Merge remote-tracking branch 'qatar/master'

* qatar/master: (22 commits)
  H.264: fix filter_mb_fast with 4:4:4 + 8x8dct
  alsa: limit buffer_size to 32768 frames.
  alsa: fallback to buffer_size/4 for period_size.
  doc: replace @pxref by @ref where appropriate
  mpeg1video: don't abort if thread_count is too high.
  segafilm: add support for videos with cri adx adpcm
  gxf: Fix 25 fps DV material in GXF being misdetected as 50 fps
  libxvid: Add const qualifier to silence compiler warning.
  H.264: improve qp_thresh check
  H.264: use fill_rectangle in CABAC decoding
  H.264: Remove redundant hl_motion_16/8 code
  H.264: merge fill_rectangle into P-SKIP MV prediction, to match B-SKIP
  H.264: faster P-SKIP decoding
  H.264: av_always_inline some more functions
  H.264: Add x86 assembly for 10-bit H.264 predict functions
  swscale: rename uv_off/uv_off2 to uv_off_px/byte.
  swscale: implement error dithering in planarCopyWrapper.
  swscale: error dithering for 16/9/10-bit to 8-bit.
  swscale: fix overflow in 16-bit vertical scaling.
  swscale: fix crash in 8-bpc bilinear output without alpha.
  ...

Conflicts:
	doc/developer.texi
	libavdevice/alsa-audio.h
	libavformat/gxf.c
	libswscale/swscale.c
	libswscale/swscale_internal.h
	libswscale/swscale_unscaled.c
	libswscale/x86/swscale_template.c
	tests/ref/lavfi/pixdesc
	tests/ref/lavfi/pixfmts_copy
	tests/ref/lavfi/pixfmts_crop
	tests/ref/lavfi/pixfmts_hflip
	tests/ref/lavfi/pixfmts_null
	tests/ref/lavfi/pixfmts_scale
	tests/ref/lavfi/pixfmts_vflip

Merged-by: Michael Niedermayer <michaelni@gmx.at>

11 files changed, 1138 insertions, 81 deletions

diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index 9265e0ab8a..1be7c2a193 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -778,24 +778,6 @@ static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t
     prefetch_motion(h, 1, pixel_shift, chroma444);
 }
 
-#define hl_motion_fn(sh, bits) \
-static av_always_inline void hl_motion_ ## bits(H264Context *h, \
-                                       uint8_t *dest_y, \
-                                       uint8_t *dest_cb, uint8_t *dest_cr, \
-                                       qpel_mc_func (*qpix_put)[16], \
-                                       h264_chroma_mc_func (*chroma_put), \
-                                       qpel_mc_func (*qpix_avg)[16], \
-                                       h264_chroma_mc_func (*chroma_avg), \
-                                       h264_weight_func *weight_op, \
-                                       h264_biweight_func *weight_avg, \
-                                       int chroma444) \
-{ \
-    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
-              qpix_avg, chroma_avg, weight_op, weight_avg, sh, chroma444); \
-}
-hl_motion_fn(0, 8);
-hl_motion_fn(1, 16);
-
 static void free_tables(H264Context *h, int free_rbsp){
     int i;
     H264Context *hx;
@@ -1443,7 +1425,7 @@ static void decode_postinit(H264Context *h, int setup_finished){
         ff_thread_finish_setup(s->avctx);
 }
 
-static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int chroma444, int simple){
+static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int chroma444, int simple){
     MpegEncContext * const s = &h->s;
     uint8_t *top_border;
     int top_idx = 1;
@@ -1518,7 +1500,7 @@ static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src
     }
 }
 
-static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
+static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
                                   uint8_t *src_cb, uint8_t *src_cr,
                                   int linesize, int uvlinesize,
                                   int xchg, int chroma444,
@@ -1876,18 +1858,11 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
             if(h->deblocking_filter)
                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
         }else if(is_h264){
-            if (pixel_shift) {
-                hl_motion_16(h, dest_y, dest_cb, dest_cr,
-                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
-                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
-                             h->h264dsp.weight_h264_pixels_tab,
-                             h->h264dsp.biweight_h264_pixels_tab, 0);
-            } else
-                hl_motion_8(h, dest_y, dest_cb, dest_cr,
-                            s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
-                            s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
-                            h->h264dsp.weight_h264_pixels_tab,
-                            h->h264dsp.biweight_h264_pixels_tab, 0);
+            hl_motion(h, dest_y, dest_cb, dest_cr,
+                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                      h->h264dsp.weight_h264_pixels_tab,
+                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 0);
         }
 
         hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
@@ -2020,18 +1995,11 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
             if(h->deblocking_filter)
                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
         }else{
-            if (pixel_shift) {
-                hl_motion_16(h, dest[0], dest[1], dest[2],
-                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
-                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
-                             h->h264dsp.weight_h264_pixels_tab,
-                             h->h264dsp.biweight_h264_pixels_tab, 1);
-            } else
-                hl_motion_8(h, dest[0], dest[1], dest[2],
-                            s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
-                            s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
-                            h->h264dsp.weight_h264_pixels_tab,
-                            h->h264dsp.biweight_h264_pixels_tab, 1);
+            hl_motion(h, dest[0], dest[1], dest[2],
+                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                      h->h264dsp.weight_h264_pixels_tab,
+                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 1);
         }
 
         for (p = 0; p < plane_count; p++)
@@ -2966,7 +2934,9 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             }
         }
     }
-    h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
+    h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
+                 - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
+                 + 6 * (h->sps.bit_depth_luma - 8);
 
 #if 0 //FMO
     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
diff --git a/libavcodec/h264.h b/libavcodec/h264.h
index 461c490276..1f20d4080d 100644
--- a/libavcodec/h264.h
+++ b/libavcodec/h264.h
@@ -766,11 +766,11 @@ static av_always_inline uint16_t pack8to16(int a, int b){
 /**
  * gets the chroma qp.
  */
-static inline int get_chroma_qp(H264Context *h, int t, int qscale){
+static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale){
     return h->pps.chroma_qp_table[t][qscale];
 }
 
-static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
+static av_always_inline void pred_pskip_motion(H264Context * const h);
 
 static void fill_decode_neighbors(H264Context *h, int mb_type){
     MpegEncContext * const s = &h->s;
@@ -1327,14 +1327,10 @@ static void av_unused decode_mb_skip(H264Context *h){
     }
     else
     {
-        int mx, my;
         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
 
         fill_decode_neighbors(h, mb_type);
-        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
-        pred_pskip_motion(h, &mx, &my);
-        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
-        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
+        pred_pskip_motion(h);
     }
 
     write_back_motion(h, mb_type);
diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c
index 325e15bc0a..ed019fe536 100644
--- a/libavcodec/h264_cabac.c
+++ b/libavcodec/h264_cabac.c
@@ -1819,8 +1819,7 @@ static av_always_inline void decode_cabac_luma_residual( H264Context *h, const u
                     }
                 }
             } else {
-                uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8+16*p] ];
-                nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
+                fill_rectangle(&h->non_zero_count_cache[scan8[4*i8x8+16*p]], 2, 2, 8, 0, 1);
             }
         }
     }
diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c
index 74c7dae523..49a4405657 100644
--- a/libavcodec/h264_loopfilter.c
+++ b/libavcodec/h264_loopfilter.c
@@ -216,7 +216,7 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
     MpegEncContext * const s = &h->s;
     int mb_xy;
     int mb_type, left_type, top_type;
-    int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
+    int qp, qp0, qp1, qpc, qpc0, qpc1;
     int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
     int chroma444 = CHROMA444;
 
