Merge remote-tracking branch 'qatar/master'
[ffmpeg.git] / libavcodec / rv34.c
index 32d8198..6927387 100644 (file)
  * RV30/40 decoder common data
  */
 
+#include "libavutil/internal.h"
+
 #include "avcodec.h"
 #include "dsputil.h"
 #include "mpegvideo.h"
 #include "golomb.h"
+#include "internal.h"
 #include "mathops.h"
 #include "rectangle.h"
+#include "thread.h"
 
 #include "rv34vlc.h"
 #include "rv34data.h"
@@ -62,8 +66,10 @@ static const int rv34_mb_type_to_lavc[12] = {
 
 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
 
+static int rv34_decode_mv(RV34DecContext *r, int block_type);
+
 /**
- * @defgroup vlc RV30/40 VLC generating functions
+ * @name RV30/40 VLC generating functions
  * @{
  */
 
@@ -125,10 +131,10 @@ static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t
 
     vlc->table = &table_data[table_offs[num]];
     vlc->table_allocated = table_offs[num + 1] - table_offs[num];
-    init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
-                    bits2, 1, 1,
-                    cw,    2, 2,
-                    syms,  2, 2, INIT_VLC_USE_NEW_STATIC);
+    ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
+                       bits2, 1, 1,
+                       cw,    2, 2,
+                       syms,  2, 2, INIT_VLC_USE_NEW_STATIC);
 }
 
 /**
@@ -169,84 +175,8 @@ static av_cold void rv34_init_tables(void)
 
 /** @} */ // 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
+ * @name RV30/40 4x4 block decoding functions
  * @{
  */
 
@@ -274,7 +204,7 @@ static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
     }
 
     for(i = 0; i < 4; i++){
-        t = modulo_three_table[code][i];
+        t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
         if(t == 1)
             cbp |= cbp_masks[get_bits1(gb)] << i;
         if(t == 2)
@@ -286,7 +216,7 @@ static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
 /**
  * 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)
+static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
 {
     if(coef){
         if(coef == esc){
@@ -299,30 +229,51 @@ static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *
         }
         if(get_bits1(gb))
             coef = -coef;
-        *dst = coef;
+        *dst = (coef*q + 8) >> 4;
     }
 }
 
 /**
  * Decode 2x2 subblock of coefficients.
  */
-static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
+static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
+{
+    int flags = modulo_three_table[code];
+
+    decode_coeff(    dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q);
+    if(is_block2){
+        decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
+        decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
+    }else{
+        decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
+        decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
+    }
+    decode_coeff(    dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
+}
+
+/**
+ * Decode a single coefficient.
+ */
+static inline void decode_subblock1(DCTELEM *dst, int code, GetBitContext *gb, VLC *vlc, int q)
 {
-    int coeffs[4];
+    int coeff = modulo_three_table[code] >> 6;
+    decode_coeff(dst, coeff, 3, gb, vlc, q);
+}
+
+static inline void decode_subblock3(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc,
+                                    int q_dc, int q_ac1, int q_ac2)
+{
+    int flags = modulo_three_table[code];
 
-    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);
+    decode_coeff(    dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q_dc);
     if(is_block2){
-        decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
-        decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
+        decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
+        decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
     }else{
-        decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
-        decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
+        decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
+        decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
     }
-    decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
+    decode_coeff(    dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
 }
 
 /**
@@ -336,64 +287,42 @@ static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2,
  *  o--o
  */
 
-static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
+static inline int rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
 {
-    int code, pattern;
+    int code, pattern, has_ac = 1;
 
     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 (modulo_three_table[code] & 0x3F) {
+        decode_subblock3(dst, code, 0, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
+    } else {
+        decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
+        if (!pattern)
+            return 0;
+        has_ac = 0;
+    }
 
     if(pattern & 4){
         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
-        decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
+        decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
     }
     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);
+        decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
     }
     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);
+        decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
     }
-
+    return has_ac || pattern;
 }
 
 /**
- * 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
-
-
-/**
- * @defgroup rv3040_bitstream RV30/40 bitstream parsing
+ * @name RV30/40 bitstream parsing
  * @{
  */
 
@@ -422,20 +351,101 @@ static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
 }
 
