float normL1_(const float* a, const float* b, int n)
{
int j = 0; float d = 0.f;
#if CV_SIMD
v_float32 v_d = vx_setzero_f32();
for (; j <= n - v_float32::nlanes; j += v_float32::nlanes)
v_d += v_absdiff(vx_load(a + j), vx_load(b + j));
d = v_reduce_sum(v_d);
#endif
for( ; j < n; j++ )
d += std::abs(a[j] - b[j]);
return d;
}
TheTest & test_float_absdiff()
{
Data<R> dataA(std::numeric_limits<LaneType>::max()),
dataB(std::numeric_limits<LaneType>::min());
dataA[0] = -1;
dataB[0] = 1;
dataA[1] = 2;
dataB[1] = -2;
R a = dataA, b = dataB;
Data<R> resC = v_absdiff(a, b);
for (int i = 0; i < R::nlanes; ++i)
{
EXPECT_EQ(dataA[i] > dataB[i] ? dataA[i] - dataB[i] : dataB[i] - dataA[i], resC[i]);
}
return *this;
}
TheTest & test_absdiff()
{
typedef typename V_RegTrait128<LaneType>::u_reg Ru;
typedef typename Ru::lane_type u_type;
Data<R> dataA(std::numeric_limits<LaneType>::max()),
dataB(std::numeric_limits<LaneType>::min());
dataA[0] = (LaneType)-1;
dataB[0] = 1;
dataA[1] = 2;
dataB[1] = (LaneType)-2;
R a = dataA, b = dataB;
Data<Ru> resC = v_absdiff(a, b);
const u_type mask = std::numeric_limits<LaneType>::is_signed ? (u_type)(1 << (sizeof(u_type)*8 - 1)) : 0;
for (int i = 0; i < Ru::nlanes; ++i)
{
u_type uA = dataA[i] ^ mask;
u_type uB = dataB[i] ^ mask;
EXPECT_EQ(uA > uB ? uA - uB : uB - uA, resC[i]);
}
return *this;
}
