Merge "Adds high bitdepth convolve, interpred & scaling"

14 files changed, 2238 insertions, 4 deletions

diff --git a/vp9/common/vp9_convolve.c b/vp9/common/vp9_convolve.c
index d8aaf32c4..ad70e59bf 100644
--- a/vp9/common/vp9_convolve.c
+++ b/vp9/common/vp9_convolve.c
@@ -282,3 +282,280 @@ void vp9_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride,
     dst += dst_stride;
   }
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
+                                uint8_t *dst8, ptrdiff_t dst_stride,
+                                const InterpKernel *x_filters,
+                                int x0_q4, int x_step_q4,
+                                int w, int h, int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_q4 = x0_q4;
+    for (x = 0; x < w; ++x) {
+      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_x[k] * x_filter[k];
+      dst[x] = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+      x_q4 += x_step_q4;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static void high_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
+                                    uint8_t *dst8, ptrdiff_t dst_stride,
+                                    const InterpKernel *x_filters,
+                                    int x0_q4, int x_step_q4,
+                                    int w, int h, int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_q4 = x0_q4;
+    for (x = 0; x < w; ++x) {
+      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_x[k] * x_filter[k];
+      dst[x] = ROUND_POWER_OF_TWO(dst[x] +
+          clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
+      x_q4 += x_step_q4;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static void high_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
+                               uint8_t *dst8, ptrdiff_t dst_stride,
+                               const InterpKernel *y_filters,
+                               int y0_q4, int y_step_q4, int w, int h,
+                               int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+  for (x = 0; x < w; ++x) {
+    int y_q4 = y0_q4;
+    for (y = 0; y < h; ++y) {
+      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = clip_pixel_high(
+          ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+      y_q4 += y_step_q4;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
+static void high_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
+                                   uint8_t *dst8, ptrdiff_t dst_stride,
+                                   const InterpKernel *y_filters,
+                                   int y0_q4, int y_step_q4, int w, int h,
+                                   int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+  for (x = 0; x < w; ++x) {
+    int y_q4 = y0_q4;
+    for (y = 0; y < h; ++y) {
+      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
+          clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
+      y_q4 += y_step_q4;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
+static void high_convolve(const uint8_t *src, ptrdiff_t src_stride,
+                          uint8_t *dst, ptrdiff_t dst_stride,
+                          const InterpKernel *const x_filters,
+                          int x0_q4, int x_step_q4,
+                          const InterpKernel *const y_filters,
+                          int y0_q4, int y_step_q4,
+                          int w, int h, int bd) {
+  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
+  // 2d filtering proceeds in 2 steps:
+  //   (1) Interpolate horizontally into an intermediate buffer, temp.
+  //   (2) Interpolate temp vertically to derive the sub-pixel result.
+  // Deriving the maximum number of rows in the temp buffer (135):
+  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
+  // --Largest block size is 64x64 pixels.
+  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
+  //   original frame (in 1/16th pixel units).
+  // --Must round-up because block may be located at sub-pixel position.
+  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
+  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
+  uint16_t temp[64 * 135];
+  int intermediate_height = (((h - 1) * y_step_q4 + 15) >> 4) + SUBPEL_TAPS;
+
+  assert(w <= 64);
+  assert(h <= 64);
+  assert(y_step_q4 <= 32);
+  assert(x_step_q4 <= 32);
+
+  if (intermediate_height < h)
+    intermediate_height = h;
+
+  high_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1),
+                      src_stride, CONVERT_TO_BYTEPTR(temp), 64,
+                      x_filters, x0_q4, x_step_q4, w,
+                      intermediate_height, bd);
+  high_convolve_vert(CONVERT_TO_BYTEPTR(temp) + 64 * (SUBPEL_TAPS / 2 - 1),
+                     64, dst, dst_stride, y_filters, y0_q4, y_step_q4,
+                     w, h, bd);
+}
+
+
+void vp9_high_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+                                uint8_t *dst, ptrdiff_t dst_stride,
+                                const int16_t *filter_x, int x_step_q4,
+                                const int16_t *filter_y, int y_step_q4,
+                                int w, int h, int bd) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+  const int x0_q4 = get_filter_offset(filter_x, filters_x);
+  (void)filter_y;
+  (void)y_step_q4;
+
+  high_convolve_horiz(src, src_stride, dst, dst_stride, filters_x,
+                      x0_q4, x_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+                                    uint8_t *dst, ptrdiff_t dst_stride,
+                                    const int16_t *filter_x, int x_step_q4,
+                                    const int16_t *filter_y, int y_step_q4,
+                                    int w, int h, int bd) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+  const int x0_q4 = get_filter_offset(filter_x, filters_x);
+  (void)filter_y;
+  (void)y_step_q4;
+
+  high_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x,
+                     x0_q4, x_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x, int x_step_q4,
+                               const int16_t *filter_y, int y_step_q4,
+                               int w, int h, int bd) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+  const int y0_q4 = get_filter_offset(filter_y, filters_y);
+  (void)filter_x;
+  (void)x_step_q4;
+
+  high_convolve_vert(src, src_stride, dst, dst_stride, filters_y,
+                     y0_q4, y_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const int16_t *filter_x, int x_step_q4,
+                                   const int16_t *filter_y, int y_step_q4,
+                                   int w, int h, int bd) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+  const int y0_q4 = get_filter_offset(filter_y, filters_y);
+  (void)filter_x;
+  (void)x_step_q4;
+
+  high_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y,
+                         y0_q4, y_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
+                          uint8_t *dst, ptrdiff_t dst_stride,
+                          const int16_t *filter_x, int x_step_q4,
+                          const int16_t *filter_y, int y_step_q4,
+                          int w, int h, int bd) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+  const int x0_q4 = get_filter_offset(filter_x, filters_x);
+
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+  const int y0_q4 = get_filter_offset(filter_y, filters_y);
+
+  high_convolve(src, src_stride, dst, dst_stride,
+                filters_x, x0_q4, x_step_q4,
+                filters_y, y0_q4, y_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
+                              uint8_t *dst, ptrdiff_t dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h, int bd) {
+  // Fixed size intermediate buffer places limits on parameters.
