new file mode 100644

@@ -0,0 +1,1550 @@

+/*

+ * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.

+ * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>

+ *

+ * This library 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 of the License, or (at your option) any later version.

+ *

+ * This library 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 this library; if not, write to the Free Software

+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA

+ */

+

+#include "avcodec.h"

+#include "bitstream.h"

+#include "golomb.h"

+#include "mpegvideo.h"

+#include "cavsdata.h"

+

+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;

+    int loop_filter_disable;

+    int alpha_offset, beta_offset;

+    int ref_flag;

+    int mbx, mby;

+    int flags;

+    int stc;

+    uint8_t *cy, *cu, *cv;

+    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;

+

+    /* 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[16], left_border_u[8], left_border_v[8];

+    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;

+    int sym_factor;

+    int direct_den[2];

+    int scale_den[2];

+    int got_keyframe;

+} AVSContext;

+

+/*****************************************************************************

+ *

+ * in-loop deblocking filter

+ *

+ ****************************************************************************/

+

+static inline int get_bs_p(vector_t *mvP, vector_t *mvQ) {

+    if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))

+        return 2;

+    if(mvP->ref != mvQ->ref)

+        return 1;

+    if( (abs(mvP->x - mvQ->x) >= 4) ||  (abs(mvP->y - mvQ->y) >= 4) )

+        return 1;

+    return 0;

+}

+

+static inline int get_bs_b(vector_t *mvP, vector_t *mvQ) {

+    if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA)) {

+        return 2;

+    } else {

+        vector_t *mvPbw = mvP + MV_BWD_OFFS;

+        vector_t *mvQbw = mvQ + MV_BWD_OFFS;

+        if( (abs(  mvP->x -   mvQ->x) >= 4) ||

+            (abs(  mvP->y -   mvQ->y) >= 4) ||

+            (abs(mvPbw->x - mvQbw->x) >= 4) ||

+            (abs(mvPbw->y - mvQbw->y) >= 4) )

+            return 1;

+    }

+    return 0;

+}

+

+/* boundary strength (bs) mapping:

+ *

+ * --4---5--

+ * 0   2   |

+ * | 6 | 7 |

+ * 1   3   |

+ * ---------

+ *

+ */

+

+#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)];

+

+static void filter_mb(AVSContext *h, enum mb_t mb_type) {

+    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*8+7];

+    h->topleft_border_v = h->top_border_v[h->mbx*8+7];

+    memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);

+    memcpy(&h->top_border_u[h->mbx* 8], h->cu +  7* h->c_stride,8);

+    memcpy(&h->top_border_v[h->mbx* 8], h->cv +  7* h->c_stride,8);

+    for(i=0;i<8;i++) {

+        h->left_border_y[i*2+0] = *(h->cy + 15 + (i*2+0)*h->l_stride);

+        h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+1)*h->l_stride);

+        h->left_border_u[i] = *(h->cu + 7 + i*h->c_stride);

+        h->left_border_v[i] = *(h->cv + 7 + i*h->c_stride);

+    }

+    if(!h->loop_filter_disable) {

+        /* clear bs */

+        *((uint64_t *)bs) = 0;

+        /* determine bs */

+        switch(mb_type) {

+        case I_8X8:

+            *((uint64_t *)bs) = 0x0202020202020202ULL;

+            break;

+        case P_8X8:

+        case P_8X16:

+            bs[2] = get_bs_p(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);

+            bs[3] = get_bs_p(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3]);

+        case P_16X8:

+            bs[6] = get_bs_p(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);

+            bs[7] = get_bs_p(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3]);

+        case P_16X16:

+        case P_SKIP:

+            bs[0] = get_bs_p(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);

+            bs[1] = get_bs_p(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);

+            bs[4] = get_bs_p(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);

+            bs[5] = get_bs_p(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);

+            break;

+        case B_SKIP:

+        case B_DIRECT:

+        case B_8X8:

+            bs[2] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);

+            bs[3] = get_bs_b(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3]);

+            bs[6] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);

+            bs[7] = get_bs_b(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3]);

+        case B_FWD_16X16:

+        case B_BWD_16X16:

+        case B_SYM_16X16:

