cavs.c

linux下的MPEG1

/*
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
 * Copyright (c) 2006  Stefan Gehrer <[e-mail]>
 *
 * 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 St, Fifth Floor, Boston, MA  02110-1301  USA
 */

/**
 * @file cavs.c
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
 * @author Stefan Gehrer <[e-mail]>
 */

#include "avcodec.h"
#include "bitstream.h"
#include "golomb.h"
#include "mpegvideo.h"
#include "cavsdata.h"

#ifdef CONFIG_CAVS_DECODER
typedef struct {
    MpegEncContext s;
    Picture picture; ///< currently decoded frame
    Picture DPB[2];  ///< reference frames
    int dist[2];     ///< temporal distances from current frame to ref frames
    int profile, level;
    int aspect_ratio;
    int mb_width, mb_height;
    int pic_type;
    int progressive;
    int pic_structure;
    int skip_mode_flag; ///< select between skip_count or one skip_flag per MB
    int loop_filter_disable;
    int alpha_offset, beta_offset;
    int ref_flag;
    int mbx, mby;      ///< macroblock coordinates
    int flags;         ///< availability flags of neighbouring macroblocks
    int stc;           ///< last start code
    uint8_t *cy, *cu, *cv; ///< current MB sample pointers
    int left_qp;
    uint8_t *top_qp;

    /** mv motion vector cache
       0:    D3  B2  B3  C2
       4:    A1  X0  X1   -
       8:    A3  X2  X3   -

       X are the vectors in the current macroblock (5,6,9,10)
       A is the macroblock to the left (4,8)
       B is the macroblock to the top (1,2)
       C is the macroblock to the top-right (3)
       D is the macroblock to the top-left (0)

       the same is repeated for backward motion vectors */
    vector_t mv[2*4*3];
    vector_t *top_mv[2];
    vector_t *col_mv;

    /** luma pred mode cache
       0:    --  B2  B3
       3:    A1  X0  X1
       6:    A3  X2  X3   */
    int pred_mode_Y[3*3];
    int *top_pred_Y;
    int l_stride, c_stride;
    int luma_scan[4];
    int qp;
    int qp_fixed;
    int cbp;
    ScanTable scantable;

    /** intra prediction is done with un-deblocked samples
     they are saved here before deblocking the MB  */
    uint8_t *top_border_y, *top_border_u, *top_border_v;
    uint8_t left_border_y[26], left_border_u[10], left_border_v[10];
    uint8_t intern_border_y[26];
    uint8_t topleft_border_y, topleft_border_u, topleft_border_v;

    void (*intra_pred_l[8])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
    void (*intra_pred_c[7])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
    uint8_t *col_type_base;
    uint8_t *col_type;

    /* scaling factors for MV prediction */
    int sym_factor;    ///< for scaling in symmetrical B block
    int direct_den[2]; ///< for scaling in direct B block
    int scale_den[2];  ///< for scaling neighbouring MVs

    int got_keyframe;
    DCTELEM *block;
} AVSContext;

/*****************************************************************************
 *
 * in-loop deblocking filter
 *
 ****************************************************************************/

static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {
    if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
        return 2;
    if( (abs(mvP->x - mvQ->x) >= 4) ||  (abs(mvP->y - mvQ->y) >= 4) )
        return 1;
    if(b){
        mvP += MV_BWD_OFFS;
        mvQ += MV_BWD_OFFS;
        if( (abs(mvP->x - mvQ->x) >= 4) ||  (abs(mvP->y - mvQ->y) >= 4) )
            return 1;
    }else{
        if(mvP->ref != mvQ->ref)
            return 1;
    }
    return 0;
}

#define SET_PARAMS                                            \
    alpha = alpha_tab[clip(qp_avg + h->alpha_offset,0,63)];   \
    beta  =  beta_tab[clip(qp_avg + h->beta_offset, 0,63)];   \
    tc    =    tc_tab[clip(qp_avg + h->alpha_offset,0,63)];

/**
 * in-loop deblocking filter for a single macroblock
 *
 * boundary strength (bs) mapping:
 *
 * --4---5--
 * 0   2   |
 * | 6 | 7 |
 * 1   3   |
 * ---------
 *
 */
static void filter_mb(AVSContext *h, enum mb_t mb_type) {
    DECLARE_ALIGNED_8(uint8_t, bs[8]);
    int qp_avg, alpha, beta, tc;
    int i;

    /* save un-deblocked lines */
    h->topleft_border_y = h->top_border_y[h->mbx*16+15];
    h->topleft_border_u = h->top_border_u[h->mbx*10+8];
    h->topleft_border_v = h->top_border_v[h->mbx*10+8];
    memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);
    memcpy(&h->top_border_u[h->mbx*10+1], h->cu +  7* h->c_stride,8);
    memcpy(&h->top_border_v[h->mbx*10+1], h->cv +  7* h->c_stride,8);
    for(i=0;i<8;i++) {
        h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+0)*h->l_stride);
        h->left_border_y[i*2+2] = *(h->cy + 15 + (i*2+1)*h->l_stride);
        h->left_border_u[i+1] = *(h->cu + 7 + i*h->c_stride);
        h->left_border_v[i+1] = *(h->cv + 7 + i*h->c_stride);
    }
    if(!h->loop_filter_disable) {
        /* determine bs */
        if(mb_type == I_8X8)
            *((uint64_t *)bs) = 0x0202020202020202ULL;
        else{
            *((uint64_t *)bs) = 0;
            if(partition_flags[mb_type] & SPLITV){
                bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
                bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
            }
            if(partition_flags[mb_type] & SPLITH){
                bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
                bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
            }
            bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
            bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
            bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
            bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
        }
        if( *((uint64_t *)bs) ) {
            if(h->flags & A_AVAIL) {
                qp_avg = (h->qp + h->left_qp + 1) >> 1;
                SET_PARAMS;
                h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
                h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
                h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
            }
            qp_avg = h->qp;
            SET_PARAMS;
            h->s.dsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
            h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,
                           bs[6],bs[7]);

            if(h->flags & B_AVAIL) {
                qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
                SET_PARAMS;
                h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
                h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
                h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
            }
        }
    }
    h->left_qp = h->qp;
    h->top_qp[h->mbx] = h->qp;
}

