M/W, middle grain

src/Makefile

@@ -29,7 +29,8 @@ NP = 5
 MPISRC = masterworker.cpp mw_addslave.cpp hierarchical.cpp  \
          perf.cpp  sequential.cpp  tensormatrix_mpi.cpp      \
          utils.cpp  utils_parall.cpp profiling.cpp mw_combined.cpp \
-	 masterworker2.cpp mw_addslave2.cpp hierarchical2.cpp 
+	 masterworker2.cpp mw_addslave2.cpp hierarchical2.cpp  \
+	 masterworker3.cpp
 
 MPIOBJ= $(MPISRC:.cpp=.o)
 

src/products.h

@@ -8,6 +8,7 @@ namespace gi = GiNaC;
 
 // internal (sequential) routines
 gi::ex one_level1_product( tensor3D_t*, int, int, int, int );
+gi::ex one_level1_product( tensor3D_t*, int, int, int );
 gi::ex one_level1_product( tensor3D_t*, int, int );
 gi::ex one_level2_product( tensor3D_t*, int, int, int, int, int );
 gi::ex two_level1_product( tensor3D_t*, int, int, int );

src/sequential.cpp

@@ -341,6 +341,91 @@ gi::ex one_level1_product( tensor3D_t* T, int size, int a4 ){
     return Tens;
 }
 
+gi::ex one_level1_product( tensor3D_t* T, int size, int a4, int a2 ){
+
+    gi::ex Tens = 0;
+    gi::ex Ti0, Ti1, Ti2;
+    gi::ex W1, W2, W3, W4, W5, W6, W7;
+    gi::ex Z1, Z2, Z6, t5, tE, t1, t12, t123, t126, t13, t134, t14, t16, t2, t23, t24, t26, t3, t4, X7Y5;
+    gi::ex TE, T1, T2, T3, T4, T5, T12, T13, T14, T16, T23, T24, T26, T123, T126, T134;
+    const char timerB[] = "B";
+    
+	int a1, a3, a5, a6;
+    int A1, A2, A3, A4, A5, A6;
+    TE = T1 = T2 = T3 = T4 = T5 = T12 = T13 = T14 = T16 = T23 = T24 = T26 = T123 = T126 = T134 = 0;
+    Ti0 = Ti1 = 0;
+    
+    int N = size/2;
+    
+    A4 = a4 + N;
+    A2 = a2 + N;
+    for( a6 = 0 ; a6 < N ; a6++ ) {
+        A6 = a6 + N;
+        
+        W1 = (*T)[a4][a2][a6];
+        W2 = (*T)[a4][A2][a6];
+        W3 = (*T)[a4][a2][A6];
+        W4 = (*T)[A4][A2][a6];
+        W5 = (*T)[a4][A2][A6];
+        W6 = (*T)[A4][a2][A6];
+        W7 = (*T)[A4][A2][A6];
+        
+        Ti1 = 0;
+        for( a1 = 0 ; a1 < N ; a1++ ) {
+            A1 = a1 + N;
+            Ti0 = TE = T12 = T13 = T14 = T16 = T23 = T24 = T26 = T1 = T2 = T3 = T4 = T5 = T123 = T126 = T134 = 0;
+            for( a5 = 0 ; a5 < N ; a5++ ) {
+                A5 = a5 + N;
+                Z1 = (*T)[a1][a5][a6];
+                Z2 = (*T)[A1][a5][a6];
+                Z6 = (*T)[A1][a5][A6];
+                t5 = W3*(*T)[a1][A5][a6];
+                tE = W4*(*T)[A1][A5][A6];
+                t1 = W3*Z2;
+                t13 = t1;
+                t2 = W5*Z1;
+                t23 = t2;
+                t3 = W3*Z1;
+                t4 = W6*Z1;
+                t12 = W5*Z2;
+                t14 = W6*Z2;
+                t134 = t14 ;
+                t16 = W1*Z6;
+                t24 = W7*Z1;
+                t26 = W2*(*T)[a1][a5][A6];
+                t123 = W5*Z2;
+                t126 = W2*Z6;
+                
+                for( a3 = 0 ; a3 < N ; a3++ ) {
+                    A3 = a3 + N;
+                    TE = TE + tE*(*T)[a1][a2][a3]*(*T)[a4][a5][A3];
+                    T5 = T5 + t5*(*T)[A1][A2][A3]*(*T)[A4][a5][a3];
+                    X7Y5 = (*T)[a1][A2][A3]*(*T)[A4][A5][a3];
+                    T1 = T1 + t1*X7Y5;
+                    T16 = T16 + t16*X7Y5;
+                    T2 = T2 + t2*(*T)[A1][a2][A3]*(*T)[A4][A5][a3];
+                    T3 = T3 + t3*(*T)[A1][A2][a3]*(*T)[A4][A5][A3];
+                    T4 = T4 + t4*(*T)[A1][A2][A3]*(*T)[a4][A5][a3];
+                    T12 = T12 + t12*(*T)[a1][a2][A3]*(*T)[A4][A5][a3];
+                    T13 = T13 + t13*(*T)[a1][A2][a3]*(*T)[A4][A5][A3];
+                    T14 = T14 + t14*(*T)[a1][A2][A3]*(*T)[a4][A5][a3];
+                    T23 = T23 + t23*(*T)[A1][a2][a3]*(*T)[A4][A5][A3];
+                    T24 = T24 + t24*(*T)[A1][a2][A3]*(*T)[a4][A5][a3];
+                    T26 = T26 + t26*(*T)[A1][a2][A3]*(*T)[A4][A5][a3];
+                    T123 = T123 + t123*(*T)[a1][a2][a3]*(*T)[A4][A5][A3];
+                    T126 = T126 + t126*(*T)[a1][a2][A3]*(*T)[A4][A5][a3];
+                    T134 = T134 + t134*(*T)[a1][A2][a3]*(*T)[a4][A5][A3];
+                }
+                Ti0 += ( 4*(TE+T12+T13+T14+T16+T23+T24+T26 - (T1 + T2 + T3 + T4 + T5 +T123 + T126 + T134)) );
+            }
+            Ti1 += Ti0;
+        }
+        Tens += Ti1;
+    }
+        
+    return Tens;
+}
+
 gi::ex one_level2_product( tensor3D_t* T, int size, int a4, int a2, int a1, int a6 ){
 
