Dynamic Dispatching for split3

modules/gapi/src/backends/fluid/gfluidcore.cpp

@@ -2516,19 +2516,11 @@ GAPI_FLUID_KERNEL(GFluidSplit3, cv::gapi::core::GSplit3, false)
 
         GAPI_Assert(3 == src.meta().chan);
         int width = src.length();
+        int w = 0;
 
-        int w = 0; // cycle counter
-
-    #if CV_SIMD128
-        for (; w <= width-16; w+=16)
-        {
-            v_uint8x16 a, b, c;
-            v_load_deinterleave(&in[3*w], a, b, c);
-            v_store(&out1[w], a);
-            v_store(&out2[w], b);
-            v_store(&out3[w], c);
-        }
-    #endif
+#if CV_SIMD
+        w = split3_simd(in, out1, out2, out3, width);
+#endif
 
         for (; w < width; w++)
         {

modules/gapi/src/backends/fluid/gfluidcore_func.dispatch.cpp

@@ -207,6 +207,13 @@ ABSDIFFC_SIMD(float)
 
 #undef ABSDIFFC_SIMD
 
+int split3_simd(const uchar in[], uchar out1[], uchar out2[],
+                uchar out3[], const int width)
+{
+    CV_CPU_DISPATCH(split3_simd, (in, out1, out2, out3, width),
+                    CV_CPU_DISPATCH_MODES_ALL);
+}
+
 } // namespace fluid
 } // namespace gapi
 } // namespace cv

modules/gapi/src/backends/fluid/gfluidcore_func.hpp

@@ -163,6 +163,9 @@ ABSDIFFC_SIMD(float)
 
 #undef ABSDIFFC_SIMD
 
+int split3_simd(const uchar in[], uchar out1[], uchar out2[],
+                uchar out3[], const int width);
+
 }  // namespace fluid
 }  // namespace gapi
 }  // namespace cv

modules/gapi/src/backends/fluid/gfluidcore_func.simd.hpp

@@ -184,6 +184,9 @@ ABSDIFFC_SIMD(float)
 
 #undef ABSDIFFC_SIMD
 
+int split3_simd(const uchar in[], uchar out1[], uchar out2[],
+                uchar out3[], const int width);
+
 #ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
 
 struct scale_tag {};
@@ -1568,6 +1571,28 @@ ABSDIFFC_SIMD(float)
 
 #undef ABSDIFFC_SIMD
 
+//-------------------------
+//
+// Fluid kernels: Split3
+//
+//-------------------------
+
+int split3_simd(const uchar in[], uchar out1[], uchar out2[],
+                uchar out3[], const int width)
+{
+    constexpr int nlanes = v_uint8::nlanes;
+    int x = 0;
+    for (; x <= width - nlanes; x += nlanes)
+    {
+        v_uint8 a, b, c;
+        v_load_deinterleave(&in[3 * x], a, b, c);
+        vx_store(&out1[x], a);
+        vx_store(&out2[x], b);
+        vx_store(&out3[x], c);
+    }
+    return x;
+}
+
 #endif  // CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
 
 CV_CPU_OPTIMIZATION_NAMESPACE_END