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, temp, 64 * 64);
+  assert(w <= 64);
+  assert(h <= 64);
+
+  vp9_high_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), 64,
+                       filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd);
+  vp9_high_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), 64, dst, dst_stride,
+                          NULL, 0, NULL, 0, w, h, bd);
+}
+
+void vp9_high_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
+                              uint8_t *dst8, ptrdiff_t dst_stride,
+                              const int16_t *filter_x, int filter_x_stride,
+                              const int16_t *filter_y, int filter_y_stride,
+                              int w, int h, int bd) {
+  int r;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  (void)filter_x;
+  (void)filter_y;
+  (void)filter_x_stride;
+  (void)filter_y_stride;
+  (void)bd;
+
+  for (r = h; r > 0; --r) {
+    vpx_memcpy(dst, src, w * sizeof(uint16_t));
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+void vp9_high_convolve_avg_c(const uint8_t *src8, ptrdiff_t src_stride,
+                             uint8_t *dst8, ptrdiff_t dst_stride,
+                             const int16_t *filter_x, int filter_x_stride,
+                             const int16_t *filter_y, int filter_y_stride,
+                             int w, int h, int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  (void)filter_x;
+  (void)filter_y;
+  (void)filter_x_stride;
+  (void)filter_y_stride;
+  (void)bd;
+
+  for (y = 0; y < h; ++y) {
+    for (x = 0; x < w; ++x) {
+      dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+#endif
diff --git a/vp9/common/vp9_convolve.h b/vp9/common/vp9_convolve.h
index 6bf71fc79..faf70b12a 100644
--- a/vp9/common/vp9_convolve.h
+++ b/vp9/common/vp9_convolve.h
@@ -23,6 +23,14 @@ typedef void (*convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
                               const int16_t *filter_y, int y_step_q4,
                               int w, int h);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef void (*high_convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const int16_t *filter_x, int x_step_q4,
+                                   const int16_t *filter_y, int y_step_q4,
+                                   int w, int h, int bd);
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index 86ae64839..28723f675 100644
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -63,6 +63,53 @@ static void build_mc_border(const uint8_t *src, int src_stride,
   } while (--b_h);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_build_mc_border(const uint8_t *src8, int src_stride,
+                                 uint16_t *dst, int dst_stride,
+                                 int x, int y, int b_w, int b_h,
+                                 int w, int h) {
+  // Get a pointer to the start of the real data for this row.
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *ref_row = src - x - y * src_stride;
+
+  if (y >= h)
+    ref_row += (h - 1) * src_stride;
+  else if (y > 0)
+    ref_row += y * src_stride;
+
+  do {
+    int right = 0, copy;
+    int left = x < 0 ? -x : 0;
+
+    if (left > b_w)
+      left = b_w;
+
+    if (x + b_w > w)
+      right = x + b_w - w;
+
+    if (right > b_w)
+      right = b_w;
+
+    copy = b_w - left - right;
+
+    if (left)
+      vpx_memset16(dst, ref_row[0], left);
+
+    if (copy)
+      memcpy(dst + left, ref_row + x + left, copy * sizeof(uint16_t));
+
+    if (right)
+      vpx_memset16(dst + left + copy, ref_row[w - 1], right);
+
+    dst += dst_stride;
+    ++y;
+
+    if (y > 0 && y < h)
+      ref_row += src_stride;
+  } while (--b_h);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void inter_predictor(const uint8_t *src, int src_stride,
                             uint8_t *dst, int dst_stride,
                             const int subpel_x,
@@ -97,6 +144,42 @@ void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
                   sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_inter_predictor(const uint8_t *src, int src_stride,
+                                 uint8_t *dst, int dst_stride,
+                                 const int subpel_x,
+                                 const int subpel_y,
+                                 const struct scale_factors *sf,
+                                 int w, int h, int ref,
+                                 const InterpKernel *kernel,
+                                 int xs, int ys, int bd) {
+  sf->high_predict[subpel_x != 0][subpel_y != 0][ref](
+      src, src_stride, dst, dst_stride,
+      kernel[subpel_x], xs, kernel[subpel_y], ys, w, h, bd);
+}
+
+void vp9_high_build_inter_predictor(const uint8_t *src, int src_stride,
+                                    uint8_t *dst, int dst_stride,
+                                    const MV *src_mv,
+                                    const struct scale_factors *sf,
+                                    int w, int h, int ref,
+                                    const InterpKernel *kernel,
+                                    enum mv_precision precision,
+                                    int x, int y, int bd) {
+  const int is_q4 = precision == MV_PRECISION_Q4;
+  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
+                     is_q4 ? src_mv->col : src_mv->col * 2 };
+  MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
+  const int subpel_x = mv.col & SUBPEL_MASK;
+  const int subpel_y = mv.row & SUBPEL_MASK;
+
+  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
+
+  high_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
+                       sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4, bd);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static INLINE int round_mv_comp_q4(int value) {
   return (value < 0 ? value - 2 : value + 2) / 4;
 }
@@ -222,8 +305,19 @@ static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
     pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride
            + (scaled_mv.col >> SUBPEL_BITS);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      high_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
+                           subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys,
+                           xd->bd);
+    } else {
+      inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
+                      subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
+    }
+#else
     inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
                     subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 }
 
@@ -393,16 +487,64 @@ static void dec_build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
           y0 < 0 || y0 > frame_height - 1 || y1 < 0 || y1 > frame_height - 1) {
         uint8_t *buf_ptr1 = ref_frame + y0 * pre_buf->stride + x0;
         // Extend the border.
