diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 7b147b424cb14076b0260bfe656ded2a8b9bd7e4..df963d1cc72e0ec5055b68426ea52104d3cb3c3b 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -384,21 +384,18 @@ static void txfrm_block_to_raster_xy(BLOCK_SIZE plane_bsize,
 }
 
 static void extend_for_intra(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize,
-                             int plane, int block, TX_SIZE tx_size) {
+                             int plane, int aoff, int loff) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
   uint8_t *const buf = pd->dst.buf;
   const int stride = pd->dst.stride;
-
-  int x, y;
-  txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
-  x = x * 4 - 1;
-  y = y * 4 - 1;
+  const int x = aoff * 4 - 1;
+  const int y = loff * 4 - 1;
   // Copy a pixel into the umv if we are in a situation where the block size
   // extends into the UMV.
   // TODO(JBB): Should be able to do the full extend in place so we don't have
   // to do this multiple times.
   if (xd->mb_to_right_edge < 0) {
-    const int bw = 4 << b_width_log2(plane_bsize);
+    const int bw = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
     const int umv_border_start = bw + (xd->mb_to_right_edge >>
                                        (3 + pd->subsampling_x));
 
@@ -409,7 +406,7 @@ static void extend_for_intra(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize,
 
   if (xd->mb_to_bottom_edge < 0) {
     if (xd->left_available || x >= 0) {
-      const int bh = 4 << b_height_log2(plane_bsize);
+      const int bh = 4 * num_4x4_blocks_high_lookup[plane_bsize];
       const int umv_border_start =
           bh + (xd->mb_to_bottom_edge >> (3 + pd->subsampling_y));
 
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index c65d209bf2ee5f6545b389eb526b35954fa62853..916cb424e730d80b68f48ece5fd326319fefef7d 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -318,7 +318,7 @@ static void predict_and_reconstruct_intra_block(int plane, int block,
   dst = &pd->dst.buf[4 * y * pd->dst.stride + 4 * x];
 
   if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0)
-    extend_for_intra(xd, plane_bsize, plane, block, tx_size);
+    extend_for_intra(xd, plane_bsize, plane, x, y);
 
   vp9_predict_intra_block(xd, block >> (tx_size << 1),
                           b_width_log2(plane_bsize), tx_size, mode,
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index b80107ea0a4291ba7d247c2c601c366e9accab0b..22ab26daf490b9f207777a08c6ee9121b889c44c 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -562,7 +562,7 @@ void vp9_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
   src_diff = &p->src_diff[4 * (j * diff_stride + i)];
 
   if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0)
-    extend_for_intra(xd, plane_bsize, plane, block, tx_size);
+    extend_for_intra(xd, plane_bsize, plane, i, j);
 
   // if (x->optimize)
   // vp9_optimize_b(plane, block, plane_bsize, tx_size, x, args->ctx);