     int a3, a5;

src/tensormatrix.h

@@ -25,6 +25,7 @@ gi::ex multiply_2levels( tensor3D_t&, int );
 // parallel
 gi::ex multiply_1level_mw( tensor3D_t&, int );
 gi::ex multiply_1level_mw2( tensor3D_t&, int );
+gi::ex multiply_1level_mw3( tensor3D_t&, int );
 gi::ex multiply_1level_mw_addslave( tensor3D_t&, int );
 gi::ex multiply_1level_mw_addslave2( tensor3D_t&, int );
 gi::ex multiply_2levels_mw_hierarch( tensor3D_t&, int );

src/tensormatrix_mpi.cpp

@@ -28,7 +28,8 @@ namespace gi = GiNaC;
    tensormatrix_mpi [N] [Function name] [Nb of foremen]
    Function names being: 
    - M/m: Master-Worker -> multiply_1level_mw
-   - n/n: Master-Worker, coarser grain -> multiply_1level_mw2
+   - n/N: Master-Worker, coarser grain -> multiply_1level_mw2
+   - o/O: Master-Worker, middle grain -> multiply_1level_mw3
    - A/a: Master-Worker, addition on a slave -> multiply_1level_mw_addslave
    - B/b: Master-Worker, coarser grain, addition on a slave -> multiply_1level_mw_addslave2
    - H/h: Hierarchical master-worker -> multiply_1level_mw_hierarch
@@ -97,6 +98,10 @@ int main( int argc, char** argv ){
             case 'n':
                 tostart = 'n';
                 break;
+            case 'O':
+            case 'o':
+                tostart = 'o';
+                break;
             case 'A':
             case 'a':
                 tostart = 'a';
@@ -154,6 +159,9 @@ int main( int argc, char** argv ){
     case 'n':
         Tpara = multiply_1level_mw2( T, N );
         break;
+    case 'o':
+        Tpara = multiply_1level_mw3( T, N );
+        break;
     case 'a':
         Tpara = multiply_1level_mw_addslave( T, N );
         break;