updated ffmpeg

git-svn-id: https://mpc-hc.svn.sourceforge.net/svnroot/mpc-hc/trunk@2200 10f7b99b-c216-0410-bff0-8a66a9350fd8

12 files changed, 214 insertions, 187 deletions

diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.c b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.c
index 2e0e1e7dc..eff067ad6 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.c
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.c
@@ -4332,10 +4332,6 @@ av_cold void attribute_align_arg dsputil_init(DSPContext* c, AVCodecContext *avc
 

     c->draw_edges = draw_edges_c;

 

-#if CONFIG_CAVS_DECODER

-    ff_cavsdsp_init(c,avctx);

-#endif

-

 #if CONFIG_MLP_DECODER || CONFIG_TRUEHD_DECODER

     ff_mlp_init(c, avctx);

 #endif

diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.h b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.h
index c2030f9b0..778d3dfc1 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.h
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/dsputil.h
@@ -327,15 +327,6 @@ typedef struct DSPContext {
     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];

     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];

 

-    /* AVS specific */

-    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];

-    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];

-    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

-    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

-    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

-    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

-    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);

-

     me_cmp_func pix_abs[2][4];

 

     /* huffyuv specific */

@@ -618,7 +609,6 @@ extern int mm_flags;
 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);

 

 void ff_dsputil_init_dwt(DSPContext *c);

-void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx);

 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);

 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);

 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);

diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp56.h b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp56.h
index 69518fa73..4b19c93b9 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp56.h
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp56.h
@@ -214,7 +214,7 @@ static av_always_inline unsigned int vp56_rac_renorm(VP56RangeCoder *c)
 
 #ifndef vp56_rac_get_prob
 #define vp56_rac_get_prob vp56_rac_get_prob
-static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)
+static av_always_inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)
 {
     unsigned int code_word = vp56_rac_renorm(c);
     unsigned int low = 1 + (((c->high - 1) * prob) >> 8);
@@ -246,7 +246,7 @@ static av_always_inline int vp56_rac_get_prob_branchy(VP56RangeCoder *c, int pro
     return 0;
 }
 
-static inline int vp56_rac_get(VP56RangeCoder *c)
+static av_always_inline int vp56_rac_get(VP56RangeCoder *c)
 {
     unsigned int code_word = vp56_rac_renorm(c);
     /* equiprobable */
@@ -265,12 +265,12 @@ static inline int vp56_rac_get(VP56RangeCoder *c)
 }
 
 // rounding is different than vp56_rac_get, is vp56_rac_get wrong?
-static inline int vp8_rac_get(VP56RangeCoder *c)
+static av_always_inline int vp8_rac_get(VP56RangeCoder *c)
 {
     return vp56_rac_get_prob(c, 128);
 }
 
-static inline int vp56_rac_gets(VP56RangeCoder *c, int bits)
+static av_unused int vp56_rac_gets(VP56RangeCoder *c, int bits)
 {
     int value = 0;
 
@@ -281,7 +281,7 @@ static inline int vp56_rac_gets(VP56RangeCoder *c, int bits)
     return value;
 }
 
-static inline int vp8_rac_get_uint(VP56RangeCoder *c, int bits)
+static av_unused int vp8_rac_get_uint(VP56RangeCoder *c, int bits)
 {
     int value = 0;
 
@@ -293,7 +293,7 @@ static inline int vp8_rac_get_uint(VP56RangeCoder *c, int bits)
 }
 
 // fixme: add 1 bit to all the calls to this?
-static inline int vp8_rac_get_sint(VP56RangeCoder *c, int bits)
+static av_unused int vp8_rac_get_sint(VP56RangeCoder *c, int bits)
 {
     int v;
 
@@ -309,21 +309,22 @@ static inline int vp8_rac_get_sint(VP56RangeCoder *c, int bits)
 }
 
 // P(7)
-static inline int vp56_rac_gets_nn(VP56RangeCoder *c, int bits)
+static av_unused int vp56_rac_gets_nn(VP56RangeCoder *c, int bits)
 {
     int v = vp56_rac_gets(c, 7) << 1;
     return v + !v;
 }
 
-static inline int vp8_rac_get_nn(VP56RangeCoder *c)
+static av_unused int vp8_rac_get_nn(VP56RangeCoder *c)
 {
     int v = vp8_rac_get_uint(c, 7) << 1;
     return v + !v;
 }
 
-static inline int vp56_rac_get_tree(VP56RangeCoder *c,
-                                    const VP56Tree *tree,
-                                    const uint8_t *probs)
+static av_always_inline
+int vp56_rac_get_tree(VP56RangeCoder *c,
+                      const VP56Tree *tree,
+                      const uint8_t *probs)
 {
     while (tree->val > 0) {
         if (vp56_rac_get_prob(c, probs[tree->prob_idx]))
@@ -339,8 +340,9 @@ static inline int vp56_rac_get_tree(VP56RangeCoder *c,
  * on a node other than the root node, needed for coeff decode where this is
  * used to save a bit after a 0 token (by disallowing EOB to immediately follow.)
  */
-static inline int vp8_rac_get_tree_with_offset(VP56RangeCoder *c, const int8_t (*tree)[2],
-                                               const uint8_t *probs, int i)
+static av_always_inline
+int vp8_rac_get_tree_with_offset(VP56RangeCoder *c, const int8_t (*tree)[2],
+                                 const uint8_t *probs, int i)
 {
     do {
         i = tree[i][vp56_rac_get_prob(c, probs[i])];
@@ -351,14 +353,15 @@ static inline int vp8_rac_get_tree_with_offset(VP56RangeCoder *c, const int8_t (
 
