/libavcodec/rv34.c - Annotate - PeerStreamer - Redmine

ffmpeg / libavcodec / rv34.c @ 2912e87a

History | View | Annotate | Download (54.2 KB)

-            a52ef6a8
+/*
  * RV30/40 decoder common data
  * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
+ *
-e87a
+ * This file is part of Libav.
-            a52ef6a8
+            Kostya Shishkov
-e87a
+ * Libav is free software; you can redistribute it and/or
-            a52ef6a8
+ * 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.
+ *
-e87a
+ * Libav is distributed in the hope that it will be useful,
-            a52ef6a8
+ * 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
-e87a
+ * License along with Libav; if not, write to the Free Software
-            a52ef6a8
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 /**
-            ba87f080
+ * @file
-            a52ef6a8
+ * RV30/40 decoder common data
  */
 #include "avcodec.h"
 #include "dsputil.h"
 #include "mpegvideo.h"
 #include "golomb.h"
-b9
+#include "mathops.h"
-            a52ef6a8
+#include "rectangle.h"
 #include "rv34vlc.h"
 #include "rv34data.h"
 #include "rv34.h"
 //#define DEBUG
-            13695827
+static inline void ZERO8x2(void* dst, int stride)
+{
     fill_rectangle(dst,                 1, 2, stride, 0, 4);
     fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
+}
-            b7eb7ef6
+            Kostya Shishkov
-            a52ef6a8
+/** translation of RV30/40 macroblock types to lavc ones */
 static const int rv34_mb_type_to_lavc[12] = {
     MB_TYPE_INTRA,
-            c0b821be
+    MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,
-            a52ef6a8
+    MB_TYPE_16x16   | MB_TYPE_L0,
     MB_TYPE_8x8     | MB_TYPE_L0,
     MB_TYPE_16x16   | MB_TYPE_L0,
     MB_TYPE_16x16   | MB_TYPE_L1,
     MB_TYPE_SKIP,
     MB_TYPE_DIRECT2 | MB_TYPE_16x16,
     MB_TYPE_16x8    | MB_TYPE_L0,
     MB_TYPE_8x16    | MB_TYPE_L0,
     MB_TYPE_16x16   | MB_TYPE_L0L1,
-            c0b821be
+    MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC
-            a52ef6a8
+};
 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
 /**
  * @defgroup vlc RV30/40 VLC generating functions
  * @{
  */
-c4b5eb8
+static const int table_offs[] = {
 ,   1818,   3622,   4144,   4698,   5234,   5804,   5868,   5900,   5932,
 ,   6252,   6316,   6348,   6380,   7674,   8944,  10274,  11668,  12250,
 ,  15846,  16372,  16962,  17512,  18148,  18180,  18212,  18244,  18308,
 ,  18628,  18660,  18692,  20036,  21314,  22648,  23968,  24614,  26384,
 ,  28736,  29366,  29938,  30608,  30640,  30672,  30704,  30768,  31024,
 ,  31120,  31184,  32570,  33898,  35236,  36644,  37286,  39020,  40802,
 ,  42052,  42692,  43348,  43380,  43412,  43444,  43476,  43604,  43668,
 ,  43732,  45100,  46430,  47778,  49160,  49802,  51550,  53340,  53972,
 ,  55348,  55994,  56122,  56154,  56186,  56218,  56346,  56410,  56442,
 ,  57878,  59290,  60636,  62036,  62682,  64460,  64524,  64588,  64716,
 ,  66076,  67466,  67978,  68542,  69064,  69648,  70296,  72010,  72074,
 ,  72202,  72330,  73572,  74936,  75454,  76030,  76566,  77176,  77822,
 ,  79646,  79678,  79742,  79870,  81180,  82536,  83064,  83672,  84242,
 ,  85576,  87384,  87448,  87480,  87544,  87672,  88982,  90340,  90902,
 ,  92182,  92846,  93488,  95246,  95278,  95310,  95374,  95502,  96878,
 ,  98848,  99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
 , 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
 , 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
 };
 static VLC_TYPE table_data[117592][2];
-            a52ef6a8
+/**
  * Generate VLC from codeword lengths.
  * @param bits   codeword lengths (zeroes are accepted)
  * @param size   length of input data
-aa73351
+ * @param vlc    output VLC
-            a52ef6a8
+ * @param insyms symbols for input codes (NULL for default ones)
-b6a4c
+ * @param num    VLC table number (for static initialization)
-            a52ef6a8
+ */
-c4b5eb8
+static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
                          const int num)
-            a52ef6a8
+            Kostya Shishkov
     int i;
     int counts[17] = {0}, codes[17];
-bfb72d9
+    uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
     uint8_t bits2[MAX_VLC_SIZE];
-            a52ef6a8
+    int maxbits = 0, realsize = 0;
     for(i = 0; i < size; i++){
         if(bits[i]){
             bits2[realsize] = bits[i];
             syms[realsize] = insyms ? insyms[i] : i;
             realsize++;
             maxbits = FFMAX(maxbits, bits[i]);
             counts[bits[i]]++;
+        }
+    }
     codes[0] = 0;
     for(i = 0; i < 16; i++)
         codes[i+1] = (codes[i] + counts[i]) << 1;
     for(i = 0; i < realsize; i++)
         cw[i] = codes[bits2[i]]++;
-c4b5eb8
+    vlc->table = &table_data[table_offs[num]];
     vlc->table_allocated = table_offs[num + 1] - table_offs[num];
-            a52ef6a8
+    init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
                     bits2, 1, 1,
                     cw,    2, 2,
-c4b5eb8
+                    syms,  2, 2, INIT_VLC_USE_NEW_STATIC);
-            a52ef6a8
+            Kostya Shishkov
 /**
  * Initialize all tables.