-        build_mc_border(buf_ptr1, pre_buf->stride, xd->mc_buf, x1 - x0 + 1,
-                        x0, y0, x1 - x0 + 1, y1 - y0 + 1, frame_width,
+#if CONFIG_VP9_HIGHBITDEPTH
+        if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+          high_build_mc_border(buf_ptr1,
+                               pre_buf->stride,
+                               xd->mc_buf_high,
+                               x1 - x0 + 1,
+                               x0,
+                               y0,
+                               x1 - x0 + 1,
+                               y1 - y0 + 1,
+                               frame_width,
+                               frame_height);
+          buf_stride = x1 - x0 + 1;
+          buf_ptr = CONVERT_TO_BYTEPTR(xd->mc_buf_high) +
+              y_pad * 3 * buf_stride + x_pad * 3;
+        } else {
+          build_mc_border(buf_ptr1,
+                          pre_buf->stride,
+                          xd->mc_buf,
+                          x1 - x0 + 1,
+                          x0,
+                          y0,
+                          x1 - x0 + 1,
+                          y1 - y0 + 1,
+                          frame_width,
+                          frame_height);
+          buf_stride = x1 - x0 + 1;
+          buf_ptr = xd->mc_buf + y_pad * 3 * buf_stride + x_pad * 3;
+        }
+#else
+        build_mc_border(buf_ptr1,
+                        pre_buf->stride,
+                        xd->mc_buf,
+                        x1 - x0 + 1,
+                        x0,
+                        y0,
+                        x1 - x0 + 1,
+                        y1 - y0 + 1,
+                        frame_width,
                         frame_height);
         buf_stride = x1 - x0 + 1;
         buf_ptr = xd->mc_buf + y_pad * 3 * buf_stride + x_pad * 3;
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       }
     }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      high_inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
+                           subpel_y, sf, w, h, ref, kernel, xs, ys, xd->bd);
+    } else {
+      inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
+                    subpel_y, sf, w, h, ref, kernel, xs, ys);
+    }
+#else
     inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
                     subpel_y, sf, w, h, ref, kernel, xs, ys);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 }
 
diff --git a/vp9/common/vp9_reconinter.h b/vp9/common/vp9_reconinter.h
index 58c596ee8..e70cc4c99 100644
--- a/vp9/common/vp9_reconinter.h
+++ b/vp9/common/vp9_reconinter.h
@@ -39,6 +39,17 @@ void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
                                enum mv_precision precision,
                                int x, int y);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_build_inter_predictor(const uint8_t *src, int src_stride,
+                                    uint8_t *dst, int dst_stride,
+                                    const MV *mv_q3,
+                                    const struct scale_factors *sf,
+                                    int w, int h, int do_avg,
+                                    const InterpKernel *kernel,
+                                    enum mv_precision precision,
+                                    int x, int y, int bd);
+#endif
+
 static INLINE int scaled_buffer_offset(int x_offset, int y_offset, int stride,
                                        const struct scale_factors *sf) {
   const int x = sf ? sf->scale_value_x(x_offset, sf) : x_offset;
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index b75ea64f0..0f52ae15c 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -607,6 +607,33 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vp9_high_dc_128_predictor_32x32/;
 
   #
+  # Sub Pixel Filters
+  #
+  add_proto qw/void vp9_high_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve_copy/;
+
+  add_proto qw/void vp9_high_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve_avg/;
+
+  add_proto qw/void vp9_high_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_horiz/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_vert/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_avg/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_avg_horiz/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_avg_vert/, "$sse2_x86_64";
+
+  #
   # dct
   #
   add_proto qw/void vp9_high_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
diff --git a/vp9/common/vp9_scale.c b/vp9/common/vp9_scale.c
index 2f58323aa..63e2b5306 100644
--- a/vp9/common/vp9_scale.c
+++ b/vp9/common/vp9_scale.c
@@ -43,9 +43,16 @@ MV32 vp9_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf) {
   return res;
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
+                                       int other_w, int other_h,
+                                       int this_w, int this_h,
+                                       int use_high) {
+#else
 void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
                                        int other_w, int other_h,
                                        int this_w, int this_h) {
+#endif
   if (!valid_ref_frame_size(other_w, other_h, this_w, this_h)) {
     sf->x_scale_fp = REF_INVALID_SCALE;
     sf->y_scale_fp = REF_INVALID_SCALE;
@@ -111,4 +118,48 @@ void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
   // 2D subpel motion always gets filtered in both directions
   sf->predict[1][1][0] = vp9_convolve8;
   sf->predict[1][1][1] = vp9_convolve8_avg;
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (use_high) {
+    if (sf->x_step_q4 == 16) {
+      if (sf->y_step_q4 == 16) {
+        // No scaling in either direction.
+        sf->high_predict[0][0][0] = vp9_high_convolve_copy;
+        sf->high_predict[0][0][1] = vp9_high_convolve_avg;
+        sf->high_predict[0][1][0] = vp9_high_convolve8_vert;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg_vert;
+        sf->high_predict[1][0][0] = vp9_high_convolve8_horiz;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg_horiz;
+      } else {
+        // No scaling in x direction. Must always scale in the y direction.
+        sf->high_predict[0][0][0] = vp9_high_convolve8_vert;
+        sf->high_predict[0][0][1] = vp9_high_convolve8_avg_vert;
+        sf->high_predict[0][1][0] = vp9_high_convolve8_vert;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg_vert;
+        sf->high_predict[1][0][0] = vp9_high_convolve8;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg;
+      }
+    } else {
+      if (sf->y_step_q4 == 16) {
+        // No scaling in the y direction. Must always scale in the x direction.
+        sf->high_predict[0][0][0] = vp9_high_convolve8_horiz;
+        sf->high_predict[0][0][1] = vp9_high_convolve8_avg_horiz;
+        sf->high_predict[0][1][0] = vp9_high_convolve8;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg;
+        sf->high_predict[1][0][0] = vp9_high_convolve8_horiz;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg_horiz;
+      } else {
+        // Must always scale in both directions.
+        sf->high_predict[0][0][0] = vp9_high_convolve8;
+        sf->high_predict[0][0][1] = vp9_high_convolve8_avg;
+        sf->high_predict[0][1][0] = vp9_high_convolve8;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg;
+        sf->high_predict[1][0][0] = vp9_high_convolve8;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg;
+      }
+    }
+    // 2D subpel motion always gets filtered in both directions.