@@ -241,10 +241,6 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
     qp1 = (qp + qp1 + 1) >> 1;
     qpc0 = (qpc + qpc0 + 1) >> 1;
     qpc1 = (qpc + qpc1 + 1) >> 1;
-    qp_thresh = 15+52 - h->slice_alpha_c0_offset;
-    if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
-       qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
-        return;
 
     if( IS_INTRA(mb_type) ) {
         static const int16_t bS4[4] = {4,4,4,4};
@@ -321,7 +317,7 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
     } else {
         LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
         int edges;
-        if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
+        if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) {
             edges = 4;
             AV_WN64A(bS[0][0], 0x0002000200020002ULL);
             AV_WN64A(bS[0][2], 0x0002000200020002ULL);
diff --git a/libavcodec/h264_mvpred.h b/libavcodec/h264_mvpred.h
index a0886d5d47..dc146a4803 100644
--- a/libavcodec/h264_mvpred.h
+++ b/libavcodec/h264_mvpred.h
@@ -35,7 +35,7 @@
 //#undef NDEBUG
 #include <assert.h>
 
-static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
+static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
     MpegEncContext *s = &h->s;
 
@@ -92,7 +92,7 @@ static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, in
  * @param mx the x component of the predicted motion vector
  * @param my the y component of the predicted motion vector
  */
-static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
+static av_always_inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
     const int index8= scan8[n];
     const int top_ref=      h->ref_cache[list][ index8 - 8 ];
     const int left_ref=     h->ref_cache[list][ index8 - 1 ];
@@ -147,7 +147,7 @@ static inline void pred_motion(H264Context * const h, int n, int part_width, int
  * @param mx the x component of the predicted motion vector
  * @param my the y component of the predicted motion vector
  */
-static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
+static av_always_inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
     if(n==0){
         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
@@ -182,7 +182,7 @@ static inline void pred_16x8_motion(H264Context * const h, int n, int list, int
  * @param mx the x component of the predicted motion vector
  * @param my the y component of the predicted motion vector
  */
-static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
+static av_always_inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
     if(n==0){
         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
@@ -213,22 +213,117 @@ static inline void pred_8x16_motion(H264Context * const h, int n, int list, int
     pred_motion(h, n, 2, list, ref, mx, my);
 }
 
-static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
-    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
-    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
+#define FIX_MV_MBAFF(type, refn, mvn, idx)\
+    if(FRAME_MBAFF){\
+        if(MB_FIELD){\
+            if(!IS_INTERLACED(type)){\
+                refn <<= 1;\
+                AV_COPY32(mvbuf[idx], mvn);\
+                mvbuf[idx][1] /= 2;\
+                mvn = mvbuf[idx];\
+            }\
+        }else{\
+            if(IS_INTERLACED(type)){\
+                refn >>= 1;\
+                AV_COPY32(mvbuf[idx], mvn);\
+                mvbuf[idx][1] <<= 1;\
+                mvn = mvbuf[idx];\
+            }\
+        }\
+    }
 
-    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
+static av_always_inline void pred_pskip_motion(H264Context * const h){
+    DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = {0};
+    DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2];
+    MpegEncContext * const s = &h->s;
+    int8_t *ref = s->current_picture.ref_index[0];
+    int16_t (*mv)[2] = s->current_picture.motion_val[0];
+    int top_ref, left_ref, diagonal_ref, match_count, mx, my;
+    const int16_t *A, *B, *C;
+    int b_stride = h->b_stride;
+
+    fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
+
+    /* To avoid doing an entire fill_decode_caches, we inline the relevant parts here.
+     * FIXME: this is a partial duplicate of the logic in fill_decode_caches, but it's
+     * faster this way.  Is there a way to avoid this duplication?
+     */
+    if(USES_LIST(h->left_type[LTOP], 0)){
+        left_ref = ref[4*h->left_mb_xy[LTOP] + 1 + (h->left_block[0]&~1)];
+        A = mv[h->mb2b_xy[h->left_mb_xy[LTOP]] + 3 + b_stride*h->left_block[0]];
+        FIX_MV_MBAFF(h->left_type[LTOP], left_ref, A, 0);
+        if(!(left_ref | AV_RN32A(A))){
+            goto zeromv;
+        }
+    }else if(h->left_type[LTOP]){
+        left_ref = LIST_NOT_USED;
+        A = zeromv;
+    }else{
+        goto zeromv;
+    }
 
-    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
-       || !( top_ref | AV_RN32A(h->mv_cache[0][ scan8[0] - 8 ]))
-       || !(left_ref | AV_RN32A(h->mv_cache[0][ scan8[0] - 1 ]))){
+    if(USES_LIST(h->top_type, 0)){
+        top_ref = ref[4*h->top_mb_xy + 2];
+        B = mv[h->mb2b_xy[h->top_mb_xy] + 3*b_stride];
+        FIX_MV_MBAFF(h->top_type, top_ref, B, 1);
+        if(!(top_ref | AV_RN32A(B))){
+            goto zeromv;
+        }
+    }else if(h->top_type){
+        top_ref = LIST_NOT_USED;
+        B = zeromv;
+    }else{
+        goto zeromv;
+    }
+
+    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
 
-        *mx = *my = 0;
-        return;
+    if(USES_LIST(h->topright_type, 0)){
+        diagonal_ref = ref[4*h->topright_mb_xy + 2];
+        C = mv[h->mb2b_xy[h->topright_mb_xy] + 3*b_stride];
+        FIX_MV_MBAFF(h->topright_type, diagonal_ref, C, 2);
+    }else if(h->topright_type){
+        diagonal_ref = LIST_NOT_USED;
+        C = zeromv;
+    }else{
+        if(USES_LIST(h->topleft_type, 0)){
+            diagonal_ref = ref[4*h->topleft_mb_xy + 1 + (h->topleft_partition & 2)];
+            C = mv[h->mb2b_xy[h->topleft_mb_xy] + 3 + b_stride + (h->topleft_partition & 2*b_stride)];
+            FIX_MV_MBAFF(h->topleft_type, diagonal_ref, C, 2);
+        }else if(h->topleft_type){
+            diagonal_ref = LIST_NOT_USED;
+            C = zeromv;
+        }else{
+            diagonal_ref = PART_NOT_AVAILABLE;
+            C = zeromv;
+        }
     }
 