  */
-            dc8a7c93
+static av_cold void rv34_init_tables(void)
-            a52ef6a8
+            Kostya Shishkov
     int i, j, k;
     for(i = 0; i < NUM_INTRA_TABLES; i++){
         for(j = 0; j < 2; j++){
-c4b5eb8
+            rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,   &intra_vlcs[i].cbppattern[j],     NULL, 19*i + 0 + j);
             rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
             rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j],  NULL, 19*i + 4 + j);
-b6a4c
+            for(k = 0; k < 4; k++){
-c4b5eb8
+                rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2],  CBP_VLC_SIZE,   &intra_vlcs[i].cbp[j][k],         rv34_cbp_code, 19*i + 6 + j*4 + k);
-b6a4c
+            Kostya Shishkov
-            a52ef6a8
+            Kostya Shishkov
-b6a4c
+        for(j = 0; j < 4; j++){
-c4b5eb8
+            rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
-b6a4c
+            Kostya Shishkov
-c4b5eb8
+        rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
-            a52ef6a8
+            Kostya Shishkov
     for(i = 0; i < NUM_INTER_TABLES; i++){
-c4b5eb8
+        rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
-b6a4c
+        for(j = 0; j < 4; j++){
-c4b5eb8
+            rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
-b6a4c
+            Kostya Shishkov
-            a52ef6a8
+        for(j = 0; j < 2; j++){
-c4b5eb8
+            rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j],  NULL, i*12 + 100 + j);
             rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
             rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j],  NULL, i*12 + 104 + j);
-            a52ef6a8
+            Kostya Shishkov
-c4b5eb8
+        rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
-            a52ef6a8
+            Kostya Shishkov
+}
 /** @} */ // vlc group
 /**
  * @defgroup transform RV30/40 inverse transform functions
  * @{
  */
 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
+{
     int i;
     for(i=0; i<4; i++){
         const int z0= 13*(block[i+8*0] +    block[i+8*2]);
         const int z1= 13*(block[i+8*0] -    block[i+8*2]);
         const int z2=  7* block[i+8*1] - 17*block[i+8*3];
         const int z3= 17* block[i+8*1] +  7*block[i+8*3];
         temp[4*i+0]= z0+z3;
         temp[4*i+1]= z1+z2;
         temp[4*i+2]= z1-z2;
         temp[4*i+3]= z0-z3;
+    }
+}
 /**
  * Real Video 3.0/4.0 inverse transform
  * Code is almost the same as in SVQ3, only scaling is different.
  */
 static void rv34_inv_transform(DCTELEM *block){
     int temp[16];
     int i;
     rv34_row_transform(temp, block);
     for(i=0; i<4; i++){
         const int z0= 13*(temp[4*0+i] +    temp[4*2+i]) + 0x200;
         const int z1= 13*(temp[4*0+i] -    temp[4*2+i]) + 0x200;
         const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
         const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
         block[i*8+0]= (z0 + z3)>>10;
         block[i*8+1]= (z1 + z2)>>10;
         block[i*8+2]= (z1 - z2)>>10;
         block[i*8+3]= (z0 - z3)>>10;
+    }
+}
 /**
  * RealVideo 3.0/4.0 inverse transform for DC block
+ *
  * Code is almost the same as rv34_inv_transform()
  * but final coefficients are multiplied by 1.5 and have no rounding.
  */
 static void rv34_inv_transform_noround(DCTELEM *block){
     int temp[16];
     int i;
     rv34_row_transform(temp, block);
     for(i=0; i<4; i++){
         const int z0= 13*(temp[4*0+i] +    temp[4*2+i]);
         const int z1= 13*(temp[4*0+i] -    temp[4*2+i]);
         const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
         const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
         block[i*8+0]= ((z0 + z3)*3)>>11;
         block[i*8+1]= ((z1 + z2)*3)>>11;
         block[i*8+2]= ((z1 - z2)*3)>>11;
         block[i*8+3]= ((z0 - z3)*3)>>11;
+    }
+}
 /** @} */ // transform
 /**
  * @defgroup block RV30/40 4x4 block decoding functions
  * @{
  */
 /**
  * Decode coded block pattern.
  */
 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
+{
     int pattern, code, cbp=0;
     int ones;
     static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
     static const int shifts[4] = { 0, 2, 8, 10 };
-ffabd4e
+    const int *curshift = shifts;
-            a52ef6a8
+    int i, t, mask;
     code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
     pattern = code & 0xF;
     code >>= 4;
     ones = rv34_count_ones[pattern];
     for(mask = 8; mask; mask >>= 1, curshift++){
         if(pattern & mask)
             cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
+    }
     for(i = 0; i < 4; i++){
         t = modulo_three_table[code][i];
         if(t == 1)
             cbp |= cbp_masks[get_bits1(gb)] << i;
         if(t == 2)
             cbp |= cbp_masks[2] << i;
+    }
     return cbp;
+}
 /**
  * Get one coefficient value from the bistream and store it.
  */
 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
+{
     if(coef){
         if(coef == esc){
             coef = get_vlc2(gb, vlc->table, 9, 2);
             if(coef > 23){
                 coef -= 23;
                 coef = 22 + ((1 << coef) | get_bits(gb, coef));
+            }
             coef += esc;
+        }
         if(get_bits1(gb))
             coef = -coef;
         *dst = coef;
+    }
+}
 /**
  * Decode 2x2 subblock of coefficients.
  */
 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
+{
     int coeffs[4];
     coeffs[0] = modulo_three_table[code][0];
     coeffs[1] = modulo_three_table[code][1];
     coeffs[2] = modulo_three_table[code][2];
     coeffs[3] = modulo_three_table[code][3];
     decode_coeff(dst  , coeffs[0], 3, gb, vlc);
     if(is_block2){
         decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
         decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
     }else{
         decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
         decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
+    }
     decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
+}
 /**
  * Decode coefficients for 4x4 block.
+ *
  * This is done by filling 2x2 subblocks with decoded coefficients
  * in this order (the same for subblocks and subblock coefficients):
  *  o--o
  *    /
  *   /
  *  o--o
  */
 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
+{
     int code, pattern;
     code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
     pattern = code & 0x7;
     code >>= 3;
     decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
     if(pattern & 4){
         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
         decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
+    }
     if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
         decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
+    }
     if(pattern & 1){
         code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
         decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
+    }
+}
 /**
  * Dequantize ordinary 4x4 block.
