Skip to content

O
opencv
Commit 042b0a58 authored by k-shinotsuka's avatar k-shinotsuka
Browse files
Options

improve LKTrackerInvoker::operator()

parent b34272f8
Hide whitespace changes
Inline Side-by-side
Showing with 7 additions and 11 deletions
modules/video/src/lkpyramid.cpp
View file @ 042b0a58
...@@ -567,18 +567,14 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const ...@@ -567,18 +567,14 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
diff0 = _mm_unpacklo_epi16(diff0, diff0); // It0 It0 It1 It1 ... diff0 = _mm_unpacklo_epi16(diff0, diff0); // It0 It0 It1 It1 ...
v00 = _mm_loadu_si128((const __m128i*)(dIptr)); // Ix0 Iy0 Ix1 Iy1 ... v00 = _mm_loadu_si128((const __m128i*)(dIptr)); // Ix0 Iy0 Ix1 Iy1 ...
v01 = _mm_loadu_si128((const __m128i*)(dIptr + 8)); v01 = _mm_loadu_si128((const __m128i*)(dIptr + 8));
v10 = _mm_mullo_epi16(v00, diff0); v10 = _mm_unpacklo_epi16(v00, v01);
v11 = _mm_mulhi_epi16(v00, diff0); v11 = _mm_unpackhi_epi16(v00, v01);
v00 = _mm_unpacklo_epi16(v10, v11); v00 = _mm_unpacklo_epi16(diff0, diff1);
v10 = _mm_unpackhi_epi16(v10, v11); v01 = _mm_unpackhi_epi16(diff0, diff1);
v00 = _mm_madd_epi16(v00, v10);
v11 = _mm_madd_epi16(v01, v11);
qb0 = _mm_add_ps(qb0, _mm_cvtepi32_ps(v00)); qb0 = _mm_add_ps(qb0, _mm_cvtepi32_ps(v00));
qb1 = _mm_add_ps(qb1, _mm_cvtepi32_ps(v10)); qb1 = _mm_add_ps(qb1, _mm_cvtepi32_ps(v11));
v10 = _mm_mullo_epi16(v01, diff1);
v11 = _mm_mulhi_epi16(v01, diff1);
v00 = _mm_unpacklo_epi16(v10, v11);
v10 = _mm_unpackhi_epi16(v10, v11);
qb0 = _mm_add_ps(qb0, _mm_cvtepi32_ps(v00));
qb1 = _mm_add_ps(qb1, _mm_cvtepi32_ps(v10));
} }
#endif #endif
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment