Add support for AMR-WB encoding via libvo-amrwbenc
[ffmpeg.git] / libavcodec / cavs.c
index bd5620f..9a73e2a 100644 (file)
@@ -2,32 +2,33 @@
  * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
  * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav 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,
+ * Libav 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
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 /**
- * @file cavs.c
+ * @file
  * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
  * @author Stefan Gehrer <stefan.gehrer@gmx.de>
  */
 
 #include "avcodec.h"
-#include "bitstream.h"
+#include "get_bits.h"
 #include "golomb.h"
+#include "mathops.h"
 #include "cavs.h"
 #include "cavsdata.h"
 
@@ -37,7 +38,7 @@
  *
  ****************************************************************************/
 
-static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {
+static inline int get_bs(cavs_vector *mvP, cavs_vector *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) )
@@ -71,8 +72,8 @@ static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {
  * ---------
  *
  */
-static void filter_mb(AVSContext *h, enum mb_t mb_type) {
-    DECLARE_ALIGNED_8(uint8_t, bs[8]);
+void ff_cavs_filter(AVSContext *h, enum cavs_mb mb_type) {
+    uint8_t bs[8];
     int qp_avg, alpha, beta, tc;
     int i;
 
@@ -92,14 +93,14 @@ static void filter_mb(AVSContext *h, enum mb_t mb_type) {
     if(!h->loop_filter_disable) {
         /* determine bs */
         if(mb_type == I_8X8)
-            *((uint64_t *)bs) = 0x0202020202020202ULL;
+            memset(bs,2,8);
         else{
-            *((uint64_t *)bs) = 0;
-            if(partition_flags[mb_type] & SPLITV){
+            memset(bs,0,8);
+            if(ff_cavs_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){
+            if(ff_cavs_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);
             }
@@ -108,26 +109,26 @@ static void filter_mb(AVSContext *h, enum mb_t mb_type) {
             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(AV_RN64(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]);
+                h->cdsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
+                h->cdsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
+                h->cdsp.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,
+            h->cdsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
+            h->cdsp.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->cdsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
+                h->cdsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
+                h->cdsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
             }
         }
     }
@@ -143,9 +144,76 @@ static void filter_mb(AVSContext *h, enum mb_t mb_type) {
  *
  ****************************************************************************/
 
+void ff_cavs_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;
+    }
+}
+
+void ff_cavs_load_intra_pred_chroma(AVSContext *h) {
+    /* extend borders by one pixel */
+    h->left_border_u[9] = h->left_border_u[8];
+    h->left_border_v[9] = h->left_border_v[8];
+    h->top_border_u[h->mbx*10+9] = h->top_border_u[h->mbx*10+8];
+    h->top_border_v[h->mbx*10+9] = h->top_border_v[h->mbx*10+8];
+    if(h->mbx && h->mby) {
+        h->top_border_u[h->mbx*10] = h->left_border_u[0] = h->topleft_border_u;
+        h->top_border_v[h->mbx*10] = h->left_border_v[0] = h->topleft_border_v;
+    } else {
+        h->left_border_u[0] = h->left_border_u[1];
+        h->left_border_v[0] = h->left_border_v[1];
+        h->top_border_u[h->mbx*10] = h->top_border_u[h->mbx*10+1];
+        h->top_border_v[h->mbx*10] = h->top_border_v[h->mbx*10+1];
+    }
+}
+
 static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
     int y;
-    uint64_t a = unaligned64(&top[1]);
+    uint64_t a = AV_RN64(&top[1]);
     for(y=0;y<8;y++) {
         *((uint64_t *)(d+y*stride)) = a;
     }
@@ -230,6 +298,26 @@ static void intra_pred_lp_top(uint8_t *d,uint8_t *top,uint8_t *left,int stride)
 