 // how probabilities are associated with decisions is different I think
 // well, the new scheme fits in the old but this way has one fewer branches per decision
-static inline int vp8_rac_get_tree(VP56RangeCoder *c, const int8_t (*tree)[2],
-                                   const uint8_t *probs)
+static av_always_inline
+int vp8_rac_get_tree(VP56RangeCoder *c, const int8_t (*tree)[2],
+                     const uint8_t *probs)
 {
     return vp8_rac_get_tree_with_offset(c, tree, probs, 0);
 }
 
 // DCTextra
-static inline int vp8_rac_get_coeff(VP56RangeCoder *c, const uint8_t *prob)
+static av_always_inline int vp8_rac_get_coeff(VP56RangeCoder *c, const uint8_t *prob)
 {
     int v = 0;
 
diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8.c b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8.c
index 6524a7141..6a1d6ac95 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8.c
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8.c
@@ -88,8 +88,8 @@ typedef struct {
     VP8FilterStrength *filter_strength;
     int mb_stride;
 
-    uint8_t *intra4x4_pred_mode;
-    uint8_t *intra4x4_pred_mode_base;
+    uint8_t *intra4x4_pred_mode_top;
+    uint8_t intra4x4_pred_mode_left[4];
     uint8_t *segmentation_map;
     int b4_stride;
 
@@ -117,6 +117,7 @@ typedef struct {
      */
     DECLARE_ALIGNED(16, uint8_t, non_zero_count_cache)[6][4];
     DECLARE_ALIGNED(16, DCTELEM, block)[6][4][16];
+    DECLARE_ALIGNED(16, DCTELEM, block_dc)[16];
     uint8_t intra4x4_pred_mode_mb[16];
 
     int chroma_pred_mode;    ///< 8x8c pred mode of the current macroblock
@@ -194,7 +195,8 @@ typedef struct {
         uint8_t golden;
         uint8_t pred16x16[4];
         uint8_t pred8x8c[3];
-        uint8_t token[4][8][3][NUM_DCT_TOKENS-1];
+        /* Padded to allow overreads */
+        uint8_t token[4][17][3][NUM_DCT_TOKENS-1];
         uint8_t mvc[2][19];
     } prob[2];
 } VP8Context;
@@ -211,20 +213,17 @@ static void vp8_decode_flush(AVCodecContext *avctx)
 
     av_freep(&s->macroblocks_base);
     av_freep(&s->filter_strength);
-    av_freep(&s->intra4x4_pred_mode_base);
+    av_freep(&s->intra4x4_pred_mode_top);
     av_freep(&s->top_nnz);
     av_freep(&s->edge_emu_buffer);
     av_freep(&s->top_border);
     av_freep(&s->segmentation_map);
 
     s->macroblocks        = NULL;
-    s->intra4x4_pred_mode = NULL;
 }
 
 static int update_dimensions(VP8Context *s, int width, int height)
 {
-    int i;
-
     if (avcodec_check_dimensions(s->avctx, width, height))
         return AVERROR_INVALIDDATA;
 
@@ -242,21 +241,16 @@ static int update_dimensions(VP8Context *s, int width, int height)
 
     s->macroblocks_base        = av_mallocz((s->mb_stride+s->mb_height*2+2)*sizeof(*s->macroblocks));
     s->filter_strength         = av_mallocz(s->mb_stride*sizeof(*s->filter_strength));
-    s->intra4x4_pred_mode_base = av_mallocz(s->b4_stride*(4*s->mb_height+1));
+    s->intra4x4_pred_mode_top  = av_mallocz(s->b4_stride*4);
     s->top_nnz                 = av_mallocz(s->mb_width*sizeof(*s->top_nnz));
     s->top_border              = av_mallocz((s->mb_width+1)*sizeof(*s->top_border));
     s->segmentation_map        = av_mallocz(s->mb_stride*s->mb_height);
 
-    if (!s->macroblocks_base || !s->filter_strength || !s->intra4x4_pred_mode_base ||
+    if (!s->macroblocks_base || !s->filter_strength || !s->intra4x4_pred_mode_top ||
         !s->top_nnz || !s->top_border || !s->segmentation_map)
         return AVERROR(ENOMEM);
 
     s->macroblocks        = s->macroblocks_base + 1;
-    s->intra4x4_pred_mode = s->intra4x4_pred_mode_base + 4 + s->b4_stride;
-
-    memset(s->intra4x4_pred_mode_base, DC_PRED, s->b4_stride);
-    for (i = 0; i < 4*s->mb_height; i++)
-        s->intra4x4_pred_mode[i*s->b4_stride-1] = DC_PRED;
 
     return 0;
 }
@@ -395,7 +389,7 @@ static void update_refs(VP8Context *s)
 static int decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
 {
     VP56RangeCoder *c = &s->c;
-    int header_size, hscale, vscale, i, j, k, l, ret;
+    int header_size, hscale, vscale, i, j, k, l, m, ret;
     int width  = s->avctx->width;
     int height = s->avctx->height;
 