  * @todo optimize
  */
 static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
+{
     int i, j;
     block[0] = (block[0] * Qdc + 8) >> 4;
     for(i = 0; i < 4; i++)
         for(j = !i; j < 4; j++)
             block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
+}
 /**
  * Dequantize 4x4 block of DC values for 16x16 macroblock.
  * @todo optimize
  */
 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
+{
     int i;
     for(i = 0; i < 3; i++)
          block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
     for(; i < 16; i++)
          block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
+}
 /** @} */ //block functions
 /**
-c254c
+ * @defgroup rv3040_bitstream RV30/40 bitstream parsing
-            a52ef6a8
+ * @{
  */
 /**
  * Decode starting slice position.
  * @todo Maybe replace with ff_h263_decode_mba() ?
  */
 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
+{
     int i;
     for(i = 0; i < 5; i++)
-            ffedb5c1
+        if(rv34_mb_max_sizes[i] >= mb_size - 1)
-            a52ef6a8
+            break;
     return rv34_mb_bits_sizes[i];
+}
 /**
  * Select VLC set for decoding from current quantizer, modifier and frame type.
  */
 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
+{
     if(mod == 2 && quant < 19) quant += 10;
     else if(mod && quant < 26) quant += 5;
     return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
                 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
+}
 /**
  * Decode quantizer difference and return modified quantizer.
  */
 static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
+{
     if(get_bits1(gb))
         return rv34_dquant_tab[get_bits1(gb)][quant];
     else
         return get_bits(gb, 5);
+}
 /** @} */ //bitstream functions
 /**
  * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
  * @{
  */
 /** macroblock partition width in 8x8 blocks */
 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
 /** macroblock partition height in 8x8 blocks */
 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
 /** availability index for subblocks */
-c4a
+static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
-            a52ef6a8
+            Kostya Shishkov
 /**
  * motion vector prediction
+ *
  * Motion prediction performed for the block by using median prediction of
  * motion vectors from the left, top and right top blocks but in corner cases
  * some other vectors may be used instead.
  */
 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
+{
     MpegEncContext *s = &r->s;
     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
     int A[2] = {0}, B[2], C[2];
     int i, j;
     int mx, my;
     int avail_index = avail_indexes[subblock_no];
     int c_off = part_sizes_w[block_type];
     mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
     if(subblock_no == 3)
         c_off = -1;
     if(r->avail_cache[avail_index - 1]){
         A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
         A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
+    }
     if(r->avail_cache[avail_index - 4]){
         B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
         B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
     }else{
         B[0] = A[0];
         B[1] = A[1];
+    }
     if(!r->avail_cache[avail_index - 4 + c_off]){
         if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
             C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
             C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
         }else{
             C[0] = A[0];
             C[1] = A[1];
+        }
     }else{
         C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
         C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
+    }
     mx = mid_pred(A[0], B[0], C[0]);
     my = mid_pred(A[1], B[1], C[1]);
     mx += r->dmv[dmv_no][0];
     my += r->dmv[dmv_no][1];
     for(j = 0; j < part_sizes_h[block_type]; j++){
         for(i = 0; i < part_sizes_w[block_type]; i++){
             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
+        }
+    }
+}
-a8c
+#define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
-            a52ef6a8
+/**
-            e640a5c4
+ * Calculate motion vector component that should be added for direct blocks.
  */
-a8c
+static int calc_add_mv(RV34DecContext *r, int dir, int val)
-            e640a5c4
+            Kostya Shishkov
-a8c
+    int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
-f621dd2
+    int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
     int mul;
-            e640a5c4
+            Kostya Shishkov
-a8c
+    if(!refdist) return 0;
-f621dd2
+    mul = (dist << 14) / refdist;
     return (val * mul + 0x2000) >> 14;
-            e640a5c4
+            Kostya Shishkov
 /**
-            a52ef6a8
+ * Predict motion vector for B-frame macroblock.
  */
 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
                                       int A_avail, int B_avail, int C_avail,
                                       int *mx, int *my)
+{
     if(A_avail + B_avail + C_avail != 3){
         *mx = A[0] + B[0] + C[0];
         *my = A[1] + B[1] + C[1];
         if(A_avail + B_avail + C_avail == 2){
             *mx /= 2;
             *my /= 2;
+        }
     }else{
         *mx = mid_pred(A[0], B[0], C[0]);
         *my = mid_pred(A[1], B[1], C[1]);
+    }
+}
 /**
  * motion vector prediction for B-frames
  */
 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
+{
     MpegEncContext *s = &r->s;
     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
     int A[2], B[2], C[2];
     int has_A = 0, has_B = 0, has_C = 0;
     int mx, my;
     int i, j;
     Picture *cur_pic = s->current_picture_ptr;
     const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
     int type = cur_pic->mb_type[mb_pos];
     memset(A, 0, sizeof(A));
     memset(B, 0, sizeof(B));
     memset(C, 0, sizeof(C));
-c4a
+    if((r->avail_cache[6-1] & type) & mask){
-            a52ef6a8
+        A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
         A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
         has_A = 1;
+    }
-c4a
+    if((r->avail_cache[6-4] & type) & mask){
-            a52ef6a8
+        B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
         B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
         has_B = 1;
+    }
-c4a
+    if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
-            a52ef6a8
+        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
         has_C = 1;
-c4a
+    }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
-            a52ef6a8
+        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
         has_C = 1;
+    }
     rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
     mx += r->dmv[dir][0];
     my += r->dmv[dir][1];
-            e640a5c4
+            Kostya Shishkov
-            a52ef6a8
+    for(j = 0; j < 2; j++){
         for(i = 0; i < 2; i++){
             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
+        }
+    }
-            b7eb7ef6
+    if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
         ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
+    }
-            a52ef6a8
+            Kostya Shishkov
-            b2153658
+/**
  * motion vector prediction - RV3 version
  */
 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
+{
     MpegEncContext *s = &r->s;
     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
     int A[2] = {0}, B[2], C[2];
-beefe
+    int i, j, k;
-            b2153658
+    int mx, my;
     int avail_index = avail_indexes[0];
     if(r->avail_cache[avail_index - 1]){
         A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
         A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
+    }
     if(r->avail_cache[avail_index - 4]){
         B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
         B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
     }else{
         B[0] = A[0];
         B[1] = A[1];
+    }
     if(!r->avail_cache[avail_index - 4 + 2]){
         if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
             C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
             C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
         }else{
             C[0] = A[0];
             C[1] = A[1];
+        }
     }else{
         C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
