mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-06-16 14:50:52 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Optimize normL2Sqr_ with AVX2

Browse Source
This commit is contained in:
Woody Chow 2017-03-16 14:20:41 +09:00
parent a83a1cafa7
commit c370cc10e9
1 changed files with 16 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
modules/core/src/stat.cpp
View File

@ -4422,7 +4422,22 @@ int normHamming(const uchar* a, const uchar* b, int n, int cellSize)
float normL2Sqr_(const float* a, const float* b, int n) float normL2Sqr_(const float* a, const float* b, int n)
{ {
int j = 0; float d = 0.f; int j = 0; float d = 0.f;
#if CV_SSE #if CV_AVX2
float CV_DECL_ALIGNED(32) buf[8];
__m256 d0 = _mm256_setzero_ps();
for( ; j <= n - 8; j += 8 )
{
__m256 t0 = _mm256_sub_ps(_mm256_loadu_ps(a + j), _mm256_loadu_ps(b + j));
#ifdef CV_FMA3
d0 = _mm256_fmadd_ps(t0, t0, d0);
#else
d0 = _mm256_add_ps(d0, _mm256_mul_ps(t0, t0));
#endif
}
_mm256_store_ps(buf, d0);
d = buf[0] + buf[1] + buf[2] + buf[3] + buf[4] + buf[5] + buf[6] + buf[7];
#elif CV_SSE
float CV_DECL_ALIGNED(16) buf[4]; float CV_DECL_ALIGNED(16) buf[4];
__m128 d0 = _mm_setzero_ps(), d1 = _mm_setzero_ps(); __m128 d0 = _mm_setzero_ps(), d1 = _mm_setzero_ps();