@@ -435,7 +429,10 @@ static int decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
             av_log_missing_feature(s->avctx, "Upscaling", 1);
 
         s->update_golden = s->update_altref = VP56_FRAME_CURRENT;
-        memcpy(s->prob->token    , vp8_token_default_probs , sizeof(s->prob->token));
+        for (i = 0; i < 4; i++)
+            for (j = 0; j < 16; j++)
+                memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]],
+                       sizeof(s->prob->token[i][j]));
         memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter, sizeof(s->prob->pred16x16));
         memcpy(s->prob->pred8x8c , vp8_pred8x8c_prob_inter , sizeof(s->prob->pred8x8c));
         memcpy(s->prob->mvc      , vp8_mv_default_prob     , sizeof(s->prob->mvc));
@@ -495,8 +492,11 @@ static int decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
         for (j = 0; j < 8; j++)
             for (k = 0; k < 3; k++)
                 for (l = 0; l < NUM_DCT_TOKENS-1; l++)
-                    if (vp56_rac_get_prob_branchy(c, vp8_token_update_probs[i][j][k][l]))
-                        s->prob->token[i][j][k][l] = vp8_rac_get_uint(c, 8);
+                    if (vp56_rac_get_prob_branchy(c, vp8_token_update_probs[i][j][k][l])) {
+                        int prob = vp8_rac_get_uint(c, 8);
+                        for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++)
+                            s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob;
+                    }
 
     if ((s->mbskip_enabled = vp8_rac_get(c)))
         s->prob->mbskip = vp8_rac_get_uint(c, 8);
@@ -643,19 +643,32 @@ const uint8_t *get_submv_prob(uint32_t left, uint32_t top)
 static av_always_inline
 int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb)
 {
-    int part_idx = mb->partitioning =
-        vp8_rac_get_tree(c, vp8_mbsplit_tree, vp8_mbsplit_prob);
-    int n, num = vp8_mbsplit_count[part_idx];
+    int part_idx;
+    int n, num;
     VP8Macroblock *top_mb  = &mb[2];
     VP8Macroblock *left_mb = &mb[-1];
     const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning],
                   *mbsplits_top = vp8_mbsplits[top_mb->partitioning],
-                  *mbsplits_cur = vp8_mbsplits[part_idx],
-                  *firstidx = vp8_mbfirstidx[part_idx];
+                  *mbsplits_cur, *firstidx;
     VP56mv *top_mv  = top_mb->bmv;
     VP56mv *left_mv = left_mb->bmv;
     VP56mv *cur_mv  = mb->bmv;
 
+    if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) {
+        if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[1])) {
+            part_idx = VP8_SPLITMVMODE_16x8 + vp56_rac_get_prob(c, vp8_mbsplit_prob[2]);
+        } else {
+            part_idx = VP8_SPLITMVMODE_8x8;
+        }
+    } else {
+        part_idx = VP8_SPLITMVMODE_4x4;
+    }
+
+    num = vp8_mbsplit_count[part_idx];
+    mbsplits_cur = vp8_mbsplits[part_idx],
+    firstidx = vp8_mbfirstidx[part_idx];
+    mb->partitioning = part_idx;
+
     for (n = 0; n < num; n++) {
         int k = firstidx[n];
         uint32_t left, above;
@@ -672,20 +685,19 @@ int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb)
 
         submv_prob = get_submv_prob(left, above);
 
-        switch (vp8_rac_get_tree(c, vp8_submv_ref_tree, submv_prob)) {
-        case VP8_SUBMVMODE_NEW4X4:
-            mb->bmv[n].y = mb->mv.y + read_mv_component(c, s->prob->mvc[0]);
-            mb->bmv[n].x = mb->mv.x + read_mv_component(c, s->prob->mvc[1]);
-            break;
-        case VP8_SUBMVMODE_ZERO4X4:
-            AV_ZERO32(&mb->bmv[n]);
-            break;
-        case VP8_SUBMVMODE_LEFT4X4:
+        if (vp56_rac_get_prob_branchy(c, submv_prob[0])) {
+            if (vp56_rac_get_prob_branchy(c, submv_prob[1])) {
+                if (vp56_rac_get_prob_branchy(c, submv_prob[2])) {
+                    mb->bmv[n].y = mb->mv.y + read_mv_component(c, s->prob->mvc[0]);
+                    mb->bmv[n].x = mb->mv.x + read_mv_component(c, s->prob->mvc[1]);
+                } else {
+                    AV_ZERO32(&mb->bmv[n]);
+                }
+            } else {
+                AV_WN32A(&mb->bmv[n], above);
+            }
+        } else {
             AV_WN32A(&mb->bmv[n], left);
-            break;
-        case VP8_SUBMVMODE_TOP4X4:
-            AV_WN32A(&mb->bmv[n], above);
-            break;
         }
     }
 