 #undef LOWPASS
 
+void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv) {
+    /* save pred modes before they get modified */
+    h->pred_mode_Y[3] =  h->pred_mode_Y[5];
+    h->pred_mode_Y[6] =  h->pred_mode_Y[8];
+    h->top_pred_Y[h->mbx*2+0] = h->pred_mode_Y[7];
+    h->top_pred_Y[h->mbx*2+1] = h->pred_mode_Y[8];
+
+    /* modify pred modes according to availability of neighbour samples */
+    if(!(h->flags & A_AVAIL)) {
+        modify_pred(ff_left_modifier_l, &h->pred_mode_Y[4] );
+        modify_pred(ff_left_modifier_l, &h->pred_mode_Y[7] );
+        modify_pred(ff_left_modifier_c, pred_mode_uv );
+    }
+    if(!(h->flags & B_AVAIL)) {
+        modify_pred(ff_top_modifier_l, &h->pred_mode_Y[4] );
+        modify_pred(ff_top_modifier_l, &h->pred_mode_Y[5] );
+        modify_pred(ff_top_modifier_c, pred_mode_uv );
+    }
+}
+
 /*****************************************************************************
  *
  * motion compensation
@@ -240,7 +328,7 @@ 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){
+                        h264_chroma_mc_func chroma_op,cavs_vector *mv){
     MpegEncContext * const s = &h->s;
     const int mx= mv->x + src_x_offset*8;
     const int my= mv->y + src_y_offset*8;
@@ -265,7 +353,7 @@ static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
           || 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,
+        s->dsp.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;
@@ -277,14 +365,14 @@ static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
     }
 
     if(emu){
-        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,
+        s->dsp.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,
+        s->dsp.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;
     }
@@ -295,7 +383,7 @@ static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int de
                         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){
+                        h264_chroma_mc_func chroma_avg, cavs_vector *mv){
     qpel_mc_func *qpix_op=  qpix_put;
     h264_chroma_mc_func chroma_op= chroma_put;
 
@@ -323,33 +411,33 @@ static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int de
     }
 }
 
-static void inter_pred(AVSContext *h, enum mb_t mb_type) {
-    if(partition_flags[mb_type] == 0){ // 16x16
+void ff_cavs_inter(AVSContext *h, enum cavs_mb mb_type) {
+    if(ff_cavs_partition_flags[mb_type] == 0){ // 16x16
         mc_part_std(h, 1, 8, 0, h->cy, h->cu, h->cv, 0, 0,
-                h->s.dsp.put_cavs_qpel_pixels_tab[0],
+                h->cdsp.put_cavs_qpel_pixels_tab[0],
                 h->s.dsp.put_h264_chroma_pixels_tab[0],
-                h->s.dsp.avg_cavs_qpel_pixels_tab[0],
+                h->cdsp.avg_cavs_qpel_pixels_tab[0],
                 h->s.dsp.avg_h264_chroma_pixels_tab[0],&h->mv[MV_FWD_X0]);
     }else{
         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->cdsp.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->cdsp.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->cdsp.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->cdsp.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->cdsp.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->cdsp.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->cdsp.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->cdsp.avg_cavs_qpel_pixels_tab[1],
                 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);
     }
 }
@@ -360,21 +448,15 @@ static void inter_pred(AVSContext *h, enum mb_t mb_type) {
  *
  ****************************************************************************/
 
-static inline void store_mvs(AVSContext *h) {
-    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];
-    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];
-    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];
-    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) {
+static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, cavs_vector *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) {
+static inline void mv_pred_median(AVSContext *h, cavs_vector *mvP,
+                        cavs_vector *mvA, cavs_vector *mvB, cavs_vector *mvC) {
     int ax, ay, bx, by, cx, cy;
     int len_ab, len_bc, len_ca, len_mid;
 
@@ -399,42 +481,13 @@ static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, v
     }
 }
 
-static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,
-                                  vector_t *col_mv) {
-    vector_t *pmv_bw = pmv_fw + MV_BWD_OFFS;
-    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];
-    const vector_t *mvP2 = NULL;
+void ff_cavs_mv(AVSContext *h, enum cavs_mv_loc nP, enum cavs_mv_loc nC,
+                enum cavs_mv_pred mode, enum cavs_block size, int ref) {
+    cavs_vector *mvP = &h->mv[nP];
+    cavs_vector *mvA = &h->mv[nP-1];
+    cavs_vector *mvB = &h->mv[nP-4];
+    cavs_vector *mvC = &h->mv[nC];
+    const cavs_vector *mvP2 = NULL;
 
     mvP->ref = ref;
     mvP->dist = h->dist[mvP->ref];
@@ -474,352 +527,89 @@ static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,
 
