SymmRowSmallVec_8u32s 1x3 asymmetric

NEON speedup: 1.95x
Auto-vect speedup: 1.17x

Test kernel: [-2, 0, 2]

modules/imgproc/src/filter.cpp

@@ -2370,9 +2370,32 @@ struct SymmRowSmallVec_8u32s
                         vst1q_s32((int32_t *)(dst + i), vmovl_s16(vget_low_s16(y0)));
                         vst1q_s32((int32_t *)(dst + i + 4), vmovl_s16(vget_high_s16(y0)));
                     }
+                }
                 else
                 {
-                    return 0;
+                    int32x4_t k32 = vdupq_n_s32(0);
+                    k32 = vld1q_lane_s32(kx + 1, k32, 1);
+
+                    int16x4_t k = vqmovn_s32(k32);
+
+                    uint8x8_t z = vdup_n_u8(0);
+
+                    for( ; i <= width - 8; i += 8, src += 8 )
+                    {
+                        uint8x8_t x0, x1;
+                        x0 = vld1_u8( (uint8_t *) (src - cn) );
+                        x1 = vld1_u8( (uint8_t *) (src + cn) );
+
+                        int16x8_t y0;
+                        int32x4_t y1, y2;
+                        y0 = vsubq_s16(vreinterpretq_s16_u16(vaddl_u8(x1, z)),
+                            vreinterpretq_s16_u16(vaddl_u8(x0, z)));
+                        y1 = vmull_lane_s16(vget_low_s16(y0), k, 1);
+                        y2 = vmull_lane_s16(vget_high_s16(y0), k, 1);
+
+                        vst1q_s32((int32_t *)(dst + i), y1);
+                        vst1q_s32((int32_t *)(dst + i + 4), y2);
+                    }
                 }
             }
             else if( _ksize == 5 )