+            bs[0] = get_bs_b(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);

+            bs[1] = get_bs_b(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);

+            bs[4] = get_bs_b(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);

+            bs[5] = get_bs_b(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);

+            break;

+        default:

+            if(mb_type & 1) { //16X8

+                bs[6] = bs[7] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);

+            } else {          //8X16

+                bs[2] = bs[3] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);

+            }

+            bs[0] = get_bs_b(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);

+            bs[1] = get_bs_b(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);

+            bs[4] = get_bs_b(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);

+            bs[5] = get_bs_b(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);

+        }

+        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:

+        memcpy(&left[1],h->left_border_y,16);

+        left[0] = left[1];

+        left[17] = left[16];

+        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))

+            left[0] = top[0] = h->topleft_border_y;

+        break;

+    case 1:

+        for(i=0;i<8;i++)

+            left[i+1] = *(h->cy + 7 + i*h->l_stride);

+        memset(&left[9],left[8],9);

+        left[0] = left[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)

+            left[0] = top[0] = h->top_border_y[h->mbx*16+7];

+        break;

+    case 2:

+        memcpy(&left[1],&h->left_border_y[8],8);

+        memset(&left[9],left[8],9);

+        memcpy(&top[1],h->cy + 7*h->l_stride,16);

+        top[17] = top[16];

+        left[0] = h->left_border_y[7];

+        top[0] = top[1];

+        if(h->flags & A_AVAIL)

+            top[0] = left[0];

+        break;

+    case 3:

+        for(i=0;i<9;i++)

+            left[i] = *(h->cy + 7 + (i+7)*h->l_stride);

+        memset(&left[9],left[8],9);

+        memcpy(&top[0],h->cy + 7 + 7*h->l_stride,9);

+        memset(&top[9],top[8],9);

+        break;

+    }

+}

+

+static inline void load_intra_pred_chroma(uint8_t *stop, uint8_t *sleft,

+                                          uint8_t stopleft, uint8_t *dtop,

+                                          uint8_t *dleft, int stride, int flags) {

+    int i;

+

+    if(flags & A_AVAIL) {

+        for(i=0; i<8; i++)

+            dleft[i+1] = sleft[i];

+        dleft[0] = dleft[1];

+        dleft[9] = dleft[8];

+    }

+    if(flags & B_AVAIL) {

+        for(i=0; i<8; i++)

+            dtop[i+1] = stop[i];

+        dtop[0] = dtop[1];

+        dtop[9] = dtop[8];

+        if(flags & A_AVAIL)

+            dleft[0] = dtop[0] = stopleft;

+    }

+}

+

+static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {

+    int y;

+    uint64_t a = *((uint64_t *)(&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 = 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 int8_t *mod_table, int *mode) {

+    int newmode = mod_table[(int)*mode];

+    if(newmode < 0) {

+        av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");

+        *mode = 0;

+    } else {

+        *mode = newmode;

+    }

+}

+

+/*****************************************************************************

+ *

+ * 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;

+    const int pic_width  = 16*h->mb_width;

+    const int pic_height = 16*h->mb_height;

+

+    if(!pic->data[0])

+        return;

+    if(mx&7) extra_width -= 3;

+    if(my&7) extra_height -= 3;

+

+    if(   full_mx < 0-extra_width

+          || full_my < 0-extra_height

+          || full_mx + 16/*FIXME*/ > pic_width + extra_width

+          || full_my + 16/*FIXME*/ > pic_height + extra_height){

+        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,

+                            16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);

+        src_y= s->edge_emu_buffer + 2 + 2*h->l_stride;

+        emu=1;

+    }

+

+    qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?