         C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
+    }
     mx = mid_pred(A[0], B[0], C[0]);
     my = mid_pred(A[1], B[1], C[1]);
     mx += r->dmv[0][0];
     my += r->dmv[0][1];
     for(j = 0; j < 2; j++){
         for(i = 0; i < 2; i++){
-beefe
+            for(k = 0; k < 2; k++){
                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
+            }
-            b2153658
+            Kostya Shishkov
+    }
+}
-ee054c2
+static const int chroma_coeffs[3] = { 0, 3, 5 };
-            f1917274
+            Kostya Shishkov
-            a52ef6a8
+/**
  * generic motion compensation function
+ *
  * @param r decoder context
  * @param block_type type of the current block
  * @param xoff horizontal offset from the start of the current block
  * @param yoff vertical offset from the start of the current block
  * @param mv_off offset to the motion vector information
  * @param width width of the current partition in 8x8 blocks
  * @param height height of the current partition in 8x8 blocks
-aa73351
+ * @param dir motion compensation direction (i.e. from the last or the next reference frame)
  * @param thirdpel motion vectors are specified in 1/3 of pixel
  * @param qpel_mc a set of functions used to perform luma motion compensation
  * @param chroma_mc a set of functions used to perform chroma motion compensation
-            a52ef6a8
+ */
 static inline void rv34_mc(RV34DecContext *r, const int block_type,
                           const int xoff, const int yoff, int mv_off,
                           const int width, const int height, int dir,
                           const int thirdpel,
                           qpel_mc_func (*qpel_mc)[16],
                           h264_chroma_mc_func (*chroma_mc))
+{
     MpegEncContext *s = &r->s;
     uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
-ee054c2
+    int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
-            a52ef6a8
+    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
     int is16x16 = 1;
     if(thirdpel){
-ee054c2
+        int chroma_mx, chroma_my;
-            f1917274
+        mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
         my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
         lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
         ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
-ee054c2
+        chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
         chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
         umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
         umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
         uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
         uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
-            a52ef6a8
+    }else{
-            f653095b
+        int cx, cy;
-            a52ef6a8
+        mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
         my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
-df3869
+        lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
         ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
-            f653095b
+        cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
         cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
         umx = cx >> 2;
         umy = cy >> 2;
         uvmx = (cx & 3) << 1;
         uvmy = (cy & 3) << 1;
         //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
         if(uvmx == 6 && uvmy == 6)
             uvmx = uvmy = 4;
-            a52ef6a8
+            Kostya Shishkov
-df3869
+    dxy = ly*4 + lx;
-            a52ef6a8
+    srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
     srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
     srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
     src_x = s->mb_x * 16 + xoff + mx;
     src_y = s->mb_y * 16 + yoff + my;
-ee054c2
+    uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
     uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
-            a52ef6a8
+    srcY += src_y * s->linesize + src_x;
     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
-            7690351c
+    if(   (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
        || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
         uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
-            a52ef6a8
+            Kostya Shishkov
         srcY -= 2 + 2*s->linesize;
-e279598
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
-            a52ef6a8
+                            src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
         srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
-e279598
+        s->dsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
-            a52ef6a8
+                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
-e279598
+        s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
-            a52ef6a8
+                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
         srcU = uvbuf;
         srcV = uvbuf + 16;
+    }
     Y = s->dest[0] + xoff      + yoff     *s->linesize;
     U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
     V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
     if(block_type == RV34_MB_P_16x8){
         qpel_mc[1][dxy](Y, srcY, s->linesize);
         Y    += 8;
         srcY += 8;
     }else if(block_type == RV34_MB_P_8x16){
         qpel_mc[1][dxy](Y, srcY, s->linesize);
         Y    += 8 * s->linesize;
         srcY += 8 * s->linesize;
+    }
     is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
     qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
     chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
     chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
+}
 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
                         const int xoff, const int yoff, int mv_off,
                         const int width, const int height, int dir)
+{
     rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
-            f1917274
+            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
-            f653095b
+                    : r->s.dsp.put_rv40_qpel_pixels_tab,
             r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
                     : r->s.dsp.put_rv40_chroma_pixels_tab);
-            a52ef6a8
+            Kostya Shishkov
 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
+{
     rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
-            f1917274
+            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
-            f653095b
+                    : r->s.dsp.put_rv40_qpel_pixels_tab,
-a2b5d
+            r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
                     : r->s.dsp.put_rv40_chroma_pixels_tab);
-            a52ef6a8
+    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
-            f1917274
+            r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
-            f653095b
+                    : r->s.dsp.avg_rv40_qpel_pixels_tab,
             r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
                     : r->s.dsp.avg_rv40_chroma_pixels_tab);
-            a52ef6a8
+            Kostya Shishkov
-c90cc7
+static void rv34_mc_2mv_skip(RV34DecContext *r)
+{
-ece410
+    int i, j;
-c90cc7
+    for(j = 0; j < 2; j++)
         for(i = 0; i < 2; i++){
              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
                     r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
-            f653095b
+                            : r->s.dsp.put_rv40_qpel_pixels_tab,
                     r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
                             : r->s.dsp.put_rv40_chroma_pixels_tab);
-c90cc7
+             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
                     r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
-            f653095b
+                            : r->s.dsp.avg_rv40_qpel_pixels_tab,
                     r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
                             : r->s.dsp.avg_rv40_chroma_pixels_tab);
-c90cc7
+            Kostya Shishkov
+}
-            a52ef6a8
+/** number of motion vectors in each macroblock type */
 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
 /**
  * Decode motion vector differences
  * and perform motion vector reconstruction and motion compensation.
