/src/ffmpeg/libavcodec/rv40.c

Source
/*
 * RV40 decoder
 * Copyright (c) 2007 Konstantin Shishkov
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * RV40 decoder
 */

#include "config.h"

#include "libavutil/imgutils.h"
#include "libavutil/thread.h"

#include "avcodec.h"
#include "codec_internal.h"
#include "mpegutils.h"
#include "mpegvideo.h"
#include "mpegvideodec.h"
#include "golomb.h"

#include "rv34.h"
#include "rv40vlc2.h"
#include "rv40data.h"

static VLCElem aic_top_vlc[23590];
static const VLCElem *aic_mode1_vlc[AIC_MODE1_NUM], *aic_mode2_vlc[AIC_MODE2_NUM];
static const VLCElem *ptype_vlc[NUM_PTYPE_VLCS],    *btype_vlc[NUM_BTYPE_VLCS];

static av_cold const VLCElem *rv40_init_table(VLCInitState *state, int nb_bits,
                                              int nb_codes, const uint8_t (*tab)[2])
{
    return ff_vlc_init_tables_from_lengths(state, nb_bits, nb_codes,
                                           &tab[0][1], 2, &tab[0][0], 2, 1,
                                           0, 0);
}

/**
 * Initialize all tables.
 */
static av_cold void rv40_init_tables(void)
{
    VLCInitState state = VLC_INIT_STATE(aic_top_vlc);
    int i;

    rv40_init_table(&state, AIC_TOP_BITS, AIC_TOP_SIZE,
                    rv40_aic_top_vlc_tab);
    for(i = 0; i < AIC_MODE1_NUM; i++){
        // Every tenth VLC table is empty
        if((i % 10) == 9) continue;
        aic_mode1_vlc[i] =
            rv40_init_table(&state, AIC_MODE1_BITS,
                            AIC_MODE1_SIZE, aic_mode1_vlc_tabs[i]);
    }
    for (unsigned i = 0; i < AIC_MODE2_NUM; i++){
        uint16_t syms[AIC_MODE2_SIZE];

        for (int j = 0; j < AIC_MODE2_SIZE; j++) {
            int first  = aic_mode2_vlc_syms[i][j] >> 4;
            int second = aic_mode2_vlc_syms[i][j] & 0xF;
            if (HAVE_BIGENDIAN)
                syms[j] = (first << 8) | second;
            else
                syms[j] = first | (second << 8);
        }
        aic_mode2_vlc[i] =
            ff_vlc_init_tables_from_lengths(&state, AIC_MODE2_BITS, AIC_MODE2_SIZE,
                                            aic_mode2_vlc_bits[i], 1,
                                            syms, 2, 2, 0, 0);
    }
    for(i = 0; i < NUM_PTYPE_VLCS; i++){
        ptype_vlc[i] =
            rv40_init_table(&state, PTYPE_VLC_BITS, PTYPE_VLC_SIZE,
                            ptype_vlc_tabs[i]);
    }
    for(i = 0; i < NUM_BTYPE_VLCS; i++){
        btype_vlc[i] =
            rv40_init_table(&state, BTYPE_VLC_BITS, BTYPE_VLC_SIZE,
                            btype_vlc_tabs[i]);
    }
}

/**
 * Get stored dimension from bitstream.
 *
 * If the width/height is the standard one then it's coded as a 3-bit index.
 * Otherwise it is coded as escaped 8-bit portions.
 */
static int get_dimension(GetBitContext *gb, const int *dim)
{
    int t   = get_bits(gb, 3);
    int val = dim[t];
    if(val < 0)
        val = dim[get_bits1(gb) - val];
    if(!val){
        do{
            if (get_bits_left(gb) < 8)
                return AVERROR_INVALIDDATA;
            t = get_bits(gb, 8);
            val += t << 2;
        }while(t == 0xFF);
    }
    return val;
}

/**
 * Get encoded picture size - usually this is called from rv40_parse_slice_header.
 */
static void rv40_parse_picture_size(GetBitContext *gb, int *w, int *h)
{
    *w = get_dimension(gb, rv40_standard_widths);
    *h = get_dimension(gb, rv40_standard_heights);
}

static int rv40_parse_slice_header(RV34DecContext *r, GetBitContext *gb, SliceInfo *si)
{
    int w = r->s.width, h = r->s.height;
    int mb_size;
    int ret;

    memset(si, 0, sizeof(SliceInfo));
    if(get_bits1(gb))
        return AVERROR_INVALIDDATA;
    si->type = get_bits(gb, 2);
    if(si->type == 1) si->type = 0;
    si->quant = get_bits(gb, 5);
    if(get_bits(gb, 2))
        return AVERROR_INVALIDDATA;
    si->vlc_set = get_bits(gb, 2);
    skip_bits1(gb);
    si->pts = get_bits(gb, 13);
    if(!si->type || !get_bits1(gb))
        rv40_parse_picture_size(gb, &w, &h);
    if ((ret = av_image_check_size(w, h, 0, r->s.avctx)) < 0)
        return ret;
    si->width  = w;
    si->height = h;
    mb_size = ((w + 15) >> 4) * ((h + 15) >> 4);
    si->start = ff_rv34_get_start_offset(gb, mb_size);

    return 0;
}

/**
 * Decode 4x4 intra types array.
 */
static int rv40_decode_intra_types(RV34DecContext *r, GetBitContext *gb, int8_t *dst)
{
    MpegEncContext *s = &r->s;
    int i, j, k, v;
    int A, B, C;
    int pattern;
    int8_t *ptr;