@@ -693,31 +705,32 @@ int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb)
 }
 
 static av_always_inline
-void decode_intra4x4_modes(VP56RangeCoder *c, uint8_t *intra4x4,
-                           int stride, int keyframe)
+void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c,
+                           int mb_x, int keyframe)
 {
-    int x, y, t, l, i;
-
+    uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
     if (keyframe) {
-        const uint8_t *ctx;
+        int x, y;
+        uint8_t* const top = s->intra4x4_pred_mode_top + 4 * mb_x;
+        uint8_t* const left = s->intra4x4_pred_mode_left;
         for (y = 0; y < 4; y++) {
             for (x = 0; x < 4; x++) {
-                t = intra4x4[x - stride];
-                l = intra4x4[x - 1];
-                ctx = vp8_pred4x4_prob_intra[t][l];
-                intra4x4[x] = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx);
+                const uint8_t *ctx;
+                ctx = vp8_pred4x4_prob_intra[top[x]][left[y]];
+                *intra4x4 = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx);
+                left[y] = top[x] = *intra4x4;
+                intra4x4++;
             }
-            intra4x4 += stride;
         }
     } else {
+        int i;
         for (i = 0; i < 16; i++)
             intra4x4[i] = vp8_rac_get_tree(c, vp8_pred4x4_tree, vp8_pred4x4_prob_inter);
     }
 }
 
 static av_always_inline
-void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
-                    uint8_t *intra4x4, uint8_t *segment)
+void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, uint8_t *segment)
 {
     VP56RangeCoder *c = &s->c;
 
@@ -731,16 +744,18 @@ void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
         mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_intra, vp8_pred16x16_prob_intra);
 
         if (mb->mode == MODE_I4x4) {
-            decode_intra4x4_modes(c, intra4x4, s->b4_stride, 1);
-        } else
-            fill_rectangle(intra4x4, 4, 4, s->b4_stride, vp8_pred4x4_mode[mb->mode], 1);
+            decode_intra4x4_modes(s, c, mb_x, 1);
+        } else {
+            const uint32_t modes = vp8_pred4x4_mode[mb->mode] * 0x01010101u;
+            AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
+            AV_WN32A(s->intra4x4_pred_mode_left, modes);
+        }
 
         s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, vp8_pred8x8c_prob_intra);
         mb->ref_frame = VP56_FRAME_CURRENT;
     } else if (vp56_rac_get_prob_branchy(c, s->prob->intra)) {
         VP56mv near[2], best;
         uint8_t cnt[4] = { 0 };
-        uint8_t p[4];
 
         // inter MB, 16.2
         if (vp56_rac_get_prob_branchy(c, s->prob->last))
@@ -752,30 +767,30 @@ void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
 
         // motion vectors, 16.3
         find_near_mvs(s, mb, mb_x, mb_y, near, &best, cnt);
-        p[0] = vp8_mode_contexts[cnt[0]][0];
-        p[1] = vp8_mode_contexts[cnt[1]][1];
-        p[2] = vp8_mode_contexts[cnt[2]][2];
-        p[3] = vp8_mode_contexts[cnt[3]][3];
-        mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_mvinter, p);
-        switch (mb->mode) {
-        case VP8_MVMODE_SPLIT:
-            clamp_mv(s, &mb->mv, &mb->mv, mb_x, mb_y);
-            mb->mv = mb->bmv[decode_splitmvs(s, c, mb) - 1];
-            break;
-        case VP8_MVMODE_ZERO:
+        if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[0]][0])) {
+            if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[1]][1])) {
+                if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[2]][2])) {
+                    if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[3]][3])) {
+                        mb->mode = VP8_MVMODE_SPLIT;
+                        clamp_mv(s, &mb->mv, &mb->mv, mb_x, mb_y);
+                        mb->mv = mb->bmv[decode_splitmvs(s, c, mb) - 1];
+                    } else {
+                        mb->mode = VP8_MVMODE_NEW;
+                        clamp_mv(s, &mb->mv, &mb->mv, mb_x, mb_y);
+                        mb->mv.y += + read_mv_component(c, s->prob->mvc[0]);
+                        mb->mv.x += + read_mv_component(c, s->prob->mvc[1]);
+                    }
+                } else {
+                    mb->mode = VP8_MVMODE_NEAR;
+                    clamp_mv(s, &mb->mv, &near[1], mb_x, mb_y);
+                }
+            } else {
+                mb->mode = VP8_MVMODE_NEAREST;
+                clamp_mv(s, &mb->mv, &near[0], mb_x, mb_y);
+            }
+        } else {
+            mb->mode = VP8_MVMODE_ZERO;
             AV_ZERO32(&mb->mv);
-            break;
-        case VP8_MVMODE_NEAREST:
-            clamp_mv(s, &mb->mv, &near[0], mb_x, mb_y);
-            break;
-        case VP8_MVMODE_NEAR:
-            clamp_mv(s, &mb->mv, &near[1], mb_x, mb_y);
-            break;
-        case VP8_MVMODE_NEW:
-            clamp_mv(s, &mb->mv, &mb->mv, mb_x, mb_y);
-            mb->mv.y += + read_mv_component(c, s->prob->mvc[0]);
-            mb->mv.x += + read_mv_component(c, s->prob->mvc[1]);
-            break;
         }
         if (mb->mode != VP8_MVMODE_SPLIT) {
             mb->partitioning = VP8_SPLITMVMODE_NONE;
@@ -786,7 +801,7 @@ void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
         mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16);
 