-    pred_motion(h, 0, 4, 0, 0, mx, my);
+    match_count= !diagonal_ref + !top_ref + !left_ref;
+    tprintf(h->s.avctx, "pred_pskip_motion match_count=%d\n", match_count);
+    if(match_count > 1){
+        mx = mid_pred(A[0], B[0], C[0]);
+        my = mid_pred(A[1], B[1], C[1]);
+    }else if(match_count==1){
+        if(!left_ref){
+            mx = A[0];
+            my = A[1];
+        }else if(!top_ref){
+            mx = B[0];
+            my = B[1];
+        }else{
+            mx = C[0];
+            my = C[1];
+        }
+    }else{
+        mx = mid_pred(A[0], B[0], C[0]);
+        my = mid_pred(A[1], B[1], C[1]);
+    }
 
+    fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
+    return;
+zeromv:
+    fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
     return;
 }
 
diff --git a/libavcodec/libxvidff.c b/libavcodec/libxvidff.c
index c09d9a8ff8..a739dc07d3 100644
--- a/libavcodec/libxvidff.c
+++ b/libavcodec/libxvidff.c
@@ -750,7 +750,7 @@ static int xvid_ff_2pass_before(struct xvid_context *ref,
 static int xvid_ff_2pass_after(struct xvid_context *ref,
                                 xvid_plg_data_t *param) {
     char *log = ref->twopassbuffer;
-    char *frame_types = " ipbs";
+    const char *frame_types = " ipbs";
     char frame_type;
 
     /* Quick bounds check */
diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index d819cc083f..170620bd95 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -575,7 +575,11 @@ void MPV_decode_defaults(MpegEncContext *s){
  */
 av_cold int MPV_common_init(MpegEncContext *s)
 {
-    int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y, threads;
+    int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y,
+        threads = (s->encoding ||
+                   (HAVE_THREADS &&
+                    s->avctx->active_thread_type & FF_THREAD_SLICE)) ?
+                  s->avctx->thread_count : 1;
 
     if(s->codec_id == CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
         s->mb_height = (s->height + 31) / 32 * 2;
@@ -589,8 +593,10 @@ av_cold int MPV_common_init(MpegEncContext *s)
 
     if((s->encoding || (s->avctx->active_thread_type & FF_THREAD_SLICE)) &&
        (s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_height))){
-        av_log(s->avctx, AV_LOG_ERROR, "too many threads\n");
-        return -1;
+        int max_threads = FFMIN(MAX_THREADS, s->mb_height);
+        av_log(s->avctx, AV_LOG_WARNING, "too many threads (%d), reducing to %d\n",
+               s->avctx->thread_count, max_threads);
+        threads = max_threads;
     }
 
     if((s->width || s->height) && av_image_check_size(s->width, s->height, 0, s->avctx))
@@ -747,8 +753,6 @@ av_cold int MPV_common_init(MpegEncContext *s)
     s->thread_context[0]= s;
 
     if (s->encoding || (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_SLICE)) {
-        threads = s->avctx->thread_count;
-
         for(i=1; i<threads; i++){
             s->thread_context[i]= av_malloc(sizeof(MpegEncContext));
             memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index d98c6a0b85..640048644c 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -42,7 +42,7 @@ DECLARE_ALIGNED(8,  const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
 DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
 {0x8000000080000000ULL, 0x8000000080000000ULL};
 
-DECLARE_ALIGNED(8,  const uint64_t, ff_pw_1  ) = 0x0001000100010001ULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_1  ) = {0x0001000100010001ULL, 0x0001000100010001ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_2  ) = {0x0002000200020002ULL, 0x0002000200020002ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_3  ) = {0x0003000300030003ULL, 0x0003000300030003ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_4  ) = {0x0004000400040004ULL, 0x0004000400040004ULL};
diff --git a/libavcodec/x86/h264_intrapred_10bit.asm b/libavcodec/x86/h264_intrapred_10bit.asm
index 5cb593ac38..d57fc79754 100644
--- a/libavcodec/x86/h264_intrapred_10bit.asm
+++ b/libavcodec/x86/h264_intrapred_10bit.asm
@@ -29,9 +29,13 @@ SECTION_RODATA
 