    for(i = 0; i < 4; i++, dst += r->intra_types_stride){
        if(!i && s->first_slice_line){
            pattern = get_vlc2(gb, aic_top_vlc, AIC_TOP_BITS, 1);
            dst[0] = (pattern >> 2) & 2;
            dst[1] = (pattern >> 1) & 2;
            dst[2] =  pattern       & 2;
            dst[3] = (pattern << 1) & 2;
            continue;
        }
        ptr = dst;
        for(j = 0; j < 4; j++){
            /* Coefficients are read using VLC chosen by the prediction pattern
             * The first one (used for retrieving a pair of coefficients) is
             * constructed from the top, top right and left coefficients
             * The second one (used for retrieving only one coefficient) is
             * top + 10 * left.
             */
            A = ptr[-r->intra_types_stride + 1]; // it won't be used for the last coefficient in a row
            B = ptr[-r->intra_types_stride];
            C = ptr[-1];
            pattern = A + B * (1 << 4) + C * (1 << 8);
            for(k = 0; k < MODE2_PATTERNS_NUM; k++)
                if(pattern == rv40_aic_table_index[k])
                    break;
            if(j < 3 && k < MODE2_PATTERNS_NUM){ //pattern is found, decoding 2 coefficients
                AV_WN16(ptr, get_vlc2(gb, aic_mode2_vlc[k], AIC_MODE2_BITS, 2));
                ptr += 2;
                j++;
            }else{
                if(B != -1 && C != -1)
                    v = get_vlc2(gb, aic_mode1_vlc[B + C*10], AIC_MODE1_BITS, 1);
                else{ // tricky decoding
                    v = 0;
                    switch(C){
                    case -1: // code 0 -> 1, 1 -> 0
                        if(B < 2)
                            v = get_bits1(gb) ^ 1;
                        break;
                    case  0:
                    case  2: // code 0 -> 2, 1 -> 0
                        v = (get_bits1(gb) ^ 1) << 1;
                        break;
                    }
                }
                *ptr++ = v;
            }
        }
    }
    return 0;
}

/**
 * Decode macroblock information.
 */
static int rv40_decode_mb_info(RV34DecContext *r)
{
    MpegEncContext *s = &r->s;
    GetBitContext *const gb = &r->gb;
    int q, i;
    int prev_type = 0;
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

    if (!r->mb_skip_run) {
        r->mb_skip_run = get_interleaved_ue_golomb(gb) + 1;
        if (r->mb_skip_run > (unsigned)s->mb_num)
            return -1;
    }

    if (--r->mb_skip_run)
         return RV34_MB_SKIP;

    if(r->avail_cache[6-4]){
        int blocks[RV34_MB_TYPES] = {0};
        int count = 0;
        if(r->avail_cache[6-1])
            blocks[r->mb_type[mb_pos - 1]]++;
        blocks[r->mb_type[mb_pos - s->mb_stride]]++;
        if(r->avail_cache[6-2])
            blocks[r->mb_type[mb_pos - s->mb_stride + 1]]++;
        if(r->avail_cache[6-5])
            blocks[r->mb_type[mb_pos - s->mb_stride - 1]]++;
        for(i = 0; i < RV34_MB_TYPES; i++){
            if(blocks[i] > count){
                count = blocks[i];
                prev_type = i;
                if(count>1)
                    break;
            }
        }
    } else if (r->avail_cache[6-1])
        prev_type = r->mb_type[mb_pos - 1];

    if(s->pict_type == AV_PICTURE_TYPE_P){
        prev_type = block_num_to_ptype_vlc_num[prev_type];
        q = get_vlc2(gb, ptype_vlc[prev_type], PTYPE_VLC_BITS, 1);
        if(q < PBTYPE_ESCAPE)
            return q;
        q = get_vlc2(gb, ptype_vlc[prev_type], PTYPE_VLC_BITS, 1);
        av_log(s->avctx, AV_LOG_ERROR, "Dquant for P-frame\n");
    }else{
        prev_type = block_num_to_btype_vlc_num[prev_type];
        q = get_vlc2(gb, btype_vlc[prev_type], BTYPE_VLC_BITS, 1);
        if(q < PBTYPE_ESCAPE)
            return q;
        q = get_vlc2(gb, btype_vlc[prev_type], BTYPE_VLC_BITS, 1);
        av_log(s->avctx, AV_LOG_ERROR, "Dquant for B-frame\n");
    }
    return 0;
}

enum RV40BlockPos{
    POS_CUR,
    POS_TOP,
    POS_LEFT,
    POS_BOTTOM,
};

#define MASK_CUR          0x0001
#define MASK_RIGHT        0x0008
#define MASK_BOTTOM       0x0010
#define MASK_TOP          0x1000
#define MASK_Y_TOP_ROW    0x000F
#define MASK_Y_LAST_ROW   0xF000
#define MASK_Y_LEFT_COL   0x1111
#define MASK_Y_RIGHT_COL  0x8888
#define MASK_C_TOP_ROW    0x0003
#define MASK_C_LAST_ROW   0x000C
#define MASK_C_LEFT_COL   0x0005
#define MASK_C_RIGHT_COL  0x000A

static const int neighbour_offs_x[4] = { 0,  0, -1, 0 };
static const int neighbour_offs_y[4] = { 0, -1,  0, 1 };

static void rv40_adaptive_loop_filter(RV34DSPContext *rdsp,
                                      uint8_t *src, int stride, int dmode,
                                      int lim_q1, int lim_p1,
                                      int alpha, int beta, int beta2,
                                      int chroma, int edge, int dir)
{
    int filter_p1, filter_q1;
    int strong;
    int lims;

    strong = rdsp->rv40_loop_filter_strength[dir](src, stride, beta, beta2,
                                                  edge, &filter_p1, &filter_q1);

    lims = filter_p1 + filter_q1 + ((lim_q1 + lim_p1) >> 1) + 1;

    if (strong) {
        rdsp->rv40_strong_loop_filter[dir](src, stride, alpha,
                                           lims, dmode, chroma);
    } else if (filter_p1 & filter_q1) {
        rdsp->rv40_weak_loop_filter[dir](src, stride, 1, 1, alpha, beta,
                                         lims, lim_q1, lim_p1);
    } else if (filter_p1 | filter_q1) {
        rdsp->rv40_weak_loop_filter[dir](src, stride, filter_p1, filter_q1,
                                         alpha, beta, lims >> 1, lim_q1 >> 1,
                                         lim_p1 >> 1);
    }
}

