George, My colleague was working on your ompi-topo bitbucket repository but it was not completed. But he found bugs in your patch attached in your previous mail and created the fixing patch. See the attached patch, which is a patch against Open MPI trunk + your patch.
His test programs are also attached. test_1 and test_2 can run with nprocs=5, and test_3 and test_4 can run with nprocs>=3. Though I'm not sure about the contents of the patch and the test programs, I can ask him if you have any questions. Regards, Takahiro Kawashima, MPI development team, Fujitsu > WHAT: Support for MPI 2.2 dist_graph > > WHY: To become [almost entierly] MPI 2.2 compliant > > WHEN: Monday July 1st > > As discussed during the last phone call, a missing functionality of the MPI > 2.2 standard (the distributed graph topology) is ready for prime-time. The > attached patch provide a minimal version (no components supporting > reordering), that will complete the topology support in Open MPI. > > It is somehow a major change compared with what we had before and it reshape > the way we deal with topologies completely. Where our topologies were mainly > storage components (they were not capable of creating the new communicator as > an example), the new version is built around a [possibly] common > representation (in mca/topo/topo.h), but the functions to attach and retrieve > the topological information are specific to each component. As a result the > ompi_create_cart and ompi_create_graph functions become useless and have been > removed. > > In addition to adding the internal infrastructure to manage the topology > information, it updates the MPI interface, and the debuggers support and > provides all Fortran interfaces. From a correctness point of view it passes > all the tests we have in ompi-tests for the cart and graph topology, and some > tests/applications for the dist_graph interface. > > I don't think there is a need for a long wait on this one so I would like to > propose a short deadline, a week from now on Monday July 1st. A patch based > on Open MPI trunk r28670 is attached below.
diff -u -r openmpi-trunk2/ompi/communicator/comm.c openmpi-trunk/ompi/communicator/comm.c
--- openmpi-trunk2/ompi/communicator/comm.c 2013-06-25 16:36:42.000000000 +0900
+++ openmpi-trunk/ompi/communicator/comm.c 2013-06-25 18:52:17.000000000 +0900
@@ -1446,6 +1446,8 @@
opal_output(0," topo-cart,");
if ( OMPI_COMM_IS_GRAPH(comm))
opal_output(0," topo-graph");
+ if ( OMPI_COMM_IS_DIST_GRAPH(comm))
+ opal_output(0," topo-dist-graph");
opal_output(0,"\n");
if (OMPI_COMM_IS_INTER(comm)) {
diff -u -r openmpi-trunk2/ompi/communicator/communicator.h openmpi-trunk/ompi/communicator/communicator.h
--- openmpi-trunk2/ompi/communicator/communicator.h 2013-06-25 16:36:42.000000000 +0900
+++ openmpi-trunk/ompi/communicator/communicator.h 2013-06-25 17:19:31.000000000 +0900
@@ -46,7 +46,7 @@
#define OMPI_COMM_INVALID 0x00000020
#define OMPI_COMM_CART 0x00000100
#define OMPI_COMM_GRAPH 0x00000200
-#define OMPI_COMM_DIST_GRAPH 0x00000300
+#define OMPI_COMM_DIST_GRAPH 0x00000400
#define OMPI_COMM_PML_ADDED 0x00001000
#define OMPI_COMM_EXTRA_RETAIN 0x00004000
--- openmpi-trunk2/ompi/mca/topo/base/topo_base_dist_graph_create_adjacent.c 2013-06-25 16:36:38.000000000 +0900
+++ openmpi-trunk/ompi/mca/topo/base/topo_base_dist_graph_create_adjacent.c 2013-06-26 11:44:40.000000000 +0900
@@ -47,18 +47,18 @@
topo->out = topo->outw = NULL;
topo->indegree = indegree;
topo->outdegree = outdegree;
- topo->weighted = !((sourceweights == MPI_UNWEIGHTED) || (destweights == MPI_UNWEIGHTED));
+ topo->weighted = !((MPI_UNWEIGHTED == sourceweights) && (MPI_UNWEIGHTED == destweights));
topo->in = (int*)malloc(sizeof(int) * topo->indegree);