 SECTION .text
 
+cextern pw_8
 cextern pw_4
+cextern pw_2
 cextern pw_1
 
+; dest, left, right, src
+; output: %1 = (t[n-1] + t[n]*2 + t[n+1] + 2) >> 2
 %macro PRED4x4_LOWPASS 4
     paddw       %2, %3
     psrlw       %2, 1
@@ -335,3 +339,930 @@ cglobal pred8x8_horizontal_10_sse2, 2,3
     dec          r2
     jg .loop
     REP_RET
+
+;-----------------------------------------------------------------------------
+; void predict_8x8_dc(pixel *src, int stride)
+;-----------------------------------------------------------------------------
+%macro MOV8 2-3
+; sort of a hack, but it works
+%if mmsize==8
+    movq    [%1+0], %2
+    movq    [%1+8], %3
+%else
+    movdqa    [%1], %2
+%endif
+%endmacro
+
+%macro PRED8x8_DC 2
+cglobal pred8x8_dc_10_%1, 2,4
+%ifdef ARCH_X86_64
+%define t0 r10
+%else
+%define t0 r0m
+%endif
+    sub         r0, r1
+    pxor        m4, m4
+    movq        m0, [r0+0]
+    movq        m1, [r0+8]
+    HADDW       m0, m2
+    mov         t0, r0
+    HADDW       m1, m2
+
+    movzx      r2d, word [r0+r1*1-2]
+    movzx      r3d, word [r0+r1*2-2]
+    lea         r0, [r0+r1*2]
+    add        r2d, r3d
+    movzx      r3d, word [r0+r1*1-2]
+    add        r2d, r3d
+    movzx      r3d, word [r0+r1*2-2]
+    add        r2d, r3d
+    lea         r0, [r0+r1*2]
+    movd        m2, r2d            ; s2
+
+    movzx      r2d, word [r0+r1*1-2]
+    movzx      r3d, word [r0+r1*2-2]
+    lea         r0, [r0+r1*2]
+    add        r2d, r3d
+    movzx      r3d, word [r0+r1*1-2]
+    add        r2d, r3d
+    movzx      r3d, word [r0+r1*2-2]
+    add        r2d, r3d
+    movd        m3, r2d            ; s3
+
+    punpcklwd   m0, m1
+    mov         r0, t0
+    punpcklwd   m2, m3
+    punpckldq   m0, m2            ; s0, s1, s2, s3
+    %2          m3, m0, 11110110b ; s2, s1, s3, s3
+    lea         r2, [r1+r1*2]
+    %2          m0, m0, 01110100b ; s0, s1, s3, s1
+    paddw       m0, m3
+    lea         r3, [r0+r1*4]
+    psrlw       m0, 2
+    pavgw       m0, m4            ; s0+s2, s1, s3, s1+s3
+%ifidn %1, sse2
+    punpcklwd   m0, m0
+    pshufd      m3, m0, 11111010b
+    punpckldq   m0, m0
+    SWAP         0,1
+%else
+    pshufw      m1, m0, 0x00
+    pshufw      m2, m0, 0x55
+    pshufw      m3, m0, 0xaa
+    pshufw      m4, m0, 0xff
+%endif
+    MOV8   r0+r1*1, m1, m2
+    MOV8   r0+r1*2, m1, m2
+    MOV8   r0+r2*1, m1, m2
+    MOV8   r0+r1*4, m1, m2
+    MOV8   r3+r1*1, m3, m4
+    MOV8   r3+r1*2, m3, m4
+    MOV8   r3+r2*1, m3, m4
+    MOV8   r3+r1*4, m3, m4
+    RET
+%endmacro
+
+INIT_MMX
+PRED8x8_DC mmxext, pshufw
+INIT_XMM
+PRED8x8_DC sse2  , pshuflw
+
+;-----------------------------------------------------------------------------
+; void pred8x8_top_dc(pixel *src, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8_TOP_DC 2
+cglobal pred8x8_top_dc_10_%1, 2,4
+    sub         r0, r1
+    movq        m0, [r0+0]
+    movq        m1, [r0+8]
+    HADDW       m0, m2
+    HADDW       m1, m3
+    lea         r2, [r1+r1*2]
+    paddw       m0, [pw_2]
+    paddw       m1, [pw_2]
+    lea         r3, [r0+r1*4]
+    psrlw       m0, 2
+    psrlw       m1, 2
+    %2          m0, m0, 0
+    %2          m1, m1, 0
+%ifidn %1, sse2
+    punpcklqdq  m0, m1
+%endif
+    MOV8   r0+r1*1, m0, m1
+    MOV8   r0+r1*2, m0, m1
+    MOV8   r0+r2*1, m0, m1
+    MOV8   r0+r1*4, m0, m1
+    MOV8   r3+r1*1, m0, m1
+    MOV8   r3+r1*2, m0, m1
+    MOV8   r3+r2*1, m0, m1
+    MOV8   r3+r1*4, m0, m1
+    RET
+%endmacro
+
+INIT_MMX
+PRED8x8_TOP_DC mmxext, pshufw
+INIT_XMM
+PRED8x8_TOP_DC sse2  , pshuflw
+
+
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_top_dc(pixel *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_TOP_DC 1
+cglobal pred8x8l_top_dc_10_%1, 4,4,6
+    sub         r0, r3
+    pxor        m7, m7
+    mova        m0, [r0-16]
+    mova        m3, [r0]
+    mova        m1, [r0+16]
+    mova        m2, m3
+    mova        m4, m3
+    PALIGNR     m2, m0, 14, m0
+    PALIGNR     m1, m4,  2, m4
+    test        r1, r1 ; top_left
+    jz .fix_lt_2
+    test        r2, r2 ; top_right
+    jz .fix_tr_1
+    jmp .body
+.fix_lt_2:
+    mova        m5, m3
+    pxor        m5, m2
+    pslldq      m5, 14
+    psrldq      m5, 14
+    pxor        m2, m5
+    test        r2, r2 ; top_right
+    jnz .body
+.fix_tr_1:
+    mova        m5, m3
+    pxor        m5, m1
+    psrldq      m5, 14
+    pslldq      m5, 14
+    pxor        m1, m5
+.body
+    lea         r1, [r3+r3*2]
+    lea         r2, [r0+r3*4]
+    PRED4x4_LOWPASS m0, m2, m1, m3
+    HADDW       m0, m1
+    paddw       m0, [pw_4]
+    psrlw       m0, 3
+    SPLATW      m0, m0, 0
+    mova [r0+r3*1], m0
+    mova [r0+r3*2], m0
+    mova [r0+r1*1], m0
+    mova [r0+r3*4], m0
+    mova [r2+r3*1], m0
+    mova [r2+r3*2], m0
+    mova [r2+r1*1], m0
+    mova [r2+r3*4], m0
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_TOP_DC sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_TOP_DC ssse3
+
+;-----------------------------------------------------------------------------
+;void pred8x8l_dc(pixel *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+;TODO: see if scalar is faster
+%macro PRED8x8L_DC 1
+cglobal pred8x8l_dc_10_%1, 4,5,8
+    sub         r0, r3
+    lea         r4, [r0+r3*2]
+    mova        m0, [r0+r3*1-16]
+    punpckhwd   m0, [r0+r3*0-16]
+    mova        m1, [r4+r3*1-16]
+    punpckhwd   m1, [r0+r3*2-16]
+    mov         r4, r0
+    punpckhdq   m1, m0
+    lea         r0, [r0+r3*4]
+    mova        m2, [r0+r3*1-16]
+    punpckhwd   m2, [r0+r3*0-16]
+    lea         r0, [r0+r3*2]
+    mova        m3, [r0+r3*1-16]
+    punpckhwd   m3, [r0+r3*0-16]
+    punpckhdq   m3, m2
+    punpckhqdq  m3, m1
+    lea         r0, [r0+r3*2]
+    mova        m0, [r0+r3*0-16]
+    mova        m1, [r4]
+    mov         r0, r4
+    mova        m4, m3
+    mova        m2, m3
+    PALIGNR     m4, m0, 14, m0
+    PALIGNR     m1, m2,  2, m2
+    test        r1, r1
+    jnz .do_left
+.fix_lt_1:
+    mova        m5, m3
+    pxor        m5, m4
+    psrldq      m5, 14
+    pslldq      m5, 12
+    pxor        m1, m5
+    jmp .do_left
+.fix_lt_2:
+    mova        m5, m3
+    pxor        m5, m2
+    pslldq      m5, 14
+    psrldq      m5, 14
+    pxor        m2, m5
+    test        r2, r2
+    jnz .body
+.fix_tr_1:
+    mova        m5, m3
+    pxor        m5, m1
+    psrldq      m5, 14
+    pslldq      m5, 14
+    pxor        m1, m5
+    jmp .body
+.do_left:
+    mova        m0, m4
+    PRED4x4_LOWPASS m2, m1, m4, m3
+    mova        m4, m0
+    mova        m7, m2
+    PRED4x4_LOWPASS m1, m3, m0, m4
+    pslldq      m1, 14
+    PALIGNR     m7, m1, 14, m3
+    mova        m0, [r0-16]
+    mova        m3, [r0]
+    mova        m1, [r0+16]
+    mova        m2, m3
+    mova        m4, m3
+    PALIGNR     m2, m0, 14, m0
+    PALIGNR     m1, m4,  2, m4
+    test        r1, r1
+    jz .fix_lt_2
+    test        r2, r2
+    jz .fix_tr_1
+.body
+    lea         r1, [r3+r3*2]
+    PRED4x4_LOWPASS m6, m2, m1, m3
+    HADDW       m7, m0
+    HADDW       m6, m0
+    lea         r2, [r0+r3*4]
+    paddw       m7, [pw_8]
+    paddw       m7, m6
+    psrlw       m7, 4
+    SPLATW      m7, m7
+    mova [r0+r3*1], m7
+    mova [r0+r3*2], m7
+    mova [r0+r1*1], m7
+    mova [r0+r3*4], m7
+    mova [r2+r3*1], m7
+    mova [r2+r3*2], m7
+    mova [r2+r1*1], m7
+    mova [r2+r3*4], m7
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_DC sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_DC ssse3
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_vertical(pixel *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_VERTICAL 1
+cglobal pred8x8l_vertical_10_%1, 4,4,6
+    sub         r0, r3
+    mova        m0, [r0-16]
+    mova        m3, [r0]
+    mova        m1, [r0+16]
+    mova        m2, m3
+    mova        m4, m3
+    PALIGNR     m2, m0, 14, m0
+    PALIGNR     m1, m4,  2, m4
+    test        r1, r1 ; top_left
+    jz .fix_lt_2
+    test        r2, r2 ; top_right
+    jz .fix_tr_1
+    jmp .body
+.fix_lt_2:
+    mova        m5, m3
+    pxor        m5, m2
+    pslldq      m5, 14
+    psrldq      m5, 14
+    pxor        m2, m5