/**
 * RV40 loop filtering function
 */
static void rv40_loop_filter(RV34DecContext *r, int row)
{
    MpegEncContext *s = &r->s;
    int mb_pos, mb_x;
    int i, j, k;
    uint8_t *Y, *C;
    int alpha, beta, betaY, betaC;
    int q;
    int mbtype[4];   ///< current macroblock and its neighbours types
    /**
     * flags indicating that macroblock can be filtered with strong filter
     * it is set only for intra coded MB and MB with DCs coded separately
     */
    int mb_strong[4];
    int clip[4];     ///< MB filter clipping value calculated from filtering strength
    /**
     * coded block patterns for luma part of current macroblock and its neighbours
     * Format:
     * LSB corresponds to the top left block,
     * each nibble represents one row of subblocks.
     */
    int cbp[4];
    /**
     * coded block patterns for chroma part of current macroblock and its neighbours
     * Format is the same as for luma with two subblocks in a row.
     */
    int uvcbp[4][2];
    /**
     * This mask represents the pattern of luma subblocks that should be filtered
     * in addition to the coded ones because they lie at the edge of
     * 8x8 block with different enough motion vectors
     */
    unsigned mvmasks[4];

    mb_pos = row * s->mb_stride;
    for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
        int mbtype = s->cur_pic.mb_type[mb_pos];
        if(IS_INTRA(mbtype) || IS_SEPARATE_DC(mbtype))
            r->cbp_luma  [mb_pos] = r->deblock_coefs[mb_pos] = 0xFFFF;
        if(IS_INTRA(mbtype))
            r->cbp_chroma[mb_pos] = 0xFF;
    }
    mb_pos = row * s->mb_stride;
    for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
        int y_h_deblock, y_v_deblock;
        int c_v_deblock[2], c_h_deblock[2];
        int clip_left;
        int avail[4];
        unsigned y_to_deblock;
        int c_to_deblock[2];

        q = s->cur_pic.qscale_table[mb_pos];
        alpha = rv40_alpha_tab[q];
        beta  = rv40_beta_tab [q];
        betaY = betaC = beta * 3;
        if(s->width * s->height <= 176*144)
            betaY += beta;

        avail[0] = 1;
        avail[1] = row;
        avail[2] = mb_x;
        avail[3] = row < s->mb_height - 1;
        for(i = 0; i < 4; i++){
            if(avail[i]){
                int pos = mb_pos + neighbour_offs_x[i] + neighbour_offs_y[i]*s->mb_stride;
                mvmasks[i] = r->deblock_coefs[pos];
                mbtype [i] = s->cur_pic.mb_type[pos];
                cbp    [i] = r->cbp_luma[pos];
                uvcbp[i][0] = r->cbp_chroma[pos] & 0xF;
                uvcbp[i][1] = r->cbp_chroma[pos] >> 4;
            }else{
                mvmasks[i] = 0;
                mbtype [i] = mbtype[0];
                cbp    [i] = 0;
                uvcbp[i][0] = uvcbp[i][1] = 0;
            }
            mb_strong[i] = IS_INTRA(mbtype[i]) || IS_SEPARATE_DC(mbtype[i]);
            clip[i] = rv40_filter_clip_tbl[mb_strong[i] + 1][q];
        }
        y_to_deblock =  mvmasks[POS_CUR]
                     | (mvmasks[POS_BOTTOM] << 16);
        /* This pattern contains bits signalling that horizontal edges of
         * the current block can be filtered.
         * That happens when either of adjacent subblocks is coded or lies on
         * the edge of 8x8 blocks with motion vectors differing by more than
         * 3/4 pel in any component (any edge orientation for some reason).
         */
        y_h_deblock =   y_to_deblock
                    | ((cbp[POS_CUR]                           <<  4) & ~MASK_Y_TOP_ROW)
                    | ((cbp[POS_TOP]        & MASK_Y_LAST_ROW) >> 12);
        /* This pattern contains bits signalling that vertical edges of
         * the current block can be filtered.
         * That happens when either of adjacent subblocks is coded or lies on
         * the edge of 8x8 blocks with motion vectors differing by more than
         * 3/4 pel in any component (any edge orientation for some reason).
         */
        y_v_deblock =   y_to_deblock
                    | ((cbp[POS_CUR]                      << 1) & ~MASK_Y_LEFT_COL)
                    | ((cbp[POS_LEFT] & MASK_Y_RIGHT_COL) >> 3);
        if(!mb_x)
            y_v_deblock &= ~MASK_Y_LEFT_COL;
        if(!row)
            y_h_deblock &= ~MASK_Y_TOP_ROW;
        if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
            y_h_deblock &= ~(MASK_Y_TOP_ROW << 16);
        /* Calculating chroma patterns is similar and easier since there is
         * no motion vector pattern for them.
         */
        for(i = 0; i < 2; i++){
            c_to_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
            c_v_deblock[i] =   c_to_deblock[i]
                           | ((uvcbp[POS_CUR] [i]                       << 1) & ~MASK_C_LEFT_COL)
                           | ((uvcbp[POS_LEFT][i]   & MASK_C_RIGHT_COL) >> 1);
            c_h_deblock[i] =   c_to_deblock[i]
                           | ((uvcbp[POS_TOP][i]    & MASK_C_LAST_ROW)  >> 2)
                           |  (uvcbp[POS_CUR][i]                        << 2);
            if(!mb_x)
                c_v_deblock[i] &= ~MASK_C_LEFT_COL;
            if(!row)
                c_h_deblock[i] &= ~MASK_C_TOP_ROW;
            if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
                c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
        }

        for(j = 0; j < 16; j += 4){
            Y = s->cur_pic.data[0] + mb_x*16 + (row*16 + j) * s->linesize;
            for(i = 0; i < 4; i++, Y += 4){
                int ij = i + j;
                int clip_cur = y_to_deblock & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
                int dither = j ? ij : i*4;