  */
 static int rv34_decode_mv(RV34DecContext *r, int block_type)
+{
     MpegEncContext *s = &r->s;
     GetBitContext *gb = &s->gb;
-a8c
+    int i, j, k, l;
-c90cc7
+    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
     int next_bt;
-            a52ef6a8
+            Kostya Shishkov
     memset(r->dmv, 0, sizeof(r->dmv));
     for(i = 0; i < num_mvs[block_type]; i++){
         r->dmv[i][0] = svq3_get_se_golomb(gb);
         r->dmv[i][1] = svq3_get_se_golomb(gb);
+    }
     switch(block_type){
     case RV34_MB_TYPE_INTRA:
     case RV34_MB_TYPE_INTRA16x16:
-            b7eb7ef6
+        ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
-            a52ef6a8
+        return 0;
     case RV34_MB_SKIP:
-            9701840b
+        if(s->pict_type == FF_P_TYPE){
-            b7eb7ef6
+            ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
-            a52ef6a8
+            rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
             break;
+        }
     case RV34_MB_B_DIRECT:
-c90cc7
+        //surprisingly, it uses motion scheme from next reference frame
         next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
-ed974
+        if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
-            b7eb7ef6
+            ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
             ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
-ed974
+        }else
-a474950
+            for(j = 0; j < 2; j++)
                 for(i = 0; i < 2; i++)
                     for(k = 0; k < 2; k++)
                         for(l = 0; l < 2; l++)
                             s->current_picture_ptr->motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][k]);
-ed974
+        if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
-c90cc7
+            rv34_mc_2mv(r, block_type);
         else
             rv34_mc_2mv_skip(r);
-            b7eb7ef6
+        ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
-            a52ef6a8
+        break;
     case RV34_MB_P_16x16:
     case RV34_MB_P_MIX16x16:
         rv34_pred_mv(r, block_type, 0, 0);
         rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
         break;
     case RV34_MB_B_FORWARD:
     case RV34_MB_B_BACKWARD:
         r->dmv[1][0] = r->dmv[0][0];
         r->dmv[1][1] = r->dmv[0][1];
-            b2153658
+        if(r->rv30)
             rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
         else
             rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
-            a52ef6a8
+        rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
         break;
     case RV34_MB_P_16x8:
     case RV34_MB_P_8x16:
         rv34_pred_mv(r, block_type, 0, 0);
         rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
         if(block_type == RV34_MB_P_16x8){
             rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
             rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
+        }
         if(block_type == RV34_MB_P_8x16){
             rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
             rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
+        }
         break;
     case RV34_MB_B_BIDIR:
         rv34_pred_mv_b  (r, block_type, 0);
         rv34_pred_mv_b  (r, block_type, 1);
         rv34_mc_2mv     (r, block_type);
         break;
     case RV34_MB_P_8x8:
         for(i=0;i< 4;i++){
             rv34_pred_mv(r, block_type, i, i);
             rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
+        }
         break;
+    }
     return 0;
+}
 /** @} */ // mv group
 /**
  * @defgroup recons Macroblock reconstruction functions
  * @{
  */
 /** mapping of RV30/40 intra prediction types to standard H.264 types */
 static const int ittrans[9] = {
  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
 };
 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
 static const int ittrans16[4] = {
  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
 };
 /**
  * Perform 4x4 intra prediction.
  */
 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
+{
     uint8_t *prev = dst - stride + 4;
     uint32_t topleft;
     if(!up && !left)
         itype = DC_128_PRED;
     else if(!up){
         if(itype == VERT_PRED) itype = HOR_PRED;
         if(itype == DC_PRED)   itype = LEFT_DC_PRED;
     }else if(!left){
         if(itype == HOR_PRED)  itype = VERT_PRED;
         if(itype == DC_PRED)   itype = TOP_DC_PRED;
         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
+    }
     if(!down){
         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
         if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
-            b0797570
+        if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
-            a52ef6a8
+            Kostya Shishkov
     if(!right && up){
         topleft = dst[-stride + 3] * 0x01010101;
-ffabd4e
+        prev = (uint8_t*)&topleft;
-            a52ef6a8
+            Kostya Shishkov
     r->h.pred4x4[itype](dst, prev, stride);
+}
 /** add_pixels_clamped for 4x4 block */
 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
+{
     int x, y;
     for(y = 0; y < 4; y++)
         for(x = 0; x < 4; x++)
             dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
+}
 static inline int adjust_pred16(int itype, int up, int left)
+{
     if(!up && !left)
         itype = DC_128_PRED8x8;
     else if(!up){
         if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
         if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
         if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
     }else if(!left){
         if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
         if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
         if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
+    }
     return itype;
+}
 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
+{
     MpegEncContext *s = &r->s;
     DSPContext *dsp = &s->dsp;
     int i, j;
     uint8_t *Y, *U, *V;
     int itype;
     int avail[6*8] = {0};
     int idx;
     // Set neighbour information.