+    test        r2, r2 ; top_right
+    jnz .body
+.fix_tr_1:
+    mova        m5, m3
+    pxor        m5, m1
+    psrldq      m5, 14
+    pslldq      m5, 14
+    pxor        m1, m5
+.body
+    lea         r1, [r3+r3*2]
+    lea         r2, [r0+r3*4]
+    PRED4x4_LOWPASS m0, m2, m1, m3
+    mova [r0+r3*1], m0
+    mova [r0+r3*2], m0
+    mova [r0+r1*1], m0
+    mova [r0+r3*4], m0
+    mova [r2+r3*1], m0
+    mova [r2+r3*2], m0
+    mova [r2+r1*1], m0
+    mova [r2+r3*4], m0
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_VERTICAL sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_VERTICAL ssse3
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_horizontal(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_HORIZONTAL 1
+cglobal pred8x8l_horizontal_10_%1, 4,4,8
+    sub         r0, r3
+    lea         r2, [r0+r3*2]
+    mova        m0, [r0+r3*1-16]
+    test        r1, r1
+    lea         r1, [r0+r3]
+    cmovnz      r1, r0
+    punpckhwd   m0, [r1+r3*0-16]
+    mova        m1, [r2+r3*1-16]
+    punpckhwd   m1, [r0+r3*2-16]
+    mov         r2, r0
+    punpckhdq   m1, m0
+    lea         r0, [r0+r3*4]
+    mova        m2, [r0+r3*1-16]
+    punpckhwd   m2, [r0+r3*0-16]
+    lea         r0, [r0+r3*2]
+    mova        m3, [r0+r3*1-16]
+    punpckhwd   m3, [r0+r3*0-16]
+    punpckhdq   m3, m2
+    punpckhqdq  m3, m1
+    lea         r0, [r0+r3*2]
+    mova        m0, [r0+r3*0-16]
+    mova        m1, [r1+r3*0-16]
+    mov         r0, r2
+    mova        m4, m3
+    mova        m2, m3
+    PALIGNR     m4, m0, 14, m0
+    PALIGNR     m1, m2,  2, m2
+    mova        m0, m4
+    PRED4x4_LOWPASS m2, m1, m4, m3
+    mova        m4, m0
+    mova        m7, m2
+    PRED4x4_LOWPASS m1, m3, m0, m4
+    pslldq      m1, 14
+    PALIGNR     m7, m1, 14, m3
+    lea         r1, [r3+r3*2]
+    punpckhwd   m3, m7, m7
+    punpcklwd   m7, m7
+    pshufd      m0, m3, 0xff
+    pshufd      m1, m3, 0xaa
+    lea         r2, [r0+r3*4]
+    pshufd      m2, m3, 0x55
+    pshufd      m3, m3, 0x00
+    pshufd      m4, m7, 0xff
+    pshufd      m5, m7, 0xaa
+    pshufd      m6, m7, 0x55
+    pshufd      m7, m7, 0x00
+    mova [r0+r3*1], m0
+    mova [r0+r3*2], m1
+    mova [r0+r1*1], m2
+    mova [r0+r3*4], m3
+    mova [r2+r3*1], m4
+    mova [r2+r3*2], m5
+    mova [r2+r1*1], m6
+    mova [r2+r3*4], m7
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_HORIZONTAL sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_HORIZONTAL ssse3
+
+;-----------------------------------------------------------------------------
+;void pred8x8l_down_left(pixel *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_DOWN_LEFT 1
+cglobal pred8x8l_down_left_10_%1, 4,4,8
+    sub         r0, r3
+    mova        m0, [r0-16]
+    mova        m3, [r0]
+    mova        m1, [r0+16]
+    mova        m2, m3
+    mova        m4, m3
+    PALIGNR     m2, m0, 14, m0
+    PALIGNR     m1, m4,  2, m4
+    test        r1, r1
+    jz .fix_lt_2
+    test        r2, r2
+    jz .fix_tr_1
+    jmp .do_top
+.fix_lt_2:
+    mova        m5, m3
+    pxor        m5, m2
+    pslldq      m5, 14
+    psrldq      m5, 14
+    pxor        m2, m5
+    test        r2, r2
+    jnz .do_top
+.fix_tr_1:
+    mova        m5, m3
+    pxor        m5, m1
+    psrldq      m5, 14
+    pslldq      m5, 14
+    pxor        m1, m5
+    jmp .do_top
+.fix_tr_2:
+    punpckhwd   m3, m3
+    pshufd      m1, m3, 0xFF
+    jmp .do_topright
+.do_top:
+    PRED4x4_LOWPASS m4, m2, m1, m3
+    mova        m7, m4
+    test        r2, r2
+    jz .fix_tr_2
+    mova        m0, [r0+16]
+    mova        m5, m0
+    mova        m2, m0
+    mova        m4, m0
+    psrldq      m5, 14
+    PALIGNR     m2, m3, 14, m3
+    PALIGNR     m5, m4,  2, m4
+    PRED4x4_LOWPASS m1, m2, m5, m0
+.do_topright:
+    lea         r1, [r3+r3*2]
+    mova        m6, m1
+    psrldq      m1, 14
+    mova        m4, m1
+    lea         r2, [r0+r3*4]
+    mova        m2, m6
+    PALIGNR     m2, m7,  2, m0
+    mova        m3, m6
+    PALIGNR     m3, m7, 14, m0
+    PALIGNR     m4, m6,  2, m0
+    mova        m5, m7
+    mova        m1, m7
+    mova        m7, m6
+    pslldq      m1, 2
+    PRED4x4_LOWPASS m0, m1, m2, m5
+    PRED4x4_LOWPASS m1, m3, m4, m7
+    mova [r2+r3*4], m1
+    mova        m2, m0
+    pslldq      m1, 2
+    psrldq      m2, 14
+    pslldq      m0, 2
+    por         m1, m2
+    mova [r2+r1*1], m1
+    mova        m2, m0
+    pslldq      m1, 2
+    psrldq      m2, 14
+    pslldq      m0, 2
+    por         m1, m2
+    mova [r2+r3*2], m1
+    mova        m2, m0
+    pslldq      m1, 2
+    psrldq      m2, 14
+    pslldq      m0, 2
+    por         m1, m2
+    mova [r2+r3*1], m1
+    mova        m2, m0
+    pslldq      m1, 2
+    psrldq      m2, 14
+    pslldq      m0, 2
+    por         m1, m2
+    mova [r0+r3*4], m1
+    mova        m2, m0
+    pslldq      m1, 2
+    psrldq      m2, 14
+    pslldq      m0, 2
+    por         m1, m2
+    mova [r0+r1*1], m1
+    mova        m2, m0
+    pslldq      m1, 2
+    psrldq      m2, 14
+    pslldq      m0, 2
+    por         m1, m2
+    mova [r0+r3*2], m1
+    pslldq      m1, 2
+    psrldq      m0, 14
+    por         m1, m0
+    mova [r0+r3*1], m1
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_DOWN_LEFT sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_DOWN_LEFT ssse3
+
+;-----------------------------------------------------------------------------
+;void pred8x8l_down_right_mxext(pixel *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_DOWN_RIGHT 1
+cglobal pred8x8l_down_right_10_%1, 4,5,8
+    sub         r0, r3
+    lea         r4, [r0+r3*2]
+    mova        m0, [r0+r3*1-16]
+    punpckhwd   m0, [r0+r3*0-16]
+    mova        m1, [r4+r3*1-16]
+    punpckhwd   m1, [r0+r3*2-16]
+    mov         r4, r0
+    punpckhdq   m1, m0
+    lea         r0, [r0+r3*4]
+    mova        m2, [r0+r3*1-16]
+    punpckhwd   m2, [r0+r3*0-16]
+    lea         r0, [r0+r3*2]
+    mova        m3, [r0+r3*1-16]
+    punpckhwd   m3, [r0+r3*0-16]
+    punpckhdq   m3, m2
+    punpckhqdq  m3, m1
+    lea         r0, [r0+r3*2]
+    mova        m0, [r0+r3*0-16]
+    mova        m1, [r4]
+    mov         r0, r4
+    mova        m4, m3
+    mova        m2, m3
+    PALIGNR     m4, m0, 14, m0
+    PALIGNR     m1, m2,  2, m2
+    test        r1, r1 ; top_left
+    jz .fix_lt_1
+.do_left:
+    mova        m0, m4
+    PRED4x4_LOWPASS m2, m1, m4, m3
+    mova        m4, m0
+    mova        m7, m2
+    mova        m6, m2
+    PRED4x4_LOWPASS m1, m3, m0, m4
+    pslldq      m1, 14
+    PALIGNR     m7, m1, 14, m3
+    mova        m0, [r0-16]
+    mova        m3, [r0]
+    mova        m1, [r0+16]
+    mova        m2, m3
+    mova        m4, m3
+    PALIGNR     m2, m0, 14, m0
+    PALIGNR     m1, m4,  2, m4
+    test        r1, r1 ; top_left
+    jz .fix_lt_2
+    test        r2, r2 ; top_right
+    jz .fix_tr_1
+.do_top:
+    PRED4x4_LOWPASS m4, m2, m1, m3
+    mova        m5, m4
+    jmp .body
+.fix_lt_1:
+    mova        m5, m3
+    pxor        m5, m4
+    psrldq      m5, 14
+    pslldq      m5, 12
+    pxor        m1, m5
+    jmp .do_left
+.fix_lt_2:
+    mova        m5, m3
+    pxor        m5, m2
+    pslldq      m5, 14
+    psrldq      m5, 14
+    pxor        m2, m5
+    test        r2, r2 ; top_right
+    jnz .do_top
+.fix_tr_1:
+    mova        m5, m3
+    pxor        m5, m1
+    psrldq      m5, 14
+    pslldq      m5, 14
+    pxor        m1, m5
+    jmp .do_top
+.body
+    lea         r1, [r3+r3*2]
+    mova        m1, m7
+    mova        m7, m5
+    mova        m5, m6
+    mova        m2, m7
+    lea         r2, [r0+r3*4]
+    PALIGNR     m2, m6,  2, m0
+    mova        m3, m7
+    PALIGNR     m3, m6, 14, m0
+    mova        m4, m7
+    psrldq      m4, 2
+    PRED4x4_LOWPASS m0, m1, m2, m5
+    PRED4x4_LOWPASS m1, m3, m4, m7
+    mova [r2+r3*4], m0
+    mova        m2, m1
+    psrldq      m0, 2
+    pslldq      m2, 14
+    psrldq      m1, 2
+    por         m0, m2
+    mova [r2+r1*1], m0
+    mova        m2, m1
+    psrldq      m0, 2
+    pslldq      m2, 14
+    psrldq      m1, 2
+    por         m0, m2
+    mova [r2+r3*2], m0
+    mova        m2, m1