                // if bottom block is coded then we can filter its top edge
                // (or bottom edge of this block, which is the same)
                if(y_h_deblock & (MASK_BOTTOM << ij)){
                    rv40_adaptive_loop_filter(&r->rdsp, Y+4*s->linesize,
                                              s->linesize, dither,
                                              y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
                                              clip_cur, alpha, beta, betaY,
                                              0, 0, 0);
                }
                // filter left block edge in ordinary mode (with low filtering strength)
                if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
                    if(!i)
                        clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
                    else
                        clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
                    rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
                                              clip_cur,
                                              clip_left,
                                              alpha, beta, betaY, 0, 0, 1);
                }
                // filter top edge of the current macroblock when filtering strength is high
                if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
                    rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
                                       clip_cur,
                                       mvmasks[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0,
                                       alpha, beta, betaY, 0, 1, 0);
                }
                // filter left block edge in edge mode (with high filtering strength)
                if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
                    clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
                    rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
                                       clip_cur,
                                       clip_left,
                                       alpha, beta, betaY, 0, 1, 1);
                }
            }
        }
        for(k = 0; k < 2; k++){
            for(j = 0; j < 2; j++){
                C = s->cur_pic.data[k + 1] + mb_x*8 + (row*8 + j*4) * s->uvlinesize;
                for(i = 0; i < 2; i++, C += 4){
                    int ij = i + j*2;
                    int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
                    if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
                        int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
                        rv40_adaptive_loop_filter(&r->rdsp, C+4*s->uvlinesize, s->uvlinesize, i*8,
                                           clip_bot,
                                           clip_cur,
                                           alpha, beta, betaC, 1, 0, 0);
                    }
                    if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
                        if(!i)
                            clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
                        else
                            clip_left = c_to_deblock[k]    & (MASK_CUR << (ij-1))  ? clip[POS_CUR]  : 0;
                        rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
                                           clip_cur,
                                           clip_left,
                                           alpha, beta, betaC, 1, 0, 1);
                    }
                    if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
                        int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
                        rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, i*8,
                                           clip_cur,
                                           clip_top,
                                           alpha, beta, betaC, 1, 1, 0);
                    }
                    if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
                        clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
                        rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
                                           clip_cur,
                                           clip_left,
                                           alpha, beta, betaC, 1, 1, 1);
                    }
                }
            }
        }
    }
}

/**
 * Initialize decoder.
 */
static av_cold int rv40_decode_init(AVCodecContext *avctx)
{
    static AVOnce init_static_once = AV_ONCE_INIT;
    RV34DecContext *r = avctx->priv_data;
    int ret;

    r->rv30 = 0;
    if ((ret = ff_rv34_decode_init(avctx)) < 0)
        return ret;
    r->parse_slice_header = rv40_parse_slice_header;
    r->decode_intra_types = rv40_decode_intra_types;
    r->decode_mb_info     = rv40_decode_mb_info;
    r->loop_filter        = rv40_loop_filter;
    r->luma_dc_quant_i = rv40_luma_dc_quant[0];
    r->luma_dc_quant_p = rv40_luma_dc_quant[1];
    ff_rv40dsp_init(&r->rdsp);
    ff_thread_once(&init_static_once, rv40_init_tables);
    return 0;
}

const FFCodec ff_rv40_decoder = {
    .p.name                = "rv40",
    CODEC_LONG_NAME("RealVideo 4.0"),
    .p.type                = AVMEDIA_TYPE_VIDEO,
    .p.id                  = AV_CODEC_ID_RV40,
    .priv_data_size        = sizeof(RV34DecContext),
    .init                  = rv40_decode_init,
    .close                 = ff_rv34_decode_end,
    FF_CODEC_DECODE_CB(ff_rv34_decode_frame),
    .p.capabilities        = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
                             AV_CODEC_CAP_FRAME_THREADS,
    .caps_internal         = FF_CODEC_CAP_INIT_CLEANUP,
    .flush                 = ff_mpeg_flush,
    UPDATE_THREAD_CONTEXT(ff_rv34_decode_update_thread_context),
};

