Adds high bit-depth frame resize functions

Change-Id: I35b015a759325d72d0da427c61a09f19f8e69697

vp9/encoder/vp9_encoder.c

@@ -2167,8 +2167,14 @@ void vp9_write_yuv_rec_frame(VP9_COMMON *cm) {
 }
 #endif
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
+                                                YV12_BUFFER_CONFIG *dst,
+                                                int bd) {
+#else
 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
                                                 YV12_BUFFER_CONFIG *dst) {
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   // TODO(dkovalev): replace YV12_BUFFER_CONFIG with vpx_image_t
   int i;
   const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
@@ -2184,15 +2190,31 @@ static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
   const int dst_heights[3] = {dst->y_crop_height, dst->uv_crop_height,
                               dst->uv_crop_height};
 
-  for (i = 0; i < MAX_MB_PLANE; ++i)
+  for (i = 0; i < MAX_MB_PLANE; ++i) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+      vp9_highbd_resize_plane(srcs[i], src_heights[i], src_widths[i],
+                              src_strides[i], dsts[i], dst_heights[i],
+                              dst_widths[i], dst_strides[i], bd);
+    } else {
+      vp9_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
+                       dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
+    }
+#else
     vp9_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
                      dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
-
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  }
   vp9_extend_frame_borders(dst);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
+                                   YV12_BUFFER_CONFIG *dst, int bd) {
+#else
 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
                                    YV12_BUFFER_CONFIG *dst) {
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   const int src_w = src->y_crop_width;
   const int src_h = src->y_crop_height;
   const int dst_w = dst->y_crop_width;
@@ -2216,10 +2238,24 @@ static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
                                      src_stride + (x / factor) * src_w / dst_w;
         uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+        if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+          vp9_high_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
+                             kernel[x_q4 & 0xf], 16 * src_w / dst_w,
+                             kernel[y_q4 & 0xf], 16 * src_h / dst_h,
+                             16 / factor, 16 / factor, bd);
+        } else {
+          vp9_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
+                        kernel[x_q4 & 0xf], 16 * src_w / dst_w,
+                        kernel[y_q4 & 0xf], 16 * src_h / dst_h,
+                        16 / factor, 16 / factor);
+        }
+#else
         vp9_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
                       kernel[x_q4 & 0xf], 16 * src_w / dst_w,
                       kernel[y_q4 & 0xf], 16 * src_h / dst_h,
                       16 / factor, 16 / factor);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       }
     }
   }
@@ -2388,9 +2424,14 @@ void vp9_scale_references(VP9_COMP *cpi) {
                                cm->subsampling_x, cm->subsampling_y,
 #if CONFIG_VP9_HIGHBITDEPTH
                                cm->use_highbitdepth,
-#endif
+#endif  // CONFIG_VP9_HIGHBITDEPTH
                                VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL);
+#if CONFIG_VP9_HIGHBITDEPTH
+      scale_and_extend_frame(ref, &cm->frame_bufs[new_fb].buf,
+                             (int)cm->bit_depth);
+#else
       scale_and_extend_frame(ref, &cm->frame_bufs[new_fb].buf);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
     } else {
       cpi->scaled_ref_idx[ref_frame - 1] = idx;
@@ -2766,7 +2807,11 @@ YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
                                           YV12_BUFFER_CONFIG *scaled) {
   if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
       cm->mi_rows * MI_SIZE != unscaled->y_height) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
+#else
     scale_and_extend_frame_nonnormative(unscaled, scaled);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
     return scaled;
   } else {
     return unscaled;