+    sf->high_predict[1][1][0] = vp9_high_convolve8;
+    sf->high_predict[1][1][1] = vp9_high_convolve8_avg;
+  }
+#endif
 }
diff --git a/vp9/common/vp9_scale.h b/vp9/common/vp9_scale.h
index ad6f5d702..2e923db9c 100644
--- a/vp9/common/vp9_scale.h
+++ b/vp9/common/vp9_scale.h
@@ -32,13 +32,23 @@ struct scale_factors {
   int (*scale_value_y)(int val, const struct scale_factors *sf);
 
   convolve_fn_t predict[2][2][2];  // horiz, vert, avg
+#if CONFIG_VP9_HIGHBITDEPTH
+  high_convolve_fn_t high_predict[2][2][2];  // horiz, vert, avg
+#endif
 };
 
 MV32 vp9_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
+                                       int other_w, int other_h,
+                                       int this_w, int this_h,
+                                       int use_high);
+#else
 void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
                                        int other_w, int other_h,
                                        int this_w, int this_h);
+#endif
 
 static INLINE int vp9_is_valid_scale(const struct scale_factors *sf) {
   return sf->x_scale_fp != REF_INVALID_SCALE &&
diff --git a/vp9/common/x86/vp9_asm_stubs.c b/vp9/common/x86/vp9_asm_stubs.c
index b6847b92e..407573aee 100644
--- a/vp9/common/x86/vp9_asm_stubs.c
+++ b/vp9/common/x86/vp9_asm_stubs.c
@@ -139,6 +139,153 @@ void vp9_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
                            filter_x, x_step_q4, filter_y, y_step_q4, w, h); \
   } \
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+
+typedef void high_filter8_1dfunction (
+  const uint16_t *src_ptr,
+  const ptrdiff_t src_pitch,
+  uint16_t *output_ptr,
+  ptrdiff_t out_pitch,
+  unsigned int output_height,
+  const int16_t *filter,
+  int bd
+);
+
+#define HIGH_FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
+  void vp9_high_convolve8_##name##_##opt(const uint8_t *src8, \
+                                         ptrdiff_t src_stride, \
+                                         uint8_t *dst8, ptrdiff_t dst_stride, \
+                                         const int16_t *filter_x, \
+                                         int x_step_q4, \
+                                         const int16_t *filter_y, \
+                                         int y_step_q4, \
+                                         int w, int h, int bd) { \
+  if (step_q4 == 16 && filter[3] != 128) { \
+    uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
+    uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
+    if (filter[0] || filter[1] || filter[2]) { \
+      while (w >= 16) { \
+        vp9_high_filter_block1d16_##dir##8_##avg##opt(src_start, \
+                                                      src_stride, \
+                                                      dst, \
+                                                      dst_stride, \
+                                                      h, \
+                                                      filter, \
+                                                      bd); \
+        src += 16; \
+        dst += 16; \
+        w -= 16; \
+      } \
+      while (w >= 8) { \
+        vp9_high_filter_block1d8_##dir##8_##avg##opt(src_start, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 8; \
+        dst += 8; \
+        w -= 8; \
+      } \
+      while (w >= 4) { \
+        vp9_high_filter_block1d4_##dir##8_##avg##opt(src_start, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 4; \
+        dst += 4; \
+        w -= 4; \
+      } \
+    } else { \
+      while (w >= 16) { \
+        vp9_high_filter_block1d16_##dir##2_##avg##opt(src, \
+                                                      src_stride, \
+                                                      dst, \
+                                                      dst_stride, \
+                                                      h, \
+                                                      filter, \
+                                                      bd); \
+        src += 16; \
+        dst += 16; \
+        w -= 16; \
+      } \
+      while (w >= 8) { \
+        vp9_high_filter_block1d8_##dir##2_##avg##opt(src, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 8; \
+        dst += 8; \
+        w -= 8; \
+      } \
+      while (w >= 4) { \
+        vp9_high_filter_block1d4_##dir##2_##avg##opt(src, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 4; \
+        dst += 4; \
+        w -= 4; \
+      } \
+    } \
+  } \
+  if (w) { \
+    vp9_high_convolve8_##name##_c(src8, src_stride, dst8, dst_stride, \
+                                  filter_x, x_step_q4, filter_y, y_step_q4, \
+                                  w, h, bd); \
+  } \
+}
+
+#define HIGH_FUN_CONV_2D(avg, opt) \
+void vp9_high_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
+                                   uint8_t *dst, ptrdiff_t dst_stride, \
+                                   const int16_t *filter_x, int x_step_q4, \
+                                   const int16_t *filter_y, int y_step_q4, \
+                                   int w, int h, int bd) { \
+  assert(w <= 64); \
+  assert(h <= 64); \
+  if (x_step_q4 == 16 && y_step_q4 == 16) { \
+    if (filter_x[0] || filter_x[1] || filter_x[2] || filter_x[3] == 128 || \
+        filter_y[0] || filter_y[1] || filter_y[2] || filter_y[3] == 128) { \
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, fdata2, 64 * 71); \
+      vp9_high_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \
+                                     CONVERT_TO_BYTEPTR(fdata2), 64, \
+                                     filter_x, x_step_q4, filter_y, y_step_q4, \
+                                     w, h + 7, bd); \
+      vp9_high_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2) + 192, \
+                                           64, dst, dst_stride, \
+                                           filter_x, x_step_q4, filter_y, \
+                                           y_step_q4, w, h, bd); \
+    } else { \
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, fdata2, 64 * 65); \
+      vp9_high_convolve8_horiz_##opt(src, src_stride, \
+                                     CONVERT_TO_BYTEPTR(fdata2), 64, \
+                                     filter_x, x_step_q4, filter_y, y_step_q4, \
+                                     w, h + 1, bd); \
+      vp9_high_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2), 64, \
