libvpx: Pull from upstream

source/libvpx/vp9/common/vp9_reconinter.c

 /*
  *  Copyright (c) 2010 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.
  */
 
@@ skipping 126 matching lines @@
            xd->mb_to_bottom_edge * (1 << (1 - ss_y)) + spel_bottom);
 
   return clamped_mv;
 }
 
 static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
                                    int bw, int bh,
                                    int x, int y, int w, int h,
                                    int mi_x, int mi_y) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  const MODE_INFO *mi = xd->mi_8x8[0];
+  const MODE_INFO *mi = xd->mi[0];
   const int is_compound = has_second_ref(&mi->mbmi);
+  const InterpKernel *kernel = vp9_get_interp_kernel(mi->mbmi.interp_filter);
   int ref;
 
   for (ref = 0; ref < 1 + is_compound; ++ref) {
     const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
     struct buf_2d *const pre_buf = &pd->pre[ref];
     struct buf_2d *const dst_buf = &pd->dst;
     uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
 
     // TODO(jkoleszar): All chroma MVs in SPLITMV mode are taken as the
     // same MV (the average of the 4 luma MVs) but we could do something
@@ skipping 27 matching lines @@
       scaled_mv.row = mv_q4.row;
       scaled_mv.col = mv_q4.col;
       xs = ys = 16;
     }
     subpel_x = scaled_mv.col & SUBPEL_MASK;
     subpel_y = scaled_mv.row & SUBPEL_MASK;
     pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride
            + (scaled_mv.col >> SUBPEL_BITS);
 
     inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
-                    subpel_x, subpel_y, sf, w, h, ref, xd->interp_kernel,
-                    xs, ys);
+                    subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
   }
 }
 
 static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
                                               int mi_row, int mi_col,
                                               int plane_from, int plane_to) {
   int plane;
   const int mi_x = mi_col * MI_SIZE;
   const int mi_y = mi_row * MI_SIZE;
   for (plane = plane_from; plane <= plane_to; ++plane) {
     const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize,
                                                         &xd->plane[plane]);
     const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
     const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
     const int bw = 4 * num_4x4_w;
     const int bh = 4 * num_4x4_h;
 
-    if (xd->mi_8x8[0]->mbmi.sb_type < BLOCK_8X8) {
+    if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
       int i = 0, x, y;
       assert(bsize == BLOCK_8X8);
       for (y = 0; y < num_4x4_h; ++y)
         for (x = 0; x < num_4x4_w; ++x)
            build_inter_predictors(xd, plane, i++, bw, bh,
                                   4 * x, 4 * y, 4, 4, mi_x, mi_y);
     } else {
       build_inter_predictors(xd, plane, 0, bw, bh,
                              0, 0, bw, bh, mi_x, mi_y);
     }
@@ skipping 11 matching lines @@
 }
 void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
                                    BLOCK_SIZE bsize) {
   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0,
                                     MAX_MB_PLANE - 1);
 }
 
 // TODO(jingning): This function serves as a placeholder for decoder prediction
 // using on demand border extension. It should be moved to /decoder/ directory.
 static void dec_build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
+                                       int bw, int bh,
                                        int x, int y, int w, int h,
                                        int mi_x, int mi_y) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  const MODE_INFO *mi = xd->mi_8x8[0];
+  const MODE_INFO *mi = xd->mi[0];
   const int is_compound = has_second_ref(&mi->mbmi);
+  const InterpKernel *kernel = vp9_get_interp_kernel(mi->mbmi.interp_filter);
   int ref;
 
   for (ref = 0; ref < 1 + is_compound; ++ref) {
     const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
     struct buf_2d *const pre_buf = &pd->pre[ref];
     struct buf_2d *const dst_buf = &pd->dst;
     uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
 
     // TODO(jkoleszar): All chroma MVs in SPLITMV mode are taken as the
     // same MV (the average of the 4 luma MVs) but we could do something
     // smarter for non-4:2:0. Just punt for now, pending the changes to get
     // rid of SPLITMV mode entirely.
     const MV mv = mi->mbmi.sb_type < BLOCK_8X8
                ? (plane == 0 ? mi->bmi[block].as_mv[ref].as_mv
                              : mi_mv_pred_q4(mi, ref))
                : mi->mbmi.mv[ref].as_mv;
+
+    // TODO(jkoleszar): This clamping is done in the incorrect place for the
+    // scaling case. It needs to be done on the scaled MV, not the pre-scaling
+    // MV. Note however that it performs the subsampling aware scaling so
+    // that the result is always q4.
+    // mv_precision precision is MV_PRECISION_Q4.
+    const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh,
+                                               pd->subsampling_x,
+                                               pd->subsampling_y);
+
     MV32 scaled_mv;
     int xs, ys, x0, y0, x0_16, y0_16, frame_width, frame_height, buf_stride,
         subpel_x, subpel_y;
     uint8_t *ref_frame, *buf_ptr;
     const YV12_BUFFER_CONFIG *ref_buf = xd->block_refs[ref]->buf;
-    const MV mv_q4 = {
-      mv.row * (1 << (1 - pd->subsampling_y)),
-      mv.col * (1 << (1 - pd->subsampling_x))
-    };
 
     // Get reference frame pointer, width and height.
     if (plane == 0) {
       frame_width = ref_buf->y_crop_width;
       frame_height = ref_buf->y_crop_height;
       ref_frame = ref_buf->y_buffer;
     } else {
       frame_width = ref_buf->uv_crop_width;
       frame_height = ref_buf->uv_crop_height;
       ref_frame = plane == 1 ? ref_buf->u_buffer : ref_buf->v_buffer;
     }
 
