int operator () (const short * src0, const uchar * mask, int * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_int32 v_sum = vx_setzero_s32();
for (; x <= len - v_int16::nlanes; x += v_int16::nlanes)
{
v_int32 v_src0, v_src1;
v_expand(vx_load(src0 + x), v_src0, v_src1);
v_sum += v_src0 + v_src1;
}
if (x <= len - v_int32::nlanes)
{
v_sum += vx_load_expand(src0 + x);
x += v_int32::nlanes;
}
if (cn == 1)
*dst += v_reduce_sum(v_sum);
else
{
int32_t CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_int32::nlanes];
v_store_aligned(ar, v_sum);
for (int i = 0; i < v_int32::nlanes; ++i)
dst[i % cn] += ar[i];
}
v_cleanup();
return x / cn;
}
TheTest & test_reduce()
{
Data<R> dataA;
R a = dataA;
EXPECT_EQ((LaneType)1, v_reduce_min(a));
EXPECT_EQ((LaneType)R::nlanes, v_reduce_max(a));
EXPECT_EQ((LaneType)((1 + R::nlanes)*R::nlanes/2), v_reduce_sum(a));
return *this;
}
TheTest & test_popcount()
{
static unsigned popcountTable[] = {0, 1, 2, 4, 5, 7, 9, 12, 13, 15, 17, 20, 22, 25, 28, 32, 33};
Data<R> dataA;
R a = dataA;
unsigned resB = (unsigned)v_reduce_sum(v_popcount(a));
EXPECT_EQ(popcountTable[R::nlanes], resB);
return *this;
}
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;
}
int operator () (const schar * src0, const uchar * mask, int * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_int32 v_sum = vx_setzero_s32();
int len0 = len & -v_int8::nlanes;
while (x < len0)
{
const int len_tmp = min(x + 256*v_int16::nlanes, len0);
v_int16 v_sum16 = vx_setzero_s16();
for (; x < len_tmp; x += v_int8::nlanes)
{
v_int16 v_src0, v_src1;
v_expand(vx_load(src0 + x), v_src0, v_src1);
v_sum16 += v_src0 + v_src1;
}
v_int32 v_half0, v_half1;
v_expand(v_sum16, v_half0, v_half1);
v_sum += v_half0 + v_half1;
}
if (x <= len - v_int16::nlanes)
{
v_int32 v_half0, v_half1;
v_expand(vx_load_expand(src0 + x), v_half0, v_half1);
v_sum += v_half0 + v_half1;
x += v_int16::nlanes;
}
if (x <= len - v_int32::nlanes)
{
v_sum += vx_load_expand_q(src0 + x);
x += v_int32::nlanes;
}
if (cn == 1)
*dst += v_reduce_sum(v_sum);
else
{
int32_t CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_int32::nlanes];
v_store_aligned(ar, v_sum);
for (int i = 0; i < v_int32::nlanes; ++i)
dst[i % cn] += ar[i];
}
v_cleanup();
return x / cn;
}
float normL2Sqr_(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_float32 t = vx_load(a + j) - vx_load(b + j);
v_d = v_muladd(t, t, v_d);
}
d = v_reduce_sum(v_d);
#endif
for( ; j < n; j++ )
{
float t = a[j] - b[j];
d += t*t;
}
return d;
}
int normHamming(const uchar* a, const uchar* b, int n)
{
CV_AVX_GUARD;
int i = 0;
int result = 0;
#if CV_AVX2
{
__m256i _r0 = _mm256_setzero_si256();
__m256i _0 = _mm256_setzero_si256();
__m256i _popcnt_table = _mm256_setr_epi8(0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4,
0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4);
__m256i _popcnt_mask = _mm256_set1_epi8(0x0F);
for(; i <= n - 32; i+= 32)
{
__m256i _a0 = _mm256_loadu_si256((const __m256i*)(a + i));
__m256i _b0 = _mm256_loadu_si256((const __m256i*)(b + i));
__m256i _xor = _mm256_xor_si256(_a0, _b0);
__m256i _popc0 = _mm256_shuffle_epi8(_popcnt_table, _mm256_and_si256(_xor, _popcnt_mask));
__m256i _popc1 = _mm256_shuffle_epi8(_popcnt_table,
_mm256_and_si256(_mm256_srli_epi16(_xor, 4), _popcnt_mask));
_r0 = _mm256_add_epi32(_r0, _mm256_sad_epu8(_0, _mm256_add_epi8(_popc0, _popc1)));
}
_r0 = _mm256_add_epi32(_r0, _mm256_shuffle_epi32(_r0, 2));
result = _mm256_extract_epi32_(_mm256_add_epi32(_r0, _mm256_permute2x128_si256(_r0, _r0, 1)), 0);
}
#endif // CV_AVX2
#if CV_POPCNT
{
# if defined CV_POPCNT_U64
for(; i <= n - 8; i += 8)
{
result += (int)CV_POPCNT_U64(*(uint64*)(a + i) ^ *(uint64*)(b + i));
}
# endif
for(; i <= n - 4; i += 4)
{
result += CV_POPCNT_U32(*(uint*)(a + i) ^ *(uint*)(b + i));
}
}
#endif // CV_POPCNT
#if CV_SIMD128
{
v_uint32x4 t = v_setzero_u32();
for(; i <= n - v_uint8x16::nlanes; i += v_uint8x16::nlanes)
{
t += v_popcount(v_load(a + i) ^ v_load(b + i));
}
result += v_reduce_sum(t);
}
#endif // CV_SIMD128
#if CV_ENABLE_UNROLLED
for(; i <= n - 4; i += 4)
{
result += popCountTable[a[i] ^ b[i]] + popCountTable[a[i+1] ^ b[i+1]] +
popCountTable[a[i+2] ^ b[i+2]] + popCountTable[a[i+3] ^ b[i+3]];
}
#endif
for(; i < n; i++)
{
result += popCountTable[a[i] ^ b[i]];
}
return result;
}