+                                           dst, dst_stride, \
+                                           filter_x, x_step_q4, filter_y, \
+                                           y_step_q4, w, h, bd); \
+    } \
+  } else { \
+    vp9_high_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
+                                filter_x, x_step_q4, filter_y, y_step_q4, w, \
+                                h, bd); \
+  } \
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 #if HAVE_AVX2 && HAVE_SSSE3
 filter8_1dfunction vp9_filter_block1d16_v8_avx2;
 filter8_1dfunction vp9_filter_block1d16_h8_avx2;
@@ -336,4 +483,75 @@ FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2);
 //                             int w, int h);
 FUN_CONV_2D(, sse2);
 FUN_CONV_2D(avg_ , sse2);
+
+#if CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
+high_filter8_1dfunction vp9_high_filter_block1d16_v8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_v8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h8_avg_sse2;
+
+high_filter8_1dfunction vp9_high_filter_block1d16_v2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_v2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h2_avg_sse2;
+
+// void vp9_high_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                                    uint8_t *dst, ptrdiff_t dst_stride,
+//                                    const int16_t *filter_x, int x_step_q4,
+//                                    const int16_t *filter_y, int y_step_q4,
+//                                    int w, int h, int bd);
+// void vp9_high_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                                   uint8_t *dst, ptrdiff_t dst_stride,
+//                                   const int16_t *filter_x, int x_step_q4,
+//                                   const int16_t *filter_y, int y_step_q4,
+//                                   int w, int h, int bd);
+// void vp9_high_convolve8_avg_horiz_sse2(const uint8_t *src,
+//                                        ptrdiff_t src_stride,
+//                                        uint8_t *dst, ptrdiff_t dst_stride,
+//                                        const int16_t *filter_x,
+//                                        int x_step_q4,
+//                                        const int16_t *filter_y,
+//                                        int y_step_q4,
+//                                        int w, int h, int bd);
+// void vp9_high_convolve8_avg_vert_sse2(const uint8_t *src,
+//                                       ptrdiff_t src_stride,
+//                                       uint8_t *dst, ptrdiff_t dst_stride,
+//                                       const int16_t *filter_x, int x_step_q4,
+//                                       const int16_t *filter_y, int y_step_q4,
+//                                       int w, int h, int bd);
+HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
+HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
+HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
+HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
+                 sse2);
+
+// void vp9_high_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                              uint8_t *dst, ptrdiff_t dst_stride,
+//                              const int16_t *filter_x, int x_step_q4,
+//                              const int16_t *filter_y, int y_step_q4,
+//                              int w, int h, int bd);
+// void vp9_high_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                                  uint8_t *dst, ptrdiff_t dst_stride,
+//                                  const int16_t *filter_x, int x_step_q4,
+//                                  const int16_t *filter_y, int y_step_q4,
+//                                  int w, int h, int bd);
+HIGH_FUN_CONV_2D(, sse2);
+HIGH_FUN_CONV_2D(avg_ , sse2);
+#endif  // CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
 #endif  // HAVE_SSE2
diff --git a/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm b/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm
new file mode 100644
index 000000000..4bdbb83f4
--- /dev/null
+++ b/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm
@@ -0,0 +1,962 @@
+;
+;  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;Note: tap3 and tap4 have to be applied and added after other taps to avoid
+;overflow.
+
+%macro HIGH_GET_FILTERS_4 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm7, [rdx]                 ;load filters
+    pshuflw     xmm0, xmm7, 0b              ;k0
+    pshuflw     xmm1, xmm7, 01010101b       ;k1
+    pshuflw     xmm2, xmm7, 10101010b       ;k2
+    pshuflw     xmm3, xmm7, 11111111b       ;k3
+    psrldq      xmm7, 8
+    pshuflw     xmm4, xmm7, 0b              ;k4
+    pshuflw     xmm5, xmm7, 01010101b       ;k5
+    pshuflw     xmm6, xmm7, 10101010b       ;k6
+    pshuflw     xmm7, xmm7, 11111111b       ;k7
+
+    punpcklwd   xmm0, xmm6
+    punpcklwd   xmm2, xmm5
+    punpcklwd   xmm3, xmm4
+    punpcklwd   xmm1, xmm7
+
+    movdqa      k0k6, xmm0
+    movdqa      k2k5, xmm2
+    movdqa      k3k4, xmm3
+    movdqa      k1k7, xmm1
+
+    movq        xmm6, rcx
+    pshufd      xmm6, xmm6, 0
+    movdqa      krd, xmm6
+
+    ;Compute max and min values of a pixel
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)      ;bps
+    movq        xmm0, rdx
+    movq        xmm1, rcx
+    pshufd      xmm0, xmm0, 0b
+    movdqa      xmm2, xmm0
+    psllw       xmm0, xmm1
+    psubw       xmm0, xmm2
+    pxor        xmm1, xmm1
+    movdqa      max, xmm0                  ;max value (for clamping)
+    movdqa      min, xmm1                  ;min value (for clamping)
+
+%endm
+
+%macro HIGH_APPLY_FILTER_4 1
+    punpcklwd   xmm0, xmm6                  ;two row in one register
+    punpcklwd   xmm1, xmm7
+    punpcklwd   xmm2, xmm5
+    punpcklwd   xmm3, xmm4
+
+    pmaddwd     xmm0, k0k6                  ;multiply the filter factors
+    pmaddwd     xmm1, k1k7
+    pmaddwd     xmm2, k2k5
+    pmaddwd     xmm3, k3k4
+
+    paddd       xmm0, xmm1                  ;sum
+    paddd       xmm0, xmm2
+    paddd       xmm0, xmm3
+
+    paddd       xmm0, krd                   ;rounding
+    psrad       xmm0, 7                     ;shift
+    packssdw    xmm0, xmm0                  ;pack to word
+
+    ;clamp the values
+    pminsw      xmm0, max
+    pmaxsw      xmm0, min
+
+%if %1
+    movq        xmm1, [rdi]
+    pavgw       xmm0, xmm1
+%endif
+    movq        [rdi], xmm0