 /**
- * Decode quantizer difference and return modified quantizer.
+ * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
  */
-static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
+static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
 {
-    if(get_bits1(gb))
-        return rv34_dquant_tab[get_bits1(gb)][quant];
-    else
-        return get_bits(gb, 5);
+    MpegEncContext *s = &r->s;
+    GetBitContext *gb = &s->gb;
+    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
+    int t;
+
+    r->is16 = get_bits1(gb);
+    if(r->is16){
+        s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA16x16;
+        r->block_type = RV34_MB_TYPE_INTRA16x16;
+        t = get_bits(gb, 2);
+        fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
+        r->luma_vlc   = 2;
+    }else{
+        if(!r->rv30){
+            if(!get_bits1(gb))
+                av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
+        }
+        s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA;
+        r->block_type = RV34_MB_TYPE_INTRA;
+        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);
+
+    return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
+}
+
+/**
+ * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
+ */
+static int rv34_decode_inter_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;
+
+    r->block_type = r->decode_mb_info(r);
+    if(r->block_type == -1)
+        return -1;
+    s->current_picture_ptr->f.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){
+        if(s->pict_type == AV_PICTURE_TYPE_P)
+            r->mb_type[mb_pos] = RV34_MB_P_16x16;
+        if(s->pict_type == AV_PICTURE_TYPE_B)
+            r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
+    }
+    r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->f.mb_type[mb_pos]);
+    rv34_decode_mv(r, r->block_type);
+    if(r->block_type == RV34_MB_SKIP){
+        fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
+        return 0;
+    }
+    r->chroma_vlc = 1;
+    r->luma_vlc   = 0;
+
+    if(IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
+        if(r->is16){
+            t = get_bits(gb, 2);
+            fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
+            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++)
+            intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
+        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);
 }
 
 /** @} */ //bitstream functions
 
 /**
- * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
+ * @name motion vector related code (prediction, reconstruction, motion compensation)
  * @{
  */
 
@@ -470,27 +480,27 @@ static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int
         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];
+        A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][0];
+        A[1] = s->current_picture_ptr->f.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];
+        B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][0];
+        B[1] = s->current_picture_ptr->f.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];
+            C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][0];
+            C[1] = s->current_picture_ptr->f.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];
+        C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][0];
+        C[1] = s->current_picture_ptr->f.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]);
@@ -498,8 +508,8 @@ static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int
     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;
+            s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
+            s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
         }
     }
 }
@@ -511,12 +521,8 @@ static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int
  */
 static int calc_add_mv(RV34DecContext *r, int dir, int val)
 {
-    int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
-    int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
-    int mul;
+    int mul = dir ? -r->weight2 : r->weight1;
 
-    if(!refdist) return 0;
-    mul = (dist << 14) / refdist;
     return (val * mul + 0x2000) >> 14;
 }
 
@@ -554,28 +560,28 @@ static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
     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];
+    int type = cur_pic->f.mb_type[mb_pos];
 
     memset(A, 0, sizeof(A));
     memset(B, 0, sizeof(B));
     memset(C, 0, sizeof(C));
     if((r->avail_cache[6-1] & type) & mask){
-        A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
-        A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
+        A[0] = cur_pic->f.motion_val[dir][mv_pos - 1][0];
+        A[1] = cur_pic->f.motion_val[dir][mv_pos - 1][1];
         has_A = 1;
     }
     if((r->avail_cache[6-4] & type) & mask){
-        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];
+        B[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][0];
+        B[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][1];
         has_B = 1;
     }
     if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
-        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];
+        C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];
+        C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][1];
         has_C = 1;
     }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
-        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];
+        C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][0];
+        C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][1];
         has_C = 1;
     }
 
@@ -586,12 +592,12 @@ static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
 
     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;
+            cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
+            cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
         }
     }
     if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
-        ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
+        ZERO8x2(cur_pic->f.motion_val[!dir][mv_pos], s->b8_stride);
     }
 }
 
@@ -608,27 +614,27 @@ static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
     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];
+        A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][0];
+        A[1] = s->current_picture_ptr->f.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];
+        B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][0];
+        B[1] = s->current_picture_ptr->f.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];
+            C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][0];
+            C[1] = s->current_picture_ptr->f.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];
+        C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][0];
+        C[1] = s->current_picture_ptr->f.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]);
@@ -637,8 +643,8 @@ static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
     for(j = 0; j < 2; j++){
         for(i = 0; i < 2; i++){
             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;
+                s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
+                s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
             }
         }
     }
@@ -664,7 +670,7 @@ static const int chroma_coeffs[3] = { 0, 3, 5 };
 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,
+                          const int thirdpel, int weighted,
                           qpel_mc_func (*qpel_mc)[16],
                           h264_chroma_mc_func (*chroma_mc))
 {
@@ -676,24 +682,24 @@ static inline void rv34_mc(RV34DecContext *r, const int block_type,
 