Coverage Report

Created: 2026-02-14 06:59

Line	Count	Source
1		/*
2		* RV40 decoder
3		* Copyright (c) 2007 Konstantin Shishkov
4		*
5		* This file is part of FFmpeg.
6		*
7		* FFmpeg is free software; you can redistribute it and/or
8		* modify it under the terms of the GNU Lesser General Public
9		* License as published by the Free Software Foundation; either
10		* version 2.1 of the License, or (at your option) any later version.
11		*
12		* FFmpeg is distributed in the hope that it will be useful,
13		* but WITHOUT ANY WARRANTY; without even the implied warranty of
14		* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15		* Lesser General Public License for more details.
16		*
17		* You should have received a copy of the GNU Lesser General Public
18		* License along with FFmpeg; if not, write to the Free Software
19		* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20		*/
21
22		/**
23		* @file
24		* RV40 decoder
25		*/
26
27		#include "config.h"
28
29		#include "libavutil/imgutils.h"
30		#include "libavutil/thread.h"
31
32		#include "avcodec.h"
33		#include "codec_internal.h"
34		#include "mpegutils.h"
35		#include "mpegvideo.h"
36		#include "mpegvideodec.h"
37		#include "golomb.h"
38
39		#include "rv34.h"
40		#include "rv40vlc2.h"
41		#include "rv40data.h"
42
43		static VLCElem aic_top_vlc[23590];
44		static const VLCElem aic_mode1_vlc[AIC_MODE1_NUM], aic_mode2_vlc[AIC_MODE2_NUM];
45		static const VLCElem ptype_vlc[NUM_PTYPE_VLCS], btype_vlc[NUM_BTYPE_VLCS];
46
47		static av_cold const VLCElem rv40_init_table(VLCInitState state, int nb_bits,
48		int nb_codes, const uint8_t (*tab)[2])
49	95	{
50	95	return ff_vlc_init_tables_from_lengths(state, nb_bits, nb_codes,
51	95	&tab[0][1], 2, &tab[0][0], 2, 1,
52	95	0, 0);
53	95	}
54
55		/**
56		* Initialize all tables.
57		*/
58		static av_cold void rv40_init_tables(void)
59	1	{
60	1	VLCInitState state = VLC_INIT_STATE(aic_top_vlc);
61	1	int i;
62
63	1	rv40_init_table(&state, AIC_TOP_BITS, AIC_TOP_SIZE,
64	1	rv40_aic_top_vlc_tab);
65	91	for(i = 0; i < AIC_MODE1_NUM; i++){
66		// Every tenth VLC table is empty
67	90	if((i % 10) == 9) continue;
68	81	aic_mode1_vlc[i] =
69	81	rv40_init_table(&state, AIC_MODE1_BITS,
70	81	AIC_MODE1_SIZE, aic_mode1_vlc_tabs[i]);
71	81	}
72	21	for (unsigned i = 0; i < AIC_MODE2_NUM; i++){
73	20	uint16_t syms[AIC_MODE2_SIZE];
74
75	1.64k	for (int j = 0; j < AIC_MODE2_SIZE; j++) {
76	1.62k	int first = aic_mode2_vlc_syms[i][j] >> 4;
77	1.62k	int second = aic_mode2_vlc_syms[i][j] & 0xF;
78	1.62k	if (HAVE_BIGENDIAN)
79	0	syms[j] = (first << 8) \| second;
80	1.62k	else
81	1.62k	syms[j] = first \| (second << 8);
82	1.62k	}
83	20	aic_mode2_vlc[i] =
84	20	ff_vlc_init_tables_from_lengths(&state, AIC_MODE2_BITS, AIC_MODE2_SIZE,
85	20	aic_mode2_vlc_bits[i], 1,
86	20	syms, 2, 2, 0, 0);
87	20	}
88	8	for(i = 0; i < NUM_PTYPE_VLCS; i++){
89	7	ptype_vlc[i] =
90	7	rv40_init_table(&state, PTYPE_VLC_BITS, PTYPE_VLC_SIZE,
91	7	ptype_vlc_tabs[i]);
92	7	}
93	7	for(i = 0; i < NUM_BTYPE_VLCS; i++){
94	6	btype_vlc[i] =
95	6	rv40_init_table(&state, BTYPE_VLC_BITS, BTYPE_VLC_SIZE,
96	6	btype_vlc_tabs[i]);
97	6	}
98	1	}
99
100		/**
101		* Get stored dimension from bitstream.
102		*
103		* If the width/height is the standard one then it's coded as a 3-bit index.
104		* Otherwise it is coded as escaped 8-bit portions.
105		*/
106		static int get_dimension(GetBitContext gb, const int dim)
107	380k	{
108	380k	int t = get_bits(gb, 3);
109	380k	int val = dim[t];
110	380k	if(val < 0)
111	9.95k	val = dim[get_bits1(gb) - val];
112	380k	if(!val){
113	742k	do{
114	742k	if (get_bits_left(gb) < 8)
115	457	return AVERROR_INVALIDDATA;
116	742k	t = get_bits(gb, 8);
117	742k	val += t << 2;
118	742k	}while(t == 0xFF);
119	14.6k	}
120	380k	return val;
121	380k	}
122
123		/**
124		* Get encoded picture size - usually this is called from rv40_parse_slice_header.
125		*/
126		static void rv40_parse_picture_size(GetBitContext gb, int w, int *h)
127	190k	{
128	190k	*w = get_dimension(gb, rv40_standard_widths);
129	190k	*h = get_dimension(gb, rv40_standard_heights);
130	190k	}
131
132		static int rv40_parse_slice_header(RV34DecContext r, GetBitContext gb, SliceInfo *si)
133	203k	{
134	203k	int w = r->s.width, h = r->s.height;
135	203k	int mb_size;
136	203k	int ret;
137
138	203k	memset(si, 0, sizeof(SliceInfo));
139	203k	if(get_bits1(gb))
140	2.91k	return AVERROR_INVALIDDATA;
141	200k	si->type = get_bits(gb, 2);
142	200k	if(si->type == 1) si->type = 0;
143	200k	si->quant = get_bits(gb, 5);
144	200k	if(get_bits(gb, 2))
145	3.01k	return AVERROR_INVALIDDATA;
146	197k	si->vlc_set = get_bits(gb, 2);
147	197k	skip_bits1(gb);
148	197k	si->pts = get_bits(gb, 13);
149	197k	if(!si->type \|\| !get_bits1(gb))
150	190k	rv40_parse_picture_size(gb, &w, &h);
151	197k	if ((ret = av_image_check_size(w, h, 0, r->s.avctx)) < 0)
152	1.09k	return ret;
153	196k	si->width = w;
154	196k	si->height = h;
155	196k	mb_size = ((w + 15) >> 4) * ((h + 15) >> 4);
156	196k	si->start = ff_rv34_get_start_offset(gb, mb_size);
157
158	196k	return 0;
159	197k	}
160
161		/**
162		* Decode 4x4 intra types array.
163		*/
164		static int rv40_decode_intra_types(RV34DecContext r, GetBitContext gb, int8_t *dst)
165	1.88M	{
166	1.88M	MpegEncContext *s = &r->s;
167	1.88M	int i, j, k, v;
168	1.88M	int A, B, C;