#undef SET_PARAMS

/*****************************************************************************
 *
 * spatial intra prediction
 *
 ****************************************************************************/

static inline void load_intra_pred_luma(AVSContext *h, uint8_t *top,
                                        uint8_t **left, int block) {
    int i;

    switch(block) {
    case 0:
        *left = h->left_border_y;
        h->left_border_y[0] = h->left_border_y[1];
        memset(&h->left_border_y[17],h->left_border_y[16],9);
        memcpy(&top[1],&h->top_border_y[h->mbx*16],16);
        top[17] = top[16];
        top[0] = top[1];
        if((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
            h->left_border_y[0] = top[0] = h->topleft_border_y;
        break;
    case 1:
        *left = h->intern_border_y;
        for(i=0;i<8;i++)
            h->intern_border_y[i+1] = *(h->cy + 7 + i*h->l_stride);
        memset(&h->intern_border_y[9],h->intern_border_y[8],9);
        h->intern_border_y[0] = h->intern_border_y[1];
        memcpy(&top[1],&h->top_border_y[h->mbx*16+8],8);
        if(h->flags & C_AVAIL)
            memcpy(&top[9],&h->top_border_y[(h->mbx + 1)*16],8);
        else
            memset(&top[9],top[8],9);
        top[17] = top[16];
        top[0] = top[1];
        if(h->flags & B_AVAIL)
            h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx*16+7];
        break;
    case 2:
        *left = &h->left_border_y[8];
        memcpy(&top[1],h->cy + 7*h->l_stride,16);
        top[17] = top[16];
        top[0] = top[1];
        if(h->flags & A_AVAIL)
            top[0] = h->left_border_y[8];
        break;
    case 3:
        *left = &h->intern_border_y[8];
        for(i=0;i<8;i++)
            h->intern_border_y[i+9] = *(h->cy + 7 + (i+8)*h->l_stride);
        memset(&h->intern_border_y[17],h->intern_border_y[16],9);
        memcpy(&top[0],h->cy + 7 + 7*h->l_stride,9);
        memset(&top[9],top[8],9);
        break;
    }
}

static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int y;
    uint64_t a = unaligned64(&top[1]);
    for(y=0;y<8;y++) {
        *((uint64_t *)(d+y*stride)) = a;
    }
}

static void intra_pred_horiz(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int y;
    uint64_t a;
    for(y=0;y<8;y++) {
        a = left[y+1] * 0x0101010101010101ULL;
        *((uint64_t *)(d+y*stride)) = a;
    }
}

static void intra_pred_dc_128(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int y;
    uint64_t a = 0x8080808080808080ULL;
    for(y=0;y<8;y++)
        *((uint64_t *)(d+y*stride)) = a;
}

static void intra_pred_plane(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y,ia;
    int ih = 0;
    int iv = 0;
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

    for(x=0; x<4; x++) {
        ih += (x+1)*(top[5+x]-top[3-x]);
        iv += (x+1)*(left[5+x]-left[3-x]);
    }
    ia = (top[8]+left[8])<<4;
    ih = (17*ih+16)>>5;
    iv = (17*iv+16)>>5;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = cm[(ia+(x-3)*ih+(y-3)*iv+16)>>5];
}

#define LOWPASS(ARRAY,INDEX)                                            \
    (( ARRAY[(INDEX)-1] + 2*ARRAY[(INDEX)] + ARRAY[(INDEX)+1] + 2) >> 2)

static void intra_pred_lp(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = (LOWPASS(top,x+1) + LOWPASS(left,y+1)) >> 1;
}

static void intra_pred_down_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = (LOWPASS(top,x+y+2) + LOWPASS(left,x+y+2)) >> 1;
}

static void intra_pred_down_right(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            if(x==y)
                d[y*stride+x] = (left[1]+2*top[0]+top[1]+2)>>2;
            else if(x>y)
                d[y*stride+x] = LOWPASS(top,x-y);
            else
                d[y*stride+x] = LOWPASS(left,y-x);
}

static void intra_pred_lp_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = LOWPASS(left,y+1);
}

static void intra_pred_lp_top(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = LOWPASS(top,x+1);
}

#undef LOWPASS

static inline void modify_pred(const int_fast8_t *mod_table, int *mode) {
    *mode = mod_table[*mode];
    if(*mode < 0) {
        av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
        *mode = 0;
    }
}

/*****************************************************************************
 *
 * motion compensation
 *
 ****************************************************************************/

static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
                        int chroma_height,int delta,int list,uint8_t *dest_y,
                        uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,
                        int src_y_offset,qpel_mc_func *qpix_op,
                        h264_chroma_mc_func chroma_op,vector_t *mv){
    MpegEncContext * const s = &h->s;
    const int mx= mv->x + src_x_offset*8;
    const int my= mv->y + src_y_offset*8;
    const int luma_xy= (mx&3) + ((my&3)<<2);
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->l_stride;
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->c_stride;
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->c_stride;
    int extra_width= 0; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
    int extra_height= extra_width;
    int emu=0;
    const int full_mx= mx>>2;
    const int full_my= my>>2;