generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

sse2 optimization of inv_sqrt_64f

Browse Source
This commit is contained in:
Ilya Lavrenov 2014-07-03 01:28:31 +04:00
parent c3329ff79d
commit e5b63d43f2
1 changed files with 12 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
modules/core/src/mathfuncs.cpp
View File

@ -348,7 +348,18 @@ static void InvSqrt_32f(const float* src, float* dst, int len)
static void InvSqrt_64f(const double* src, double* dst, int len) static void InvSqrt_64f(const double* src, double* dst, int len)
{ {
for( int i = 0; i < len; i++ ) int i = 0;
#if CV_SSE2
if (USE_SSE2)
{
__m128d v_1 = _mm_set1_pd(1.0);
for ( ; i <= len - 2; i += 2)
_mm_storeu_pd(dst + i, _mm_div_pd(v_1, _mm_sqrt_pd(_mm_loadu_pd(src + i))));
}
#endif
for( ; i < len; i++ )
dst[i] = 1/std::sqrt(src[i]); dst[i] = 1/std::sqrt(src[i]);
} }