169	1.88M	int pattern;
170	1.88M	int8_t *ptr;
171
172	9.41M	for(i = 0; i < 4; i++, dst += r->intra_types_stride){
173	7.53M	if(!i && s->first_slice_line){
174	32.3k	pattern = get_vlc2(gb, aic_top_vlc, AIC_TOP_BITS, 1);
175	32.3k	dst[0] = (pattern >> 2) & 2;
176	32.3k	dst[1] = (pattern >> 1) & 2;
177	32.3k	dst[2] = pattern & 2;
178	32.3k	dst[3] = (pattern << 1) & 2;
179	32.3k	continue;
180	32.3k	}
181	7.50M	ptr = dst;
182	34.1M	for(j = 0; j < 4; j++){
183		/* Coefficients are read using VLC chosen by the prediction pattern
184		* The first one (used for retrieving a pair of coefficients) is
185		* constructed from the top, top right and left coefficients
186		* The second one (used for retrieving only one coefficient) is
187		* top + 10 * left.
188		*/
189	26.6M	A = ptr[-r->intra_types_stride + 1]; // it won't be used for the last coefficient in a row
190	26.6M	B = ptr[-r->intra_types_stride];
191	26.6M	C = ptr[-1];
192	26.6M	pattern = A + B * (1 << 4) + C * (1 << 8);
193	521M	for(k = 0; k < MODE2_PATTERNS_NUM; k++)
194	498M	if(pattern == rv40_aic_table_index[k])
195	3.53M	break;
196	26.6M	if(j < 3 && k < MODE2_PATTERNS_NUM){ //pattern is found, decoding 2 coefficients
197	3.40M	AV_WN16(ptr, get_vlc2(gb, aic_mode2_vlc[k], AIC_MODE2_BITS, 2));
198	3.40M	ptr += 2;
199	3.40M	j++;
200	23.2M	}else{
201	23.2M	if(B != -1 && C != -1)
202	22.4M	v = get_vlc2(gb, aic_mode1_vlc[B + C*10], AIC_MODE1_BITS, 1);
203	785k	else{ // tricky decoding
204	785k	v = 0;
205	785k	switch(C){
206	785k	case -1: // code 0 -> 1, 1 -> 0
207	785k	if(B < 2)
208	778k	v = get_bits1(gb) ^ 1;
209	785k	break;
210	0	case 0:
211	0	case 2: // code 0 -> 2, 1 -> 0
212	0	v = (get_bits1(gb) ^ 1) << 1;
213	0	break;
214	785k	}
215	785k	}
216	23.2M	*ptr++ = v;
217	23.2M	}
218	26.6M	}
219	7.50M	}
220	1.88M	return 0;
221	1.88M	}
222
223		/**
224		* Decode macroblock information.
225		*/
226		static int rv40_decode_mb_info(RV34DecContext *r)
227	3.68M	{
228	3.68M	MpegEncContext *s = &r->s;
229	3.68M	GetBitContext *const gb = &r->gb;
230	3.68M	int q, i;
231	3.68M	int prev_type = 0;
232	3.68M	int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
233
234	3.68M	if (!r->mb_skip_run) {
235	636k	r->mb_skip_run = get_interleaved_ue_golomb(gb) + 1;
236	636k	if (r->mb_skip_run > (unsigned)s->mb_num)
237	7.80k	return -1;
238	636k	}
239
240	3.67M	if (--r->mb_skip_run)
241	3.05M	return RV34_MB_SKIP;
242
243	627k	if(r->avail_cache[6-4]){
244	596k	int blocks[RV34_MB_TYPES] = {0};
245	596k	int count = 0;
246	596k	if(r->avail_cache[6-1])
247	552k	blocks[r->mb_type[mb_pos - 1]]++;
248	596k	blocks[r->mb_type[mb_pos - s->mb_stride]]++;
249	596k	if(r->avail_cache[6-2])
250	554k	blocks[r->mb_type[mb_pos - s->mb_stride + 1]]++;
251	596k	if(r->avail_cache[6-5])
252	552k	blocks[r->mb_type[mb_pos - s->mb_stride - 1]]++;
253	2.71M	for(i = 0; i < RV34_MB_TYPES; i++){
254	2.69M	if(blocks[i] > count){
255	684k	count = blocks[i];
256	684k	prev_type = i;
257	684k	if(count>1)
258	574k	break;
259	684k	}
260	2.69M	}
261	596k	} else if (r->avail_cache[6-1])
262	26.8k	prev_type = r->mb_type[mb_pos - 1];
263
264	627k	if(s->pict_type == AV_PICTURE_TYPE_P){
265	491k	prev_type = block_num_to_ptype_vlc_num[prev_type];
266	491k	q = get_vlc2(gb, ptype_vlc[prev_type], PTYPE_VLC_BITS, 1);
267	491k	if(q < PBTYPE_ESCAPE)
268	476k	return q;
269	14.3k	q = get_vlc2(gb, ptype_vlc[prev_type], PTYPE_VLC_BITS, 1);
270	14.3k	av_log(s->avctx, AV_LOG_ERROR, "Dquant for P-frame\n");
271	136k	}else{
272	136k	prev_type = block_num_to_btype_vlc_num[prev_type];
273	136k	q = get_vlc2(gb, btype_vlc[prev_type], BTYPE_VLC_BITS, 1);
274	136k	if(q < PBTYPE_ESCAPE)
275	132k	return q;
276	3.90k	q = get_vlc2(gb, btype_vlc[prev_type], BTYPE_VLC_BITS, 1);
277	3.90k	av_log(s->avctx, AV_LOG_ERROR, "Dquant for B-frame\n");
278	3.90k	}
279	18.2k	return 0;
280	627k	}
281
282		enum RV40BlockPos{
283		POS_CUR,
284		POS_TOP,
285		POS_LEFT,
286		POS_BOTTOM,
287		};
288
289	657M	#define MASK_CUR 0x0001
290	12.1M	#define MASK_RIGHT 0x0008
291	143M	#define MASK_BOTTOM 0x0010
292	10.9M	#define MASK_TOP 0x1000
293	9.94M	#define MASK_Y_TOP_ROW 0x000F
294	6.55M	#define MASK_Y_LAST_ROW 0xF000
295	7.10M	#define MASK_Y_LEFT_COL 0x1111
296	6.55M	#define MASK_Y_RIGHT_COL 0x8888
297	6.77M	#define MASK_C_TOP_ROW 0x0003
298	13.1M	#define MASK_C_LAST_ROW 0x000C
299	14.2M	#define MASK_C_LEFT_COL 0x0005
300	13.1M	#define MASK_C_RIGHT_COL 0x000A
301
302		static const int neighbour_offs_x[4] = { 0, 0, -1, 0 };
303		static const int neighbour_offs_y[4] = { 0, -1, 0, 1 };
304
305		static void rv40_adaptive_loop_filter(RV34DSPContext *rdsp,
306		uint8_t *src, int stride, int dmode,
307		int lim_q1, int lim_p1,
308		int alpha, int beta, int beta2,
309		int chroma, int edge, int dir)
310	141M	{
311	141M	int filter_p1, filter_q1;
312	141M	int strong;
313	141M	int lims;
314
315	141M	strong = rdsp->rv40_loop_filter_strength[dir](src, stride, beta, beta2,
316	141M	edge, &filter_p1, &filter_q1);
317
318	141M	lims = filter_p1 + filter_q1 + ((lim_q1 + lim_p1) >> 1) + 1;
319
320	141M	if (strong) {
321	26.8M	rdsp->rv40_strong_loop_filter[dir](src, stride, alpha,
322	26.8M	lims, dmode, chroma);
323	114M	} else if (filter_p1 & filter_q1) {
324	67.7M	rdsp->rv40_weak_loop_filter[dir](src, stride, 1, 1, alpha, beta,
325	67.7M	lims, lim_q1, lim_p1);