+    psrldq      m0, 2
+    pslldq      m2, 14
+    psrldq      m1, 2
+    por         m0, m2
+    mova [r2+r3*1], m0
+    mova        m2, m1
+    psrldq      m0, 2
+    pslldq      m2, 14
+    psrldq      m1, 2
+    por         m0, m2
+    mova [r0+r3*4], m0
+    mova        m2, m1
+    psrldq      m0, 2
+    pslldq      m2, 14
+    psrldq      m1, 2
+    por         m0, m2
+    mova [r0+r1*1], m0
+    mova        m2, m1
+    psrldq      m0, 2
+    pslldq      m2, 14
+    psrldq      m1, 2
+    por         m0, m2
+    mova [r0+r3*2], m0
+    psrldq      m0, 2
+    pslldq      m1, 14
+    por         m0, m1
+    mova [r0+r3*1], m0
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_DOWN_RIGHT sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_DOWN_RIGHT ssse3
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_vertical_right(pixel *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_VERTICAL_RIGHT 1
+cglobal pred8x8l_vertical_right_10_%1, 4,5,8
+    sub         r0, r3
+    lea         r4, [r0+r3*2]
+    mova        m0, [r0+r3*1-16]
+    punpckhwd   m0, [r0+r3*0-16]
+    mova        m1, [r4+r3*1-16]
+    punpckhwd   m1, [r0+r3*2-16]
+    mov         r4, r0
+    punpckhdq   m1, m0
+    lea         r0, [r0+r3*4]
+    mova        m2, [r0+r3*1-16]
+    punpckhwd   m2, [r0+r3*0-16]
+    lea         r0, [r0+r3*2]
+    mova        m3, [r0+r3*1-16]
+    punpckhwd   m3, [r0+r3*0-16]
+    punpckhdq   m3, m2
+    punpckhqdq  m3, m1
+    lea         r0, [r0+r3*2]
+    mova        m0, [r0+r3*0-16]
+    mova        m1, [r4]
+    mov         r0, r4
+    mova        m4, m3
+    mova        m2, m3
+    PALIGNR     m4, m0, 14, m0
+    PALIGNR     m1, m2,  2, m2
+    test        r1, r1
+    jz .fix_lt_1
+    jmp .do_left
+.fix_lt_1:
+    mova        m5, m3
+    pxor        m5, m4
+    psrldq      m5, 14
+    pslldq      m5, 12
+    pxor        m1, m5
+    jmp .do_left
+.fix_lt_2:
+    mova        m5, m3
+    pxor        m5, m2
+    pslldq      m5, 14
+    psrldq      m5, 14
+    pxor        m2, m5
+    test        r2, r2
+    jnz .do_top
+.fix_tr_1:
+    mova        m5, m3
+    pxor        m5, m1
+    psrldq      m5, 14
+    pslldq      m5, 14
+    pxor        m1, m5
+    jmp .do_top
+.do_left:
+    mova        m0, m4
+    PRED4x4_LOWPASS m2, m1, m4, m3
+    mova        m7, m2
+    mova        m0, [r0-16]
+    mova        m3, [r0]
+    mova        m1, [r0+16]
+    mova        m2, m3
+    mova        m4, m3
+    PALIGNR     m2, m0, 14, m0
+    PALIGNR     m1, m4,  2, m4
+    test        r1, r1
+    jz .fix_lt_2
+    test        r2, r2
+    jz .fix_tr_1
+.do_top
+    PRED4x4_LOWPASS m6, m2, m1, m3
+    lea         r1, [r3+r3*2]
+    mova        m2, m6
+    mova        m3, m6
+    PALIGNR     m3, m7, 14, m0
+    PALIGNR     m6, m7, 12, m1
+    mova        m4, m3
+    pavgw       m3, m2
+    lea         r2, [r0+r3*4]
+    PRED4x4_LOWPASS m0, m6, m2, m4
+    mova [r0+r3*1], m3
+    mova [r0+r3*2], m0
+    mova        m5, m0
+    mova        m6, m3
+    mova        m1, m7
+    mova        m2, m1
+    pslldq      m2, 2
+    mova        m3, m1
+    pslldq      m3, 4
+    PRED4x4_LOWPASS m0, m1, m3, m2
+    PALIGNR     m6, m0, 14, m2
+    mova [r0+r1*1], m6
+    pslldq      m0, 2
+    PALIGNR     m5, m0, 14, m1
+    mova [r0+r3*4], m5
+    pslldq      m0, 2
+    PALIGNR     m6, m0, 14, m2
+    mova [r2+r3*1], m6
+    pslldq      m0, 2
+    PALIGNR     m5, m0, 14, m1
+    mova [r2+r3*2], m5
+    pslldq      m0, 2
+    PALIGNR     m6, m0, 14, m2
+    mova [r2+r1*1], m6
+    pslldq      m0, 2
+    PALIGNR     m5, m0, 14, m1
+    mova [r2+r3*4], m5
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_VERTICAL_RIGHT sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_VERTICAL_RIGHT ssse3
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_horizontal_up(pixel *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_HORIZONTAL_UP 1
+cglobal pred8x8l_horizontal_up_10_%1, 4,4,8
+    sub         r0, r3
+    lea         r2, [r0+r3*2]
+    mova        m0, [r0+r3*1-16]
+    test        r1, r1
+    lea         r1, [r0+r3]
+    cmovnz      r1, r0
+    punpckhwd   m0, [r1+r3*0-16]
+    mova        m1, [r2+r3*1-16]
+    punpckhwd   m1, [r0+r3*2-16]
+    mov         r2, r0
+    punpckhdq   m1, m0
+    lea         r0, [r0+r3*4]
+    mova        m2, [r0+r3*1-16]
+    punpckhwd   m2, [r0+r3*0-16]
+    lea         r0, [r0+r3*2]
+    mova        m3, [r0+r3*1-16]
+    punpckhwd   m3, [r0+r3*0-16]
+    punpckhdq   m3, m2
+    punpckhqdq  m3, m1
+    lea         r0, [r0+r3*2]
+    mova        m0, [r0+r3*0-16]
+    mova        m1, [r1+r3*0-16]
+    mov         r0, r2
+    mova        m4, m3
+    mova        m2, m3
+    PALIGNR     m4, m0, 14, m0
+    PALIGNR     m1, m2,  2, m2
+    mova        m0, m4
+    PRED4x4_LOWPASS m2, m1, m4, m3
+    mova        m4, m0
+    mova        m7, m2
+    PRED4x4_LOWPASS m1, m3, m0, m4
+    pslldq      m1, 14
+    PALIGNR     m7, m1, 14, m3
+    lea         r1, [r3+r3*2]
+    pshufd      m0, m7, 00011011b ; l6 l7 l4 l5 l2 l3 l0 l1
+    pslldq      m7, 14             ; l7 .. .. .. .. .. .. ..
+    mova        m2, m0
+    pslld       m0, 16
+    psrld       m2, 16
+    por         m2, m0            ; l7 l6 l5 l4 l3 l2 l1 l0
+    mova        m3, m2
+    mova        m4, m2
+    mova        m5, m2
+    psrldq      m2, 2
+    psrldq      m3, 4
+    lea         r2, [r0+r3*4]
+    por         m2, m7            ; l7 l7 l6 l5 l4 l3 l2 l1
+    punpckhwd   m7, m7
+    por         m3, m7            ; l7 l7 l7 l6 l5 l4 l3 l2
+    pavgw       m4, m2
+    PRED4x4_LOWPASS m1, m3, m5, m2
+    mova        m5, m4
+    punpcklwd   m4, m1            ; p4 p3 p2 p1
+    punpckhwd   m5, m1            ; p8 p7 p6 p5
+    mova        m6, m5
+    mova        m7, m5
+    mova        m0, m5
+    PALIGNR     m5, m4, 4, m1
+    pshufd      m1, m6, 11111001b
+    PALIGNR     m6, m4, 8, m2
+    pshufd      m2, m7, 11111110b
+    PALIGNR     m7, m4, 12, m3
+    pshufd      m3, m0, 11111111b
+    mova [r0+r3*1], m4
+    mova [r0+r3*2], m5
+    mova [r0+r1*1], m6
+    mova [r0+r3*4], m7
+    mova [r2+r3*1], m0
+    mova [r2+r3*2], m1
+    mova [r2+r1*1], m2
+    mova [r2+r3*4], m3
+    RET
+%endmacro
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_HORIZONTAL_UP sse2
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_HORIZONTAL_UP ssse3
+
+
+
+;-----------------------------------------------------------------------------
+; void pred16x16_vertical(pixel *src, int stride)
+;-----------------------------------------------------------------------------
+%macro MOV16 3-5
+    mova [%1+     0], %2
+    mova [%1+mmsize], %3
+%if mmsize==8
+    mova [%1+    16], %4
+    mova [%1+    24], %5
+%endif
+%endmacro
+
+%macro PRED16x16_VERTICAL 1
+cglobal pred16x16_vertical_10_%1, 2,3
+    sub   r0, r1
+    mov   r2, 8
+    mova  m0, [r0+ 0]
+    mova  m1, [r0+mmsize]
+%if mmsize==8
+    mova  m2, [r0+16]
+    mova  m3, [r0+24]
+%endif
+.loop:
+    MOV16 r0+r1*1, m0, m1, m2, m3
+    MOV16 r0+r1*2, m0, m1, m2, m3
+    lea   r0, [r0+r1*2]
+    dec   r2
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_MMX
+PRED16x16_VERTICAL mmxext
+INIT_XMM
+PRED16x16_VERTICAL sse2
+
+;-----------------------------------------------------------------------------
+; void pred16x16_horizontal(pixel *src, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED16x16_HORIZONTAL 1
+cglobal pred16x16_horizontal_10_%1, 2,3
+    mov    r2, 8
+.vloop:
+    movd   m0, [r0+r1*0-4]
+    movd   m1, [r0+r1*1-4]
+    SPLATW m0, m0, 1
+    SPLATW m1, m1, 1
+    MOV16  r0+r1*0, m0, m0, m0, m0
+    MOV16  r0+r1*1, m1, m1, m1, m1
+    lea    r0, [r0+r1*2]
+    dec    r2
+    jge .vloop
+    REP_RET
+%endmacro
+
+INIT_MMX
+PRED16x16_HORIZONTAL mmxext
+INIT_XMM
+PRED16x16_HORIZONTAL sse2
diff --git a/libavcodec/x86/h264_intrapred_init.c b/libavcodec/x86/h264_intrapred_init.c
index d71f71e861..2ccc584e91 100644
--- a/libavcodec/x86/h264_intrapred_init.c
+++ b/libavcodec/x86/h264_intrapred_init.c
@@ -43,9 +43,41 @@ PRED4x4(horizontal_down, 10, avx)
 #define PRED8x8(TYPE, DEPTH, OPT) \
 void ff_pred8x8_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
 