     if(thirdpel){
         int chroma_mx, chroma_my;
-        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;
-        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;
+        mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
+        my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
+        lx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
+        ly = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
+        chroma_mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
+        chroma_my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
         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];
     }else{
         int cx, cy;
-        mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
-        my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
-        lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
-        ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
-        cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
-        cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
+        mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] >> 2;
+        my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] >> 2;
+        lx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] & 3;
+        ly = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] & 3;
+        cx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
+        cy = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
         umx = cx >> 2;
         umy = cy >> 2;
         uvmx = (cx & 3) << 1;
@@ -702,10 +708,18 @@ static inline void rv34_mc(RV34DecContext *r, const int block_type,
         if(uvmx == 6 && uvmy == 6)
             uvmx = uvmy = 4;
     }
+
+    if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
+        /* wait for the referenced mb row to be finished */
+        int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
+        AVFrame *f = dir ? &s->next_picture_ptr->f : &s->last_picture_ptr->f;
+        ff_thread_await_progress(f, mb_row, 0);
+    }
+
     dxy = ly*4 + lx;
-    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];
+    srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];
+    srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];
+    srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];
     src_x = s->mb_x * 16 + xoff + mx;
     src_y = s->mb_y * 16 + yoff + my;
     uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
@@ -713,9 +727,10 @@ static inline void rv34_mc(RV34DecContext *r, const int block_type,
     srcY += src_y * s->linesize + src_x;
     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
-    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;
+    if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
+       (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;
 
         srcY -= 2 + 2*s->linesize;
         s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
@@ -728,9 +743,15 @@ static inline void rv34_mc(RV34DecContext *r, const int block_type,
         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(!weighted){
+        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;
+    }else{
+        Y = r->tmp_b_block_y [dir]     +  xoff     +  yoff    *s->linesize;
+        U = r->tmp_b_block_uv[dir*2]   + (xoff>>1) + (yoff>>1)*s->uvlinesize;
+        V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
+    }
 
     if(block_type == RV34_MB_P_16x8){
         qpel_mc[1][dxy](Y, srcY, s->linesize);
@@ -751,43 +772,70 @@ 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,
-            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
-                    : 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);
+    rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
+            r->rdsp.put_pixels_tab,
+            r->rdsp.put_chroma_pixels_tab);
+}
+
+static void rv4_weight(RV34DecContext *r)
+{
+    r->rdsp.rv40_weight_pixels_tab[0](r->s.dest[0],
+                                      r->tmp_b_block_y[0],
+                                      r->tmp_b_block_y[1],
+                                      r->weight1,
+                                      r->weight2,
+                                      r->s.linesize);
+    r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[1],
+                                      r->tmp_b_block_uv[0],
+                                      r->tmp_b_block_uv[2],
+                                      r->weight1,
+                                      r->weight2,
+                                      r->s.uvlinesize);
+    r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[2],
+                                      r->tmp_b_block_uv[1],
+                                      r->tmp_b_block_uv[3],
+                                      r->weight1,
+                                      r->weight2,
+                                      r->s.uvlinesize);
 }
 
 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
 {
-    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
-            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
-                    : 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);
-    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
-            r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
-                    : 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);
+    int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
+
+    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
+            r->rdsp.put_pixels_tab,
+            r->rdsp.put_chroma_pixels_tab);
+    if(!weighted){
+        rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
+                r->rdsp.avg_pixels_tab,
+                r->rdsp.avg_chroma_pixels_tab);
+    }else{
+        rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
+                r->rdsp.put_pixels_tab,
+                r->rdsp.put_chroma_pixels_tab);
+        rv4_weight(r);
+    }
 }
 
 static void rv34_mc_2mv_skip(RV34DecContext *r)
 {
     int i, j;
+    int weighted = !r->rv30 && r->weight1 != 8192;
+
     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
-                            : 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);
+                     weighted,
+                     r->rdsp.put_pixels_tab,
+                     r->rdsp.put_chroma_pixels_tab);
              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
-                            : 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);
+                     weighted,
+                     weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
+                     weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
         }
+    if(weighted)
+        rv4_weight(r);
 }
 
 /** number of motion vectors in each macroblock type */
@@ -813,31 +861,35 @@ static int rv34_decode_mv(RV34DecContext *r, int block_type)
     switch(block_type){
     case RV34_MB_TYPE_INTRA:
     case RV34_MB_TYPE_INTRA16x16:
-        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->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
         return 0;
     case RV34_MB_SKIP:
-        if(s->pict_type == FF_P_TYPE){
-            ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
+        if(s->pict_type == AV_PICTURE_TYPE_P){
+            ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
             rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
             break;
         }
     case RV34_MB_B_DIRECT:
         //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];
+        /* wait for the current mb row to be finished */
+        if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
+            ff_thread_await_progress(&s->next_picture_ptr->f, FFMAX(0, s->mb_y-1), 0);
+
+        next_bt = s->next_picture_ptr->f.mb_type[s->mb_x + s->mb_y * s->mb_stride];
         if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
-            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);
+            ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
+            ZERO8x2(s->current_picture_ptr->f.motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
         }else
             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]);
+                            s->current_picture_ptr->f.motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][k]);
         if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
             rv34_mc_2mv(r, block_type);
         else
             rv34_mc_2mv_skip(r);
-        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->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
         break;
     case RV34_MB_P_16x16:
     case RV34_MB_P_MIX16x16:
@@ -885,7 +937,7 @@ static int rv34_decode_mv(RV34DecContext *r, int block_type)
 /** @} */ // mv group
 