         if (mb->mode == MODE_I4x4)
-            decode_intra4x4_modes(c, intra4x4, 4, 0);
+            decode_intra4x4_modes(s, c, mb_x, 0);
 
         s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, s->prob->pred8x8c);
         mb->ref_frame = VP56_FRAME_CURRENT;
@@ -806,37 +821,30 @@ void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
  * @return 0 if no coeffs were decoded
  *         otherwise, the index of the last coeff decoded plus one
  */
-static int decode_block_coeffs(VP56RangeCoder *c, DCTELEM block[16],
-                               uint8_t probs[8][3][NUM_DCT_TOKENS-1],
-                               int i, int zero_nhood, int16_t qmul[2])
+static int decode_block_coeffs_internal(VP56RangeCoder *c, DCTELEM block[16],
+                                        uint8_t probs[8][3][NUM_DCT_TOKENS-1],
+                                        int i, uint8_t *token_prob, int16_t qmul[2])
 {
-    uint8_t *token_prob;
-    int nonzero = 0;
-    int coeff;
-
+    goto skip_eob;
     do {
-        token_prob = probs[vp8_coeff_band[i]][zero_nhood];
-
+        int coeff;
         if (!vp56_rac_get_prob_branchy(c, token_prob[0]))   // DCT_EOB
-            return nonzero;
+            return i;
 
 skip_eob:
         if (!vp56_rac_get_prob_branchy(c, token_prob[1])) { // DCT_0
-            zero_nhood = 0;
-            token_prob = probs[vp8_coeff_band[++i]][0];
-            if (i < 16)
-                goto skip_eob;
-            return nonzero; // invalid input; blocks should end with EOB
+            if (++i == 16)
+                return i; // invalid input; blocks should end with EOB
+            token_prob = probs[i][0];
+            goto skip_eob;
         }
 
         if (!vp56_rac_get_prob_branchy(c, token_prob[2])) { // DCT_1
             coeff = 1;
-            zero_nhood = 1;
+            token_prob = probs[i+1][1];
         } else {
-            zero_nhood = 2;
-
             if (!vp56_rac_get_prob_branchy(c, token_prob[3])) { // DCT 2,3,4
-                coeff = vp56_rac_get_prob(c, token_prob[4]);
+                coeff = vp56_rac_get_prob_branchy(c, token_prob[4]);
                 if (coeff)
                     coeff += vp56_rac_get_prob(c, token_prob[5]);
                 coeff += 2;
@@ -858,36 +866,49 @@ skip_eob:
                     coeff += vp8_rac_get_coeff(c, vp8_dct_cat_prob[cat]);
                 }
             }
+            token_prob = probs[i+1][2];
         }
-
-        // todo: full [16] qmat? load into register?
         block[zigzag_scan[i]] = (vp8_rac_get(c) ? -coeff : coeff) * qmul[!!i];
-        nonzero = ++i;
-    } while (i < 16);
+    } while (++i < 16);
+
+    return i;
+}
 
-    return nonzero;
+static av_always_inline
+int decode_block_coeffs(VP56RangeCoder *c, DCTELEM block[16],
+                        uint8_t probs[8][3][NUM_DCT_TOKENS-1],
+                        int i, int zero_nhood, int16_t qmul[2])
+{
+    uint8_t *token_prob = probs[i][zero_nhood];
+    if (!vp56_rac_get_prob_branchy(c, token_prob[0]))   // DCT_EOB
+        return 0;
+    return decode_block_coeffs_internal(c, block, probs, i, token_prob, qmul);
 }
 
 static av_always_inline
 void decode_mb_coeffs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
                       uint8_t t_nnz[9], uint8_t l_nnz[9])
 {
-    LOCAL_ALIGNED_16(DCTELEM, dc,[16]);
     int i, x, y, luma_start = 0, luma_ctx = 3;
     int nnz_pred, nnz, nnz_total = 0;
     int segment = s->segment;
+    int block_dc = 0;
 
     if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
-        AV_ZERO128(dc);
-        AV_ZERO128(dc+8);
         nnz_pred = t_nnz[8] + l_nnz[8];
 