     if(r->avail_cache[1])
-c4a
+        avail[0] = 1;
-            a52ef6a8
+    if(r->avail_cache[2])
-c4a
+        avail[1] = avail[2] = 1;
-            a52ef6a8
+    if(r->avail_cache[3])
-c4a
+        avail[3] = avail[4] = 1;
-            a52ef6a8
+    if(r->avail_cache[4])
-c4a
+        avail[5] = 1;
     if(r->avail_cache[5])
-            a52ef6a8
+        avail[8] = avail[16] = 1;
-c4a
+    if(r->avail_cache[9])
-            a52ef6a8
+        avail[24] = avail[32] = 1;
     Y = s->dest[0];
     U = s->dest[1];
     V = s->dest[2];
     if(!is16){
         for(j = 0; j < 4; j++){
             idx = 9 + j*8;
             for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
                 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
                 avail[idx] = 1;
                 if(cbp & 1)
                     rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
+            }
             Y += s->linesize * 4 - 4*4;
-c601b5
+            intra_types += r->intra_types_stride;
-            a52ef6a8
+            Kostya Shishkov
-c601b5
+        intra_types -= r->intra_types_stride * 4;
-c4a
+        fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
-            a52ef6a8
+        for(j = 0; j < 2; j++){
-c4a
+            idx = 6 + j*4;
-            a52ef6a8
+            for(i = 0; i < 2; i++, cbp >>= 1, idx++){
-c601b5
+                rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
                 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
-            a52ef6a8
+                r->avail_cache[idx] = 1;
                 if(cbp & 0x01)
                     rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
                 if(cbp & 0x10)
                     rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
+            }
+        }
     }else{
         itype = ittrans16[intra_types[0]];
-c4a
+        itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
-            a52ef6a8
+        r->h.pred16x16[itype](Y, s->linesize);
-a0fa8c
+        dsp->add_pixels_clamped(s->block[0], Y,     s->linesize);
         dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
         Y += s->linesize * 8;
         dsp->add_pixels_clamped(s->block[2], Y,     s->linesize);
         dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
-            a52ef6a8
+            Kostya Shishkov
         itype = ittrans16[intra_types[0]];
         if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
-c4a
+        itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
-            a52ef6a8
+        r->h.pred8x8[itype](U, s->uvlinesize);
         dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
         r->h.pred8x8[itype](V, s->uvlinesize);
         dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
+    }
+}
 /** @} */ // recons group
 /**
  * @addtogroup bitstream
  * Decode macroblock header and return CBP in case of success, -1 otherwise.
  */
 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
+{
     MpegEncContext *s = &r->s;
     GetBitContext *gb = &s->gb;
     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
     int i, t;
     if(!r->si.type){
         r->is16 = get_bits1(gb);
         if(!r->is16 && !r->rv30){
             if(!get_bits1(gb))
                 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
+        }
         s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
         r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
     }else{
         r->block_type = r->decode_mb_info(r);
         if(r->block_type == -1)
             return -1;
         s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
         r->mb_type[mb_pos] = r->block_type;
         if(r->block_type == RV34_MB_SKIP){
-            9701840b
+            if(s->pict_type == FF_P_TYPE)
-            a52ef6a8
+                r->mb_type[mb_pos] = RV34_MB_P_16x16;
-            9701840b
+            if(s->pict_type == FF_B_TYPE)
-            a52ef6a8
+                r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
+        }
         r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
         rv34_decode_mv(r, r->block_type);
         if(r->block_type == RV34_MB_SKIP){
-c601b5
+            fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
-            a52ef6a8
+            return 0;
+        }
         r->chroma_vlc = 1;
         r->luma_vlc   = 0;
+    }
     if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
         if(r->is16){
             t = get_bits(gb, 2);
-c601b5
+            fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
-            a52ef6a8
+            r->luma_vlc   = 2;
         }else{
             if(r->decode_intra_types(r, gb, intra_types) < 0)
                 return -1;
             r->luma_vlc   = 1;
+        }
         r->chroma_vlc = 0;
         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
     }else{
         for(i = 0; i < 16; i++)
-c601b5
+            intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
-            a52ef6a8
+        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
         if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
             r->is16 = 1;
             r->chroma_vlc = 1;
             r->luma_vlc   = 2;
             r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
+        }
+    }
     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
+}
 /**
  * @addtogroup recons
  * @{
  */
 /**
  * mask for retrieving all bits in coded block pattern
  * corresponding to one 8x8 block
  */
-            cd982791
+#define LUMA_CBP_BLOCK_MASK 0x33
-            a52ef6a8
+            Kostya Shishkov
 #define U_CBP_MASK 0x0F0000
 #define V_CBP_MASK 0xF00000
 static void rv34_apply_differences(RV34DecContext *r, int cbp)
+{
     static const int shifts[4] = { 0, 2, 8, 10 };
     MpegEncContext *s = &r->s;
     int i;
     for(i = 0; i < 4; i++)
-            cd982791
+        if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
-            a52ef6a8
+            s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
     if(cbp & U_CBP_MASK)
         s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
     if(cbp & V_CBP_MASK)
         s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
+}
-            e1223112
+static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
+{
     int d;
     d = motion_val[0][0] - motion_val[-step][0];
     if(d < -3 || d > 3)
         return 1;
     d = motion_val[0][1] - motion_val[-step][1];
     if(d < -3 || d > 3)
         return 1;
     return 0;
+}
 static int rv34_set_deblock_coef(RV34DecContext *r)
+{
     MpegEncContext *s = &r->s;
-ecc97
+    int hmvmask = 0, vmvmask = 0, i, j;
-            e1223112
+    int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
-b76ce9
+    int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
-            e1223112
+    for(j = 0; j < 16; j += 8){
         for(i = 0; i < 2; i++){
             if(is_mv_diff_gt_3(motion_val + i, 1))
-ecc97
+                vmvmask |= 0x11 << (j + i*2);
-e16cde4
+            if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
-ecc97
+                hmvmask |= 0x03 << (j + i*2);
-            e1223112
+            Kostya Shishkov
         motion_val += s->b8_stride;
+    }
-ecc97
+    if(s->first_slice_line)
         hmvmask &= ~0x000F;
     if(!s->mb_x)
         vmvmask &= ~0x1111;
-bc319c
+    if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
         vmvmask |= (vmvmask & 0x4444) >> 1;
         hmvmask |= (hmvmask & 0x0F00) >> 4;
         if(s->mb_x)
             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
         if(!s->first_slice_line)
             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
+    }
     return hmvmask | vmvmask;
-            e1223112
+            Kostya Shishkov
-            a52ef6a8
+static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
+{
     MpegEncContext *s = &r->s;
     GetBitContext *gb = &s->gb;
     int cbp, cbp2;
     int i, blknum, blkoff;
     DCTELEM block16[64];
     int luma_dc_quant;
     int dist;
     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
     // Calculate which neighbours are available. Maybe it's worth optimizing too.
