diff --git a/src/qsched.c b/src/qsched.c
index 9f47dafb5d06ef8b7cce9508ec38d3d270a3104e..25a92611b173898be735b7d47cb649e9799c5c8a 100644
--- a/src/qsched.c
+++ b/src/qsched.c
@@ -865,7 +865,7 @@ void qsched_partition_compute_costs( struct qsched *s, idx_t *res_costs)
}
}
-void qsched_partition_build_nodelist(struct qsched *s, idx_t *nodelist, idx_t *noderef, idx_t *node_count_input, idx_t *res_costs, idx_t *pos_in_nodelist)
+void qsched_partition_build_nodelist(struct qsched *s, idx_t *nodelist, long long int *noderef, idx_t *node_count_input, idx_t *res_costs, idx_t *pos_in_nodelist)
{
idx_t node_count = *node_count_input;
struct res *r;
@@ -936,7 +936,7 @@ void qsched_partition_build_nodelist(struct qsched *s, idx_t *nodelist, idx_t *n
}
void qsched_partition_build_edgelist(struct qsched *s, idx_t **edge_vwgts, idx_t **edge_lists, idx_t *edge_counts, idx_t *edge_sizes,
- idx_t node_count, idx_t *noderef, idx_t *pos_in_nodelist )
+ idx_t node_count, long long int *noderef, idx_t *pos_in_nodelist )
{
int i, j, k, l;
struct task *t;
@@ -2341,6 +2341,7 @@ for(i = 0; i < count; i++)
task_data[3] = (int)(s->res[getindex(t->uses[j],s)].ID & 0xFFFFFFFF);
task_data[4] = sends_added;
+
/* Create the send task. */
send_task_id = tsched_addtask(&ts, task_type_send , 0 , task_data , 5 * sizeof(int) , s->res[getindex(t->uses[j],s)].size );
/* The send task needs to lock the resource. */
@@ -2473,6 +2474,7 @@ for(i = 0; i < count; i++)
task_data[3] = (int)(s->res[getindex(t->locks[j],s)].ID & 0xFFFFFFFF);
task_data[4] = sends_added;
+
/* Create the send task. */
send_task_id = tsched_addtask(&ts, task_type_send , 0 , task_data , 5 * sizeof(int) , s->res[getindex(t->locks[j],s)].size );
/* The send task needs to lock the resource. */
@@ -2505,7 +2507,8 @@ for(i = 0; i < count; i++)
task_data[2] = (int)(s->res[getindex(t->locks[j],s)].ID >> 32);
task_data[3] = (int)(s->res[getindex(t->locks[j],s)].ID & 0xFFFFFFFF);
task_data[4] = sends_added;
-
+
+
/* Create the send task. */
send_task_id = tsched_addtask(&ts, task_type_send , 0 , task_data , 5 * sizeof(int) , s->res[getindex(t->locks[j],s)].size );
/* The send task needs to lock the resource. */
@@ -2634,6 +2637,7 @@ for(i = 0; i < count; i++)
task_data[3] = (int)(s->res[getindex(t->uses[j],s)].ID & 0xFFFFFFFF);
task_data[4] = sends_added;
+
/* Create the send task. */
send_task_id = tsched_addtask(&ts, task_type_send , 0 , task_data , 5 * sizeof(int) , s->res[getindex(t->uses[j],s)].size );
/* The send task needs to lock the resource. */
@@ -2807,12 +2811,12 @@ void qsched_partition( struct qsched *s){
//Build a nodelist of the highest level hierarchical resources which are locked or used.
