summaryrefslogtreecommitdiff
path: root/vpx_dsp/arm/sad_neon.c
diff options
context:
space:
mode:
authorJohann <johannkoenig@google.com>2017-06-28 14:11:35 -0700
committerJohann <johannkoenig@google.com>2017-07-06 07:43:09 -0700
commit63bdc574e52951bcda3e6f19c4d37b25452bcdaf (patch)
treea3b1eb1dec8a01744f5dd5982447610f3cee0d20 /vpx_dsp/arm/sad_neon.c
parent6bac3f80ee21b754e17d0f32ddc18d1bbe9dfc6a (diff)
downloadlibvpx-63bdc574e52951bcda3e6f19c4d37b25452bcdaf.tar
libvpx-63bdc574e52951bcda3e6f19c4d37b25452bcdaf.tar.gz
libvpx-63bdc574e52951bcda3e6f19c4d37b25452bcdaf.tar.bz2
libvpx-63bdc574e52951bcda3e6f19c4d37b25452bcdaf.zip
sad neon: avg for 8x[4,8,16]
BUG=webm:1425 Change-Id: If2ab51e3050e078b0011b174efe41fcb65a15f44
Diffstat (limited to 'vpx_dsp/arm/sad_neon.c')
-rw-r--r--vpx_dsp/arm/sad_neon.c43
1 files changed, 43 insertions, 0 deletions
diff --git a/vpx_dsp/arm/sad_neon.c b/vpx_dsp/arm/sad_neon.c
index c227ddc9d..e7aaf81c5 100644
--- a/vpx_dsp/arm/sad_neon.c
+++ b/vpx_dsp/arm/sad_neon.c
@@ -114,6 +114,49 @@ uint32_t vpx_sad8x16_neon(const uint8_t *src, int src_stride,
return horizontal_add_16x8(abs);
}
+static INLINE uint16x8_t sad8x_avg(const uint8_t *a, int a_stride,
+ const uint8_t *b, int b_stride,
+ const uint8_t *c, const int height) {
+ int i;
+ uint16x8_t abs = vdupq_n_u16(0);
+
+ for (i = 0; i < height; ++i) {
+ const uint8x8_t a_u8 = vld1_u8(a);
+ const uint8x8_t b_u8 = vld1_u8(b);
+ const uint8x8_t c_u8 = vld1_u8(c);
+ const uint8x8_t avg = vrhadd_u8(b_u8, c_u8);
+ a += a_stride;
+ b += b_stride;
+ c += 8;
+ abs = vabal_u8(abs, a_u8, avg);
+ }
+ return abs;
+}
+
+uint32_t vpx_sad8x4_avg_neon(const uint8_t *src, int src_stride,
+ const uint8_t *ref, int ref_stride,
+ const uint8_t *second_pred) {
+ const uint16x8_t abs =
+ sad8x_avg(src, src_stride, ref, ref_stride, second_pred, 4);
+ return horizontal_add_16x8(abs);
+}
+
+uint32_t vpx_sad8x8_avg_neon(const uint8_t *src, int src_stride,
+ const uint8_t *ref, int ref_stride,
+ const uint8_t *second_pred) {
+ const uint16x8_t abs =
+ sad8x_avg(src, src_stride, ref, ref_stride, second_pred, 8);
+ return horizontal_add_16x8(abs);
+}
+
+uint32_t vpx_sad8x16_avg_neon(const uint8_t *src, int src_stride,
+ const uint8_t *ref, int ref_stride,
+ const uint8_t *second_pred) {
+ const uint16x8_t abs =
+ sad8x_avg(src, src_stride, ref, ref_stride, second_pred, 16);
+ return horizontal_add_16x8(abs);
+}
+
static INLINE uint16x8_t sad16x(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
const int height) {