use universal intrinsic in VBLAS

 - brush up v_reduce_sum of SSE version

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

@@ -1101,6 +1101,15 @@ OPENCV_HAL_IMPL_SSE_REDUCE_OP_8(int16x8, short, max, epi16, (short)-32768)
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_8(int16x8, short, min, epi16, (short)-32768)
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_8_SUM(int16x8, short, 16)
 
+#define OPENCV_HAL_IMPL_SSE_REDUCE_OP_4_SUM(_Tpvec, scalartype, regtype, suffix, cast_from, cast_to, extract) \
+inline scalartype v_reduce_sum(const _Tpvec& a) \
+{ \
+    regtype val = a.val; \
+    val = _mm_add_##suffix(val, cast_to(_mm_srli_si128(cast_from(val), 8))); \
+    val = _mm_add_##suffix(val, cast_to(_mm_srli_si128(cast_from(val), 4))); \
+    return (scalartype)_mm_cvt##extract(val); \
+}
+
 #define OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(_Tpvec, scalartype, func, scalar_func) \
 inline scalartype v_reduce_##func(const _Tpvec& a) \
 { \
@@ -1111,13 +1120,14 @@ inline scalartype v_reduce_##func(const _Tpvec& a) \
     return scalar_func(s0, s1); \
 }
 
-OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_uint32x4, unsigned, sum, OPENCV_HAL_ADD)
+OPENCV_HAL_IMPL_SSE_REDUCE_OP_4_SUM(v_uint32x4, unsigned, __m128i, epi32, OPENCV_HAL_NOP, OPENCV_HAL_NOP, si128_si32)
+OPENCV_HAL_IMPL_SSE_REDUCE_OP_4_SUM(v_int32x4, int, __m128i, epi32, OPENCV_HAL_NOP, OPENCV_HAL_NOP, si128_si32)
+OPENCV_HAL_IMPL_SSE_REDUCE_OP_4_SUM(v_float32x4, float, __m128, ps, _mm_castps_si128, _mm_castsi128_ps, ss_f32)
+
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_uint32x4, unsigned, max, std::max)
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_uint32x4, unsigned, min, std::min)
-OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_int32x4, int, sum, OPENCV_HAL_ADD)
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_int32x4, int, max, std::max)
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_int32x4, int, min, std::min)
-OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_float32x4, float, sum, OPENCV_HAL_ADD)
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_float32x4, float, max, std::max)
 OPENCV_HAL_IMPL_SSE_REDUCE_OP_4(v_float32x4, float, min, std::min)