Sync high bit-depth temporal filter

Change-Id: Ifdcfb91416be8189569f703bee9be253d7b3d9b6

diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index 7e4c61f..015dbc0 100644 (file)
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -205,15 +205,34 @@ void vp9_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
 
   for (i = 0, k = 0; i < block_height; i++) {
     for (j = 0; j < block_width; j++, k++) {
-      int src_byte = frame1[byte];
-      int pixel_value = *frame2++;
-
-      modifier   = src_byte - pixel_value;
-      // This is an integer approximation of:
-      // float coeff = (3.0 * modifer * modifier) / pow(2, strength);
-      // modifier =  (int)roundf(coeff > 16 ? 0 : 16-coeff);
-      modifier *= modifier;
+      int pixel_value = *frame2;
+      int diff_sse[9] = { 0 };
+      int idx, idy, index = 0;
+
+      for (idy = -1; idy <= 1; ++idy) {
+        for (idx = -1; idx <= 1; ++idx) {
+          int row = i + idy;
+          int col = j + idx;
+
+          if (row >= 0 && row < (int)block_height &&
+              col >= 0 && col < (int)block_width) {
+            int diff = frame1[byte + idy * (int)stride + idx] -
+                frame2[idy * (int)block_width + idx];
+            diff_sse[index] = diff * diff;
+            ++index;
+          }
+        }
+      }
+      assert(index > 0);
+
+      modifier = 0;
+      for (idx = 0; idx < 9; ++idx)
+        modifier += diff_sse[idx];
+
       modifier *= 3;
+      modifier /= index;
+
+      ++frame2;
       modifier += rounding;
       modifier >>= strength;
 
@@ -406,56 +425,58 @@ static void temporal_filter_iterate_c(VP9_COMP *cpi,
           if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
             int adj_strength = strength + 2 * (mbd->bd - 8);
             // Apply the filter (YUV)
-            vp9_highbd_temporal_filter_apply(f->y_buffer + mb_y_offset,
-                                             f->y_stride,
-                                             predictor, 16, 16, adj_strength,
-                                             filter_weight,
-                                             accumulator, count);
-            vp9_highbd_temporal_filter_apply(f->u_buffer + mb_uv_offset,
-                                             f->uv_stride, predictor + 256,
-                                             mb_uv_width, mb_uv_height,
-                                             adj_strength,
-                                             filter_weight, accumulator + 256,
-                                             count + 256);
-            vp9_highbd_temporal_filter_apply(f->v_buffer + mb_uv_offset,
-                                             f->uv_stride, predictor + 512,
-                                             mb_uv_width, mb_uv_height,
-                                             adj_strength, filter_weight,
-                                             accumulator + 512, count + 512);
+            vp9_highbd_temporal_filter_apply_c(f->y_buffer + mb_y_offset,
+                                               f->y_stride,
+                                               predictor, 16, 16, adj_strength,
+                                               filter_weight,
+                                               accumulator, count);
+            vp9_highbd_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
+                                               f->uv_stride, predictor + 256,
+                                               mb_uv_width, mb_uv_height,
+                                               adj_strength,
+                                               filter_weight, accumulator + 256,
+                                               count + 256);
+            vp9_highbd_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
+                                               f->uv_stride, predictor + 512,
+                                               mb_uv_width, mb_uv_height,
+                                               adj_strength, filter_weight,
+                                               accumulator + 512, count + 512);
           } else {
             // Apply the filter (YUV)
-            vp9_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
+            vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
+                                        predictor, 16, 16,
+                                        strength, filter_weight,
+                                        accumulator, count);
+            vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
+                                        f->uv_stride,
+                                        predictor + 256,
+                                        mb_uv_width, mb_uv_height, strength,
+                                        filter_weight, accumulator + 256,
+                                        count + 256);
+            vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
+                                        f->uv_stride,
+                                        predictor + 512,
+                                        mb_uv_width, mb_uv_height, strength,
+                                        filter_weight, accumulator + 512,
+                                        count + 512);
+          }
+#else
+          // Apply the filter (YUV)
+          // TODO(jingning): Need SIMD optimization for this.
+          vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
                                       predictor, 16, 16,
                                       strength, filter_weight,
                                       accumulator, count);
-            vp9_temporal_filter_apply(f->u_buffer + mb_uv_offset, f->uv_stride,
+          vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset, f->uv_stride,
                                       predictor + 256,
                                       mb_uv_width, mb_uv_height, strength,
                                       filter_weight, accumulator + 256,
                                       count + 256);
-            vp9_temporal_filter_apply(f->v_buffer + mb_uv_offset, f->uv_stride,
+          vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset, f->uv_stride,
                                       predictor + 512,
                                       mb_uv_width, mb_uv_height, strength,
                                       filter_weight, accumulator + 512,
                                       count + 512);
-          }
-#else
-          // Apply the filter (YUV)
-          // TODO(jingning): Need SIMD optimization for this.
-          vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
-                                    predictor, 16, 16,
-                                    strength, filter_weight,
-                                    accumulator, count);
-          vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset, f->uv_stride,
-                                    predictor + 256,
-                                    mb_uv_width, mb_uv_height, strength,
-                                    filter_weight, accumulator + 256,
-                                    count + 256);
-          vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset, f->uv_stride,
-                                    predictor + 512,
-                                    mb_uv_width, mb_uv_height, strength,
-                                    filter_weight, accumulator + 512,
-                                    count + 512);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
         }
       }