-    // Get block position in current frame.
-    x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
-    y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
+    if (vp9_is_scaled(sf)) {
+      // Co-ordinate of containing block to pixel precision.
+      int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x));
+      int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y));
 
-    // Precision of x0_16 and y0_16 is 1/16th pixel.
-    x0_16 = x0 << SUBPEL_BITS;
-    y0_16 = y0 << SUBPEL_BITS;
+      // Co-ordinate of the block to 1/16th pixel precision.
+      x0_16 = (x_start + x) << SUBPEL_BITS;
+      y0_16 = (y_start + y) << SUBPEL_BITS;
 
-    if (vp9_is_scaled(sf)) {
+      // Co-ordinate of current block in reference frame
+      // to 1/16th pixel precision.
+      x0_16 = sf->scale_value_x(x0_16, sf);
+      y0_16 = sf->scale_value_y(y0_16, sf);
+
+      // Map the top left corner of the block into the reference frame.
+      x0 = sf->scale_value_x(x_start + x, sf);
+      y0 = sf->scale_value_y(y_start + y, sf);
+
+      // Scale the MV and incorporate the sub-pixel offset of the block
+      // in the reference frame.
       scaled_mv = vp9_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
       xs = sf->x_step_q4;
       ys = sf->y_step_q4;
-      // Map the top left corner of the block into the reference frame.
-      x0 = sf->scale_value_x(x0, sf);
-      y0 = sf->scale_value_y(y0, sf);
-      x0_16 = sf->scale_value_x(x0_16, sf);
-      y0_16 = sf->scale_value_y(y0_16, sf);
     } else {
+      // Co-ordinate of containing block to pixel precision.
+      x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
+      y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
+
+      // Co-ordinate of the block to 1/16th pixel precision.
+      x0_16 = x0 << SUBPEL_BITS;
+      y0_16 = y0 << SUBPEL_BITS;
+
       scaled_mv.row = mv_q4.row;
       scaled_mv.col = mv_q4.col;
       xs = ys = 16;
     }
     subpel_x = scaled_mv.col & SUBPEL_MASK;
     subpel_y = scaled_mv.row & SUBPEL_MASK;
 
     // Calculate the top left corner of the best matching block in the reference frame.
     x0 += scaled_mv.col >> SUBPEL_BITS;
     y0 += scaled_mv.row >> SUBPEL_BITS;
@@ skipping 23 matching lines @@
         y0 -= VP9_INTERP_EXTEND - 1;
         y1 += VP9_INTERP_EXTEND;
         y_pad = 1;
       }
 
       // Skip border extension if block is inside the frame.
       if (x0 < 0 || x0 > frame_width - 1 || x1 < 0 || x1 > frame_width ||
           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,
-                        x0, y0, x1 - x0, y1 - y0, frame_width, frame_height);
-        buf_stride = x1 - x0;
+        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;
       }
     }
 
     inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
-                    subpel_y, sf, w, h, ref, xd->interp_kernel, xs, ys);
+                    subpel_y, sf, w, h, ref, kernel, xs, ys);
   }
 }
 
 void vp9_dec_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
                                        BLOCK_SIZE bsize) {
   int plane;
   const int mi_x = mi_col * MI_SIZE;
   const int mi_y = mi_row * MI_SIZE;
   for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
     const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize,
                                                         &xd->plane[plane]);
     const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
     const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
     const int bw = 4 * num_4x4_w;
     const int bh = 4 * num_4x4_h;
 
-    if (xd->mi_8x8[0]->mbmi.sb_type < BLOCK_8X8) {
+    if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
       int i = 0, x, y;
       assert(bsize == BLOCK_8X8);
       for (y = 0; y < num_4x4_h; ++y)
         for (x = 0; x < num_4x4_w; ++x)
-          dec_build_inter_predictors(xd, plane, i++,
+          dec_build_inter_predictors(xd, plane, i++, bw, bh,
                                      4 * x, 4 * y, 4, 4, mi_x, mi_y);
     } else {
-      dec_build_inter_predictors(xd, plane, 0,
+      dec_build_inter_predictors(xd, plane, 0, bw, bh,
                                  0, 0, bw, bh, mi_x, mi_y);
     }
   }
 }
+
+void vp9_setup_dst_planes(MACROBLOCKD *xd,
+                          const YV12_BUFFER_CONFIG *src,
+                          int mi_row, int mi_col) {
+  uint8_t *const buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+                               src->alpha_buffer};
+  const int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+                          src->alpha_stride};
+  int i;
+
+  for (i = 0; i < MAX_MB_PLANE; ++i) {
+    struct macroblockd_plane *const pd = &xd->plane[i];
+    setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL,
+                     pd->subsampling_x, pd->subsampling_y);
+  }
+}
+
+void vp9_setup_pre_planes(MACROBLOCKD *xd, int idx,
+                          const YV12_BUFFER_CONFIG *src,
+                          int mi_row, int mi_col,
+                          const struct scale_factors *sf) {
+  if (src != NULL) {
+    int i;
+    uint8_t *const buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+                                 src->alpha_buffer};
+    const int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+                            src->alpha_stride};
+
+    for (i = 0; i < MAX_MB_PLANE; ++i) {
+      struct macroblockd_plane *const pd = &xd->plane[i];
+      setup_pred_plane(&pd->pre[idx], buffers[i], strides[i], mi_row, mi_col,
+                       sf, pd->subsampling_x, pd->subsampling_y);
+    }
+  }
+}