Upstream version 10.39.225.0

[platform/framework/web/crosswalk.git] / src / third_party / libvpx / source / libvpx / vp9 / encoder / arm / neon / vp9_sad_neon.c

diff --git a/src/third_party/libvpx/source/libvpx/vp9/encoder/arm/neon/vp9_sad_neon.c b/src/third_party/libvpx/source/libvpx/vp9/encoder/arm/neon/vp9_sad_neon.c
index fe40b54..c4cd856 100644 (file)
--- a/src/third_party/libvpx/source/libvpx/vp9/encoder/arm/neon/vp9_sad_neon.c
+++ b/src/third_party/libvpx/source/libvpx/vp9/encoder/arm/neon/vp9_sad_neon.c
@@ -26,9 +26,8 @@ static INLINE unsigned int horizontal_long_add_16x8(const uint16x8_t vec_lo,
                                 vreinterpret_u32_u64(vget_high_u64(b)));
   return vget_lane_u32(c, 0);
 }
-static INLINE unsigned int horizontal_add_16x8(const uint16x8_t vec_lo,
-                                               const uint16x8_t vec_hi) {
-  const uint32x4_t a = vpaddlq_u16(vaddq_u16(vec_lo, vec_hi));
+static INLINE unsigned int horizontal_add_16x8(const uint16x8_t vec_16x8) {
+  const uint32x4_t a = vpaddlq_u16(vec_16x8);
   const uint64x2_t b = vpaddlq_u32(a);
   const uint32x2_t c = vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
                                 vreinterpret_u32_u64(vget_high_u64(b)));
@@ -93,7 +92,7 @@ unsigned int vp9_sad32x32_neon(const uint8_t *src, int src_stride,
     vec_accum_hi = vabal_u8(vec_accum_hi, vget_high_u8(vec_src_16),
                             vget_high_u8(vec_ref_16));
   }
-  return horizontal_add_16x8(vec_accum_lo, vec_accum_hi);
+  return horizontal_add_16x8(vaddq_u16(vec_accum_lo, vec_accum_hi));
 }
 
 unsigned int vp9_sad16x16_neon(const uint8_t *src, int src_stride,
@@ -112,5 +111,20 @@ unsigned int vp9_sad16x16_neon(const uint8_t *src, int src_stride,
     vec_accum_hi = vabal_u8(vec_accum_hi, vget_high_u8(vec_src),
                             vget_high_u8(vec_ref));
   }
-  return horizontal_add_16x8(vec_accum_lo, vec_accum_hi);
+  return horizontal_add_16x8(vaddq_u16(vec_accum_lo, vec_accum_hi));
+}
+
+unsigned int vp9_sad8x8_neon(const uint8_t *src, int src_stride,
+                             const uint8_t *ref, int ref_stride) {
+  int i;
+  uint16x8_t vec_accum = vdupq_n_u16(0);
+
+  for (i = 0; i < 8; ++i) {
+    const uint8x8_t vec_src = vld1_u8(src);
+    const uint8x8_t vec_ref = vld1_u8(ref);
+    src += src_stride;
+    ref += ref_stride;
+    vec_accum = vabal_u8(vec_accum, vec_src, vec_ref);
+  }
+  return horizontal_add_16x8(vec_accum);
 }