(mode < 0 || mode > 6)
        return -1;
    
    if(!(h->top_samples_available&0x8000)){
        mode= top[ mode ];
        if(mode<0){
            av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);

static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];

    tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);

    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
       

    }
    return 0;
}

/**
 * instantaneos decoder refresh.
 */
static void idr(H264Context *h){
    int i;

    for(i=0; i<h->long_ref_count; i++){

    first_mb_in_slice= get_ue_golomb(&s->gb);

    h->slice_type= get_ue_golomb(&s->gb);
    if(h->slice_type > 9){
        av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
    }
    if(h->slice_type > 4){
        h->slice_type -= 5;
        h->slice_type_fixed=1;
    }else

    s->mb_height= h->sps.mb_height;
    
    h->b_stride=  s->mb_width*4;
    h->b8_stride= s->mb_width*2;

    s->mb_x = first_mb_in_slice % s->mb_width;
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
    
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
    if(h->sps.frame_mbs_only_flag)
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
    else

            s->avctx->frame_rate = h->sps.time_scale;
            s->avctx->frame_rate_base = h->sps.num_units_in_tick;
        }
    }

    if(first_mb_in_slice == 0){
        frame_start(h);
    }

    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);

                return -1;
            }
        }
    }

    if(first_mb_in_slice == 0){
        fill_default_ref_list(h);
    }

    decode_ref_pic_list_reordering(h);

#if 0 //FMO
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
        slice_group_change_cycle= get_bits(&s->gb, ?);
#endif

    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
        av_log(h->s.avctx, AV_LOG_DEBUG, "mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
               first_mb_in_slice, 
               av_get_pict_type_char(h->slice_type),
               pps_id, h->frame_num,
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
               h->ref_count[0], h->ref_count[1],
               s->qscale,

    }else if(!IS_DIRECT(mb_type)){
        int list, mx, my, i;
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
        if(IS_16X16(mb_type)){
            for(list=0; list<2; list++){
                if(h->ref_count[0]>0){
                    if(IS_DIR(mb_type, 0, list)){
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
                    }
                }

            }

            if( ++s->mb_x >= s->mb_width ) {
                s->mb_x = 0;
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
                if( ++s->mb_y >= s->mb_height ) {
                    tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                }
            }

            if( eos || s->mb_y >= s->mb_height ) {
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
                return 0;
            }
#if 0
            /* TODO test over-reading in cabac code */

                    }
                }
            }

            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

                    return 0;
                }else{

/**
 * finds the end of the current frame in the bitstream.
 * @return the position of the first byte of the next frame, or -1
 */
static int find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size){
    int i;
    uint32_t state;
//printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
//    mb_addr= pc->mb_addr - 1;
    state= pc->state;
    //FIXME this will fail with slices
    for(i=0; i<buf_size; i++){
        state= (state<<8) | buf[i];
        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
            if(pc->frame_start_found){
                pc->state=-1; 
                pc->frame_start_found= 0;
                return i-3;
            }
            pc->frame_start_found= 1;
        }
    }
    
    pc->state= state;
    return END_NOT_FOUND;
}

static int h264_parse(AVCodecParserContext *s,
                      AVCodecContext *avctx,
                      uint8_t **poutbuf, int *poutbuf_size, 
                      const uint8_t *buf, int buf_size)
{
    ParseContext *pc = s->priv_data;
    int next;
    
    next= find_frame_end(pc, buf, buf_size);

    if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
        *poutbuf = NULL;
        *poutbuf_size = 0;
        return buf_size;

    int i;
    for(i=0; i<32; i++){
        printf("%X ", buf[i]);
    }
#endif
    for(;;){
        int consumed;
        int dst_length;
        int bit_length;
        uint8_t *ptr;

    if (buf_size == 0) {
        return 0;
    }
    
    if(s->flags&CODEC_FLAG_TRUNCATED){
        int next= find_frame_end(&s->parse_context, buf, buf_size);
        
        if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
            return buf_size;
//printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
    }

    avctx->frame_number = s->picture_number - 1;
#endif
#if 0
    /* dont output the last pic after seeking */
    if(s->last_picture_ptr || s->low_delay)
    //Note this isnt a issue as a IDR pic should flush teh buffers
#endif
        *data_size = sizeof(AVFrame);
    return get_consumed_bytes(s, buf_index, buf_size);
}
#if 0

    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
};

AVCodecParser h264_parser = {
    { CODEC_ID_H264 },
    sizeof(ParseContext),
    NULL,
    h264_parse,
    ff_parse_close,
};

2392:299f2d85d27d

#endif
}

/**
 * fill a rectangle.
 * @param h height of the rectangle, should be a constant
 * @param w width of the rectangle, should be a constant
 * @param size the size of val (1 or 4), should be a constant
 */
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
    uint8_t *p= (uint8_t*)vp;
    assert(size==1 || size==4);

static inline int check_intra_pred_mode(H264Context *h, int mode){
    MpegEncContext * const s = &h->s;
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
    
    if(mode < 0 || mode > 6) {
        av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
        return -1;
    }
    
    if(!(h->top_samples_available&0x8000)){
        mode= top[ mode ];
        if(mode<0){
            av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);

static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];