if( NULL == topo->in ) {
goto bail_out;
}
memcpy( topo->in, sources, sizeof(int) * topo->indegree );
- if( sourceweights == MPI_UNWEIGHTED ) {
+ if( MPI_UNWEIGHTED != sourceweights ) {
topo->inw = (int*)malloc(sizeof(int) * topo->indegree);
if( NULL == topo->inw ) {
goto bail_out;
}
- memcpy( topo->in, sourceweights, sizeof(int) * topo->indegree );
+ memcpy( topo->inw, sourceweights, sizeof(int) * topo->indegree );
}
topo->out = (int*)malloc(sizeof(int) * topo->outdegree);
if( NULL == topo->out ) {
@@ -66,12 +66,12 @@
}
memcpy( topo->out, destinations, sizeof(int) * topo->outdegree );
topo->outw = NULL;
- if( destweights == MPI_UNWEIGHTED ) {
+ if( MPI_UNWEIGHTED != destweights ) {
topo->outw = (int*)malloc(sizeof(int) * topo->outdegree);
if( NULL == topo->outw ) {
goto bail_out;
}
- memcpy( topo->out, destweights, sizeof(int) * topo->outdegree );
+ memcpy( topo->outw, destweights, sizeof(int) * topo->outdegree );
}
(*newcomm)->c_topo = module;
(*newcomm)->c_topo->mtc.dist_graph = topo;
@@ -82,15 +82,16 @@
bail_out:
if( NULL != topo->in ) free(topo->in);
- if( sourceweights == MPI_UNWEIGHTED ) {
+ if( MPI_UNWEIGHTED != sourceweights ) {
if( NULL != topo->inw ) free(topo->inw);
}
if( NULL != topo->out ) free(topo->out);
- if( destweights == MPI_UNWEIGHTED ) {
+ if( MPI_UNWEIGHTED != destweights ) {
if( NULL != topo->outw ) free(topo->outw);
}
if( NULL != topo ) {
free(topo);
}
+ ompi_comm_free(newcomm);
return err;
}
diff -u -r openmpi-trunk2/ompi/mca/topo/base/topo_base_dist_graph_create.c openmpi-trunk/ompi/mca/topo/base/topo_base_dist_graph_create.c
--- openmpi-trunk2/ompi/mca/topo/base/topo_base_dist_graph_create.c 2013-06-25 16:36:38.000000000 +0900
+++ openmpi-trunk/ompi/mca/topo/base/topo_base_dist_graph_create.c 2013-06-25 18:39:36.000000000 +0900
@@ -102,7 +102,7 @@
position = pos[targets[index]].in + idx[targets[index]].in;
if( MPI_UNWEIGHTED != weights ) {
position *= 2;
- rin[position + 1] = weights[i];
+ rin[position + 1] = weights[index];
}
rin[position + 0] = nodes[i];
idx[targets[index]].in++;
@@ -130,19 +130,15 @@
topo->in = (int*)malloc(sizeof(int) * topo->indegree);
topo->out = (int*)malloc(sizeof(int) * topo->outdegree);
if( (NULL == topo->in) || (NULL == topo->out) ) {
- if( NULL != topo->in ) free(topo->in);
- if( NULL != topo->out ) free(topo->out);
err = OMPI_ERR_OUT_OF_RESOURCE;
goto bail_out;
}
topo->inw = NULL;
topo->outw = NULL;
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
topo->inw = (int*)malloc(sizeof(int) * topo->indegree);
topo->outw = (int*)malloc(sizeof(int) * topo->outdegree);
if( (NULL == topo->inw) || (NULL == topo->outw) ) {
- if( NULL != topo->inw ) free(topo->inw);
- if( NULL != topo->outw ) free(topo->outw);
err = OMPI_ERR_OUT_OF_RESOURCE;
goto bail_out;
}
@@ -153,7 +149,7 @@
int position;
if( 0 != (count = cnt[i].in) ) {
position = pos[i].in;
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
count *= 2; /* don't forget the weights */
position *= 2;
}
@@ -164,7 +160,7 @@
}
if( 0 != (count = cnt[i].out) ) {
position = pos[i].out;
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
count *= 2; /* don't forget the weights */
position *= 2;
}
@@ -181,7 +177,7 @@
*/
count = topo->indegree;
temp = topo->in;
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
count *= 2; /* don't forget the weights */
temp = (int*)malloc(count*sizeof(int)); /* Allocate an array big enough to hold the edges and their weights */
if( NULL == temp ) {
@@ -195,15 +191,15 @@
MPI_ANY_SOURCE, MCA_TOPO_BASE_TAG_DIST_EDGE_IN,
comm, &status ));
how_much = status._ucount / int_size;
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
for( j = 0; j < ((int)how_much >> 1); j++, current_pos++ ) {