idx_t *nodelist;
idx_t node_count=0;
- idx_t *noderef;
+ long long int *noderef;
idx_t *pos_in_nodelist;
nodelist = (idx_t *) calloc( s->res_ranks[s->count_ranks], sizeof(idx_t) );
if(nodelist == NULL)
error("Failed to allocate nodelist");
- noderef = (idx_t *) calloc(s->res_ranks[s->count_ranks], sizeof(idx_t) );
+ noderef = (long long int *) calloc(s->res_ranks[s->count_ranks], sizeof(long long int) );
if(noderef == NULL)
error("Failed to allocate noderef");
pos_in_nodelist = (idx_t *) malloc(s->res_ranks[s->count_ranks] * sizeof(idx_t) );
@@ -2896,7 +2900,7 @@ for(i = 0; i < node_count; i++)
message("qsched_partition_build_edgelist took %lli (= %e) ticks\n", toc-tic, (float)(toc-tic));
#endif
- free(pos_in_nodelist);
+ // free(pos_in_nodelist);
//This might work but for safety I've made all nodes compute the edgelist anyway.
/*#if IDXTYPEWIDTH == 32
@@ -2972,14 +2976,15 @@ for(i = 0; i < node_count; i++)
objval[1] = 0;
objval[2] = 0;
idx_t *nodeIDs;
- nodeIDs = (idx_t *)malloc( sizeof(idx_t) * node_count );
+ nodeIDs = (idx_t *)calloc( node_count, sizeof(idx_t) );
if(nodeIDs == NULL)
error("Failed to allocate nodeIDs");
idx_t temp_count_ranks = s->count_ranks;
tic = getticks();
- if( METIS_PartGraphKway(&node_count, &one, edgelist_pos, edgelist_new, nodelist, NULL, edgelist_vwgt, &temp_count_ranks, NULL, NULL,options, objval, nodeIDs) != METIS_OK)
- error("Failed to partition\n");
-
+ if(s->count_ranks > 1) {
+ if( METIS_PartGraphKway(&node_count, &one, edgelist_pos, edgelist_new, nodelist, NULL, edgelist_vwgt, &temp_count_ranks, NULL, NULL,options, objval, nodeIDs) != METIS_OK)
+ error("Failed to partition\n");
+ }
toc = getticks();
message("METIS_PartGraphKway took %lli (= %e) ticks\n", toc-tic, (float)(toc-tic));
//TODO Check the costs.
@@ -3022,8 +3027,8 @@ for(i = 0; i < node_count; i++)
error("A local resource has no data associated with it.");
MPI_Isend(s->res[noderef[i]].data, s->res[noderef[i]].size, MPI_BYTE, nodeIDs[i], i, s->comm, &reqs[reqnr] );
reqnr++;
- s->res[noderef[i]].node = nodeIDs[i];
}
+ s->res[noderef[i]].node = nodeIDs[i];
}
}
@@ -3050,8 +3055,8 @@ for(i = 0; i < node_count; i++)
for(j = 0; j < s->res_ranks[s->count_ranks]; j++)
{
struct res *temp = &s->res[j];
- if(temp->num_lockers == 0 && temp->num_users == 0)
- continue;
+ // if(temp->num_lockers == 0 && temp->num_users == 0)
+ // continue;
struct res *parent;
int offset = temp->offset;
while(temp->parent != -1)
@@ -3747,6 +3752,18 @@ void qsched_enqueue ( struct qsched *s , struct task *t ) {
struct res *resource = &s->res[getindex(resID, s)];
int err;
// printf("Sending tag = %i\n", tag);
+ if(resource->data == NULL)
+ {
+ message("resource->parent = %lli", resource->parent);
+ message("from = %i", from);
+ message("t->node= %i", t->node);
+ if(resource->parent != -1)
+ {
+ struct res *temp = &s->res[getindex(resource->parent, s)];
+ message("parent->node = %i", temp->node);
+ }
+ error("We're sending a resource that has data = NULL, resource->ID = %lli and resource-node = %i", resource->ID, resource->node);
+ }
if ((err = MPI_Isend(resource->data, resource->size,
MPI_BYTE, to, tag, s->comm,
&t->req)) != MPI_SUCCESS) {