+    if(!square){

+        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);

+    }

+

+    if(emu){

+        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,

+                            9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

+        src_cb= s->edge_emu_buffer;

+    }

+    chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7);

+

+    if(emu){

+        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,

+                            9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

+        src_cr= s->edge_emu_buffer;

+    }

+    chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx&7, my&7);

+}

+

+static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int delta,

+                        uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,

+                        int x_offset, int y_offset,qpel_mc_func *qpix_put,

+                        h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,

+                        h264_chroma_mc_func chroma_avg, vector_t *mv){

+    qpel_mc_func *qpix_op=  qpix_put;

+    h264_chroma_mc_func chroma_op= chroma_put;

+

+    dest_y  += 2*x_offset + 2*y_offset*h->l_stride;

+    dest_cb +=   x_offset +   y_offset*h->c_stride;

+    dest_cr +=   x_offset +   y_offset*h->c_stride;

+    x_offset += 8*h->mbx;

+    y_offset += 8*h->mby;

+

+    if(mv->ref >= 0){

+        Picture *ref= &h->DPB[mv->ref];

+        mc_dir_part(h, ref, square, chroma_height, delta, 0,

+                    dest_y, dest_cb, dest_cr, x_offset, y_offset,

+                    qpix_op, chroma_op, mv);

+

+        qpix_op=  qpix_avg;

+        chroma_op= chroma_avg;

+    }

+

+    if((mv+MV_BWD_OFFS)->ref >= 0){

+        Picture *ref= &h->DPB[0];

+        mc_dir_part(h, ref, square, chroma_height, delta, 1,

+                    dest_y, dest_cb, dest_cr, x_offset, y_offset,

+                    qpix_op, chroma_op, mv+MV_BWD_OFFS);

+    }

+}

+

+static void inter_pred(AVSContext *h) {

+    /* always do 8x8 blocks TODO: are larger blocks worth it? */

+    mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,

+                h->s.dsp.put_cavs_qpel_pixels_tab[1],

+                h->s.dsp.put_h264_chroma_pixels_tab[1],

+                h->s.dsp.avg_cavs_qpel_pixels_tab[1],

+                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);

+    mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,

+                h->s.dsp.put_cavs_qpel_pixels_tab[1],

+                h->s.dsp.put_h264_chroma_pixels_tab[1],

+                h->s.dsp.avg_cavs_qpel_pixels_tab[1],

+                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);

+    mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,

+                h->s.dsp.put_cavs_qpel_pixels_tab[1],

+                h->s.dsp.put_h264_chroma_pixels_tab[1],

+                h->s.dsp.avg_cavs_qpel_pixels_tab[1],

+                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);

+    mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,

+                h->s.dsp.put_cavs_qpel_pixels_tab[1],

+                h->s.dsp.put_h264_chroma_pixels_tab[1],

+                h->s.dsp.avg_cavs_qpel_pixels_tab[1],

+                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);

+    /* set intra prediction modes to default values */

+    h->pred_mode_Y[3] =  h->pred_mode_Y[6] = INTRA_L_LP;

+    h->top_pred_Y[h->mbx*2+0] = h->top_pred_Y[h->mbx*2+1] = INTRA_L_LP;

+}

+

+/*****************************************************************************

+ *

+ * motion vector prediction

+ *

+ ****************************************************************************/

+

+static inline void veccpy(vector_t *dst, vector_t *src) {

+    *((uint64_t *)dst) = *((uint64_t *)src);

+}

+

+static inline void set_mvs(vector_t *mv, enum block_t size) {

+    switch(size) {

+    case BLK_16X16:

+        veccpy(mv+MV_STRIDE  ,mv);

+        veccpy(mv+MV_STRIDE+1,mv);

+    case BLK_16X8:

+        veccpy(mv          +1,mv);

+        break;

+    case BLK_8X16:

+        veccpy(mv+MV_STRIDE  ,mv);

+        break;

+    }

+}

+

+static inline void store_mvs(AVSContext *h) {

+    veccpy(&h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0], &h->mv[MV_FWD_X0]);

+    veccpy(&h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1], &h->mv[MV_FWD_X1]);

+    veccpy(&h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2], &h->mv[MV_FWD_X2]);

+    veccpy(&h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3], &h->mv[MV_FWD_X3]);

+}

+

+static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, vector_t *src, int distp) {

+    int den = h->scale_den[src->ref];

+

+    *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;

+    *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;

+}

+

+static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {

+    int ax, ay, bx, by, cx, cy;

+    int len_ab, len_bc, len_ca, len_mid;

+

+    /* scale candidates according to their temporal span */

+    scale_mv(h, &ax, &ay, mvA, mvP->dist);