 /**
- * @defgroup recons Macroblock reconstruction functions
+ * @name Macroblock reconstruction functions
  * @{
  */
 /** mapping of RV30/40 intra prediction types to standard H.264 types */
@@ -923,21 +975,12 @@ static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int
         if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
     }
     if(!right && up){
-        topleft = dst[-stride + 3] * 0x01010101;
+        topleft = dst[-stride + 3] * 0x01010101u;
         prev = (uint8_t*)&topleft;
     }
     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)
@@ -954,15 +997,35 @@ static inline int adjust_pred16(int itype, int up, int left)
     return itype;
 }
 
-static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
+static inline void rv34_process_block(RV34DecContext *r,
+                                      uint8_t *pdst, int stride,
+                                      int fc, int sc, int q_dc, int q_ac)
 {
     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;
+    DCTELEM *ptr = s->block[0];
+    int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
+                                   fc, sc, q_dc, q_ac, q_ac);
+    if(has_ac){
+        r->rdsp.rv34_idct_add(pdst, stride, ptr);
+    }else{
+        r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
+        ptr[0] = 0;
+    }
+}
+
+static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
+{
+    LOCAL_ALIGNED_16(DCTELEM, block16, [16]);
+    MpegEncContext *s    = &r->s;
+    GetBitContext  *gb   = &s->gb;
+    int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
+                    q_ac = rv34_qscale_tab[s->qscale];
+    uint8_t        *dst  = s->dest[0];
+    DCTELEM        *ptr  = s->block[0];
+    int       avail[6*8] = {0};
+    int i, j, itype, has_ac;
+
+    memset(block16, 0, 16 * sizeof(*block16));
 
     // Set neighbour information.
     if(r->avail_cache[1])
@@ -978,151 +1041,118 @@ static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int c
     if(r->avail_cache[9])
         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;
-            intra_types += r->intra_types_stride;
-        }
-        intra_types -= r->intra_types_stride * 4;
-        fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
-        for(j = 0; j < 2; j++){
-            idx = 6 + j*4;
-            for(i = 0; i < 2; i++, cbp >>= 1, idx++){
-                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]);
-                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);
-            }
+    has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
+    if(has_ac)
+        r->rdsp.rv34_inv_transform(block16);
+    else
+        r->rdsp.rv34_inv_transform_dc(block16);
+
+    itype = ittrans16[intra_types[0]];
+    itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
+    r->h.pred16x16[itype](dst, s->linesize);
+
+    for(j = 0; j < 4; j++){
+        for(i = 0; i < 4; i++, cbp >>= 1){
+            int dc = block16[i + j*4];
+
+            if(cbp & 1){
+                has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
+            }else
+                has_ac = 0;
+
+            if(has_ac){
+                ptr[0] = dc;
+                r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
+            }else
+                r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
         }
-    }else{
-        itype = ittrans16[intra_types[0]];
-        itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
-        r->h.pred16x16[itype](Y, s->linesize);
-        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);
-
-        itype = ittrans16[intra_types[0]];
-        if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
-        itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
-        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);
+
+        dst += 4*s->linesize;
     }
-}
 
-/** @} */ // recons group
+    itype = ittrans16[intra_types[0]];
+    if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
+    itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
 
-/**
- * @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;
+    q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
+    q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
 
-    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){
-            if(s->pict_type == FF_P_TYPE)
-                r->mb_type[mb_pos] = RV34_MB_P_16x16;
-            if(s->pict_type == FF_B_TYPE)
-                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){
-            fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
-            return 0;
+    for(j = 1; j < 3; j++){
+        dst = s->dest[j];
+        r->h.pred8x8[itype](dst, s->uvlinesize);
+        for(i = 0; i < 4; i++, cbp >>= 1){
+            uint8_t *pdst;
+            if(!(cbp & 1)) continue;
+            pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
+
+            rv34_process_block(r, pdst, s->uvlinesize,
+                               r->chroma_vlc, 1, q_dc, q_ac);
         }
-        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);
-            fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
-            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++)
-            intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
-        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);
+}
+
+static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
+{
+    MpegEncContext *s   = &r->s;
+    uint8_t        *dst = s->dest[0];
+    int      avail[6*8] = {0};
+    int i, j, k;
+    int idx, q_ac, q_dc;
+
+    // Set neighbour information.
+    if(r->avail_cache[1])
+        avail[0] = 1;
+    if(r->avail_cache[2])
+        avail[1] = avail[2] = 1;
+    if(r->avail_cache[3])
+        avail[3] = avail[4] = 1;
+    if(r->avail_cache[4])
+        avail[5] = 1;
+    if(r->avail_cache[5])
+        avail[8] = avail[16] = 1;
+    if(r->avail_cache[9])
+        avail[24] = avail[32] = 1;
+
+    q_ac = rv34_qscale_tab[s->qscale];
+    for(j = 0; j < 4; j++){
+        idx = 9 + j*8;
+        for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
+            rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
+            avail[idx] = 1;
+            if(!(cbp & 1)) continue;
+
+            rv34_process_block(r, dst, s->linesize,
+                               r->luma_vlc, 0, q_ac, q_ac);
         }
+        dst += s->linesize * 4 - 4*4;
+        intra_types += r->intra_types_stride;
     }
 