 /*****************************************************************************
  *
- * residual data decoding
+ * macroblock level
  *
  ****************************************************************************/
 
-/** kth-order exponential golomb code */
-static inline int get_ue_code(GetBitContext *gb, int order) {
-    if(order) {
-        int ret = get_ue_golomb(gb) << order;
-        return ret + get_bits(gb,order);
-    }
-    return get_ue_golomb(gb);
-}
-
 /**
- * decode coefficients from one 8x8 block, dequantize, inverse transform
- *  and add them to sample block
- * @param r pointer to 2D VLC table
- * @param esc_golomb_order escape codes are k-golomb with this order k
- * @param qp quantizer
- * @param dst location of sample block
- * @param stride line stride in frame buffer
+ * initialise predictors for motion vectors and intra prediction
  */
-static int decode_residual_block(AVSContext *h, GetBitContext *gb,
-                                 const dec_2dvlc_t *r, int esc_golomb_order,
-                                 int qp, uint8_t *dst, int stride) {
-    int i, level_code, esc_code, level, run, mask;
-    DCTELEM level_buf[64];
-    uint8_t run_buf[64];
-    DCTELEM *block = h->block;
-
-    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) + 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 {
-            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;
-    }
-    if(dequant(h,level_buf, run_buf, block, dequant_mul[qp],
-               dequant_shift[qp], i))
-        return -1;
-    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_dec,0, chroma_qp[h->qp],
-                              h->cu,h->c_stride);
-    if(h->cbp & (1<<5))
-        decode_residual_block(h,&h->s.gb,chroma_dec,0, chroma_qp[h->qp],
-                              h->cv,h->c_stride);
-}
-
-static inline int decode_residual_inter(AVSContext *h) {
-    int block;
-
-    /* get coded block pattern */
-    int cbp= get_ue_golomb(&h->s.gb);
-    if(cbp > 63){
-        av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");
-        return -1;
-    }
-    h->cbp = cbp_tab[cbp][1];
-
-    /* get quantizer */
-    if(h->cbp && !h->qp_fixed)
-        h->qp = (h->qp + get_se_golomb(&h->s.gb)) & 63;
-    for(block=0;block<4;block++)
-        if(h->cbp & (1<<block))
-            decode_residual_block(h,&h->s.gb,inter_dec,0,h->qp,
-                                  h->cy + h->luma_scan[block], h->l_stride);
-    decode_residual_chroma(h);
-
-    return 0;
-}
-
-/*****************************************************************************
- *
- * macroblock level
- *
- ****************************************************************************/
-
-static int decode_mb_i(AVSContext *h, int cbp_code) {
-    GetBitContext *gb = &h->s.gb;
-    int block, pred_mode_uv;
-    uint8_t top[18];
-    uint8_t *left = NULL;
-    uint8_t *d;
-
-    init_mb(h);
-
-    /* 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];
-        predpred = FFMIN(nA,nB);
-        if(predpred == NOT_AVAIL) // if either is not available
-            predpred = INTRA_L_LP;
-        if(!get_bits1(gb)){
-            int rem_mode= get_bits(gb, 2);
-            predpred = rem_mode + (rem_mode >= predpred);
-        }
-        h->pred_mode_Y[pos] = predpred;
-    }
-    pred_mode_uv = get_ue_golomb(gb);
-    if(pred_mode_uv > 6) {
-        av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");
-        return -1;
-    }
-    modify_mb_i(h, &pred_mode_uv);
-
-    /* get coded block pattern */
-    if(h->pic_type == FF_I_TYPE)
-        cbp_code = get_ue_golomb(gb);
-    if(cbp_code > 63){
-        av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");
-        return -1;
-    }
-    h->cbp = cbp_tab[cbp_code][0];
-    if(h->cbp && !h->qp_fixed)
-        h->qp = (h->qp + get_se_golomb(gb)) & 63; //qp_delta
-
-    /* luma intra prediction interleaved with residual decode/transform/add */
-    for(block=0;block<4;block++) {
-        d = h->cy + h->luma_scan[block];
-        load_intra_pred_luma(h, top, &left, block);
-        h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
-            (d, top, left, h->l_stride);
-        if(h->cbp & (1<<block))
-            decode_residual_block(h,gb,intra_dec,1,h->qp,d,h->l_stride);
-    }
-
-    /* chroma intra prediction */
-    load_intra_pred_chroma(h);
-    h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
-                                  h->left_border_u, h->c_stride);
-    h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
-                                  h->left_border_v, h->c_stride);
-
-    decode_residual_chroma(h);
-    filter_mb(h,I_8X8);
-    set_mv_intra(h);
-    return 0;
-}
-
-static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
-    GetBitContext *gb = &h->s.gb;
-    int ref[4];
-
-    init_mb(h);
-    switch(mb_type) {
-    case P_SKIP:
-        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
-        break;
-    case P_16X16:
-        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
-        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_16X16,ref[0]);
-        break;
-    case P_16X8:
-        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
-        ref[2] = h->ref_flag ? 0 : get_bits1(gb);
-        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,      BLK_16X8, ref[0]);
-        mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT,     BLK_16X8, ref[2]);
-        break;
-    case P_8X16:
-        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
-        ref[1] = h->ref_flag ? 0 : get_bits1(gb);
-        mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT,     BLK_8X16, ref[0]);
-        mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
-        break;
-    case P_8X8:
-        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
-        ref[1] = h->ref_flag ? 0 : get_bits1(gb);
-        ref[2] = h->ref_flag ? 0 : get_bits1(gb);
-        ref[3] = h->ref_flag ? 0 : get_bits1(gb);
-        mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN,   BLK_8X8, ref[0]);
-        mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_8X8, ref[1]);
-        mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN,   BLK_8X8, ref[2]);
-        mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN,   BLK_8X8, ref[3]);
-    }
-    inter_pred(h, mb_type);
-    set_intra_mode_default(h);
-    store_mvs(h);
-    if(mb_type != P_SKIP)
-        decode_residual_inter(h);
-    filter_mb(h,mb_type);
-    *h->col_type = mb_type;
-}
-
-static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
-    int block;
-    enum sub_mb_t sub_type[4];
-    int flags;
-
-    init_mb(h);
+void ff_cavs_init_mb(AVSContext *h) {
+    int i;
 