         // decode DC values and do hadamard
-        nnz = decode_block_coeffs(c, dc, s->prob->token[1], 0, nnz_pred,
+        nnz = decode_block_coeffs(c, s->block_dc, s->prob->token[1], 0, nnz_pred,
                                   s->qmat[segment].luma_dc_qmul);
         l_nnz[8] = t_nnz[8] = !!nnz;
-        nnz_total += nnz;
-        s->vp8dsp.vp8_luma_dc_wht(s->block, dc);
+        if (nnz) {
+            nnz_total += nnz;
+            block_dc = 1;
+            if (nnz == 1)
+                s->vp8dsp.vp8_luma_dc_wht_dc(s->block, s->block_dc);
+            else
+                s->vp8dsp.vp8_luma_dc_wht(s->block, s->block_dc);
+        }
         luma_start = 1;
         luma_ctx = 0;
     }
@@ -898,8 +919,8 @@ void decode_mb_coeffs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
             nnz_pred = l_nnz[y] + t_nnz[x];
             nnz = decode_block_coeffs(c, s->block[y][x], s->prob->token[luma_ctx], luma_start,
                                       nnz_pred, s->qmat[segment].luma_qmul);
-            // nnz+luma_start may be one more than the actual last index, but we don't care
-            s->non_zero_count_cache[y][x] = nnz + luma_start;
+            // nnz+block_dc may be one more than the actual last index, but we don't care
+            s->non_zero_count_cache[y][x] = nnz + block_dc;
             t_nnz[x] = l_nnz[y] = !!nnz;
             nnz_total += nnz;
         }
@@ -982,7 +1003,7 @@ int check_intra_pred_mode(int mode, int mb_x, int mb_y)
 
 static av_always_inline
 void intra_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
-                   uint8_t *intra4x4, int mb_x, int mb_y)
+                   int mb_x, int mb_y)
 {
     int x, y, mode, nnz, tr;
 
@@ -998,7 +1019,7 @@ void intra_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
         s->hpc.pred16x16[mode](dst[0], s->linesize);
     } else {
         uint8_t *ptr = dst[0];
-        int stride = s->keyframe ? s->b4_stride : 4;
+        uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
 
         // all blocks on the right edge of the macroblock use bottom edge
         // the top macroblock for their topright edge
@@ -1033,7 +1054,7 @@ void intra_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
             }
 
             ptr   += 4*s->linesize;
-            intra4x4 += stride;
+            intra4x4 += 4;
         }
     }
 
@@ -1520,11 +1541,12 @@ static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
     // top edge of 127 for intra prediction
     memset(s->top_border, 127, (s->mb_width+1)*sizeof(*s->top_border));
     memset(s->ref_count, 0, sizeof(s->ref_count));
+    if (s->keyframe)
+        memset(s->intra4x4_pred_mode_top, DC_PRED, s->b4_stride*4);
 
     for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
         VP56RangeCoder *c = &s->coeff_partition[mb_y & (s->num_coeff_partitions-1)];
         VP8Macroblock *mb = s->macroblocks + (s->mb_height - mb_y - 1)*2;
-        uint8_t *intra4x4 = s->intra4x4_pred_mode + 4*mb_y*s->b4_stride;
         uint8_t *segment_map = s->segmentation_map + mb_y*s->mb_stride;
         int mb_xy = mb_y * s->mb_stride;
         uint8_t *dst[3] = {
@@ -1534,6 +1556,7 @@ static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
         };
 
         memset(s->left_nnz, 0, sizeof(s->left_nnz));
+        AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101);
 
         // left edge of 129 for intra prediction
         if (!(avctx->flags & CODEC_FLAG_EMU_EDGE))
@@ -1544,14 +1567,13 @@ static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
             memset(s->top_border, 129, sizeof(*s->top_border));
 
         for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
-            uint8_t *intra4x4_mb = s->keyframe ? intra4x4 + 4*mb_x : s->intra4x4_pred_mode_mb;
             uint8_t *segment_mb = segment_map+mb_x;
 
             /* Prefetch the current frame, 4 MBs ahead */
             s->dsp.prefetch(dst[0] + (mb_x&3)*4*s->linesize + 64, s->linesize, 4);
             s->dsp.prefetch(dst[1] + (mb_x&7)*s->uvlinesize + 64, dst[2] - dst[1], 2);
 
-            decode_mb_mode(s, mb, mb_x, mb_y, intra4x4_mb, segment_mb);
+            decode_mb_mode(s, mb, mb_x, mb_y, segment_mb);
 
             prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);
 
@@ -1559,7 +1581,7 @@ static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
                 decode_mb_coeffs(s, c, mb, s->top_nnz[mb_x], s->left_nnz);
 
             if (mb->mode <= MODE_I4x4)
-                intra_predict(s, dst, mb, intra4x4_mb, mb_x, mb_y);
+                intra_predict(s, dst, mb, mb_x, mb_y);
             else
                 inter_predict(s, dst, mb, mb_x, mb_y);
 
diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8data.h b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8data.h
index 1bdac16be..a64b7d5b4 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8data.h
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8data.h
@@ -55,13 +55,6 @@ enum inter_mvmode {
     VP8_MVMODE_SPLIT
 };
 
-enum inter_submvmode {
-    VP8_SUBMVMODE_LEFT4X4,
-    VP8_SUBMVMODE_TOP4X4,
-    VP8_SUBMVMODE_ZERO4X4,
-    VP8_SUBMVMODE_NEW4X4
-};
-
 enum inter_splitmvmode {
     VP8_SPLITMVMODE_16x8 = 0,    ///< 2 16x8 blocks (vertical)
     VP8_SPLITMVMODE_8x16,        ///< 2 8x16 blocks (horizontal)
@@ -111,13 +104,6 @@ static const int vp8_mode_contexts[6][4] = {
     { 234, 188, 128,  28 },
 };
 