-    return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
-}
+    intra_types -= r->intra_types_stride * 4;
 
-/**
- * @addtogroup recons
- * @{
- */
-/**
- * mask for retrieving all bits in coded block pattern
- * corresponding to one 8x8 block
- */
-#define LUMA_CBP_BLOCK_MASK 0x33
+    q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
+    q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
 
-#define U_CBP_MASK 0x0F0000
-#define V_CBP_MASK 0xF00000
+    for(k = 0; k < 2; k++){
+        dst = s->dest[1+k];
+        fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
 
+        for(j = 0; j < 2; j++){
+            int* acache = r->avail_cache + 6 + j*4;
+            for(i = 0; i < 2; i++, cbp >>= 1, acache++){
+                int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
+                rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
+                acache[0] = 1;
 
-static void rv34_apply_differences(RV34DecContext *r, int cbp)
-{
-    static const int shifts[4] = { 0, 2, 8, 10 };
-    MpegEncContext *s = &r->s;
-    int i;
+                if(!(cbp&1)) continue;
+
+                rv34_process_block(r, dst + 4*i, s->uvlinesize,
+                                   r->chroma_vlc, 1, q_dc, q_ac);
+            }
 
-    for(i = 0; i < 4; i++)
-        if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
-            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);
+            dst += 4*s->uvlinesize;
+        }
+    }
 }
 
 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
@@ -1142,7 +1172,7 @@ static int rv34_set_deblock_coef(RV34DecContext *r)
     MpegEncContext *s = &r->s;
     int hmvmask = 0, vmvmask = 0, i, j;
     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
-    int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
+    int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];
     for(j = 0; j < 16; j += 8){
         for(i = 0; i < 2; i++){
             if(is_mv_diff_gt_3(motion_val + i, 1))
@@ -1167,16 +1197,17 @@ static int rv34_set_deblock_coef(RV34DecContext *r)
     return hmvmask | vmvmask;
 }
 
-static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
+static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
 {
-    MpegEncContext *s = &r->s;
-    GetBitContext *gb = &s->gb;
+    MpegEncContext *s   = &r->s;
+    GetBitContext  *gb  = &s->gb;
+    uint8_t        *dst = s->dest[0];
+    DCTELEM        *ptr = s->block[0];
+    int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;
     int cbp, cbp2;
-    int i, blknum, blkoff;
-    DCTELEM block16[64];
-    int luma_dc_quant;
+    int q_dc, q_ac, has_ac;
+    int i, j;
     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));
@@ -1184,65 +1215,136 @@ static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
     if(s->mb_x && dist)
         r->avail_cache[5] =
-        r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
+        r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
     if(dist >= s->mb_width)
         r->avail_cache[2] =
-        r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
+        r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
-        r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
+        r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
     if(s->mb_x && dist > s->mb_width)
-        r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
+        r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
 
     s->qscale = r->si.quant;
-    cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
+    cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
     r->cbp_luma  [mb_pos] = cbp;
     r->cbp_chroma[mb_pos] = cbp >> 16;
-    if(s->pict_type == FF_I_TYPE)
-        r->deblock_coefs[mb_pos] = 0xFFFF;
-    else
-        r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
-    s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
+    r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
+    s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
 
     if(cbp == -1)
         return -1;
 