+    scale_mv(h, &bx, &by, mvB, mvP->dist);

+    scale_mv(h, &cx, &cy, mvC, mvP->dist);

+    /* find the geometrical median of the three candidates */

+    len_ab = abs(ax - bx) + abs(ay - by);

+    len_bc = abs(bx - cx) + abs(by - cy);

+    len_ca = abs(cx - ax) + abs(cy - ay);

+    len_mid = mid_pred(len_ab, len_bc, len_ca);

+    if(len_mid == len_ab) {

+        mvP->x = cx;

+        mvP->y = cy;

+    } else if(len_mid == len_bc) {

+        mvP->x = ax;

+        mvP->y = ay;

+    } else {

+        mvP->x = bx;

+        mvP->y = by;

+    }

+}

+

+static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,

+                                  vector_t *pmv_bw, vector_t *col_mv) {

+    int den = h->direct_den[col_mv->ref];

+    int m = col_mv->x >> 31;

+

+    pmv_fw->dist = h->dist[1];

+    pmv_bw->dist = h->dist[0];

+    pmv_fw->ref = 1;

+    pmv_bw->ref = 0;

+    /* scale the co-located motion vector according to its temporal span */

+    pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;

+    pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);

+    m = col_mv->y >> 31;

+    pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;

+    pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);

+}

+

+static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {

+    vector_t *dst = src + MV_BWD_OFFS;

+

+    /* backward mv is the scaled and negated forward mv */

+    dst->x = -((src->x * h->sym_factor + 256) >> 9);

+    dst->y = -((src->y * h->sym_factor + 256) >> 9);

+    dst->ref = 0;

+    dst->dist = h->dist[0];

+    set_mvs(dst, size);

+}

+

+static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,

+                    enum mv_pred_t mode, enum block_t size, int ref) {

+    vector_t *mvP = &h->mv[nP];

+    vector_t *mvA = &h->mv[nP-1];

+    vector_t *mvB = &h->mv[nP-4];

+    vector_t *mvC = &h->mv[nC];

+    int mvAref = mvA->ref;

+    int mvBref = mvB->ref;

+    int mvCref;

+

+    mvP->ref = ref;

+    mvP->dist = h->dist[mvP->ref];

+    if(mvC->ref == NOT_AVAIL)

+        mvC = &h->mv[nP-5]; // set to top-left (mvD)

+    mvCref = mvC->ref;

+    if(mode == MV_PRED_PSKIP) {

+        if((mvAref == NOT_AVAIL) || (mvBref == NOT_AVAIL) ||

+           ((mvA->x | mvA->y | mvA->ref) == 0)  ||

+           ((mvB->x | mvB->y | mvB->ref) == 0) ) {

+            mvP->x = mvP->y = 0;

+            set_mvs(mvP,size);

+            return;

+        }

+    }

+    /* if there is only one suitable candidate, take it */

+    if((mvAref >= 0) && (mvBref < 0) && (mvCref < 0)) {

+        mvP->x = mvA->x;

+        mvP->y = mvA->y;

+    } else if((mvAref < 0) && (mvBref >= 0) && (mvCref < 0)) {

+        mvP->x = mvB->x;

+        mvP->y = mvB->y;

+    } else if((mvAref < 0) && (mvBref < 0) && (mvCref >= 0)) {

+        mvP->x = mvC->x;

+        mvP->y = mvC->y;

+    } else {

+        switch(mode) {

+        case MV_PRED_LEFT:

+            if(mvAref == mvP->ref) {

+                mvP->x = mvA->x;

+                mvP->y = mvA->y;

+            } else

+                mv_pred_median(h, mvP, mvA, mvB, mvC);

+            break;

+        case MV_PRED_TOP:

+            if(mvBref == mvP->ref) {

+                mvP->x = mvB->x;

+                mvP->y = mvB->y;

+            } else

+                mv_pred_median(h, mvP, mvA, mvB, mvC);

+            break;

+        case MV_PRED_TOPRIGHT:

+            if(mvCref == mvP->ref) {

+                mvP->x = mvC->x;

+                mvP->y = mvC->y;

+            } else

+                mv_pred_median(h, mvP, mvA, mvB, mvC);

+            break;