-    /* reset all MVs */
-    h->mv[MV_FWD_X0] = ff_cavs_dir_mv;
-    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
-    h->mv[MV_BWD_X0] = ff_cavs_dir_mv;
-    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
-    switch(mb_type) {
-    case B_SKIP:
-    case B_DIRECT:
-        if(!(*h->col_type)) {
-            /* intra MB at co-location, do in-plane prediction */
-            mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
-            mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
-        } else
-            /* direct prediction from co-located P MB, block-wise */
-            for(block=0;block<4;block++)
-                mv_pred_direct(h,&h->mv[mv_scan[block]],
-                            &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);
-        break;
-    case B_FWD_16X16:
-        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
-        break;
-    case B_SYM_16X16:
-        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
-        mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
-        break;
-    case B_BWD_16X16:
-        mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
-        break;
-    case B_8X8:
-        for(block=0;block<4;block++)
-            sub_type[block] = get_bits(&h->s.gb,2);
-        for(block=0;block<4;block++) {
-            switch(sub_type[block]) {
-            case B_SUB_DIRECT:
-                if(!(*h->col_type)) {
-                    /* intra MB at co-location, do in-plane prediction */
-                    mv_pred(h, mv_scan[block], mv_scan[block]-3,
-                            MV_PRED_BSKIP, BLK_8X8, 1);
-                    mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
-                            mv_scan[block]-3+MV_BWD_OFFS,
-                            MV_PRED_BSKIP, BLK_8X8, 0);
-                } else
-                    mv_pred_direct(h,&h->mv[mv_scan[block]],
-                                   &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
-                break;
-            case B_SUB_FWD:
-                mv_pred(h, mv_scan[block], mv_scan[block]-3,
-                        MV_PRED_MEDIAN, BLK_8X8, 1);
-                break;
-            case B_SUB_SYM:
-                mv_pred(h, mv_scan[block], mv_scan[block]-3,
-                        MV_PRED_MEDIAN, BLK_8X8, 1);
-                mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
-                break;
-            }
-        }
-        for(block=0;block<4;block++) {
-            if(sub_type[block] == B_SUB_BWD)
-                mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
-                        mv_scan[block]+MV_BWD_OFFS-3,
-                        MV_PRED_MEDIAN, BLK_8X8, 0);
-        }
-        break;
-    default:
-        assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
-        flags = partition_flags[mb_type];
-        if(mb_type & 1) { /* 16x8 macroblock types */
-            if(flags & FWD0)
-                mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,  BLK_16X8, 1);
-            if(flags & SYM0)
-                mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
-            if(flags & FWD1)
-                mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
-            if(flags & SYM1)
-                mv_pred_sym(h, &h->mv[MV_FWD_X2], BLK_16X8);
-            if(flags & BWD0)
-                mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP,  BLK_16X8, 0);
-            if(flags & BWD1)
-                mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
-        } else {          /* 8x16 macroblock types */
-            if(flags & FWD0)
-                mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
-            if(flags & SYM0)
-                mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
-            if(flags & FWD1)
-                mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
-            if(flags & SYM1)
-                mv_pred_sym(h, &h->mv[MV_FWD_X1], BLK_8X16);
-            if(flags & BWD0)
-                mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
-            if(flags & BWD1)
-                mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
-        }
+    /* copy predictors from top line (MB B and C) into cache */
+    for(i=0;i<3;i++) {
+        h->mv[MV_FWD_B2+i] = h->top_mv[0][h->mbx*2+i];
+        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)) {
+        h->mv[MV_FWD_B2] = ff_cavs_un_mv;
+        h->mv[MV_FWD_B3] = ff_cavs_un_mv;
+        h->mv[MV_BWD_B2] = ff_cavs_un_mv;
+        h->mv[MV_BWD_B3] = ff_cavs_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)) {
+        h->mv[MV_FWD_C2] = ff_cavs_un_mv;
+        h->mv[MV_BWD_C2] = ff_cavs_un_mv;
+    }
+    /* clear top-left predictors if MB D is not available */
+    if(!(h->flags & D_AVAIL)) {
+        h->mv[MV_FWD_D3] = ff_cavs_un_mv;
+        h->mv[MV_BWD_D3] = ff_cavs_un_mv;
     }
-    inter_pred(h, mb_type);
-    set_intra_mode_default(h);
-    if(mb_type != B_SKIP)
-        decode_residual_inter(h);
-    filter_mb(h,mb_type);
 }
 
