permitted handling of userdefined elementwise functions with non-iden…

…tity alpha, since this is needed when scaling for symmetric permutations
solomonik · Feb 10, 2016 · 2dde4d4 · 2dde4d4
1 parent e2c9991
commit 2dde4d4
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 8 deletions.
diff --git a/examples/dft.cxx b/examples/dft.cxx
@@ -47,6 +47,8 @@ int test_dft(int64_t n,
                std::complex<double> (0.0, 0.0), "ik");*/
   DFT["ik"] = .5*DFT["ij"]*IDFT["jk"];
 
+  Scalar< std::complex<double> > ss(wrld);
+  ss[""] = ((Function< std::complex<double>, std::complex<double>, std::complex<double> >)([](std::complex<double> a, std::complex<double> b){ return a+b; }))(DFT["ij"],DFT["ij"]);
 
   DFT.read_local(&np, &idx, &data);
   int pass = 1;

diff --git a/src/contraction/sym_seq_ctr.cxx b/src/contraction/sym_seq_ctr.cxx
@@ -248,18 +248,20 @@ printf("HERE1\n");
         }
         CTF_FLOPS_ADD(2*(imax-imin));
       } else {
-        ASSERT(0);
-        assert(0);
+        //ASSERT(0);
+        //assert(0);
         //printf("HERTE alpha = %d\n",*(int*)alpha);
         for (int i=imin; i<imax; i++){
           char tmp[sr_C->el_size];
-          sr_C->mul(A+offsets_A[0][i],
+          func->apply_f(A+offsets_A[0][i], 
+                        B+offsets_B[0][i], 
+                        tmp);
+          sr_C->mul(tmp,
                     alpha,
                     tmp);
-          func->acc_f(tmp, 
-                      B+offsets_B[0][i], 
-                      C+offsets_C[0][i],
-                      sr_C);
+          sr_C->add(tmp, 
+                    C+offsets_C[0][i], 
+                    C+offsets_C[0][i]);
         }
         CTF_FLOPS_ADD(3*(imax-imin));
       }

diff --git a/src/interface/fun_term.cxx b/src/interface/fun_term.cxx
@@ -104,7 +104,8 @@ namespace CTF_int {
       ASSERT(0);
       assert(0);
     }
-    contraction c(opA.parent, opA.idx_map, opB.parent, opB.idx_map, NULL, output.parent, output.idx_map, output.scale, func);
+    contraction c(opA.parent, opA.idx_map, opB.parent, opB.idx_map, output.sr->mulid(), output.parent, output.idx_map, output.scale, func);
+    //contraction c(opA.parent, opA.idx_map, opB.parent, opB.idx_map, NULL, output.parent, output.idx_map, output.scale, func);
     c.execute();
 //    if (scl != NULL) cdealloc(scl);
   }