+%endm
+
+%macro HIGH_GET_FILTERS 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm7, [rdx]                 ;load filters
+    pshuflw     xmm0, xmm7, 0b              ;k0
+    pshuflw     xmm1, xmm7, 01010101b       ;k1
+    pshuflw     xmm2, xmm7, 10101010b       ;k2
+    pshuflw     xmm3, xmm7, 11111111b       ;k3
+    pshufhw     xmm4, xmm7, 0b              ;k4
+    pshufhw     xmm5, xmm7, 01010101b       ;k5
+    pshufhw     xmm6, xmm7, 10101010b       ;k6
+    pshufhw     xmm7, xmm7, 11111111b       ;k7
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+    punpcklwd   xmm0, xmm1
+    punpckhwd   xmm6, xmm7
+    punpckhwd   xmm2, xmm5
+    punpckhwd   xmm3, xmm4
+
+    movdqa      k0k1, xmm0                  ;store filter factors on stack
+    movdqa      k6k7, xmm6
+    movdqa      k2k5, xmm2
+    movdqa      k3k4, xmm3
+
+    movq        xmm6, rcx
+    pshufd      xmm6, xmm6, 0
+    movdqa      krd, xmm6                   ;rounding
+
+    ;Compute max and min values of a pixel
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)       ;bps
+    movq        xmm0, rdx
+    movq        xmm1, rcx
+    pshufd      xmm0, xmm0, 0b
+    movdqa      xmm2, xmm0
+    psllw       xmm0, xmm1
+    psubw       xmm0, xmm2
+    pxor        xmm1, xmm1
+    movdqa      max, xmm0                  ;max value (for clamping)
+    movdqa      min, xmm1                  ;min value (for clamping)
+%endm
+
+%macro LOAD_VERT_8 1
+    movdqu      xmm0, [rsi + %1]            ;0
+    movdqu      xmm1, [rsi + rax + %1]      ;1
+    movdqu      xmm6, [rsi + rdx * 2 + %1]  ;6
+    lea         rsi,  [rsi + rax]
+    movdqu      xmm7, [rsi + rdx * 2 + %1]  ;7
+    movdqu      xmm2, [rsi + rax + %1]      ;2
+    movdqu      xmm3, [rsi + rax * 2 + %1]  ;3
+    movdqu      xmm4, [rsi + rdx + %1]      ;4
+    movdqu      xmm5, [rsi + rax * 4 + %1]  ;5
+%endm
+
+%macro HIGH_APPLY_FILTER_8 2
+    movdqu      temp, xmm4
+    movdqa      xmm4, xmm0
+    punpcklwd   xmm0, xmm1
+    punpckhwd   xmm4, xmm1
+    movdqa      xmm1, xmm6
+    punpcklwd   xmm6, xmm7
+    punpckhwd   xmm1, xmm7
+    movdqa      xmm7, xmm2
+    punpcklwd   xmm2, xmm5
+    punpckhwd   xmm7, xmm5
+
+    movdqu      xmm5, temp
+    movdqu      temp, xmm4
+    movdqa      xmm4, xmm3
+    punpcklwd   xmm3, xmm5
+    punpckhwd   xmm4, xmm5
+    movdqu      xmm5, temp
+
+    pmaddwd     xmm0, k0k1
+    pmaddwd     xmm5, k0k1
+    pmaddwd     xmm6, k6k7
+    pmaddwd     xmm1, k6k7
+    pmaddwd     xmm2, k2k5
+    pmaddwd     xmm7, k2k5
+    pmaddwd     xmm3, k3k4
+    pmaddwd     xmm4, k3k4
+
+    paddd       xmm0, xmm6
+    paddd       xmm0, xmm2
+    paddd       xmm0, xmm3
+    paddd       xmm5, xmm1
+    paddd       xmm5, xmm7
+    paddd       xmm5, xmm4
+
+    paddd       xmm0, krd                   ;rounding
+    paddd       xmm5, krd
+    psrad       xmm0, 7                     ;shift
+    psrad       xmm5, 7
+    packssdw    xmm0, xmm5                  ;pack back to word
+
+    ;clamp the values
+    pminsw      xmm0, max
+    pmaxsw      xmm0, min
+
+%if %1
+    movdqu      xmm1, [rdi + %2]
+    pavgw       xmm0, xmm1
+%endif
+    movdqu      [rdi + %2], xmm0
+%endm
+
+;void vp9_filter_block1d4_v8_sse2
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d4_v8_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movq        xmm0, [rsi]                 ;load src: row 0
+    movq        xmm1, [rsi + rax]           ;1
+    movq        xmm6, [rsi + rdx * 2]       ;6
+    lea         rsi,  [rsi + rax]
+    movq        xmm7, [rsi + rdx * 2]       ;7
+    movq        xmm2, [rsi + rax]           ;2
+    movq        xmm3, [rsi + rax * 2]       ;3
+    movq        xmm4, [rsi + rdx]           ;4
+    movq        xmm5, [rsi + rax * 4]       ;5
+
+    HIGH_APPLY_FILTER_4 0
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d8_v8_sse2
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d8_v8_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 0, 0
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d16_v8_sse2
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d16_v8_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 0, 0
+    sub         rsi, rax
+
+    LOAD_VERT_8 16
+    HIGH_APPLY_FILTER_8 0, 16
+    add         rdi, rbx
+
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d4_v8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movq        xmm0, [rsi]                 ;load src: row 0
+    movq        xmm1, [rsi + rax]           ;1
+    movq        xmm6, [rsi + rdx * 2]       ;6
+    lea         rsi,  [rsi + rax]
+    movq        xmm7, [rsi + rdx * 2]       ;7
+    movq        xmm2, [rsi + rax]           ;2
+    movq        xmm3, [rsi + rax * 2]       ;3
+    movq        xmm4, [rsi + rdx]           ;4
+    movq        xmm5, [rsi + rax * 4]       ;5
+
+    HIGH_APPLY_FILTER_4 1
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d8_v8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 1, 0
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_v8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 1, 0
+    sub         rsi, rax
+
+    LOAD_VERT_8 16
+    HIGH_APPLY_FILTER_8 1, 16
+    add         rdi, rbx
+
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d4_h8_sse2
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d4_h8_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm4,   [rsi + 2]
+    movdqa      xmm1, xmm0
+    movdqa      xmm6, xmm4
+    movdqa      xmm7, xmm4
+    movdqa      xmm2, xmm0
+    movdqa      xmm3, xmm0
+    movdqa      xmm5, xmm4
+
+    psrldq      xmm1, 2
+    psrldq      xmm6, 4
+    psrldq      xmm7, 6
+    psrldq      xmm2, 4
+    psrldq      xmm3, 6
+    psrldq      xmm5, 2
+
+    HIGH_APPLY_FILTER_4 0
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d8_h8_sse2
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d8_h8_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 0, 0
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d16_h8_sse2
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d16_h8_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 0, 0
+
+    movdqu      xmm0,   [rsi + 10]           ;load src
+    movdqu      xmm1,   [rsi + 12]
+    movdqu      xmm2,   [rsi + 14]
+    movdqu      xmm3,   [rsi + 16]
+    movdqu      xmm4,   [rsi + 18]
+    movdqu      xmm5,   [rsi + 20]
+    movdqu      xmm6,   [rsi + 22]
+    movdqu      xmm7,   [rsi + 24]
+
+    HIGH_APPLY_FILTER_8 0, 16
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d4_h8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm4,   [rsi + 2]