-/*****************************************************************************
- *
- * slice level
- *
- ****************************************************************************/
+/**
+ * save predictors for later macroblocks and increase
+ * macroblock address
+ * @return 0 if end of frame is reached, 1 otherwise
+ */
+int ff_cavs_next_mb(AVSContext *h) {
+    int i;
 
-static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
-    if(h->stc > 0xAF)
-        av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
-    h->mby = h->stc;
-    if((h->mby == 0) && (!h->qp_fixed)){
-        h->qp_fixed = get_bits1(gb);
-        h->qp = get_bits(gb,6);
-    }
-    /* inter frame or second slice can have weighting params */
-    if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
-        if(get_bits1(gb)) { //slice_weighting_flag
-            av_log(h->s.avctx, AV_LOG_ERROR,
-                   "weighted prediction not yet supported\n");
+    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)
+        h->mv[i] = h->mv[i+2];
+    /* copy bottom mvs from cache to top line */
+    h->top_mv[0][h->mbx*2+0] = h->mv[MV_FWD_X2];
+    h->top_mv[0][h->mbx*2+1] = h->mv[MV_FWD_X3];
+    h->top_mv[1][h->mbx*2+0] = h->mv[MV_BWD_X2];
+    h->top_mv[1][h->mbx*2+1] = h->mv[MV_BWD_X3];
+    /* next MB address */
+    h->mbidx++;
+    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)
+            h->mv[i] = ff_cavs_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;
         }
-    return 0;
-}
-
-static inline void check_for_slice(AVSContext *h) {
-    GetBitContext *gb = &h->s.gb;
-    int align;
-    align = (-get_bits_count(gb)) & 7;
-    if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
-        get_bits_long(gb,24+align);
-        h->stc = get_bits(gb,8);
-        decode_slice_header(h,gb);
     }
+    return 1;
 }
 