+PRED8x8(dc, 10, mmxext)
+PRED8x8(dc, 10, sse2)
+PRED8x8(top_dc, 10, mmxext)
+PRED8x8(top_dc, 10, sse2)
 PRED8x8(vertical, 10, sse2)
 PRED8x8(horizontal, 10, sse2)
 
+#define PRED8x8L(TYPE, DEPTH, OPT)\
+void ff_pred8x8l_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int has_topleft, int has_topright, int stride);
+
+PRED8x8L(dc, 10, sse2)
+PRED8x8L(dc, 10, ssse3)
+PRED8x8L(top_dc, 10, sse2)
+PRED8x8L(top_dc, 10, ssse3)
+PRED8x8L(vertical, 10, sse2)
+PRED8x8L(vertical, 10, ssse3)
+PRED8x8L(horizontal, 10, sse2)
+PRED8x8L(horizontal, 10, ssse3)
+PRED8x8L(down_left, 10, sse2)
+PRED8x8L(down_left, 10, ssse3)
+PRED8x8L(down_right, 10, sse2)
+PRED8x8L(down_right, 10, ssse3)
+PRED8x8L(vertical_right, 10, sse2)
+PRED8x8L(vertical_right, 10, ssse3)
+PRED8x8L(horizontal_up, 10, sse2)
+PRED8x8L(horizontal_up, 10, ssse3)
+
+#define PRED16x16(TYPE, DEPTH, OPT)\
+void ff_pred16x16_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
+
+PRED16x16(vertical, 10, mmxext)
+PRED16x16(vertical, 10, sse2)
+PRED16x16(horizontal, 10, mmxext)
+PRED16x16(horizontal, 10, sse2)
+
 void ff_pred16x16_vertical_mmx     (uint8_t *src, int stride);
 void ff_pred16x16_vertical_sse     (uint8_t *src, int stride);
 void ff_pred16x16_horizontal_mmx   (uint8_t *src, int stride);
@@ -253,6 +285,12 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth
         if (mm_flags & AV_CPU_FLAG_MMX2) {
             h->pred4x4[DC_PRED             ] = ff_pred4x4_dc_10_mmxext;
             h->pred4x4[HOR_UP_PRED         ] = ff_pred4x4_horizontal_up_10_mmxext;
+
+            h->pred8x8[DC_PRED8x8          ] = ff_pred8x8_dc_10_mmxext;
+            h->pred8x8[TOP_DC_PRED8x8      ] = ff_pred8x8_top_dc_10_mmxext;
+
+            h->pred16x16[VERT_PRED8x8      ] = ff_pred16x16_vertical_10_mmxext;
+            h->pred16x16[HOR_PRED8x8       ] = ff_pred16x16_horizontal_10_mmxext;
         }
         if (mm_flags & AV_CPU_FLAG_SSE2) {
             h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_sse2;
@@ -261,13 +299,33 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth
             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_sse2;
             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_sse2;
 
