From: James Zern <jzern@google.com>
Date: Sat, 9 Sep 2017 02:20:07 +0000 (-0700)
Subject: intrapred: sync highbd_d63_predictor w/d63_
X-Git-Tag: v1.7.0~179
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=fb40b5d7a76978e6fcedf5997fd7a45571aa6b0d;p=libvpx

intrapred: sync highbd_d63_predictor w/d63_

8/16/32: ~6%/~18%/~33% faster

previously:
7012ba639 vp9_reconintra: simplify d63_predictor

BUG=webm:1411

Change-Id: Ie775f3a4f7fd74df44754e65686d826a51c2cdc2
---

diff --git a/vpx_dsp/intrapred.c b/vpx_dsp/intrapred.c
index 4c640e5d2..400e632e9 100644
--- a/vpx_dsp/intrapred.c
+++ b/vpx_dsp/intrapred.c
@@ -489,15 +489,19 @@ static INLINE void highbd_d63_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
                                         const uint16_t *above,
                                         const uint16_t *left, int bd) {
   int r, c;
+  int size;
   (void)left;
   (void)bd;
-  for (r = 0; r < bs; ++r) {
-    for (c = 0; c < bs; ++c) {
-      dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
-                            above[(r >> 1) + c + 2])
-                     : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
-    }
-    dst += stride;
+  for (c = 0; c < bs; ++c) {
+    dst[c] = AVG2(above[c], above[c + 1]);
+    dst[stride + c] = AVG3(above[c], above[c + 1], above[c + 2]);
+  }
+  for (r = 2, size = bs - 2; r < bs; r += 2, --size) {
+    memcpy(dst + (r + 0) * stride, dst + (r >> 1), size * sizeof(*dst));
+    vpx_memset16(dst + (r + 0) * stride + size, above[bs - 1], bs - size);
+    memcpy(dst + (r + 1) * stride, dst + stride + (r >> 1),
+           size * sizeof(*dst));
+    vpx_memset16(dst + (r + 1) * stride + size, above[bs - 1], bs - size);
   }
 }