-static const int8_t vp8_pred16x16_tree_mvinter[4][2] = {
-    { -VP8_MVMODE_ZERO,      1 },           // '0'
-     { -VP8_MVMODE_NEAREST,  2 },           // '10'
-      { -VP8_MVMODE_NEAR,    3 },           // '110'
-       { -VP8_MVMODE_NEW, -VP8_MVMODE_SPLIT } // '1110', '1111'
-};
-
 static const uint8_t vp8_mbsplits[5][16] = {
     {  0,  0,  0,  0,  0,  0,  0,  0,
        1,  1,  1,  1,  1,  1,  1,  1  },
@@ -137,12 +123,6 @@ static const uint8_t vp8_mbfirstidx[4][16] = {
        8,  9, 10, 11, 12, 13, 14, 15 }
 };
 
-static const int8_t vp8_mbsplit_tree[3][2] = {
-    { -VP8_SPLITMVMODE_4x4,  1 },           // '0' - 16 individual MVs
-     { -VP8_SPLITMVMODE_8x8,  2 },          // '10' - quarter-based MVs
-      { -VP8_SPLITMVMODE_16x8,              // '110' - top/bottom MVs
-        -VP8_SPLITMVMODE_8x16 }             // '111' - left/right MVs
-};
 static const uint8_t vp8_mbsplit_count[4] = {   2,   2,   4,  16 };
 static const uint8_t vp8_mbsplit_prob[3]  = { 110, 111, 150 };
 
@@ -154,12 +134,6 @@ static const uint8_t vp8_submv_prob[5][3] = {
     { 208,   1,   1 }
 };
 
-static const int8_t vp8_submv_ref_tree[3][2] = {
-    { -VP8_SUBMVMODE_LEFT4X4, 1 },          // '0'
-     { -VP8_SUBMVMODE_TOP4X4, 2 },          // '10'
-      { -VP8_SUBMVMODE_ZERO4X4, -VP8_SUBMVMODE_NEW4X4 } // '110', '111'
-};
-
 static const uint8_t vp8_pred16x16_prob_intra[4] = { 145, 156, 163, 128 };
 static const uint8_t vp8_pred16x16_prob_inter[4] = { 112,  86, 140,  37 };
 
@@ -327,6 +301,20 @@ static const uint8_t vp8_coeff_band[16] =
     0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7
 };
 
+/* Inverse of vp8_coeff_band: mappings of bands to coefficient indexes.
+ * Each list is -1-terminated. */
+static const int8_t vp8_coeff_band_indexes[8][10] =
+{
+    {0, -1},
+    {1, -1},
+    {2, -1},
+    {3, -1},
+    {5, -1},
+    {6, -1},
+    {4, 7, 8, 9, 10, 11, 12, 13, 14, -1},
+    {15, -1}
+};
+
 static const uint8_t vp8_dct_cat1_prob[] = { 159, 0 };
 static const uint8_t vp8_dct_cat2_prob[] = { 165, 145, 0 };
 static const uint8_t vp8_dct_cat3_prob[] = { 173, 148, 140, 0 };
diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.c b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.c
index 0c61d9252..5c83b6df8 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.c
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.c
@@ -46,14 +46,30 @@ static void vp8_luma_dc_wht_c(DCTELEM block[4][4][16], DCTELEM dc[16])
         t1 = dc[i*4+1] + dc[i*4+2];
         t2 = dc[i*4+1] - dc[i*4+2];
         t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding
-
-        *block[i][0] = (t0 + t1) >> 3;
-        *block[i][1] = (t3 + t2) >> 3;
-        *block[i][2] = (t0 - t1) >> 3;
-        *block[i][3] = (t3 - t2) >> 3;
+        dc[i*4+0] = 0;
+        dc[i*4+1] = 0;
+        dc[i*4+2] = 0;
+        dc[i*4+3] = 0;
+
+        block[i][0][0] = (t0 + t1) >> 3;
+        block[i][1][0] = (t3 + t2) >> 3;
+        block[i][2][0] = (t0 - t1) >> 3;
+        block[i][3][0] = (t3 - t2) >> 3;
     }
 }
 
+static void vp8_luma_dc_wht_dc_c(DCTELEM block[4][4][16], DCTELEM dc[16])
+{
+    int i, val = (dc[0] + 3) >> 3;
+    dc[0] = 0;
+
+    for (i = 0; i < 4; i++) {
+        block[i][0][0] = val;
+        block[i][1][0] = val;
+        block[i][2][0] = val;
+        block[i][3][0] = val;
+    }
+}
 
