Merge branch 'master' of ../ctf-g

examples/ccsdt_map_test.cxx

@@ -25,15 +25,17 @@ int ccsdt_map_test(int const     n,
   MPI_Comm_rank(MPI_COMM_WORLD, &rank);
   MPI_Comm_size(MPI_COMM_WORLD, &num_pes);
 
-  int shapeAS6[] = {NS,AS,NS,NS,NS,NS};
+  int shapeAS6[] = {AS,AS,NS,AS,AS,NS};
   int mmmnnn[] = {m,m,m,n,n,n};
+  int shapeNS4[] = {NS,NS,NS,NS};
+  int mnmn[] = {m,n,m,n};
 
   //* Creates distributed tensors initialized with zeros
-  CTF_Matrix W(n, n, NS, dw, "W", 1);
+  CTF_Tensor W(4, mnmn, shapeNS4, dw, "W", 1);
   CTF_Tensor T(6, mmmnnn, shapeAS6, dw, "T", 1);
   CTF_Tensor Z(6, mmmnnn, shapeAS6, dw, "Z", 1);
 
-  Z["abcijk"] += W["mj"]*T["abcimk"];
+  Z["612745"] += W["6307"]*T["012345"];
 
   return 1;
 } 

src/dist_tensor/dist_tensor_map.cxx

@@ -156,8 +156,8 @@ int dist_tensor<dtype>::map_tensor_pair( const int      tid_A,
   uint64_t size;
   uint64_t min_size = UINT64_MAX;
   /* Attempt to map to all possible permutations of processor topology */
-  for (i=global_comm.rank; i<2*(int)topovec.size(); i+=global_comm.np){
-//  for (i=global_comm.rank*topovec.size(); i<(int)topovec.size(); i++){
+//  for (i=global_comm.rank; i<2*(int)topovec.size(); i+=global_comm.np){
+  for (i=global_comm.rank*topovec.size(); i<(int)topovec.size(); i++){
     clear_mapping(tsr_A);
     clear_mapping(tsr_B);
     set_padding(tsr_A);
@@ -544,7 +544,7 @@ int dist_tensor<dtype>::
   if (!check_self_mapping(tid_B, idx_B))
     pass = 0;
   if (pass == 0)
-    DPRINTF(3,"failed confirmation here\n");
+    DPRINTF(4,"failed confirmation here\n");
 
   for (i=0; i<ndim_tot; i++){
     iA = idx_arr[2*i];
@@ -552,7 +552,7 @@ int dist_tensor<dtype>::
     if (iA != -1 && iB != -1) {
       if (!comp_dim_map(&tsr_A->edge_map[iA], &tsr_B->edge_map[iB])){
         pass = 0;
-        DPRINTF(3,"failed confirmation here i=%d\n",i);
+        DPRINTF(4,"failed confirmation here i=%d\n",i);
       }
     }
     if (iA != -1) {
@@ -1186,7 +1186,7 @@ int dist_tensor<dtype>::map_tensors(CTF_ctr_type_t const *      type,
       if (comm_vol < bcomm_vol) {
         bcomm_vol = comm_vol;
         bmemuse = memuse;
-        bnvirt = nvirt;
+        bnvirt = 1;//nvirt;
         btopo = 6*i+j;      
       }  
       delete sctr;

src/dist_tensor/dt_aux_permute.hxx

@@ -2546,9 +2546,6 @@ int cyclic_reshuffle(int const          ndim,
   old_nvirt = 1;
   old_np = 1;
   new_np = 1;
-  if (ord_glb_comm.rank == 0){
-    DPRINTF(3,"is_cyclic = %d, was_cyclic = %d\n",is_cyclic,was_cyclic);
-  }
   idx_lyr = ord_glb_comm.rank;
   for (i=0; i<ndim; i++) {
     buf_lda[i] = nbuf;
@@ -2953,7 +2950,7 @@ void block_reshuffle(int const        ndim,
         blk_idx += ( idx[i] + new_rank[i]*new_virt_dim[i])                 *phase_lda[i];
         prc_idx += ((idx[i] + new_rank[i]*new_virt_dim[i])/old_virt_dim[i])*old_pe_lda[i];
       }
-      DPRINTF(3,"proc %d receiving blk %d (loc %d, size "PRId64") from proc %d\n", 
+      DPRINTF(4,"proc %d receiving blk %d (loc %d, size "PRId64") from proc %d\n", 
               glb_comm.rank, blk_idx, loc_idx, blk_sz, prc_idx);
       MPI_Irecv(tsr_cyclic_data+loc_idx*blk_sz, blk_sz*sizeof(dtype), 
                 MPI_CHAR, prc_idx, blk_idx, glb_comm.cm, reqs+loc_idx);
@@ -2980,7 +2977,7 @@ void block_reshuffle(int const        ndim,
         blk_idx += ( idx[i] + old_rank[i]*old_virt_dim[i])                 *phase_lda[i];
         prc_idx += ((idx[i] + old_rank[i]*old_virt_dim[i])/new_virt_dim[i])*new_pe_lda[i];
       }
-      DPRINTF(3,"proc %d sending blk %d (loc %d) to proc %d\n", 
+      DPRINTF(4,"proc %d sending blk %d (loc %d) to proc %d\n", 
               glb_comm.rank, blk_idx, loc_idx, prc_idx);
       MPI_Isend(tsr_data+loc_idx*blk_sz, blk_sz*sizeof(dtype), 
                 MPI_CHAR, prc_idx, blk_idx, glb_comm.cm, reqs+num_new_virt+loc_idx);