     memset(r->avail_cache, 0, sizeof(r->avail_cache));
-c4a
+    fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
-            a52ef6a8
+    dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
     if(s->mb_x && dist)
-c4a
+        r->avail_cache[5] =
         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
-            a52ef6a8
+    if(dist >= s->mb_width)
-c4a
+        r->avail_cache[2] =
         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
-            a52ef6a8
+    if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
-c4a
+        r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
-            a52ef6a8
+    if(s->mb_x && dist > s->mb_width)
-c4a
+        r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
-            a52ef6a8
+            Kostya Shishkov
     s->qscale = r->si.quant;
     cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
-            bb39171b
+    r->cbp_luma  [mb_pos] = cbp;
     r->cbp_chroma[mb_pos] = cbp >> 16;
-            e1223112
+    if(s->pict_type == FF_I_TYPE)
-c3fca64
+        r->deblock_coefs[mb_pos] = 0xFFFF;
-            e1223112
+    else
-c3fca64
+        r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
-            bb39171b
+    s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
-            a52ef6a8
+            Kostya Shishkov
     if(cbp == -1)
         return -1;
-ce2f9c
+    luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
-            a52ef6a8
+    if(r->is16){
         memset(block16, 0, sizeof(block16));
         rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
         rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
         rv34_inv_transform_noround(block16);
+    }
     for(i = 0; i < 16; i++, cbp >>= 1){
         if(!r->is16 && !(cbp & 1)) continue;
         blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
         blkoff = ((i & 1) << 2) + ((i & 4) << 3);
         if(cbp & 1)
             rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
-ce2f9c
+        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
-            a52ef6a8
+        if(r->is16) //FIXME: optimize
             s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
         rv34_inv_transform(s->block[blknum] + blkoff);
+    }
     if(r->block_type == RV34_MB_P_MIX16x16)
         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
     for(; i < 24; i++, cbp >>= 1){
         if(!(cbp & 1)) continue;
         blknum = ((i & 4) >> 2) + 4;
         blkoff = ((i & 1) << 2) + ((i & 2) << 4);
         rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
         rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
         rv34_inv_transform(s->block[blknum] + blkoff);
+    }
-            bb39171b
+    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
-            a52ef6a8
+        rv34_output_macroblock(r, intra_types, cbp2, r->is16);
     else
         rv34_apply_differences(r, cbp2);
     return 0;
+}
 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
+{
     int bits;
     if(s->mb_y >= s->mb_height)
         return 1;
     if(!s->mb_num_left)
         return 1;
     if(r->s.mb_skip_run > 1)
         return 0;
     bits = r->bits - get_bits_count(&s->gb);
     if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
         return 1;
     return 0;
+}
 static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
+{
     return si1->type   != si2->type  ||
            si1->start  >= si2->start ||
            si1->width  != si2->width ||
-a8c
+           si1->height != si2->height||
            si1->pts    != si2->pts;
-            a52ef6a8
+            Kostya Shishkov
-b17c53
+static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
-            a52ef6a8
+            Kostya Shishkov
     MpegEncContext *s = &r->s;
     GetBitContext *gb = &s->gb;
     int mb_pos;
     int res;
     init_get_bits(&r->s.gb, buf, buf_size*8);
     res = r->parse_slice_header(r, gb, &r->si);
     if(res < 0){
         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
         return -1;
+    }
     if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
         if(s->width != r->si.width || s->height != r->si.height){
             av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
             MPV_common_end(s);
-            d90aeeaf
+            s->width  = r->si.width;
             s->height = r->si.height;
             avcodec_set_dimensions(s->avctx, s->width, s->height);
-            a52ef6a8
+            if(MPV_common_init(s) < 0)
                 return -1;
-c601b5
+            r->intra_types_stride = s->mb_width*4 + 4;
             r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
             r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
-            a52ef6a8
+            r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
-            b008e6b8
+            r->cbp_luma   = av_realloc(r->cbp_luma,   r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
             r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
-            def4f63b
+            r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
-            a52ef6a8
+            Kostya Shishkov
-            9701840b
+        s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
-            a52ef6a8
+        if(MPV_frame_start(s, s->avctx) < 0)
             return -1;
         ff_er_frame_start(s);
-a8c
+        r->cur_pts = r->si.pts;
         if(s->pict_type != FF_B_TYPE){
             r->last_pts = r->next_pts;
             r->next_pts = r->cur_pts;
+        }
-            a52ef6a8
+        s->mb_x = s->mb_y = 0;
+    }
     r->si.end = end;
     s->qscale = r->si.quant;
     r->bits = buf_size*8;
     s->mb_num_left = r->si.end - r->si.start;
     r->s.mb_skip_run = 0;
     mb_pos = s->mb_x + s->mb_y * s->mb_width;
     if(r->si.start != mb_pos){
         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
         s->mb_x = r->si.start % s->mb_width;
         s->mb_y = r->si.start / s->mb_width;
+    }
-c601b5
+    memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
-            a52ef6a8
+    s->first_slice_line = 1;
     s->resync_mb_x= s->mb_x;
     s->resync_mb_y= s->mb_y;
     ff_init_block_index(s);
     while(!check_slice_end(r, s)) {
         ff_update_block_index(s);
         s->dsp.clear_blocks(s->block[0]);
-c601b5
+        if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){
-            a52ef6a8
+            ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_ERROR|DC_ERROR|MV_ERROR);
             return -1;
+        }
         if (++s->mb_x == s->mb_width) {
             s->mb_x = 0;
             s->mb_y++;
             ff_init_block_index(s);
-c601b5
+            memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
-            de8cac16
+            Kostya Shishkov
             if(r->loop_filter && s->mb_y >= 2)
                 r->loop_filter(r, s->mb_y - 2);
-            a52ef6a8
+            Kostya Shishkov
         if(s->mb_x == s->resync_mb_x)
             s->first_slice_line=0;
         s->mb_num_left--;
+    }
     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_END|DC_END|MV_END);
-            ccd425e7
+    return s->mb_y == s->mb_height;
-            a52ef6a8
+            Kostya Shishkov
 /** @} */ // recons group end
 /**
  * Initialize decoder.