+            h->pred8x8[DC_PRED8x8          ] = ff_pred8x8_dc_10_sse2;
+            h->pred8x8[TOP_DC_PRED8x8      ] = ff_pred8x8_top_dc_10_sse2;
             h->pred8x8[VERT_PRED8x8        ] = ff_pred8x8_vertical_10_sse2;
             h->pred8x8[HOR_PRED8x8         ] = ff_pred8x8_horizontal_10_sse2;
+
+            h->pred8x8l[VERT_PRED           ] = ff_pred8x8l_vertical_10_sse2;
+            h->pred8x8l[HOR_PRED            ] = ff_pred8x8l_horizontal_10_sse2;
+            h->pred8x8l[DC_PRED             ] = ff_pred8x8l_dc_10_sse2;
+            h->pred8x8l[TOP_DC_PRED         ] = ff_pred8x8l_top_dc_10_sse2;
+            h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_sse2;
+            h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_sse2;
+            h->pred8x8l[VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_10_sse2;
+            h->pred8x8l[HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_10_sse2;
+
+            h->pred16x16[VERT_PRED8x8      ] = ff_pred16x16_vertical_10_sse2;
+            h->pred16x16[HOR_PRED8x8       ] = ff_pred16x16_horizontal_10_sse2;
         }
         if (mm_flags & AV_CPU_FLAG_SSSE3) {
             h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_ssse3;
             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_ssse3;
             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_ssse3;
+
+            h->pred8x8l[VERT_PRED           ] = ff_pred8x8l_vertical_10_ssse3;
+            h->pred8x8l[HOR_PRED            ] = ff_pred8x8l_horizontal_10_ssse3;
+            h->pred8x8l[DC_PRED             ] = ff_pred8x8l_dc_10_ssse3;
+            h->pred8x8l[TOP_DC_PRED         ] = ff_pred8x8l_top_dc_10_ssse3;
+            h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_ssse3;
         }
 #if HAVE_AVX
         if (mm_flags & AV_CPU_FLAG_AVX) {
diff --git a/libavcodec/x86/x86util.asm b/libavcodec/x86/x86util.asm
index b8ca348b68..09c2d54f87 100644
--- a/libavcodec/x86/x86util.asm
+++ b/libavcodec/x86/x86util.asm
@@ -528,6 +528,14 @@
 %endif
 %endmacro
 
+%macro SPLATD 2-3 0
+%if mmsize == 16
+    pshufd %1, %2, (%3)*0x55
+%else
+    pshufw %1, %2, (%3)*0x11 + ((%3)+1)*0x44
+%endif
+%endmacro
+
 %macro CLIPW 3 ;(dst, min, max)
     pmaxsw %1, %2
     pminsw %1, %3