+        default:

+            mv_pred_median(h, mvP, mvA, mvB, mvC);

+            break;

+        }

+    }

+    if(mode < MV_PRED_PSKIP) {

+        mvP->x += get_se_golomb(&h->s.gb);

+        mvP->y += get_se_golomb(&h->s.gb);

+    }

+    set_mvs(mvP,size);

+}

+

+/*****************************************************************************

+ *

+ * residual data decoding

+ *

+ ****************************************************************************/

+

+/* kth-order exponential golomb code */

+static inline int get_ue_code(GetBitContext *gb, int order) {

+    if(order)

+        return (get_ue_golomb(gb) << order) + get_bits(gb,order);

+    return get_ue_golomb(gb);

+}

+

+static int decode_residual_block(AVSContext *h, GetBitContext *gb,

+                                 const residual_vlc_t *r, int esc_golomb_order,

+                                 int qp, uint8_t *dst, int stride) {

+    int i,pos = -1;

+    int level_code, esc_code, level, run, mask;

+    int level_buf[64];

+    int run_buf[64];

+    int dqm = dequant_mul[qp];

+    int dqs = dequant_shift[qp];

+    int dqa = 1 << (dqs - 1);

+    const uint8_t *scantab = ff_zigzag_direct;

+    DCTELEM block[64];

+

+    memset(block,0,64*sizeof(DCTELEM));

+    for(i=0;i<65;i++) {

+        level_code = get_ue_code(gb,r->golomb_order);

+        if(level_code >= ESCAPE_CODE) {

+            run = (level_code - ESCAPE_CODE) >> 1;

+            esc_code = get_ue_code(gb,esc_golomb_order);

+            level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);

+            while(level > r->inc_limit)

+                r++;

+            mask = -(level_code & 1);

+            level = (level^mask) - mask;

+        } else {

+            if(level_code < 0)

+                return -1;

+            level = r->rltab[level_code][0];

+            if(!level) //end of block signal

+                break;

+            run   = r->rltab[level_code][1];

+            r += r->rltab[level_code][2];

+        }

+        level_buf[i] = level;

+        run_buf[i] = run;

+    }

+    /* inverse scan and dequantization */

+    for(i=i-1;i>=0;i--) {

+        pos += 1 + run_buf[i];

+        if(pos > 63) {

+            av_log(h->s.avctx, AV_LOG_ERROR,

+                   "position out of block bounds at pic %d MB(%d,%d)\n",

+                   h->picture.poc, h->mbx, h->mby);

+            return -1;

+        }

+        block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;

+    }

+    h->s.dsp.cavs_idct8_add(dst,block,stride);

+    return 0;

+}

+

+

+static inline void decode_residual_chroma(AVSContext *h) {

+    if(h->cbp & (1<<4))

+        decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],

+                              h->cu,h->c_stride);

+    if(h->cbp & (1<<5))

+        decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],

+                              h->cv,h->c_stride);

+}

+

+static inline void decode_residual_inter(AVSContext *h) {

+    int block;

+

+    /* get coded block pattern */

+    h->cbp = cbp_tab[get_ue_golomb(&h->s.gb)][1];

+    /* get quantizer */

+    if(h->cbp && !h->qp_fixed)

+        h->qp += get_se_golomb(&h->s.gb);

+    for(block=0;block<4;block++)

+        if(h->cbp & (1<<block))

+            decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,

+                                  h->cy + h->luma_scan[block], h->l_stride);

+    decode_residual_chroma(h);

+}

+

+/*****************************************************************************

+ *

+ * macroblock level

+ *

+ ****************************************************************************/

+

+static inline void init_mb(AVSContext *h) {

+    int i;

+

+    /* copy predictors from top line (MB B and C) into cache */

+    for(i=0;i<3;i++) {

+        veccpy(&h->mv[MV_FWD_B2+i],&h->top_mv[0][h->mbx*2+i]);

+        veccpy(&h->mv[MV_BWD_B2+i],&h->top_mv[1][h->mbx*2+i]);

+    }

+    h->pred_mode_Y[1] = h->top_pred_Y[h->mbx*2+0];

+    h->pred_mode_Y[2] = h->top_pred_Y[h->mbx*2+1];