vp9/encoder/vp9_resize.c

@@ -312,7 +312,7 @@ static void interpolate(const uint8_t *const input, int inlength,
 static void down2_symeven(const uint8_t *const input, int length,
                           uint8_t *output) {
   // Actual filter len = 2 * filter_len_half.
-  static const int16_t *filter = vp9_down2_symeven_half_filter;
+  const int16_t *filter = vp9_down2_symeven_half_filter;
   const int filter_len_half = sizeof(vp9_down2_symeven_half_filter) / 2;
   int i, j;
   uint8_t *optr = output;
@@ -368,7 +368,7 @@ static void down2_symeven(const uint8_t *const input, int length,
 static void down2_symodd(const uint8_t *const input, int length,
                          uint8_t *output) {
   // Actual filter len = 2 * filter_len_half - 1.
-  static const int16_t *filter = vp9_down2_symodd_half_filter;
+  const int16_t *filter = vp9_down2_symodd_half_filter;
   const int filter_len_half = sizeof(vp9_down2_symodd_half_filter) / 2;
   int i, j;
   uint8_t *optr = output;
@@ -529,6 +529,302 @@ void vp9_resize_plane(const uint8_t *const input,
   free(arrbuf);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void highbd_interpolate(const uint16_t *const input, int inlength,
+                               uint16_t *output, int outlength, int bd) {
+  const int64_t delta =
+      (((uint64_t)inlength << 32) + outlength / 2) / outlength;
+  const int64_t offset = inlength > outlength ?
+      (((int64_t)(inlength - outlength) << 31) + outlength / 2) / outlength :
+      -(((int64_t)(outlength - inlength) << 31) + outlength / 2) / outlength;
+  uint16_t *optr = output;
+  int x, x1, x2, sum, k, int_pel, sub_pel;
+  int64_t y;
+
+  const interp_kernel *interp_filters =
+      choose_interp_filter(inlength, outlength);
+
+  x = 0;
+  y = offset;
+  while ((y >> INTERP_PRECISION_BITS) < (INTERP_TAPS / 2 - 1)) {
+    x++;
+    y += delta;
+  }
+  x1 = x;
+  x = outlength - 1;
+  y = delta * x + offset;
+  while ((y >> INTERP_PRECISION_BITS) +
+         (int64_t)(INTERP_TAPS / 2) >= inlength) {
+    x--;
+    y -= delta;
+  }
+  x2 = x;
+  if (x1 > x2) {
+    for (x = 0, y = offset; x < outlength; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k) {
+        const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
+        sum += filter[k] *
+            input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
+      }
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+  } else {
+    // Initial part.
+    for (x = 0, y = offset; x < x1; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k)
+        sum += filter[k] *
+            input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0 ?
+                   0 : int_pel - INTERP_TAPS / 2 + 1 + k)];
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+    // Middle part.
+    for (; x <= x2; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k)
+        sum += filter[k] * input[int_pel - INTERP_TAPS / 2 + 1 + k];
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+    // End part.
+    for (; x < outlength; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k)
+        sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >=
+                                  inlength ?  inlength - 1 :
+                                  int_pel - INTERP_TAPS / 2 + 1 + k)];
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+  }
+}
+
+static void highbd_down2_symeven(const uint16_t *const input, int length,
+                                 uint16_t *output, int bd) {
+  // Actual filter len = 2 * filter_len_half.
+  static const int16_t *filter = vp9_down2_symeven_half_filter;
+  const int filter_len_half = sizeof(vp9_down2_symeven_half_filter) / 2;
+  int i, j;
+  uint16_t *optr = output;
+  int l1 = filter_len_half;
+  int l2 = (length - filter_len_half);
+  l1 += (l1 & 1);
+  l2 += (l2 & 1);
+  if (l1 > l2) {
+    // Short input length.
+    for (i = 0; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] +
+                input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  } else {
+    // Initial part.
+    for (i = 0; i < l1; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // Middle part.
+    for (; i < l2; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[i - j] + input[i + 1 + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // End part.
+    for (; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[i - j] +
+                input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  }
+}
+
+static void highbd_down2_symodd(const uint16_t *const input, int length,
+                              uint16_t *output, int bd) {
+  // Actual filter len = 2 * filter_len_half - 1.
+  static const int16_t *filter = vp9_down2_symodd_half_filter;
+  const int filter_len_half = sizeof(vp9_down2_symodd_half_filter) / 2;
+  int i, j;
+  uint16_t *optr = output;
+  int l1 = filter_len_half - 1;
+  int l2 = (length - filter_len_half + 1);
+  l1 += (l1 & 1);
+  l2 += (l2 & 1);
+  if (l1 > l2) {
+    // Short input length.
+    for (i = 0; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] +
+                input[(i + j >= length ? length - 1 : i + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  } else {
+    // Initial part.
+    for (i = 0; i < l1; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // Middle part.
+    for (; i < l2; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[i - j] + input[i + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // End part.
+    for (; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  }
+}
+
+static void highbd_resize_multistep(const uint16_t *const input,
+                                    int length,
+                                    uint16_t *output,
+                                    int olength,
+                                    uint16_t *buf,
+                                    int bd) {
+  int steps;
+  if (length == olength) {
+    memcpy(output, input, sizeof(uint16_t) * length);
+    return;
+  }
+  steps = get_down2_steps(length, olength);
+
+  if (steps > 0) {
+    int s;
+    uint16_t *out = NULL;
+    uint16_t *tmpbuf = NULL;
+    uint16_t *otmp, *otmp2;
+    int filteredlength = length;
+    if (!tmpbuf) {
+      tmpbuf = (uint16_t *)malloc(sizeof(uint16_t) * length);
+      otmp = tmpbuf;
+    } else {
+      otmp = buf;
+    }
+    otmp2 = otmp + get_down2_length(length, 1);
+    for (s = 0; s < steps; ++s) {
+      const int proj_filteredlength = get_down2_length(filteredlength, 1);
+      const uint16_t *const in = (s == 0 ? input : out);
+      if (s == steps - 1 && proj_filteredlength == olength)
+        out = output;
+      else
+        out = (s & 1 ? otmp2 : otmp);
+      if (filteredlength & 1)
+        highbd_down2_symodd(in, filteredlength, out, bd);
+      else
+        highbd_down2_symeven(in, filteredlength, out, bd);
+      filteredlength = proj_filteredlength;
+    }
+    if (filteredlength != olength) {
+      highbd_interpolate(out, filteredlength, output, olength, bd);
+    }
+    if (tmpbuf)
+      free(tmpbuf);
+  } else {
+    highbd_interpolate(input, length, output, olength, bd);
+  }
+}
+
+static void highbd_fill_col_to_arr(uint16_t *img, int stride, int len,
+                                   uint16_t *arr) {
+  int i;
+  uint16_t *iptr = img;
+  uint16_t *aptr = arr;
+  for (i = 0; i < len; ++i, iptr += stride) {
+    *aptr++ = *iptr;
+  }
+}
+
+static void highbd_fill_arr_to_col(uint16_t *img, int stride, int len,
+                                   uint16_t *arr) {
+  int i;
+  uint16_t *iptr = img;
+  uint16_t *aptr = arr;
+  for (i = 0; i < len; ++i, iptr += stride) {
+    *iptr = *aptr++;
+  }
+}
+
+void vp9_highbd_resize_plane(const uint8_t *const input,
+                             int height,
+                             int width,
+                             int in_stride,
+                             uint8_t *output,
+                             int height2,
+                             int width2,
+                             int out_stride,
+                             int bd) {
+  int i;
+  uint16_t *intbuf = (uint16_t *)malloc(sizeof(uint16_t) * width2 * height);
+  uint16_t *tmpbuf = (uint16_t *)malloc(sizeof(uint16_t) *
+                                        (width < height ? height : width));
+  uint16_t *arrbuf = (uint16_t *)malloc(sizeof(uint16_t) * (height + height2));
+  for (i = 0; i < height; ++i) {
+    highbd_resize_multistep(CONVERT_TO_SHORTPTR(input + in_stride * i), width,
+                            intbuf + width2 * i, width2, tmpbuf, bd);
+  }
+  for (i = 0; i < width2; ++i) {
+    highbd_fill_col_to_arr(intbuf + i, width2, height, arrbuf);
+    highbd_resize_multistep(arrbuf, height, arrbuf + height, height2, tmpbuf,
+                            bd);
+    highbd_fill_arr_to_col(CONVERT_TO_SHORTPTR(output + i), out_stride, height2,
+                           arrbuf + height);
+  }
+  free(intbuf);
+  free(tmpbuf);
+  free(arrbuf);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 void vp9_resize_frame420(const uint8_t *const y,
                          int y_stride,
                          const uint8_t *const u, const uint8_t *const v,
@@ -574,3 +870,51 @@ void vp9_resize_frame444(const uint8_t *const y, int y_stride,
   vp9_resize_plane(v, height, width, uv_stride,
                    ov, oheight, owidth, ouv_stride);
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_resize_frame420(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u, const uint8_t *const v,
+                                int uv_stride,
+                                int height, int width,
+                                uint8_t *oy, int oy_stride,
+                                uint8_t *ou, uint8_t *ov, int ouv_stride,
+                                int oheight, int owidth, int bd) {
+  vp9_highbd_resize_plane(y, height, width, y_stride,
+                          oy, oheight, owidth, oy_stride, bd);
+  vp9_highbd_resize_plane(u, height / 2, width / 2, uv_stride,
+                          ou, oheight / 2, owidth / 2, ouv_stride, bd);
+  vp9_highbd_resize_plane(v, height / 2, width / 2, uv_stride,
+                          ov, oheight / 2, owidth / 2, ouv_stride, bd);
+}
+
+void vp9_highbd_resize_frame422(const uint8_t *const y, int y_stride,
+                                const uint8_t *const u, const uint8_t *const v,
+                                int uv_stride,
+                                int height, int width,
+                                uint8_t *oy, int oy_stride,
+                                uint8_t *ou, uint8_t *ov, int ouv_stride,
+                                int oheight, int owidth, int bd) {
+  vp9_highbd_resize_plane(y, height, width, y_stride,
+                          oy, oheight, owidth, oy_stride, bd);
+  vp9_highbd_resize_plane(u, height, width / 2, uv_stride,
+                          ou, oheight, owidth / 2, ouv_stride, bd);
+  vp9_highbd_resize_plane(v, height, width / 2, uv_stride,
+                          ov, oheight, owidth / 2, ouv_stride, bd);
+}
+
+void vp9_highbd_resize_frame444(const uint8_t *const y, int y_stride,
+                                const uint8_t *const u, const uint8_t *const v,
+                                int uv_stride,
+                                int height, int width,
+                                uint8_t *oy, int oy_stride,
+                                uint8_t *ou, uint8_t *ov, int ouv_stride,
+                                int oheight, int owidth, int bd) {
+  vp9_highbd_resize_plane(y, height, width, y_stride,
+                          oy, oheight, owidth, oy_stride, bd);
+  vp9_highbd_resize_plane(u, height, width, uv_stride,
+                          ou, oheight, owidth, ouv_stride, bd);
+  vp9_highbd_resize_plane(v, height, width, uv_stride,
+                          ov, oheight, owidth, ouv_stride, bd);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH

vp9/encoder/vp9_resize.h

@@ -65,4 +65,60 @@ void vp9_resize_frame444(const uint8_t *const y,
                          int oheight,
                          int owidth);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_resize_plane(const uint8_t *const input,
+                             int height,
+                             int width,
+                             int in_stride,
+                             uint8_t *output,
+                             int height2,
+                             int width2,
+                             int out_stride,
+                             int bd);
+void vp9_highbd_resize_frame420(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u,
+                                const uint8_t *const v,
+                                int uv_stride,
+                                int height,
+                                int width,
+                                uint8_t *oy,
+                                int oy_stride,
+                                uint8_t *ou,
+                                uint8_t *ov,
+                                int ouv_stride,
+                                int oheight,
+                                int owidth,
+                                int bd);
+void vp9_highbd_resize_frame422(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u,
+                                const uint8_t *const v,
+                                int uv_stride,
+                                int height,
+                                int width,
+                                uint8_t *oy,
+                                int oy_stride,
+                                uint8_t *ou,
+                                uint8_t *ov,
+                                int ouv_stride,
+                                int oheight,
+                                int owidth,
+                                int bd);
+void vp9_highbd_resize_frame444(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u,
+                                const uint8_t *const v,
+                                int uv_stride,
+                                int height,
+                                int width,
+                                uint8_t *oy,
+                                int oy_stride,
+                                uint8_t *ou,
+                                uint8_t *ov,
+                                int ouv_stride,
+                                int oheight,
+                                int owidth,
+                                int bd);
+#endif    // CONFIG_VP9_HIGHBITDEPTH
 #endif    // VP9_ENCODER_VP9_RESIZE_H_