+    movdqa      xmm1, xmm0
+    movdqa      xmm6, xmm4
+    movdqa      xmm7, xmm4
+    movdqa      xmm2, xmm0
+    movdqa      xmm3, xmm0
+    movdqa      xmm5, xmm4
+
+    psrldq      xmm1, 2
+    psrldq      xmm6, 4
+    psrldq      xmm7, 6
+    psrldq      xmm2, 4
+    psrldq      xmm3, 6
+    psrldq      xmm5, 2
+
+    HIGH_APPLY_FILTER_4 1
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d8_h8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 1, 0
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_h8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 1, 0
+
+    movdqu      xmm0,   [rsi + 10]           ;load src
+    movdqu      xmm1,   [rsi + 12]
+    movdqu      xmm2,   [rsi + 14]
+    movdqu      xmm3,   [rsi + 16]
+    movdqu      xmm4,   [rsi + 18]
+    movdqu      xmm5,   [rsi + 20]
+    movdqu      xmm6,   [rsi + 22]
+    movdqu      xmm7,   [rsi + 24]
+
+    HIGH_APPLY_FILTER_8 1, 16
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm b/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm
new file mode 100644
index 000000000..b7d4a61ff
--- /dev/null
+++ b/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm
@@ -0,0 +1,494 @@
+;
+;  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro HIGH_GET_PARAM_4 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm3, [rdx]                 ;load filters
+    pshuflw     xmm4, xmm3, 11111111b       ;k3
+    psrldq      xmm3, 8
+    pshuflw     xmm3, xmm3, 0b              ;k4
+    punpcklwd   xmm4, xmm3                  ;k3k4
+
+    movq        xmm3, rcx                   ;rounding
+    pshufd      xmm3, xmm3, 0
+
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)       ;bps
+    movq        xmm5, rdx
+    movq        xmm2, rcx
+    pshufd      xmm5, xmm5, 0b
+    movdqa      xmm1, xmm5
+    psllw       xmm5, xmm2
+    psubw       xmm5, xmm1                  ;max value (for clamping)
+    pxor        xmm2, xmm2                  ;min value (for clamping)
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+%endm
+
+%macro HIGH_APPLY_FILTER_4 1
+
+    punpcklwd   xmm0, xmm1                  ;two row in one register
+    pmaddwd     xmm0, xmm4                  ;multiply the filter factors
+
+    paddd       xmm0, xmm3                  ;rounding
+    psrad       xmm0, 7                     ;shift
+    packssdw    xmm0, xmm0                  ;pack to word
+
+    ;clamp the values
+    pminsw      xmm0, xmm5
+    pmaxsw      xmm0, xmm2
+
+%if %1
+    movq        xmm1, [rdi]
+    pavgw       xmm0, xmm1
+%endif
+
+    movq        [rdi], xmm0
+    lea         rsi, [rsi + 2*rax]
+    lea         rdi, [rdi + 2*rdx]
+    dec         rcx
+%endm
+
+%if ARCH_X86_64
+%macro HIGH_GET_PARAM 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm6, [rdx]                 ;load filters
+
+    pshuflw     xmm7, xmm6, 11111111b       ;k3
+    pshufhw     xmm6, xmm6, 0b              ;k4
+    psrldq      xmm6, 8
+    punpcklwd   xmm7, xmm6                  ;k3k4k3k4k3k4k3k4
+
+    movq        xmm4, rcx                   ;rounding
+    pshufd      xmm4, xmm4, 0
+
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)       ;bps
+    movq        xmm8, rdx
+    movq        xmm5, rcx
+    pshufd      xmm8, xmm8, 0b
+    movdqa      xmm1, xmm8
+    psllw       xmm8, xmm5
+    psubw       xmm8, xmm1                  ;max value (for clamping)
+    pxor        xmm5, xmm5                  ;min value (for clamping)
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+%endm
+
+%macro HIGH_APPLY_FILTER_8 1
+    movdqa      xmm6, xmm0
+    punpckhwd   xmm6, xmm1
+    punpcklwd   xmm0, xmm1
+    pmaddwd     xmm6, xmm7
+    pmaddwd     xmm0, xmm7
+
+    paddd       xmm6, xmm4                  ;rounding
+    paddd       xmm0, xmm4                  ;rounding
+    psrad       xmm6, 7                     ;shift
+    psrad       xmm0, 7                     ;shift
+    packssdw    xmm0, xmm6                  ;pack back to word
+
+    ;clamp the values
+    pminsw      xmm0, xmm8
+    pmaxsw      xmm0, xmm5
+
+%if %1
+    movdqu      xmm1, [rdi]
+    pavgw       xmm0, xmm1
+%endif
+    movdqu      [rdi], xmm0                 ;store the result
+
+    lea         rsi, [rsi + 2*rax]
+    lea         rdi, [rdi + 2*rdx]
+    dec         rcx
+%endm
+
+%macro HIGH_APPLY_FILTER_16 1
+    movdqa      xmm9, xmm0
+    movdqa      xmm6, xmm2
+    punpckhwd   xmm9, xmm1
+    punpckhwd   xmm6, xmm3
+    punpcklwd   xmm0, xmm1
+    punpcklwd   xmm2, xmm3
+
+    pmaddwd     xmm9, xmm7
+    pmaddwd     xmm6, xmm7
+    pmaddwd     xmm0, xmm7
+    pmaddwd     xmm2, xmm7
+
+    paddd       xmm9, xmm4                  ;rounding
+    paddd       xmm6, xmm4
+    paddd       xmm0, xmm4
+    paddd       xmm2, xmm4
+
+    psrad       xmm9, 7                     ;shift
+    psrad       xmm6, 7
+    psrad       xmm0, 7
+    psrad       xmm2, 7
+
+    packssdw    xmm0, xmm9                  ;pack back to word
+    packssdw    xmm2, xmm6                  ;pack back to word
+
+    ;clamp the values
+    pminsw      xmm0, xmm8
+    pmaxsw      xmm0, xmm5
+    pminsw      xmm2, xmm8
+    pmaxsw      xmm2, xmm5
+
+%if %1
+    movdqu      xmm1, [rdi]
+    movdqu      xmm3, [rdi + 16]
+    pavgw       xmm0, xmm1
+    pavgw       xmm2, xmm3
+%endif
+    movdqu      [rdi], xmm0               ;store the result
+    movdqu      [rdi + 16], xmm2          ;store the result
+
+    lea         rsi, [rsi + 2*rax]
+    lea         rdi, [rdi + 2*rdx]
+    dec         rcx
+%endm
+%endif
+
+global sym(vp9_high_filter_block1d4_v2_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movq        xmm0, [rsi]                 ;load src
+    movq        xmm1, [rsi + 2*rax]
+
+    HIGH_APPLY_FILTER_4 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_v2_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;0
+    movdqu      xmm1, [rsi + 2*rax]         ;1
+
+    HIGH_APPLY_FILTER_8 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_v2_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu        xmm0, [rsi]               ;0
+    movdqu        xmm2, [rsi + 16]
+    movdqu        xmm1, [rsi + 2*rax]       ;1
+    movdqu        xmm3, [rsi + 2*rax + 16]
+
+    HIGH_APPLY_FILTER_16 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
+
+global sym(vp9_high_filter_block1d4_v2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movq        xmm0, [rsi]                 ;load src
+    movq        xmm1, [rsi + 2*rax]
+