 #define MUL_20091(a) ((((a)*20091) >> 16) + (a))
 #define MUL_35468(a)  (((a)*35468) >> 16)
@@ -476,6 +492,7 @@ VP8_BILINEAR(4)
 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
 {
     dsp->vp8_luma_dc_wht    = vp8_luma_dc_wht_c;
+    dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
     dsp->vp8_idct_add       = vp8_idct_add_c;
     dsp->vp8_idct_dc_add    = vp8_idct_dc_add_c;
     dsp->vp8_idct_dc_add4y  = vp8_idct_dc_add4y_c;
diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.h b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.h
index 47b1a9077..ee5c7ec0f 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.h
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/vp8dsp.h
@@ -31,6 +31,7 @@ typedef void (*vp8_mc_func)(uint8_t *dst/*align 8*/, int dstStride, uint8_t *src
 
 typedef struct VP8DSPContext {
     void (*vp8_luma_dc_wht)(DCTELEM block[4][4][16], DCTELEM dc[16]);
+    void (*vp8_luma_dc_wht_dc)(DCTELEM block[4][4][16], DCTELEM dc[16]);
     void (*vp8_idct_add)(uint8_t *dst, DCTELEM block[16], int stride);
     void (*vp8_idct_dc_add)(uint8_t *dst, DCTELEM block[16], int stride);
     void (*vp8_idct_dc_add4y)(uint8_t *dst, DCTELEM block[4][16], int stride);
diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.c b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.c
index cd4e46219..616943146 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.c
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.c
@@ -2725,9 +2725,6 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
                 c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;

 #endif

 

-            if (CONFIG_CAVS_DECODER)

-                ff_cavsdsp_init_mmx2(c, avctx);

-

             if (CONFIG_VC1_DECODER)

                 ff_vc1dsp_init_mmx(c, avctx);

 

@@ -2788,9 +2785,6 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 

             c->avg_rv40_chroma_pixels_tab[0]= avg_rv40_chroma_mc8_3dnow;

             c->avg_rv40_chroma_pixels_tab[1]= avg_rv40_chroma_mc4_3dnow;

-

-            if (CONFIG_CAVS_DECODER)

-                ff_cavsdsp_init_3dnow(c, avctx);

         }

 

 

diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.h b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.h
index 13067dffc..33dafed1f 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.h
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/dsputil_mmx.h
@@ -162,8 +162,6 @@ void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size
 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 
-void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx);
-void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx);
 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp56_arith.h b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp56_arith.h
index 95f96e365..233fd7abe 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp56_arith.h
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp56_arith.h
@@ -26,7 +26,7 @@
 

 #if HAVE_FAST_CMOV

 #define vp56_rac_get_prob vp56_rac_get_prob

-static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)

+static av_always_inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)

 {

     unsigned int code_word = vp56_rac_renorm(c);

     unsigned int high = c->high;

diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp-init.c b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp-init.c
index dd7dc696e..40fd0e4e3 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp-init.c
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp-init.c
@@ -224,6 +224,7 @@ extern void ff_vp8_idct_dc_add4y_mmx(uint8_t *dst, DCTELEM block[4][16], int str
 extern void ff_vp8_idct_dc_add4y_sse2(uint8_t *dst, DCTELEM block[4][16], int stride);
 extern void ff_vp8_idct_dc_add4uv_mmx(uint8_t *dst, DCTELEM block[2][16], int stride);
 extern void ff_vp8_luma_dc_wht_mmx(DCTELEM block[4][4][16], DCTELEM dc[16]);
+extern void ff_vp8_luma_dc_wht_sse(DCTELEM block[4][4][16], DCTELEM dc[16]);
 extern void ff_vp8_idct_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
 extern void ff_vp8_idct_add_sse(uint8_t *dst, DCTELEM block[16], int stride);
 
@@ -337,6 +338,7 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
 
     if (mm_flags & FF_MM_SSE) {
         c->vp8_idct_add                         = ff_vp8_idct_add_sse;
+        c->vp8_luma_dc_wht                      = ff_vp8_luma_dc_wht_sse;
         c->put_vp8_epel_pixels_tab[0][0][0]     =
         c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_sse;
     }
diff --git a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp.asm b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp.asm
index 4f430d80c..6999e87b6 100644
--- a/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp.asm
+++ b/src/filters/transform/MPCVideoDec/ffmpeg/libavcodec/x86/vp8dsp.asm
@@ -1186,12 +1186,23 @@ VP8_IDCT_ADD sse
     SWAP %1, %4, %3
 %endmacro
 
-INIT_MMX
-cglobal vp8_luma_dc_wht_mmx, 2,3
+%macro VP8_DC_WHT 1
+cglobal vp8_luma_dc_wht_%1, 2,3
     movq          m0, [r1]
     movq          m1, [r1+8]
     movq          m2, [r1+16]
     movq          m3, [r1+24]
+%ifidn %1, sse
+    xorps      xmm0, xmm0
+    movaps  [r1+ 0], xmm0
+    movaps  [r1+16], xmm0
+%else
+    pxor         m4, m4
+    movq    [r1+ 0], m4
+    movq    [r1+ 8], m4
+    movq    [r1+16], m4
+    movq    [r1+24], m4
+%endif
     HADAMARD4_1D  0, 1, 2, 3
     TRANSPOSE4x4W 0, 1, 2, 3, 4
     paddw         m0, [pw_3]
@@ -1203,6 +1214,11 @@ cglobal vp8_luma_dc_wht_mmx, 2,3
     SCATTER_WHT   0, 1, 0
     SCATTER_WHT   2, 3, 2
     RET
+%endmacro
+
+INIT_MMX
+VP8_DC_WHT mmx
+VP8_DC_WHT sse
 
 ;-----------------------------------------------------------------------------
 ; void vp8_h/v_loop_filter_simple_<opt>(uint8_t *dst, int stride, int flim);