A version of Gradient working for merge

llvm/test/VISC/parboil/benchmarks/pipeline/src/Gradient/main.cc

@@ -177,7 +177,8 @@ void horizontal(float* Is, size_t bytesIs,
     //G[gloc] = Gx*Gx + Gy*Gy;
   }
   Gx[gloc] = Val;
-  __visc__return(m);
+  //__visc__return(m);
+  __visc__return(bytesSx);
 }
 
 void WrapperHorizontal(float* Is, size_t bytesIs,
@@ -251,7 +252,8 @@ void vertical(float* Is, size_t bytesIs,
     //G[gloc] = Gx*Gx + Gy*Gy;
   }
   Gy[gloc] = Val;
-  __visc__return(n);
+  //__visc__return(n);
+  __visc__return(bytesSy);
 }
 
 void WrapperVertical(float* Is, size_t bytesIs,
@@ -280,7 +282,8 @@ void WrapperVertical(float* Is, size_t bytesIs,
 void squareRoot(float* Gx, size_t bytesGx,
               float* Gy, size_t bytesGy,
               float* G, size_t bytesG,
-              long m, long n) {
+              long m, long n,
+              size_t bytesSx, size_t bytesSy) {
 
   __visc__hint(visc::DEVICE);
   __visc__attributes(2, Gx, Gy, 1, G);
@@ -294,13 +297,15 @@ void squareRoot(float* Gx, size_t bytesGx,
     G[index] = __visc__sqrt(Gx[index]*Gx[index]+Gy[index]*Gy[index]);
   }
 
-  __visc__return(m);
+  //__visc__return(m);
+  __visc__return(bytesG);
 }
 
 void WrapperSquareRoot(float* Gx, size_t bytesGx,
             float* Gy, size_t bytesGy,
             float* G, size_t bytesG,
-            long m, long n) {
+            long m, long n,
+            size_t bytesSx, size_t bytesSy) {
 
   __visc__hint(visc::CPU_TARGET);
   __visc__attributes(2,  Gx, Gy, 1, G);
@@ -314,6 +319,8 @@ void WrapperSquareRoot(float* Gx, size_t bytesGx,
   __visc__bindIn(Node, 5, 5, 0); // Bind bytesG
   __visc__bindIn(Node, 6, 6, 0); // Bind m
   __visc__bindIn(Node, 7, 7, 0); // Bind n
+  __visc__bindIn(Node, 8, 8, 0); // Bind dummy 1
+  __visc__bindIn(Node, 9, 9, 0); // Bind dummy 2
 
   __visc__bindOut(Node, 0, 0, 0); // bind output m
 
@@ -355,18 +362,22 @@ void Gradient(float* Is, size_t bytesIs,// 0
 
   // Linear combination
   __visc__bindIn(SNode, 6, 0, 1); // Bind Gx
-  __visc__bindIn(SNode, 7, 1, 1); // Bind bytesGx
+  __visc__bindIn(SNode, 7, 1, 1); // Bind bytesGx [Alt: Pass as edge from horizontal]
   __visc__bindIn(SNode, 8, 2, 1); // Bind Gy
-  __visc__bindIn(SNode, 9, 3, 1); // Bind bytesGy
+  __visc__bindIn(SNode, 9, 3, 1); // Bind bytesGy [Alt: Pass as edge from vertical]
   __visc__bindIn(SNode, 10, 4, 1); // Bind G
   __visc__bindIn(SNode, 11, 5, 1); // Bind bytesG
-  //__visc__bindIn(SNode, 12, 6, 0); // Bind m [Pass as edge from dilate]
-  //__visc__bindIn(SNode, 13, 7, 0); // Bind n [Pass as edge from erode]
+  __visc__bindIn(SNode, 12, 6, 1); // Bind m [Alt: Pass as edge from horizontal]
+  __visc__bindIn(SNode, 13, 7, 1); // Bind n [Alt: Pass as edge from vertical]
 
-  __visc__edge(HNode, SNode, 0, 0, 6, 1); // Get m from dilate
-  __visc__edge(VNode, SNode, 0, 0, 7, 1); // Get n from erode
+  //__visc__edge(HNode, SNode, 0, 0, 6, 1); // Get m from horizontal
+  //__visc__edge(VNode, SNode, 0, 0, 7, 1); // Get n from vertical
+  //__visc__edge(HNode, SNode, 0, 0, 1, 1); // Get bytesGx from horizontal
+  //__visc__edge(VNode, SNode, 0, 0, 3, 1); // Get bytesGy from vertical
+  __visc__edge(HNode, SNode, 0, 0, 8, 1); // Get bytes from horizontal
+  __visc__edge(VNode, SNode, 0, 0, 9, 1); // Get bytesGy from vertical
 
-  __visc__bindOut(SNode, 0, 0, 1); // bind output m
+  __visc__bindOut(SNode, 0, 0, 1); // bind output bytesG
 
 }