Fix for SSE2 intrinsics problem in the part of saturation arithmetic processing during 32s->16u packed conversion -

for some big negative values less than -INT_MAX+32767 the sign of the numbers is lost due to overflow that leads to
incorrect saturation to MAX value, instead of zero.
The issue is not reproduced with CV_ENABLED_INTRINSICS=OFF

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

@@ -494,7 +494,12 @@ void v_rshr_pack_store(ushort* ptr, const v_uint32x4& a)
 inline v_uint16x8 v_pack_u(const v_int32x4& a, const v_int32x4& b)
 {
     __m128i delta32 = _mm_set1_epi32(32768);
-    __m128i r = _mm_packs_epi32(_mm_sub_epi32(a.val, delta32), _mm_sub_epi32(b.val, delta32));
+
+    // preliminary saturate negative values to zero
+    __m128i a1 = _mm_and_si128(a.val, _mm_cmpgt_epi32(a.val, _mm_set1_epi32(0)));
+    __m128i b1 = _mm_and_si128(b.val, _mm_cmpgt_epi32(b.val, _mm_set1_epi32(0)));
+
+    __m128i r = _mm_packs_epi32(_mm_sub_epi32(a1, delta32), _mm_sub_epi32(b1, delta32));
     return v_uint16x8(_mm_sub_epi16(r, _mm_set1_epi16(-32768)));
 }
 

modules/core/src/convert.cpp

@@ -453,9 +453,9 @@ struct Cvt_SIMD<int, uchar>
             {
                 v_int32x4 v_src1 = v_load(src + x), v_src2 = v_load(src + x + cWidth);
                 v_int32x4 v_src3 = v_load(src + x + cWidth * 2), v_src4 = v_load(src + x + cWidth * 3);
-                v_uint16x8 v_dst1 = v_pack_u(v_src1, v_src2);
-                v_uint16x8 v_dst2 = v_pack_u(v_src3, v_src4);
-                v_store(dst + x, v_pack(v_dst1, v_dst2));
+                v_int16x8 v_dst1 = v_pack(v_src1, v_src2);
+                v_int16x8 v_dst2 = v_pack(v_src3, v_src4);
+                v_store(dst + x, v_pack_u(v_dst1, v_dst2));
             }
         }
         return x;