 /*****************************************************************************
@@ -828,7 +618,7 @@ static inline void check_for_slice(AVSContext *h) {
  *
  ****************************************************************************/
 
-static void init_pic(AVSContext *h) {
+void ff_cavs_init_pic(AVSContext *h) {
     int i;
 
     /* clear some predictors */
@@ -846,140 +636,10 @@ static void init_pic(AVSContext *h) {
     h->c_stride = h->picture.linesize[1];
     h->luma_scan[2] = 8*h->l_stride;
     h->luma_scan[3] = 8*h->l_stride+8;
-    h->mbx = h->mby = 0;
+    h->mbx = h->mby = h->mbidx = 0;
     h->flags = 0;
 }
 
-static int decode_pic(AVSContext *h) {
-    MpegEncContext *s = &h->s;
-    int skip_count;
-    enum mb_t mb_type;
-
-    if (!s->context_initialized) {
-        s->avctx->idct_algo = FF_IDCT_CAVS;
-        if (MPV_common_init(s) < 0)
-            return -1;
-        ff_init_scantable(s->dsp.idct_permutation,&h->scantable,ff_zigzag_direct);
-    }
-    get_bits(&s->gb,16);//bbv_dwlay
-    if(h->stc == PIC_PB_START_CODE) {
-        h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
-        if(h->pic_type > FF_B_TYPE) {
-            av_log(s->avctx, AV_LOG_ERROR, "illegal picture type\n");
-            return -1;
-        }
-        /* make sure we have the reference frames we need */
-        if(!h->DPB[0].data[0] ||
-          (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
-            return -1;
-    } else {
-        h->pic_type = FF_I_TYPE;
-        if(get_bits1(&s->gb))
-            get_bits(&s->gb,16);//time_code
-    }
-    /* release last B frame */
-    if(h->picture.data[0])
-        s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);
-
-    s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
-    init_pic(h);
-    h->picture.poc = get_bits(&s->gb,8)*2;
-
-    /* get temporal distances and MV scaling factors */
-    if(h->pic_type != FF_B_TYPE) {
-        h->dist[0] = (h->picture.poc - h->DPB[0].poc  + 512) % 512;
-    } else {
-        h->dist[0] = (h->DPB[0].poc  - h->picture.poc + 512) % 512;
-    }
-    h->dist[1] = (h->picture.poc - h->DPB[1].poc  + 512) % 512;
-    h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
-    h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
-    if(h->pic_type == FF_B_TYPE) {
-        h->sym_factor = h->dist[0]*h->scale_den[1];
-    } else {
-        h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
-        h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
-    }
-
-    if(s->low_delay)
-        get_ue_golomb(&s->gb); //bbv_check_times
-    h->progressive             = get_bits1(&s->gb);
-    if(h->progressive)
-        h->pic_structure = 1;
-    else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
-        get_bits1(&s->gb);     //advanced_pred_mode_disable
-    skip_bits1(&s->gb);        //top_field_first
-    skip_bits1(&s->gb);        //repeat_first_field
-    h->qp_fixed                = get_bits1(&s->gb);
-    h->qp                      = get_bits(&s->gb,6);
-    if(h->pic_type == FF_I_TYPE) {
-        if(!h->progressive && !h->pic_structure)
-            skip_bits1(&s->gb);//what is this?
-        skip_bits(&s->gb,4);   //reserved bits
-    } else {