- topo->in[current_pos] = temp[2 * j + 0];
- topo->inw[current_pos] = temp[2 * j + 1];
+ topo->in[current_pos] = temp[2 * j + 0 + (count - left_over)];
+ topo->inw[current_pos] = temp[2 * j + 1 + (count - left_over)];
}
}
left_over -= how_much;
}
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
free(temp);
}
@@ -213,7 +209,7 @@
*/
count = topo->outdegree;
temp = topo->out;
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
count *= 2; /* don't forget the weights */
temp = (int*)malloc(count*sizeof(int)); /* Allocate an array big enough to hold the edges and their weights */
if( NULL == temp ) {
@@ -228,15 +224,15 @@
comm, &status ));
how_much = status._ucount / int_size;
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
for( j = 0; j < ((int)how_much >> 1); j++, current_pos++ ) {
- topo->out[current_pos] = temp[2 * j + 0];
- topo->outw[current_pos] = temp[2 * j + 1];
+ topo->out[current_pos] = temp[2 * j + 0 + (count - left_over)];
+ topo->outw[current_pos] = temp[2 * j + 1 + (count - left_over)];
}
}
left_over -= how_much;
}
- if (weights != MPI_UNWEIGHTED) {
+ if (MPI_UNWEIGHTED != weights) {
free(temp);
}
@@ -255,6 +251,18 @@
free(cnt);
}
if( NULL != topo ) {
+ if ( NULL != topo->in ) {
+ free(topo->in);
+ }
+ if ( NULL != topo->out ) {
+ free(topo->out);
+ }
+ if ( NULL != topo->inw ) {
+ free(topo->inw);
+ }
+ if ( NULL != topo->outw ) {
+ free(topo->outw);
+ }
free(topo);
}
return err;
diff -u -r openmpi-trunk2/ompi/mca/topo/base/topo_base_dist_graph_neighbors.c openmpi-trunk/ompi/mca/topo/base/topo_base_dist_graph_neighbors.c
--- openmpi-trunk2/ompi/mca/topo/base/topo_base_dist_graph_neighbors.c 2013-06-25 16:36:38.000000000 +0900
+++ openmpi-trunk/ompi/mca/topo/base/topo_base_dist_graph_neighbors.c 2013-06-26 08:44:25.000000000 +0900
@@ -28,17 +28,17 @@
if (!OMPI_COMM_IS_DIST_GRAPH(comm)) {
return OMPI_ERR_NOT_FOUND;
- } else if (maxindegree < dg->indegree || maxoutdegree < dg->outdegree) {
+ } else if (maxindegree > dg->indegree || maxoutdegree > dg->outdegree) {
return OMPI_ERR_BAD_PARAM;
}
- for (i = 0; i < dg->indegree; ++i) {
+ for (i = 0; (i < dg->indegree) && (i < maxindegree); ++i) {
sources[i] = dg->in[i];
if (NULL != dg->inw) {
sourceweights[i] = dg->inw[i];
}
}
- for (i = 0; i < dg->outdegree; ++i) {
+ for (i = 0; (i < dg->outdegree) && (i < maxoutdegree); ++i) {
destinations[i] = dg->out[i];
if (NULL != dg->outw) {
destweights[i] = dg->outw[i];
diff -u -r openmpi-trunk2/ompi/mpi/c/dist_graph_create_adjacent.c openmpi-trunk/ompi/mpi/c/dist_graph_create_adjacent.c
--- openmpi-trunk2/ompi/mpi/c/dist_graph_create_adjacent.c 2013-06-25 16:36:39.000000000 +0900
+++ openmpi-trunk/ompi/mpi/c/dist_graph_create_adjacent.c 2013-06-26 11:49:22.000000000 +0900
@@ -63,18 +63,19 @@
}
comm_size = ompi_comm_size(comm_old);
for (i = 0; i < indegree; ++i) {
- if (sources[i] < 0 || sources[i] >= comm_size ||
- sourceweights[i] < 0) {
+ if (sources[i] < 0 || sources[i] >= comm_size) {
+ return OMPI_ERRHANDLER_INVOKE(comm_old, MPI_ERR_ARG,
+ FUNC_NAME);
+ } else if (MPI_UNWEIGHTED != sourceweights && sourceweights[i] < 0) {
return OMPI_ERRHANDLER_INVOKE(comm_old, MPI_ERR_ARG,
"MPI_Dist_create_graph adjacent 3");
}
}
for (i = 0; i < outdegree; ++i) {
- if (destinations[i] < 0 || destinations[i] >= comm_size ||
- destweights[i] < 0) {
- opal_output(0, "MPI_DCGA %d, %d",
- destinations[i],
- destweights[i]);
+ if (destinations[i] < 0 || destinations[i] >= comm_size) {
+ return OMPI_ERRHANDLER_INVOKE(comm_old, MPI_ERR_ARG,
+ FUNC_NAME);
+ } else if (MPI_UNWEIGHTED != destweights && destweights[i] < 0) {
return OMPI_ERRHANDLER_INVOKE(comm_old, MPI_ERR_ARG,
"MPI_Dist_create_graph_adjacent 4\n");
}
dist-graph-test.tar.gz
Description: dist-graph-test.tar.gz
