From 0f513752a03b87af3216275ea7bc7a4212ba4213 Mon Sep 17 00:00:00 2001
From: Jingning Han <jingning@google.com>
Date: Mon, 16 May 2016 18:27:20 -0700
Subject: [PATCH] Rework sub8x8 chroma component inter predictor

This commit makes the sub8x8 chroma component inter predictor
operate at 2x2 block level. This allows one to use the actual motion
vector associated with each individal pixel block. It improves the
compression performance

lowres  0.40%
midres  0.25%
hdres   0.15%

Change-Id: Ia40e07cc7fde463dbf660018850e024932136c4f
---
 vp10/common/reconinter.c | 68 ++++++++++++++++++++++++++++++++++++++++
 vp10/common/reconinter.h |  3 +-
 2 files changed, 70 insertions(+), 1 deletion(-)

diff --git a/vp10/common/reconinter.c b/vp10/common/reconinter.c
index d6ac4bb8d..ca2537bec 100644
--- a/vp10/common/reconinter.c
+++ b/vp10/common/reconinter.c
@@ -759,6 +759,74 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
   const int is_compound = has_second_ref(&mi->mbmi);
   int ref;
 
+#if CONFIG_DUAL_FILTER
+  if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0) {
+    int blk_num = 1 << (pd->subsampling_x + pd->subsampling_y);
+    int chr_idx;
+    int x_base = x;
+    int y_base = y;
+    int x_step = w >> pd->subsampling_x;
+    int y_step = h >> pd->subsampling_y;
+
+    for (chr_idx = 0; chr_idx < blk_num; ++chr_idx) {
+      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 *dst = dst_buf->buf;
+        const MV mv = mi->bmi[chr_idx].as_mv[ref].as_mv;
+        const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh,
+                                                   pd->subsampling_x,
+                                                   pd->subsampling_y);
+        uint8_t *pre;
+        MV32 scaled_mv;
+        int xs, ys, subpel_x, subpel_y;
+        const int is_scaled = vp10_is_scaled(sf);
+
+        x = x_base + (chr_idx & 0x01) * x_step;
+        y = y_base + (chr_idx >> 1) * y_step;
+
+        dst += dst_buf->stride * y + x;
+
+        if (is_scaled) {
+          pre = pre_buf->buf + scaled_buffer_offset(x, y, pre_buf->stride, sf);
+          scaled_mv = vp10_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
+          xs = sf->x_step_q4;
+          ys = sf->y_step_q4;
+        } else {
+          pre = pre_buf->buf + y * pre_buf->stride + x;
+          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);
+
+    #if CONFIG_EXT_INTER
+        if (ref && is_interinter_wedge_used(mi->mbmi.sb_type) &&
+            mi->mbmi.use_wedge_interinter)
+          vp10_make_masked_inter_predictor(
+              pre, pre_buf->stride, dst, dst_buf->stride,
+              subpel_x, subpel_y, sf, w, h,
+              mi->mbmi.interp_filter, xs, ys,
+    #if CONFIG_SUPERTX
+              wedge_offset_x, wedge_offset_y,
+    #endif  // CONFIG_SUPERTX
+              xd);
+        else
+    #endif  // CONFIG_EXT_INTER
+          vp10_make_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
+                                    subpel_x, subpel_y, sf, x_step, y_step, ref,
+                                    mi->mbmi.interp_filter, xs, ys, xd);
+      }
+    }
+    return;
+  }
+#endif
+
   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];
diff --git a/vp10/common/reconinter.h b/vp10/common/reconinter.h
index c4a09783e..65b97f8bb 100644
--- a/vp10/common/reconinter.h
+++ b/vp10/common/reconinter.h
@@ -44,7 +44,8 @@ static INLINE void inter_predictor(const uint8_t *src, int src_stride,
 
 #if CONFIG_DUAL_FILTER
   if (interp_filter_params_x.taps == SUBPEL_TAPS &&
-      interp_filter_params_y.taps == SUBPEL_TAPS) {
+      interp_filter_params_y.taps == SUBPEL_TAPS &&
+      w > 2 && h > 2) {
     const int16_t *kernel_x =
         vp10_get_interp_filter_subpel_kernel(interp_filter_params_x, subpel_x);
     const int16_t *kernel_y =
-- 
GitLab