  */
-a6fff9
+av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
-            a52ef6a8
+            Kostya Shishkov
     RV34DecContext *r = avctx->priv_data;
     MpegEncContext *s = &r->s;
     MPV_decode_defaults(s);
     s->avctx= avctx;
     s->out_format = FMT_H263;
     s->codec_id= avctx->codec_id;
     s->width = avctx->width;
     s->height = avctx->height;
     r->s.avctx = avctx;
     avctx->flags |= CODEC_FLAG_EMU_EDGE;
     r->s.flags |= CODEC_FLAG_EMU_EDGE;
     avctx->pix_fmt = PIX_FMT_YUV420P;
     avctx->has_b_frames = 1;
     s->low_delay = 0;
     if (MPV_common_init(s) < 0)
         return -1;
     ff_h264_pred_init(&r->h, CODEC_ID_RV40);
-c601b5
+    r->intra_types_stride = 4*s->mb_stride + 4;
     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
-            a52ef6a8
+            Kostya Shishkov
     r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
-            b008e6b8
+    r->cbp_luma   = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
     r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
-            def4f63b
+    r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
-            b008e6b8
+            Kostya Shishkov
-c4b5eb8
+    if(!intra_vlcs[0].cbppattern[0].bits)
         rv34_init_tables();
-            a52ef6a8
+            Kostya Shishkov
     return 0;
+}
-b17c53
+static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
-            a52ef6a8
+            Kostya Shishkov
     if(avctx->slice_count) return avctx->slice_offset[n];
     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
+}
 int ff_rv34_decode_frame(AVCodecContext *avctx,
                             void *data, int *data_size,
-a00bbad
+                            AVPacket *avpkt)
-            a52ef6a8
+            Kostya Shishkov
-a00bbad
+    const uint8_t *buf = avpkt->data;
     int buf_size = avpkt->size;
-            a52ef6a8
+    RV34DecContext *r = avctx->priv_data;
     MpegEncContext *s = &r->s;
     AVFrame *pict = data;
     SliceInfo si;
     int i;
     int slice_count;
-b17c53
+    const uint8_t *slices_hdr = NULL;
-            a52ef6a8
+    int last = 0;
     /* no supplementary picture */
     if (buf_size == 0) {
         /* special case for last picture */
         if (s->low_delay==0 && s->next_picture_ptr) {
             *pict= *(AVFrame*)s->next_picture_ptr;
             s->next_picture_ptr= NULL;
             *data_size = sizeof(AVFrame);
+        }
         return 0;
+    }
     if(!avctx->slice_count){
         slice_count = (*buf++) + 1;
         slices_hdr = buf + 4;
         buf += 8 * slice_count;
     }else
         slice_count = avctx->slice_count;
-f1f11
+    //parse first slice header to check whether this frame can be decoded
     if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){
         av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
         return -1;
+    }
     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
-dd3974
+    if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
-f1f11
+        av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
         return -1;
+    }
     if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == FF_B_TYPE)
         return -1;
-fa70598
+    /* skip b frames if we are in a hurry */
     if(avctx->hurry_up && si.type==FF_B_TYPE) return buf_size;
     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==FF_B_TYPE)
        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=FF_I_TYPE)
        ||  avctx->skip_frame >= AVDISCARD_ALL)
         return buf_size;
     /* skip everything if we are in a hurry>=5 */
     if(avctx->hurry_up>=5)
         return buf_size;
-f1f11
+            Kostya Shishkov
-            a52ef6a8
+    for(i=0; i<slice_count; i++){
         int offset= get_slice_offset(avctx, slices_hdr, i);
         int size;
         if(i+1 == slice_count)
             size= buf_size - offset;
         else
             size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
-d522
+        if(offset > buf_size){
             av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
             break;
+        }
-            a52ef6a8
+        r->si.end = s->mb_width * s->mb_height;
         if(i+1 < slice_count){
             init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8);
             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
                 if(i+2 < slice_count)
                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
                 else
                     size = buf_size - offset;
             }else
                 r->si.end = si.start;
+        }
         last = rv34_decode_slice(r, r->si.end, buf + offset, size);
         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
         if(last)
             break;
+    }
     if(last){
         if(r->loop_filter)
-            de8cac16
+            r->loop_filter(r, s->mb_height - 1);
-            a52ef6a8
+        ff_er_frame_end(s);
         MPV_frame_end(s);
-            9701840b
+        if (s->pict_type == FF_B_TYPE || s->low_delay) {
-            a52ef6a8
+            *pict= *(AVFrame*)s->current_picture_ptr;
         } else if (s->last_picture_ptr != NULL) {
             *pict= *(AVFrame*)s->last_picture_ptr;
+        }
         if(s->last_picture_ptr || s->low_delay){
             *data_size = sizeof(AVFrame);
             ff_print_debug_info(s, pict);
+        }
         s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
+    }
     return buf_size;
+}
-a6fff9
+av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
-            a52ef6a8
+            Kostya Shishkov
     RV34DecContext *r = avctx->priv_data;
     MPV_common_end(&r->s);
     av_freep(&r->intra_types_hist);
     r->intra_types = NULL;
     av_freep(&r->mb_type);
-            ccfe8ae3
+    av_freep(&r->cbp_luma);
     av_freep(&r->cbp_chroma);
     av_freep(&r->deblock_coefs);
-            a52ef6a8
+            Kostya Shishkov
     return 0;
+}