summaryrefslogtreecommitdiff
path: root/vp9/encoder
diff options
context:
space:
mode:
authorJingning Han <jingning@google.com>2015-02-23 12:55:50 -0800
committerJingning Han <jingning@google.com>2015-02-24 11:48:38 -0800
commitf87e315e1e34de158581cbf7bc96248542d4b3fb (patch)
tree7d911a8ba17507b122b09f50f2606b03c931b0fc /vp9/encoder
parenta28a8cb72626d8bea3543052e9bf825e95666c7e (diff)
downloadlibvpx-f87e315e1e34de158581cbf7bc96248542d4b3fb.tar
libvpx-f87e315e1e34de158581cbf7bc96248542d4b3fb.tar.gz
libvpx-f87e315e1e34de158581cbf7bc96248542d4b3fb.tar.bz2
libvpx-f87e315e1e34de158581cbf7bc96248542d4b3fb.zip
Re-distribute hierarchical vector match pattern
This commit modifies the hierarchical vector match patter. It avoids repeated SAD computation at same points. The function vp9_vector_sad_sse2 is called 12 times per 64x64 block, instead of 15 times as before. The effective coverage remains the same. Change-Id: I91ad9d27d40db8963c907d02af84e10702136994
Diffstat (limited to 'vp9/encoder')
-rw-r--r--vp9/encoder/vp9_encodeframe.c25
1 files changed, 19 insertions, 6 deletions
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 39dedf6cb..6787faa8a 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -528,10 +528,10 @@ static int vector_match(int16_t *ref, int16_t *src) {
}
center = offset;
- for (d = -8; d <= 8; d += 4) {
+ for (d = -8; d <= 8; d += 16) {
int this_pos = offset + d;
// check limit
- if (this_pos < 0 || this_pos > 64 || this_pos == 32)
+ if (this_pos < 0 || this_pos > 64)
continue;
this_sad = vp9_vector_sad(&ref[this_pos], src, 64);
if (this_sad < best_sad) {
@@ -541,10 +541,10 @@ static int vector_match(int16_t *ref, int16_t *src) {
}
offset = center;
- for (d = -4; d <= 4; d += 2) {
+ for (d = -4; d <= 4; d += 8) {
int this_pos = offset + d;
// check limit
- if (this_pos < 0 || this_pos > 64 || this_pos == 32)
+ if (this_pos < 0 || this_pos > 64)
continue;
this_sad = vp9_vector_sad(&ref[this_pos], src, 64);
if (this_sad < best_sad) {
@@ -554,10 +554,23 @@ static int vector_match(int16_t *ref, int16_t *src) {
}
offset = center;
- for (d = -2; d <= 2; d += 1) {
+ for (d = -2; d <= 2; d += 4) {
int this_pos = offset + d;
// check limit
- if (this_pos < 0 || this_pos > 64 || this_pos == 32)
+ if (this_pos < 0 || this_pos > 64)
+ continue;
+ this_sad = vp9_vector_sad(&ref[this_pos], src, 64);
+ if (this_sad < best_sad) {
+ best_sad = this_sad;
+ center = this_pos;
+ }
+ }
+ offset = center;
+
+ for (d = -1; d <= 1; d += 2) {
+ int this_pos = offset + d;
+ // check limit
+ if (this_pos < 0 || this_pos > 64)
continue;
this_sad = vp9_vector_sad(&ref[this_pos], src, 64);
if (this_sad < best_sad) {