-        if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
-            h->ref_flag        = get_bits1(&s->gb);
-        skip_bits(&s->gb,4);   //reserved bits
-        h->skip_mode_flag      = get_bits1(&s->gb);
-    }
-    h->loop_filter_disable     = get_bits1(&s->gb);
-    if(!h->loop_filter_disable && get_bits1(&s->gb)) {
-        h->alpha_offset        = get_se_golomb(&s->gb);
-        h->beta_offset         = get_se_golomb(&s->gb);
-    } else {
-        h->alpha_offset = h->beta_offset  = 0;
-    }
-    check_for_slice(h);
-    if(h->pic_type == FF_I_TYPE) {
-        do {
-            decode_mb_i(h, 0);
-        } while(next_mb(h));
-    } else if(h->pic_type == FF_P_TYPE) {
-        do {
-            if(h->skip_mode_flag) {
-                skip_count = get_ue_golomb(&s->gb);
-                while(skip_count--) {
-                    decode_mb_p(h,P_SKIP);
-                    if(!next_mb(h))
-                        goto done;
-                }
-                mb_type = get_ue_golomb(&s->gb) + P_16X16;
-            } else
-                mb_type = get_ue_golomb(&s->gb) + P_SKIP;
-            if(mb_type > P_8X8) {
-                decode_mb_i(h, mb_type - P_8X8 - 1);
-            } else
-                decode_mb_p(h,mb_type);
-        } while(next_mb(h));
-    } else { /* FF_B_TYPE */
-        do {
-            if(h->skip_mode_flag) {
-                skip_count = get_ue_golomb(&s->gb);
-                while(skip_count--) {
-                    decode_mb_b(h,B_SKIP);
-                    if(!next_mb(h))
-                        goto done;
-                }
-                mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
-            } else
-                mb_type = get_ue_golomb(&s->gb) + B_SKIP;
-            if(mb_type > B_8X8) {
-                decode_mb_i(h, mb_type - B_8X8 - 1);
-            } else
-                decode_mb_b(h,mb_type);
-        } while(next_mb(h));
-    }
- done:
-    if(h->pic_type != FF_B_TYPE) {
-        if(h->DPB[1].data[0])
-            s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
-        memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
-        memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
-        memset(&h->picture,0,sizeof(Picture));
-    }
-    return 0;
-}
-
 /*****************************************************************************
  *
  * headers and interface
@@ -991,135 +651,28 @@ static int decode_pic(AVSContext *h) {
  * this data has to be stored for one complete row of macroblocks
  * and this storage space is allocated here
  */
-static void init_top_lines(AVSContext *h) {
+void ff_cavs_init_top_lines(AVSContext *h) {
     /* alloc top line of predictors */
     h->top_qp       = av_malloc( h->mb_width);
-    h->top_mv[0]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
-    h->top_mv[1]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
+    h->top_mv[0]    = av_malloc((h->mb_width*2+1)*sizeof(cavs_vector));
+    h->top_mv[1]    = av_malloc((h->mb_width*2+1)*sizeof(cavs_vector));
     h->top_pred_Y   = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
     h->top_border_y = av_malloc((h->mb_width+1)*16);
     h->top_border_u = av_malloc((h->mb_width)*10);
     h->top_border_v = av_malloc((h->mb_width)*10);
 
