Minor improvement to SSE code in HOGDescriptor::computeGradient, replace emulation of _mm_mullo_epi32 with constant multiplicand 3 with two _mm_add_epi32. OpenCV issue #11161

7 years ago · 333e53eee8
parent 24acbecd33
commit 333e53eee8
1 changed files with 1 additions and 5 deletions
--- a/modules/objdetect/src/hog.cpp
+++ b/modules/objdetect/src/hog.cpp
@ -323,14 +323,10 @@ void HOGDescriptor::computeGradient(const Mat& img, Mat& grad, Mat& qangle,
        int end = gradsize.width + 2;
        xmap -= 1, x = 0;
 #if CV_SSE2
-        __m128i ithree = _mm_set1_epi32(3);
        for ( ; x <= end - 4; x += 4)
        {
-            //emulation of _mm_mullo_epi32
            __m128i mul_res = _mm_loadu_si128((const __m128i*)(xmap + x));
-            __m128i tmp1 = _mm_mul_epu32(ithree, mul_res);
-            __m128i tmp2 = _mm_mul_epu32( _mm_srli_si128(ithree,4), _mm_srli_si128(mul_res,4));
-            mul_res = _mm_unpacklo_epi32(_mm_shuffle_epi32(tmp1, _MM_SHUFFLE (0,0,2,0)), _mm_shuffle_epi32(tmp2, _MM_SHUFFLE (0,0,2,0)));
+            mul_res = _mm_add_epi32(_mm_add_epi32(mul_res, mul_res), mul_res); // multiply by 3
            _mm_storeu_si128((__m128i*)(xmap + x), mul_res);
        }
 #elif CV_NEON