-    luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
+    if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
+        if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
+        else        rv34_output_intra(r, intra_types, cbp);
+        return 0;
+    }
+
     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);
-        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
-        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)
+        // Only for RV34_MB_P_MIX16x16
+        LOCAL_ALIGNED_16(DCTELEM, block16, [16]);
+        memset(block16, 0, 16 * sizeof(*block16));
+        q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
+        q_ac = rv34_qscale_tab[s->qscale];
+        if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
+            r->rdsp.rv34_inv_transform(block16);
+        else
+            r->rdsp.rv34_inv_transform_dc(block16);
+
+        q_ac = rv34_qscale_tab[s->qscale];
+
+        for(j = 0; j < 4; j++){
+            for(i = 0; i < 4; i++, cbp >>= 1){
+                int      dc   = block16[i + j*4];
+
+                if(cbp & 1){
+                    has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
+                }else
+                    has_ac = 0;
+
+                if(has_ac){
+                    ptr[0] = dc;
+                    r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
+                }else
+                    r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
+            }
+
+            dst += 4*s->linesize;
+        }
+
         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);
-    }
-    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
-        rv34_output_macroblock(r, intra_types, cbp2, r->is16);
-    else
-        rv34_apply_differences(r, cbp2);
+    }else{
+        q_ac = rv34_qscale_tab[s->qscale];
+
+        for(j = 0; j < 4; j++){
+            for(i = 0; i < 4; i++, cbp >>= 1){
+                if(!(cbp & 1)) continue;
+
+                rv34_process_block(r, dst + 4*i, s->linesize,
+                                   r->luma_vlc, 0, q_ac, q_ac);
+            }
+            dst += 4*s->linesize;
+        }
+    }
+
+    q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
+    q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
+
+    for(j = 1; j < 3; j++){
+        dst = s->dest[j];
+        for(i = 0; i < 4; i++, cbp >>= 1){
+            uint8_t *pdst;
+            if(!(cbp & 1)) continue;
+            pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
+
+            rv34_process_block(r, pdst, s->uvlinesize,
+                               r->chroma_vlc, 1, q_dc, q_ac);
+        }
+    }
 
     return 0;
 }
 
+static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
+{
+    MpegEncContext *s = &r->s;
+    int cbp, 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));
+    fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
+    dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
+    if(s->mb_x && dist)
+        r->avail_cache[5] =
+        r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
+    if(dist >= s->mb_width)
+        r->avail_cache[2] =
+        r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
+    if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
+        r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
+    if(s->mb_x && dist > s->mb_width)
+        r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
+
+    s->qscale = r->si.quant;
+    cbp = rv34_decode_intra_mb_header(r, intra_types);
+    r->cbp_luma  [mb_pos] = cbp;
+    r->cbp_chroma[mb_pos] = cbp >> 16;
+    r->deblock_coefs[mb_pos] = 0xFFFF;
+    s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
+
+    if(cbp == -1)
+        return -1;
+
+    if(r->is16){
+        rv34_output_i16x16(r, intra_types, cbp);
+        return 0;
+    }
+
+    rv34_output_intra(r, intra_types, cbp);
+    return 0;
+}
+
 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
 {
     int bits;
@@ -1252,21 +1354,59 @@ static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
         return 1;
     if(r->s.mb_skip_run > 1)
         return 0;
-    bits = r->bits - get_bits_count(&s->gb);
+    bits = get_bits_left(&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)
+
+static void rv34_decoder_free(RV34DecContext *r)
+{
+    av_freep(&r->intra_types_hist);
+    r->intra_types = NULL;
+    av_freep(&r->tmp_b_block_base);
+    av_freep(&r->mb_type);
+    av_freep(&r->cbp_luma);
+    av_freep(&r->cbp_chroma);
+    av_freep(&r->deblock_coefs);
+}
+
+
+static int rv34_decoder_alloc(RV34DecContext *r)
+{
+    r->intra_types_stride = r->s.mb_width * 4 + 4;
+
+    r->cbp_chroma       = av_malloc(r->s.mb_stride * r->s.mb_height *
+                                    sizeof(*r->cbp_chroma));
+    r->cbp_luma         = av_malloc(r->s.mb_stride * r->s.mb_height *
+                                    sizeof(*r->cbp_luma));
+    r->deblock_coefs    = av_malloc(r->s.mb_stride * r->s.mb_height *
+                                    sizeof(*r->deblock_coefs));
+    r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
+                                    sizeof(*r->intra_types_hist));
+    r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *
+                                     sizeof(*r->mb_type));
+
+    if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&
+          r->intra_types_hist && r->mb_type)) {
+        rv34_decoder_free(r);
+        return AVERROR(ENOMEM);
+    }
+
+    r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
+
+    return 0;
+}
+
+
+static int rv34_decoder_realloc(RV34DecContext *r)
 {
-    return si1->type   != si2->type  ||
-           si1->start  >= si2->start ||
-           si1->width  != si2->width ||
-           si1->height != si2->height||
-           si1->pts    != si2->pts;
+    rv34_decoder_free(r);
+    return rv34_decoder_alloc(r);
 }
 
+
 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
 {
     MpegEncContext *s = &r->s;
@@ -1282,37 +1422,67 @@ static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int
     }
 
     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);
+        if (s->width != r->si.width || s->height != r->si.height) {
+            int err;
+
+            av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
+                   r->si.width, r->si.height);
+            ff_MPV_common_end(s);
             s->width  = r->si.width;
             s->height = r->si.height;
             avcodec_set_dimensions(s->avctx, s->width, s->height);
-            if(MPV_common_init(s) < 0)
-                return -1;
-            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;
-            r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
-            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));
-            r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
+            if ((err = ff_MPV_common_init(s)) < 0)
+                return err;
+            if ((err = rv34_decoder_realloc(r)) < 0)
+                return err;
         }
-        s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
-        if(MPV_frame_start(s, s->avctx) < 0)
+        s->pict_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
+        if(ff_MPV_frame_start(s, s->avctx) < 0)
             return -1;
         ff_er_frame_start(s);
+        if (!r->tmp_b_block_base) {
+            int i;
+
+            r->tmp_b_block_base = av_malloc(s->linesize * 48);
+            for (i = 0; i < 2; i++)
+                r->tmp_b_block_y[i] = r->tmp_b_block_base + i * 16 * s->linesize;
+            for (i = 0; i < 4; i++)
+                r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
+                                       + (i >> 1) * 8 * s->uvlinesize + (i & 1) * 16;
+        }
         r->cur_pts = r->si.pts;
-        if(s->pict_type != FF_B_TYPE){
+        if(s->pict_type != AV_PICTURE_TYPE_B){
             r->last_pts = r->next_pts;
             r->next_pts = r->cur_pts;
+        }else{
+            int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
+            int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);
+            int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);
+
+            if(!refdist){
+                r->weight1 = r->weight2 = 8192;
+            }else{
+                r->weight1 = (dist0 << 14) / refdist;
+                r->weight2 = (dist1 << 14) / refdist;
+            }
         }
         s->mb_x = s->mb_y = 0;
+        ff_thread_finish_setup(s->avctx);
+    } else {
+        int slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
+
+        if (slice_type != s->pict_type) {
+            av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
+            return AVERROR_INVALIDDATA;
+        }
+        if (s->width != r->si.width || s->height != r->si.height) {
+            av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
+            return AVERROR_INVALIDDATA;
+        }
     }
 
     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;
 
@@ -1324,16 +1494,19 @@ static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int
     }
     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
     s->first_slice_line = 1;
-    s->resync_mb_x= s->mb_x;
-    s->resync_mb_y= s->mb_y;
+    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]);
 
-        if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){
-            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);
+        if(r->si.type)
+            res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
+        else
+            res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
+        if(res < 0){
+            ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
             return -1;
         }
         if (++s->mb_x == s->mb_width) {
@@ -1346,12 +1519,17 @@ static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int
 
             if(r->loop_filter && s->mb_y >= 2)
                 r->loop_filter(r, s->mb_y - 2);
+
+            if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
+                ff_thread_report_progress(&s->current_picture_ptr->f,
+                                          s->mb_y - 2, 0);
+
         }
         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);
+    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
 
     return s->mb_y == s->mb_height;
 }
@@ -1365,13 +1543,14 @@ av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
 {
     RV34DecContext *r = avctx->priv_data;
     MpegEncContext *s = &r->s;
+    int ret;
 
-    MPV_decode_defaults(s);
-    s->avctx= avctx;
+    ff_MPV_decode_defaults(s);
+    s->avctx      = avctx;
     s->out_format = FMT_H263;
-    s->codec_id= avctx->codec_id;
+    s->codec_id   = avctx->codec_id;
 
-    s->width = avctx->width;
+    s->width  = avctx->width;
     s->height = avctx->height;
 
     r->s.avctx = avctx;
@@ -1381,20 +1560,22 @@ av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
     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, 8);
+    if ((ret = ff_MPV_common_init(s)) < 0)
+        return ret;
 
-    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;
+    ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8, 1);
 
-    r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
+#if CONFIG_RV30_DECODER
+    if (avctx->codec_id == CODEC_ID_RV30)
+        ff_rv30dsp_init(&r->rdsp, &r->s.dsp);
+#endif
+#if CONFIG_RV40_DECODER
+    if (avctx->codec_id == CODEC_ID_RV40)
+        ff_rv40dsp_init(&r->rdsp, &r->s.dsp);
+#endif
 
-    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));
-    r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
+    if ((ret = rv34_decoder_alloc(r)) < 0)
+        return ret;
 
     if(!intra_vlcs[0].cbppattern[0].bits)
         rv34_init_tables();
@@ -1402,6 +1583,58 @@ av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
     return 0;
 }
 