+    HIGH_APPLY_FILTER_4 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_v2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;0
+    movdqu      xmm1, [rsi + 2*rax]         ;1
+
+    HIGH_APPLY_FILTER_8 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_v2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu        xmm0, [rsi]               ;0
+    movdqu        xmm1, [rsi + 2*rax]       ;1
+    movdqu        xmm2, [rsi + 16]
+    movdqu        xmm3, [rsi + 2*rax + 16]
+
+    HIGH_APPLY_FILTER_16 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
+
+global sym(vp9_high_filter_block1d4_h2_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqa      xmm1, xmm0
+    psrldq      xmm1, 2
+
+    HIGH_APPLY_FILTER_4 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_h2_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqu      xmm1, [rsi + 2]
+
+    HIGH_APPLY_FILTER_8 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_h2_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0,   [rsi]               ;load src
+    movdqu      xmm1,   [rsi + 2]
+    movdqu      xmm2,   [rsi + 16]
+    movdqu      xmm3,   [rsi + 18]
+
+    HIGH_APPLY_FILTER_16 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
+
+global sym(vp9_high_filter_block1d4_h2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqa      xmm1, xmm0
+    psrldq      xmm1, 2
+
+    HIGH_APPLY_FILTER_4 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_h2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqu      xmm1, [rsi + 2]
+
+    HIGH_APPLY_FILTER_8 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_h2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0,   [rsi]               ;load src
+    movdqu      xmm1,   [rsi + 2]
+    movdqu      xmm2,   [rsi + 16]
+    movdqu      xmm3,   [rsi + 18]
+
+    HIGH_APPLY_FILTER_16 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index f99fa7a58..499fb4f71 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -1265,10 +1265,18 @@ static size_t read_uncompressed_header(VP9Decoder *pbi,
 
       for (i = 0; i < REFS_PER_FRAME; ++i) {
         RefBuffer *const ref_buf = &cm->frame_refs[i];
+#if CONFIG_VP9_HIGHBITDEPTH
+        vp9_setup_scale_factors_for_frame(&ref_buf->sf,
+                                          ref_buf->buf->y_crop_width,
+                                          ref_buf->buf->y_crop_height,
+                                          cm->width, cm->height,
+                                          cm->use_highbitdepth);
+#else
         vp9_setup_scale_factors_for_frame(&ref_buf->sf,
                                           ref_buf->buf->y_crop_width,
                                           ref_buf->buf->y_crop_height,
                                           cm->width, cm->height);
+#endif
         if (vp9_is_scaled(&ref_buf->sf))
           vp9_extend_frame_borders(ref_buf->buf);
       }
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 9545ba0f3..9a61b2a93 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2767,10 +2767,17 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
     RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - 1];
     ref_buf->buf = buf;
     ref_buf->idx = idx;
+#if CONFIG_VP9_HIGHBITDEPTH
+    vp9_setup_scale_factors_for_frame(&ref_buf->sf,
+                                      buf->y_crop_width, buf->y_crop_height,
+                                      cm->width, cm->height,
+                                      (buf->flags & YV12_FLAG_HIGHBITDEPTH) ?
+                                          1 : 0);
+#else
     vp9_setup_scale_factors_for_frame(&ref_buf->sf,
                                       buf->y_crop_width, buf->y_crop_height,
                                       cm->width, cm->height);
-
+#endif
     if (vp9_is_scaled(&ref_buf->sf))
       vp9_extend_frame_borders(buf);
   }
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index ff026666b..6a532b243 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -454,12 +454,20 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
     // In spatial svc the scaling factors might be less then 1/2. So we will use
     // non-normative scaling.
     int frame_used = 0;
+#if CONFIG_VP9_HIGHBITDEPTH
+    vp9_setup_scale_factors_for_frame(&sf,
+                                      get_frame_new_buffer(cm)->y_crop_width,
+                                      get_frame_new_buffer(cm)->y_crop_height,
+                                      get_frame_new_buffer(cm)->y_crop_width,
+                                      get_frame_new_buffer(cm)->y_crop_height,
+                                      cm->use_highbitdepth);
+#else
     vp9_setup_scale_factors_for_frame(&sf,
                                       get_frame_new_buffer(cm)->y_crop_width,
                                       get_frame_new_buffer(cm)->y_crop_height,
                                       get_frame_new_buffer(cm)->y_crop_width,
                                       get_frame_new_buffer(cm)->y_crop_height);
-
+#endif
     for (frame = 0; frame < frames_to_blur; ++frame) {
       if (cm->mi_cols * MI_SIZE != frames[frame]->y_width ||
           cm->mi_rows * MI_SIZE != frames[frame]->y_height) {
@@ -481,11 +489,20 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
     }
   } else {
     // ARF is produced at the native frame size and resized when coded.
+#if CONFIG_VP9_HIGHBITDEPTH
+    vp9_setup_scale_factors_for_frame(&sf,
+                                      frames[0]->y_crop_width,
+                                      frames[0]->y_crop_height,
+                                      frames[0]->y_crop_width,
+                                      frames[0]->y_crop_height,
+                                      cm->use_highbitdepth);
+#else
     vp9_setup_scale_factors_for_frame(&sf,
                                       frames[0]->y_crop_width,
                                       frames[0]->y_crop_height,
                                       frames[0]->y_crop_width,
                                       frames[0]->y_crop_height);
+#endif
   }
 
   temporal_filter_iterate_c(cpi, frames, frames_to_blur,
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index e88060c64..07a3be8db 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -91,6 +91,8 @@ endif
 
 ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_intrapred_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_8t_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_bilinear_sse2.asm
 endif
 
 # common (c)