fix for AVX2 version of v_reduce_min intrinsic

modules/core/include/opencv2/core/hal/intrin_avx.hpp

@@ -1083,7 +1083,7 @@ OPENCV_HAL_IMPL_AVX_REDUCE_8(v_int32x8,  int,      max, _mm_max_epi32)
         __m128 v1 = _v256_extract_high(a.val);                        \
         v0 = intrin(v0, v1);                                          \
         v0 = intrin(v0, _mm_permute_ps(v0, _MM_SHUFFLE(0, 0, 3, 2))); \
-        v0 = intrin(v0, _mm_permute_ps(v0, _MM_SHUFFLE(0, 0, 0, 3))); \
+        v0 = intrin(v0, _mm_permute_ps(v0, _MM_SHUFFLE(0, 0, 0, 1))); \
         return _mm_cvtss_f32(v0);                                     \
     }
 

modules/core/test/test_intrin_utils.hpp

@@ -772,6 +772,9 @@ template<typename R> struct TheTest
         EXPECT_EQ((LaneType)1, v_reduce_min(a));
         EXPECT_EQ((LaneType)R::nlanes, v_reduce_max(a));
         EXPECT_EQ((LaneType)((1 + R::nlanes)*R::nlanes/2), v_reduce_sum(a));
+        dataA[0] += R::nlanes;
+        R an = dataA;
+        EXPECT_EQ((LaneType)2, v_reduce_min(an));
         return *this;
     }