mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-07-24 14:06:27 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #27334 from fengyuentau:4x/imgproc/compareHist_chisqr_simd

Browse Source 
imgproc: vectorize mode CHISQR and CHISQR_ALT in compareHist
This commit is contained in:
Alexander Smorkalov 2025-05-21 07:07:57 +03:00 committed by GitHub
commit 79a5e5276a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 40 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
modules/imgproc/src/histogram.cpp
View File

@ -2048,6 +2048,46 @@ double cv::compareHist( InputArray _H1, InputArray _H2, int method )
if( (method == CV_COMP_CHISQR) || (method == CV_COMP_CHISQR_ALT))
{
#if CV_SIMD_64F || CV_SIMD_SCALABLE_64F
v_float64 v_eps = vx_setall_f64(DBL_EPSILON);
v_float64 v_one = vx_setall_f64(1.f);
v_float64 v_zero = vx_setzero_f64();
v_float64 v_res = vx_setzero_f64();
for ( ; j <= len - VTraits<v_float32>::vlanes(); j += VTraits<v_float32>::vlanes())
{
v_float32 v_h1 = vx_load(h1 + j), v_h2 = vx_load(h2 + j);
v_float64 v_h1_l = v_cvt_f64(v_h1), v_h1_h = v_cvt_f64_high(v_h1);
v_float64 v_h2_l = v_cvt_f64(v_h2), v_h2_h = v_cvt_f64_high(v_h2);
v_float64 v_a_l, v_a_h;
v_a_l = v_sub(v_h1_l, v_h2_l);
v_a_h = v_sub(v_h1_h, v_h2_h);
v_float64 v_b_l, v_b_h;
if (method == CV_COMP_CHISQR)
{
v_b_l = v_h1_l;
v_b_h = v_h1_h;
}
else
{
v_b_l = v_add(v_h1_l, v_h2_l);
v_b_h = v_add(v_h1_h, v_h2_h);
}
// low part
auto v_res_l = v_mul(v_mul(v_a_l, v_a_l), v_div(v_one, v_b_l));
auto mask = v_gt(v_abs(v_b_l), v_eps);
v_res_l = v_select(mask, v_res_l, v_zero);
v_res = v_add(v_res, v_res_l);
// high part
auto v_res_h = v_mul(v_mul(v_a_h, v_a_h), v_div(v_one, v_b_h));
mask = v_gt(v_abs(v_b_h), v_eps);
v_res_h = v_select(mask, v_res_h, v_zero);
v_res = v_add(v_res, v_res_h);
}
result += v_reduce_sum(v_res);
#endif
for( ; j < len; j++ )
{
double a = h1[j] - h2[j];