template<bool align> SIMD_INLINE v128_u8 BlurRow(const Buffer & buffer, size_t offset)
{
v128_u16 lo = DivideBy16(BinomialSum(Load<align>(buffer.src0 + offset), Load<align>(buffer.src1 + offset), Load<align>(buffer.src2 + offset)));
offset += HA;
v128_u16 hi = DivideBy16(BinomialSum(Load<align>(buffer.src0 + offset), Load<align>(buffer.src1 + offset), Load<align>(buffer.src2 + offset)));
return vec_pack(lo, hi);
}
template<bool abs,int sequence> SIMD_INLINE v8i16 SobelDx(v16u8 a[3][3])
{
v8i16 d0,d1,d2;
d0 = ExtendSub<sequence>(a[0][2],a[0][0]);
d1 = ExtendSub<sequence>(a[1][2],a[1][0]);
d2 = ExtendSub<sequence>(a[2][2],a[2][0]);
v8i16 b = BinomialSum(d0,d1,d2);
return ConditionalAbs<abs>(b);
}
template<bool align> SIMD_INLINE void BlurCol(v128_u8 a[3], uint16_t * b)
{
Store<align>(b, BinomialSum(UnpackLoU8(a[0]), UnpackLoU8(a[1]), UnpackLoU8(a[2])));
Store<align>(b + HA, BinomialSum(UnpackHiU8(a[0]), UnpackHiU8(a[1]), UnpackHiU8(a[2])));
}
template <bool compensation> SIMD_INLINE v128_u8 ReduceRow(const v128_u16 lo[3], const v128_u16 hi[3])
{
return vec_packsu(
DivideBy16<compensation>(BinomialSum(lo[0], lo[1], lo[2])),
DivideBy16<compensation>(BinomialSum(hi[0], hi[1], hi[2])));
}
template<bool align> SIMD_INLINE v128_u16 ReduceColBody(const uint8_t * p)
{
const v128_u8 t0 = Load<false>(p - 1);
const v128_u8 t1 = Load<align>(p);
return BinomialSum(vec_mule(t0, K8_01), vec_mule(t1, K8_01), vec_mulo(t1, K8_01));
}
template<bool align> SIMD_INLINE v128_u16 ReduceColNose(const uint8_t * p)
{
const v128_u8 t = Load<align>(p);
return BinomialSum(vec_mule(LoadBeforeFirst<1>(t), K8_01), vec_mule(t, K8_01), vec_mulo(t, K8_01));
}