326	67.7M	} else if (filter_p1 \| filter_q1) {
327	4.71M	rdsp->rv40_weak_loop_filter[dir](src, stride, filter_p1, filter_q1,
328	4.71M	alpha, beta, lims >> 1, lim_q1 >> 1,
329	4.71M	lim_p1 >> 1);
330	4.71M	}
331	141M	}
332
333		/**
334		* RV40 loop filtering function
335		*/
336		static void rv40_loop_filter(RV34DecContext *r, int row)
337	550k	{
338	550k	MpegEncContext *s = &r->s;
339	550k	int mb_pos, mb_x;
340	550k	int i, j, k;
341	550k	uint8_t Y, C;
342	550k	int alpha, beta, betaY, betaC;
343	550k	int q;
344	550k	int mbtype[4]; ///< current macroblock and its neighbours types
345		/**
346		* flags indicating that macroblock can be filtered with strong filter
347		* it is set only for intra coded MB and MB with DCs coded separately
348		*/
349	550k	int mb_strong[4];
350	550k	int clip[4]; ///< MB filter clipping value calculated from filtering strength
351		/**
352		* coded block patterns for luma part of current macroblock and its neighbours
353		* Format:
354		* LSB corresponds to the top left block,
355		* each nibble represents one row of subblocks.
356		*/
357	550k	int cbp[4];
358		/**
359		* coded block patterns for chroma part of current macroblock and its neighbours
360		* Format is the same as for luma with two subblocks in a row.
361		*/
362	550k	int uvcbp[4][2];
363		/**
364		* This mask represents the pattern of luma subblocks that should be filtered
365		* in addition to the coded ones because they lie at the edge of
366		* 8x8 block with different enough motion vectors
367		*/
368	550k	unsigned mvmasks[4];
369
370	550k	mb_pos = row * s->mb_stride;
371	7.10M	for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
372	6.55M	int mbtype = s->cur_pic.mb_type[mb_pos];
373	6.55M	if(IS_INTRA(mbtype) \|\| IS_SEPARATE_DC(mbtype))
374	2.74M	r->cbp_luma [mb_pos] = r->deblock_coefs[mb_pos] = 0xFFFF;
375	6.55M	if(IS_INTRA(mbtype))
376	2.72M	r->cbp_chroma[mb_pos] = 0xFF;
377	6.55M	}
378	550k	mb_pos = row * s->mb_stride;
379	7.10M	for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
380	6.55M	int y_h_deblock, y_v_deblock;
381	6.55M	int c_v_deblock[2], c_h_deblock[2];
382	6.55M	int clip_left;
383	6.55M	int avail[4];
384	6.55M	unsigned y_to_deblock;
385	6.55M	int c_to_deblock[2];
386
387	6.55M	q = s->cur_pic.qscale_table[mb_pos];
388	6.55M	alpha = rv40_alpha_tab[q];
389	6.55M	beta = rv40_beta_tab [q];
390	6.55M	betaY = betaC = beta * 3;
391	6.55M	if(s->width * s->height <= 176*144)
392	277k	betaY += beta;
393
394	6.55M	avail[0] = 1;
395	6.55M	avail[1] = row;
396	6.55M	avail[2] = mb_x;
397	6.55M	avail[3] = row < s->mb_height - 1;
398	32.7M	for(i = 0; i < 4; i++){
399	26.2M	if(avail[i]){
400	25.0M	int pos = mb_pos + neighbour_offs_x[i] + neighbour_offs_y[i]*s->mb_stride;
401	25.0M	mvmasks[i] = r->deblock_coefs[pos];
402	25.0M	mbtype [i] = s->cur_pic.mb_type[pos];
403	25.0M	cbp [i] = r->cbp_luma[pos];
404	25.0M	uvcbp[i][0] = r->cbp_chroma[pos] & 0xF;
405	25.0M	uvcbp[i][1] = r->cbp_chroma[pos] >> 4;
406	25.0M	}else{
407	1.16M	mvmasks[i] = 0;
408	1.16M	mbtype [i] = mbtype[0];
409	1.16M	cbp [i] = 0;
410	1.16M	uvcbp[i][0] = uvcbp[i][1] = 0;
411	1.16M	}
412	26.2M	mb_strong[i] = IS_INTRA(mbtype[i]) \|\| IS_SEPARATE_DC(mbtype[i]);
413	26.2M	clip[i] = rv40_filter_clip_tbl[mb_strong[i] + 1][q];
414	26.2M	}
415	6.55M	y_to_deblock = mvmasks[POS_CUR]
416	6.55M	\| (mvmasks[POS_BOTTOM] << 16);
417		/* This pattern contains bits signalling that horizontal edges of
418		* the current block can be filtered.
419		* That happens when either of adjacent subblocks is coded or lies on
420		* the edge of 8x8 blocks with motion vectors differing by more than
421		* 3/4 pel in any component (any edge orientation for some reason).
422		*/
423	6.55M	y_h_deblock = y_to_deblock
424	6.55M	\| ((cbp[POS_CUR] << 4) & ~MASK_Y_TOP_ROW)
425	6.55M	\| ((cbp[POS_TOP] & MASK_Y_LAST_ROW) >> 12);
426		/* This pattern contains bits signalling that vertical edges of
427		* the current block can be filtered.
428		* That happens when either of adjacent subblocks is coded or lies on
429		* the edge of 8x8 blocks with motion vectors differing by more than
430		* 3/4 pel in any component (any edge orientation for some reason).
431		*/
432	6.55M	y_v_deblock = y_to_deblock
433	6.55M	\| ((cbp[POS_CUR] << 1) & ~MASK_Y_LEFT_COL)
434	6.55M	\| ((cbp[POS_LEFT] & MASK_Y_RIGHT_COL) >> 3);
435	6.55M	if(!mb_x)
436	550k	y_v_deblock &= ~MASK_Y_LEFT_COL;
437	6.55M	if(!row)
438	196k	y_h_deblock &= ~MASK_Y_TOP_ROW;
439	6.55M	if(row == s->mb_height - 1 \|\| (mb_strong[POS_CUR] \| mb_strong[POS_BOTTOM]))
440	3.19M	y_h_deblock &= ~(MASK_Y_TOP_ROW << 16);
441		/* Calculating chroma patterns is similar and easier since there is
442		* no motion vector pattern for them.
443		*/
444	19.6M	for(i = 0; i < 2; i++){
445	13.1M	c_to_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) \| uvcbp[POS_CUR][i];
446	13.1M	c_v_deblock[i] = c_to_deblock[i]
447	13.1M	\| ((uvcbp[POS_CUR] [i] << 1) & ~MASK_C_LEFT_COL)
448	13.1M	\| ((uvcbp[POS_LEFT][i] & MASK_C_RIGHT_COL) >> 1);
449	13.1M	c_h_deblock[i] = c_to_deblock[i]
450	13.1M	\| ((uvcbp[POS_TOP][i] & MASK_C_LAST_ROW) >> 2)
451	13.1M	\| (uvcbp[POS_CUR][i] << 2);
452	13.1M	if(!mb_x)
453	1.10M	c_v_deblock[i] &= ~MASK_C_LEFT_COL;
454	13.1M	if(!row)
455	392k	c_h_deblock[i] &= ~MASK_C_TOP_ROW;
