5 files changed, 33 insertions, 9 deletions

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index dfb86d010..c22bf85e3 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -4129,6 +4129,10 @@ static void encode_nonrd_sb_row(VP9_COMP *cpi, ThreadData *td,
     (*(cpi->row_mt_sync_read_ptr))(&tile_data->row_mt_sync, sb_row,
                                    sb_col_in_tile);
 
+    if (cpi->use_skin_detection) {
+      vp9_compute_skin_sb(cpi, BLOCK_16X16, mi_row, mi_col);
+    }
+
     x->source_variance = UINT_MAX;
     vp9_zero(x->pred_mv);
     vp9_rd_cost_init(&dummy_rdc);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index aac4b51b3..badf791e8 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -3557,7 +3557,6 @@ static void encode_without_recode_loop(VP9_COMP *cpi, size_t *size,
       cpi->oxcf.content != VP9E_CONTENT_SCREEN &&
       cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
     cpi->use_skin_detection = 1;
-    vp9_compute_skin_map(cpi, BLOCK_16X16);
   }
 
   vp9_set_quantizer(cm, q);
diff --git a/vp9/encoder/vp9_noise_estimate.c b/vp9/encoder/vp9_noise_estimate.c
index 57f4fc078..1d77767dc 100644
--- a/vp9/encoder/vp9_noise_estimate.c
+++ b/vp9/encoder/vp9_noise_estimate.c
@@ -172,6 +172,7 @@ void vp9_update_noise_estimate(VP9_COMP *const cpi) {
     int mi_row, mi_col;
     int num_low_motion = 0;
     int frame_low_motion = 1;
+    if (cpi->use_skin_detection) vp9_compute_skin_map(cpi, BLOCK_16X16);
     for (mi_row = 0; mi_row < cm->mi_rows; mi_row++) {
       for (mi_col = 0; mi_col < cm->mi_cols; mi_col++) {
         int bl_index = mi_row * cm->mi_cols + mi_col;
diff --git a/vp9/encoder/vp9_skin_detection.c b/vp9/encoder/vp9_skin_detection.c
index 6b36a000a..c1f0293b9 100644
--- a/vp9/encoder/vp9_skin_detection.c
+++ b/vp9/encoder/vp9_skin_detection.c
@@ -37,8 +37,9 @@ int vp9_compute_skin_block(const uint8_t *y, const uint8_t *u, const uint8_t *v,
   }
 }
 
-void vp9_compute_skin_map(VP9_COMP *const cpi, BLOCK_SIZE bsize) {
-  int mi_row, mi_col, num_bl;
+void vp9_compute_skin_sb(VP9_COMP *const cpi, BLOCK_SIZE bsize, int mi_row,
+                         int mi_col) {
+  int i, j, num_bl;
   VP9_COMMON *const cm = &cpi->common;
   const uint8_t *src_y = cpi->Source->y_buffer;
   const uint8_t *src_u = cpi->Source->u_buffer;
@@ -50,13 +51,17 @@ void vp9_compute_skin_map(VP9_COMP *const cpi, BLOCK_SIZE bsize) {
   const int shy = (y_bsize == 8) ? 3 : 4;
   const int shuv = shy - 1;
   const int fac = y_bsize / 8;
-  // Loop through blocks and set skin map based on center pixel of block.
-  // Ignore rightmost/bottom boundary blocks.
-  for (mi_row = 0; mi_row < cm->mi_rows - 1; mi_row += fac) {
+  const int y_shift = src_ystride * (mi_row << 3) + (mi_col << 3);
+  const int uv_shift = src_uvstride * (mi_row << 2) + (mi_col << 2);
+  src_y += y_shift;
+  src_u += uv_shift;
+  src_v += uv_shift;
+
+  for (i = mi_row; i < VPXMIN(mi_row + 7, cm->mi_rows - 1); i += fac) {
     num_bl = 0;
-    for (mi_col = 0; mi_col < cm->mi_cols - 1; mi_col += fac) {
+    for (j = mi_col; j < VPXMIN(mi_col + 7, cm->mi_cols - 1); j += fac) {
       int consec_zeromv = 0;
-      int bl_index = mi_row * cm->mi_cols + mi_col;
+      int bl_index = i * cm->mi_cols + j;
       int bl_index1 = bl_index + 1;
       int bl_index2 = bl_index + cm->mi_cols;
       int bl_index3 = bl_index2 + 1;
@@ -67,7 +72,7 @@ void vp9_compute_skin_map(VP9_COMP *const cpi, BLOCK_SIZE bsize) {
                                VPXMIN(cpi->consec_zero_mv[bl_index1],
                                       VPXMIN(cpi->consec_zero_mv[bl_index2],
                                              cpi->consec_zero_mv[bl_index3])));
-      cpi->skin_map[mi_row * cm->mi_cols + mi_col] =
+      cpi->skin_map[bl_index] =
           vp9_compute_skin_block(src_y, src_u, src_v, src_ystride, src_uvstride,
                                  bsize, consec_zeromv, 0);
       num_bl++;
@@ -81,6 +86,18 @@ void vp9_compute_skin_map(VP9_COMP *const cpi, BLOCK_SIZE bsize) {
   }
 }
 
+void vp9_compute_skin_map(VP9_COMP *const cpi, BLOCK_SIZE bsize) {
+  int mi_row, mi_col;
+  VP9_COMMON *const cm = &cpi->common;
+  // Loop through blocks and set skin map based on center pixel of block.
+  // Ignore rightmost/bottom boundary blocks.
+  for (mi_row = 0; mi_row < cm->mi_rows - 1; mi_row += MI_BLOCK_SIZE) {
+    for (mi_col = 0; mi_col < cm->mi_cols - 1; mi_col += MI_BLOCK_SIZE) {
+      vp9_compute_skin_sb(cpi, bsize, mi_row, mi_col);
+    }
+  }
+}
+
 #ifdef OUTPUT_YUV_SKINMAP
 // For viewing skin map on input source.
 void vp9_output_skin_map(VP9_COMP *const cpi, FILE *yuv_skinmap_file) {
diff --git a/vp9/encoder/vp9_skin_detection.h b/vp9/encoder/vp9_skin_detection.h
index 6b5f149c9..c24d691b2 100644
--- a/vp9/encoder/vp9_skin_detection.h
+++ b/vp9/encoder/vp9_skin_detection.h
@@ -25,6 +25,9 @@ int vp9_compute_skin_block(const uint8_t *y, const uint8_t *u, const uint8_t *v,
                            int stride, int strideuv, int bsize,
                            int consec_zeromv, int curr_motion_magn);
 
+void vp9_compute_skin_sb(struct VP9_COMP *const cpi, BLOCK_SIZE bsize,
+                         int mi_row, int mi_col);
+
 void vp9_compute_skin_map(struct VP9_COMP *const cpi, BLOCK_SIZE bsize);
 
 #ifdef OUTPUT_YUV_SKINMAP