+    /* clear top predictors if MB B is not available */

+    if(!(h->flags & B_AVAIL)) {

+        veccpy(&h->mv[MV_FWD_B2],(vector_t *)&un_mv);

+        veccpy(&h->mv[MV_FWD_B3],(vector_t *)&un_mv);

+        veccpy(&h->mv[MV_BWD_B2],(vector_t *)&un_mv);

+        veccpy(&h->mv[MV_BWD_B3],(vector_t *)&un_mv);

+        h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;

+        h->flags &= ~(C_AVAIL|D_AVAIL);

+    } else if(h->mbx) {

+        h->flags |= D_AVAIL;

+    }

+    if(h->mbx == h->mb_width-1) //MB C not available

+        h->flags &= ~C_AVAIL;

+    /* clear top-right predictors if MB C is not available */

+    if(!(h->flags & C_AVAIL)) {

+        veccpy(&h->mv[MV_FWD_C2],(vector_t *)&un_mv);

+        veccpy(&h->mv[MV_BWD_C2],(vector_t *)&un_mv);

+    }

+    /* clear top-left predictors if MB D is not available */

+    if(!(h->flags & D_AVAIL)) {

+        veccpy(&h->mv[MV_FWD_D3],(vector_t *)&un_mv);

+        veccpy(&h->mv[MV_BWD_D3],(vector_t *)&un_mv);

+    }

+    /* set pointer for co-located macroblock type */

+    h->col_type = &h->col_type_base[h->mby*h->mb_width + h->mbx];

+}

+

+static inline void check_for_slice(AVSContext *h);

+

+static inline int next_mb(AVSContext *h) {

+    int i;

+

+    h->flags |= A_AVAIL;

+    h->cy += 16;

+    h->cu += 8;

+    h->cv += 8;

+    /* copy mvs as predictors to the left */

+    for(i=0;i<=20;i+=4)

+        veccpy(&h->mv[i],&h->mv[i+2]);

+    /* copy bottom mvs from cache to top line */

+    veccpy(&h->top_mv[0][h->mbx*2+0],&h->mv[MV_FWD_X2]);

+    veccpy(&h->top_mv[0][h->mbx*2+1],&h->mv[MV_FWD_X3]);

+    veccpy(&h->top_mv[1][h->mbx*2+0],&h->mv[MV_BWD_X2]);

+    veccpy(&h->top_mv[1][h->mbx*2+1],&h->mv[MV_BWD_X3]);

+    /* next MB address */

+    h->mbx++;

+    if(h->mbx == h->mb_width) { //new mb line

+        h->flags = B_AVAIL|C_AVAIL;

+        /* clear left pred_modes */

+        h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;

+        /* clear left mv predictors */

+        for(i=0;i<=20;i+=4)

+            veccpy(&h->mv[i],(vector_t *)&un_mv);

+        h->mbx = 0;

+        h->mby++;

+        /* re-calculate sample pointers */

+        h->cy = h->picture.data[0] + h->mby*16*h->l_stride;

+        h->cu = h->picture.data[1] + h->mby*8*h->c_stride;

+        h->cv = h->picture.data[2] + h->mby*8*h->c_stride;

+        if(h->mby == h->mb_height) { //frame end

+            return 0;

+        } else {

+            //check_for_slice(h);

+        }

+    }

+    return 1;

+}

+

+static void decode_mb_i(AVSContext *h, int is_i_pic) {

+    GetBitContext *gb = &h->s.gb;

+    int block, pred_mode_uv;

+    uint8_t top[18];

+    uint8_t left[18];

+    uint8_t *d;

+

+    /* get intra prediction modes from stream */

+    for(block=0;block<4;block++) {

+        int nA,nB,predpred;

+        int pos = scan3x3[block];

+

+        nA = h->pred_mode_Y[pos-1];

+        nB = h->pred_mode_Y[pos-3];

+        if((nA == NOT_AVAIL) || (nB == NOT_AVAIL))

+            predpred = 2;

+        else

+            predpred = FFMIN(nA,nB);

+        if(get_bits1(gb))

+            h->pred_mode_Y[pos] = predpred;

+        else {