456	13.1M	if(row == s->mb_height - 1 \|\| (mb_strong[POS_CUR] \| mb_strong[POS_BOTTOM]))
457	6.38M	c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
458	13.1M	}
459
460	32.7M	for(j = 0; j < 16; j += 4){
461	26.2M	Y = s->cur_pic.data[0] + mb_x16 + (row16 + j) * s->linesize;
462	131M	for(i = 0; i < 4; i++, Y += 4){
463	104M	int ij = i + j;
464	104M	int clip_cur = y_to_deblock & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
465	104M	int dither = j ? ij : i*4;
466
467		// if bottom block is coded then we can filter its top edge
468		// (or bottom edge of this block, which is the same)
469	104M	if(y_h_deblock & (MASK_BOTTOM << ij)){
470	38.2M	rv40_adaptive_loop_filter(&r->rdsp, Y+4*s->linesize,
471	38.2M	s->linesize, dither,
472	38.2M	y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
473	38.2M	clip_cur, alpha, beta, betaY,
474	38.2M	0, 0, 0);
475	38.2M	}
476		// filter left block edge in ordinary mode (with low filtering strength)
477	104M	if(y_v_deblock & (MASK_CUR << ij) && (i \|\| !(mb_strong[POS_CUR] \| mb_strong[POS_LEFT]))){
478	38.3M	if(!i)
479	2.01M	clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
480	36.3M	else
481	36.3M	clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
482	38.3M	rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
483	38.3M	clip_cur,
484	38.3M	clip_left,
485	38.3M	alpha, beta, betaY, 0, 0, 1);
486	38.3M	}
487		// filter top edge of the current macroblock when filtering strength is high
488	104M	if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] \| mb_strong[POS_TOP])){
489	10.9M	rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
490	10.9M	clip_cur,
491	10.9M	mvmasks[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0,
492	10.9M	alpha, beta, betaY, 0, 1, 0);
493	10.9M	}
494		// filter left block edge in edge mode (with high filtering strength)
495	104M	if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] \| mb_strong[POS_LEFT])){
496	10.1M	clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
497	10.1M	rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
498	10.1M	clip_cur,
499	10.1M	clip_left,
500	10.1M	alpha, beta, betaY, 0, 1, 1);
501	10.1M	}
502	104M	}
503	26.2M	}
504	19.6M	for(k = 0; k < 2; k++){
505	39.3M	for(j = 0; j < 2; j++){
506	26.2M	C = s->cur_pic.data[k + 1] + mb_x8 + (row8 + j4) s->uvlinesize;
507	78.7M	for(i = 0; i < 2; i++, C += 4){
508	52.4M	int ij = i + j*2;
509	52.4M	int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
510	52.4M	if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
511	11.4M	int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
512	11.4M	rv40_adaptive_loop_filter(&r->rdsp, C+4s->uvlinesize, s->uvlinesize, i8,
513	11.4M	clip_bot,
514	11.4M	clip_cur,
515	11.4M	alpha, beta, betaC, 1, 0, 0);
516	11.4M	}
517	52.4M	if((c_v_deblock[k] & (MASK_CUR << ij)) && (i \|\| !(mb_strong[POS_CUR] \| mb_strong[POS_LEFT]))){
518	11.4M	if(!i)
519	287k	clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
520	11.1M	else
521	11.1M	clip_left = c_to_deblock[k] & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
522	11.4M	rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
523	11.4M	clip_cur,
524	11.4M	clip_left,
525	11.4M	alpha, beta, betaC, 1, 0, 1);
526	11.4M	}
527	52.4M	if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] \| mb_strong[POS_TOP])){
528	10.8M	int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
529	10.8M	rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, i*8,
530	10.8M	clip_cur,
531	10.8M	clip_top,
532	10.8M	alpha, beta, betaC, 1, 1, 0);
533	10.8M	}
534	52.4M	if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] \| mb_strong[POS_LEFT])){
535	10.0M	clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
536	10.0M	rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
537	10.0M	clip_cur,
538	10.0M	clip_left,
539	10.0M	alpha, beta, betaC, 1, 1, 1);
540	10.0M	}
541	52.4M	}
542	26.2M	}
543	13.1M	}
544	6.55M	}
545	550k	}
546
547		/**
548		* Initialize decoder.
549		*/
550		static av_cold int rv40_decode_init(AVCodecContext *avctx)
551	7.55k	{
552	7.55k	static AVOnce init_static_once = AV_ONCE_INIT;
553	7.55k	RV34DecContext *r = avctx->priv_data;
554	7.55k	int ret;
555
556	7.55k	r->rv30 = 0;
557	7.55k	if ((ret = ff_rv34_decode_init(avctx)) < 0)
558	0	return ret;
559	7.55k	r->parse_slice_header = rv40_parse_slice_header;
560	7.55k	r->decode_intra_types = rv40_decode_intra_types;
561	7.55k	r->decode_mb_info = rv40_decode_mb_info;
562	7.55k	r->loop_filter = rv40_loop_filter;
563	7.55k	r->luma_dc_quant_i = rv40_luma_dc_quant[0];
564	7.55k	r->luma_dc_quant_p = rv40_luma_dc_quant[1];
565	7.55k	ff_rv40dsp_init(&r->rdsp);
566	7.55k	ff_thread_once(&init_static_once, rv40_init_tables);
567	7.55k	return 0;
568	7.55k	}
569
570		const FFCodec ff_rv40_decoder = {
571		.p.name = "rv40",
572		CODEC_LONG_NAME("RealVideo 4.0"),
573		.p.type = AVMEDIA_TYPE_VIDEO,
574		.p.id = AV_CODEC_ID_RV40,
575		.priv_data_size = sizeof(RV34DecContext),
576		.init = rv40_decode_init,
577		.close = ff_rv34_decode_end,
578		FF_CODEC_DECODE_CB(ff_rv34_decode_frame),
579		.p.capabilities = AV_CODEC_CAP_DR1 \| AV_CODEC_CAP_DELAY \|
580		AV_CODEC_CAP_FRAME_THREADS,
581		.caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
582		.flush = ff_mpeg_flush,
583		UPDATE_THREAD_CONTEXT(ff_rv34_decode_update_thread_context),
584		};