    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);

    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
       

    }
    return 0;
}

/**
 * instantaneous decoder refresh.
 */
static void idr(H264Context *h){
    int i;

    for(i=0; i<h->long_ref_count; i++){

    first_mb_in_slice= get_ue_golomb(&s->gb);

    h->slice_type= get_ue_golomb(&s->gb);
    if(h->slice_type > 9){
        av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
        return -1;
    }
    if(h->slice_type > 4){
        h->slice_type -= 5;
        h->slice_type_fixed=1;
    }else

    s->mb_height= h->sps.mb_height;
    
    h->b_stride=  s->mb_width*4;
    h->b8_stride= s->mb_width*2;

    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
    s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
    
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
    if(h->sps.frame_mbs_only_flag)
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
    else

            s->avctx->frame_rate = h->sps.time_scale;
            s->avctx->frame_rate_base = h->sps.num_units_in_tick;
        }
    }

    if(h->slice_num == 0){
        frame_start(h);
    }

    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);

                return -1;
            }
        }
    }

    if(h->slice_num == 0){
        fill_default_ref_list(h);
    }

    decode_ref_pic_list_reordering(h);

#if 0 //FMO
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
        slice_group_change_cycle= get_bits(&s->gb, ?);
#endif

    h->slice_num++;

    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
               h->slice_num, first_mb_in_slice, 
               av_get_pict_type_char(h->slice_type),
               pps_id, h->frame_num,
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
               h->ref_count[0], h->ref_count[1],
               s->qscale,

    }else if(!IS_DIRECT(mb_type)){
        int list, mx, my, i;
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
        if(IS_16X16(mb_type)){
            for(list=0; list<2; list++){
                if(h->ref_count[list]>0){
                    if(IS_DIR(mb_type, 0, list)){
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
                    }
                }

            }

            if( ++s->mb_x >= s->mb_width ) {
                s->mb_x = 0;
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
                ++s->mb_y;
            }

            if( eos || s->mb_y >= s->mb_height ) {
                tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
                return 0;
            }
#if 0
            /* TODO test over-reading in cabac code */

                    }
                }
            }

            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
                tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

                    return 0;
                }else{

/**
 * finds the end of the current frame in the bitstream.
 * @return the position of the first byte of the next frame, or -1
 */
static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
    int i;
    uint32_t state;
    ParseContext *pc = &(h->s.parse_context);
//printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
//    mb_addr= pc->mb_addr - 1;
    state= pc->state;
    for(i=0; i<=buf_size; i++){
        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
            tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
            if(pc->frame_start_found){
                // If there isn't one more byte in the buffer
                // the test on first_mb_in_slice cannot be done yet
                // do it at next call.
                if (i >= buf_size) break;
                if (buf[i] & 0x80) {
                    // first_mb_in_slice is 0, probably the first nal of a new
                    // slice
                    tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
                    pc->state=-1; 
                    pc->frame_start_found= 0;
                    return i-4;
                }
            }
            pc->frame_start_found = 1;
        }
        if (i<buf_size)
            state= (state<<8) | buf[i];
    }
    
    pc->state= state;
    return END_NOT_FOUND;
}

static int h264_parse(AVCodecParserContext *s,
                      AVCodecContext *avctx,
                      uint8_t **poutbuf, int *poutbuf_size, 
                      const uint8_t *buf, int buf_size)
{
    H264Context *h = s->priv_data;
    ParseContext *pc = &h->s.parse_context;
    int next;
    
    next= find_frame_end(h, buf, buf_size);

    if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
        *poutbuf = NULL;
        *poutbuf_size = 0;
        return buf_size;

    int i;
    for(i=0; i<32; i++){
        printf("%X ", buf[i]);
    }
#endif
    h->slice_num = 0;
    for(;;){
        int consumed;
        int dst_length;
        int bit_length;
        uint8_t *ptr;

    if (buf_size == 0) {
        return 0;
    }
    
    if(s->flags&CODEC_FLAG_TRUNCATED){
        int next= find_frame_end(h, buf, buf_size);
        
        if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
            return buf_size;
//printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
    }

    avctx->frame_number = s->picture_number - 1;
#endif
#if 0
    /* dont output the last pic after seeking */
    if(s->last_picture_ptr || s->low_delay)
    //Note this isnt a issue as a IDR pic should flush the buffers
#endif
        *data_size = sizeof(AVFrame);
    return get_consumed_bytes(s, buf_index, buf_size);
}
#if 0

    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
};

AVCodecParser h264_parser = {
    { CODEC_ID_H264 },
    sizeof(H264Context),
    NULL,
    h264_parse,
    ff_parse_close,
};