+int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)
+{
+    int err;
+    RV34DecContext *r = avctx->priv_data;
+
+    r->s.avctx = avctx;
+
+    if (avctx->internal->is_copy) {
+        r->tmp_b_block_base = NULL;
+        if ((err = ff_MPV_common_init(&r->s)) < 0)
+            return err;
+        if ((err = rv34_decoder_alloc(r)) < 0)
+            return err;
+    }
+    return 0;
+}
+
+int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
+{
+    RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
+    MpegEncContext * const s = &r->s, * const s1 = &r1->s;
+    int err;
+
+    if (dst == src || !s1->context_initialized)
+        return 0;
+
+    if (s->height != s1->height || s->width != s1->width) {
+        ff_MPV_common_end(s);
+        s->height = s1->height;
+        s->width  = s1->width;
+        if ((err = ff_MPV_common_init(s)) < 0)
+            return err;
+        if ((err = rv34_decoder_realloc(r)) < 0)
+            return err;
+    }
+
+    if ((err = ff_mpeg_update_thread_context(dst, src)))
+        return err;
+
+    r->cur_pts  = r1->cur_pts;
+    r->last_pts = r1->last_pts;
+    r->next_pts = r1->next_pts;
+
+    memset(&r->si, 0, sizeof(r->si));
+
+    /* necessary since it is it the condition checked for in decode_slice
+     * to call ff_MPV_frame_start. cmp. comment at the end of decode_frame */
+    s->current_picture_ptr = NULL;
+
+    return 0;
+}
+
 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
 {
     if(avctx->slice_count) return avctx->slice_offset[n];
@@ -1427,8 +1660,8 @@ int ff_rv34_decode_frame(AVCodecContext *avctx,
     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;
+            *pict = *(AVFrame*)s->next_picture_ptr;
+            s->next_picture_ptr = NULL;
 
             *data_size = sizeof(AVFrame);
         }
@@ -1439,50 +1672,52 @@ int ff_rv34_decode_frame(AVCodecContext *avctx,
         slice_count = (*buf++) + 1;
         slices_hdr = buf + 4;
         buf += 8 * slice_count;
+        buf_size -= 1 + 8 * slice_count;
     }else
         slice_count = avctx->slice_count;
 
     //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;
+    if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||
+       get_slice_offset(avctx, slices_hdr, 0) > buf_size){
+        av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
+        return AVERROR_INVALIDDATA;
     }
-    init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
+    init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);
     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
-    if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == FF_B_TYPE)
-        return -1;
-#if FF_API_HURRY_UP
-    /* skip b frames if we are in a hurry */
-    if(avctx->hurry_up && si.type==FF_B_TYPE) return buf_size;
-#endif
-    if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==FF_B_TYPE)
-       || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=FF_I_TYPE)
+    if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) &&
+        si.type == AV_PICTURE_TYPE_B) {
+        av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
+               "reference data.\n");
+        return AVERROR_INVALIDDATA;
+    }
+    if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
+       || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
        ||  avctx->skip_frame >= AVDISCARD_ALL)
-        return buf_size;
-#if FF_API_HURRY_UP
-    /* skip everything if we are in a hurry>=5 */
-    if(avctx->hurry_up>=5)
-        return buf_size;
-#endif
+        return avpkt->size;
 
-    for(i=0; i<slice_count; i++){
-        int offset= get_slice_offset(avctx, slices_hdr, i);
+    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;
+            size = buf_size - offset;
         else
-            size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
+            size = get_slice_offset(avctx, slices_hdr, i+1) - offset;
 
-        if(offset > buf_size){
-            av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
+        if(offset < 0 || offset > buf_size){
+            av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
             break;
         }
 
         r->si.end = s->mb_width * s->mb_height;
         if(i+1 < slice_count){
+            if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||
+                get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {
+                av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
+                break;
+            }
             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)
@@ -1492,44 +1727,46 @@ int ff_rv34_decode_frame(AVCodecContext *avctx,
             }else
                 r->si.end = si.start;
         }
+        if (size < 0 || size > buf_size - offset) {
+            av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");
+            break;
+        }
         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(last && s->current_picture_ptr){
         if(r->loop_filter)
             r->loop_filter(r, s->mb_height - 1);
         ff_er_frame_end(s);
-        MPV_frame_end(s);
-        if (s->pict_type == FF_B_TYPE || s->low_delay) {
-            *pict= *(AVFrame*)s->current_picture_ptr;
+        ff_MPV_frame_end(s);
+
+        if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
+            ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
+
+        if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
+            *pict = *(AVFrame*)s->current_picture_ptr;
         } else if (s->last_picture_ptr != NULL) {
-            *pict= *(AVFrame*)s->last_picture_ptr;
+            *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...)
+        s->current_picture_ptr = NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
     }
-    return buf_size;
+    return avpkt->size;
 }
 
 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
 {
     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);
-    av_freep(&r->cbp_luma);
-    av_freep(&r->cbp_chroma);
-    av_freep(&r->deblock_coefs);
+    ff_MPV_common_end(&r->s);
+    rv34_decoder_free(r);
 
     return 0;
 }