     /* alloc space for co-located MVs and types */
-    h->col_mv       = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
+    h->col_mv       = av_malloc( h->mb_width*h->mb_height*4*sizeof(cavs_vector));
     h->col_type_base = av_malloc(h->mb_width*h->mb_height);
     h->block        = av_mallocz(64*sizeof(DCTELEM));
 }
 
-static int decode_seq_header(AVSContext *h) {
-    MpegEncContext *s = &h->s;
-    int frame_rate_code;
-
-    h->profile =         get_bits(&s->gb,8);
-    h->level =           get_bits(&s->gb,8);
-    skip_bits1(&s->gb); //progressive sequence
-    s->width =           get_bits(&s->gb,14);
-    s->height =          get_bits(&s->gb,14);
-    skip_bits(&s->gb,2); //chroma format
-    skip_bits(&s->gb,3); //sample_precision
-    h->aspect_ratio =    get_bits(&s->gb,4);
-    frame_rate_code =    get_bits(&s->gb,4);
-    skip_bits(&s->gb,18);//bit_rate_lower
-    skip_bits1(&s->gb);  //marker_bit
-    skip_bits(&s->gb,12);//bit_rate_upper
-    s->low_delay =       get_bits1(&s->gb);
-    h->mb_width  = (s->width  + 15) >> 4;
-    h->mb_height = (s->height + 15) >> 4;
-    h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
-    h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
-    h->s.avctx->width  = s->width;
-    h->s.avctx->height = s->height;
-    if(!h->top_qp)
-        init_top_lines(h);
-    return 0;
-}
-
-static void cavs_flush(AVCodecContext * avctx) {
-    AVSContext *h = avctx->priv_data;
-    h->got_keyframe = 0;
-}
-
-static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
-                             uint8_t * buf, int buf_size) {
-    AVSContext *h = avctx->priv_data;
-    MpegEncContext *s = &h->s;
-    int input_size;
-    const uint8_t *buf_end;
-    const uint8_t *buf_ptr;
-    AVFrame *picture = data;
-    uint32_t stc;
-
-    s->avctx = avctx;
-
-    if (buf_size == 0) {
-        if(!s->low_delay && h->DPB[0].data[0]) {
-            *data_size = sizeof(AVPicture);
-            *picture = *(AVFrame *) &h->DPB[0];
-        }
-        return 0;
-    }
-
-    buf_ptr = buf;
-    buf_end = buf + buf_size;
-    for(;;) {
-        buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
-        if(stc & 0xFFFFFE00)
-            return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
-        input_size = (buf_end - buf_ptr)*8;
-        switch(stc) {
-        case CAVS_START_CODE:
-            init_get_bits(&s->gb, buf_ptr, input_size);
-            decode_seq_header(h);
-            break;
-        case PIC_I_START_CODE:
-            if(!h->got_keyframe) {
-                if(h->DPB[0].data[0])
-                    avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
-                if(h->DPB[1].data[0])
-                    avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
-                h->got_keyframe = 1;
-            }
-        case PIC_PB_START_CODE:
-            *data_size = 0;
-            if(!h->got_keyframe)
-                break;
-            init_get_bits(&s->gb, buf_ptr, input_size);
-            h->stc = stc;
-            if(decode_pic(h))
-                break;
-            *data_size = sizeof(AVPicture);
-            if(h->pic_type != FF_B_TYPE) {
-                if(h->DPB[1].data[0]) {
-                    *picture = *(AVFrame *) &h->DPB[1];
-                } else {
-                    *data_size = 0;
-                }
-            } else
-                *picture = *(AVFrame *) &h->picture;
-            break;
-        case EXT_START_CODE:
-            //mpeg_decode_extension(avctx,buf_ptr, input_size);
-            break;
-        case USER_START_CODE:
-            //mpeg_decode_user_data(avctx,buf_ptr, input_size);
-            break;
-        default:
-            if (stc >= SLICE_MIN_START_CODE &&
-                stc <= SLICE_MAX_START_CODE) {
-                init_get_bits(&s->gb, buf_ptr, input_size);
-                decode_slice_header(h, &s->gb);
-            }
-            break;
-        }
-    }
-}
-
-static int cavs_decode_init(AVCodecContext * avctx) {
+av_cold int ff_cavs_init(AVCodecContext *avctx) {
     AVSContext *h = avctx->priv_data;
     MpegEncContext * const s = &h->s;
 
     MPV_decode_defaults(s);
+    ff_cavsdsp_init(&h->cdsp, avctx);
     s->avctx = avctx;
 
     avctx->pix_fmt= PIX_FMT_YUV420P;
@@ -1146,7 +699,7 @@ static int cavs_decode_init(AVCodecContext * avctx) {
     return 0;
 }
 
-static int cavs_decode_end(AVCodecContext * avctx) {
+av_cold int ff_cavs_end(AVCodecContext *avctx) {
     AVSContext *h = avctx->priv_data;
 
     av_free(h->top_qp);
@@ -1161,16 +714,3 @@ static int cavs_decode_end(AVCodecContext * avctx) {
     av_free(h->block);
     return 0;
 }
-
-AVCodec cavs_decoder = {
-    "cavs",
-    CODEC_TYPE_VIDEO,
-    CODEC_ID_CAVS,
-    sizeof(AVSContext),
-    cavs_decode_init,
-    NULL,
-    cavs_decode_end,
-    cavs_decode_frame,
-    CODEC_CAP_DR1 | CODEC_CAP_DELAY,
-    .flush= cavs_flush,
-};