This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via f46354acff66cb0cf5ccc134399d9f5b1bc46b31 (commit) via 7a9210fff1f9545215bea4a2a67161e88c8dcab9 (commit) via 687dd1dc8d9d64cdda9926d48d75ebfc70b81301 (commit) via 0f0f20a601889ae480958bc95112d0eee686e822 (commit) from b270ae2b734be3e34e1906cfa9cb8e56a89d7b53 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/f46354acff66cb0cf5ccc134399d9f5b1b... commit f46354acff66cb0cf5ccc134399d9f5b1bc46b31 Author: Pavan Balaji <[email protected]> Date: Wed Jan 29 15:14:44 2014 -0600 Bug-fix: deal with wrap around PMI process mapping strings If the PMI process mapping string wraps around to node 0, we were creating a bad node list of which processes are local and which are not. This patch provides a hacky fix for this case by only repeating the part of the PMI mapping string from the point where it wrapped around. The patch is hacky because it assumes that seeing a start node ID of 0 means a wrap around. This is not necessarily true. A user-defined node list can use the node ID 0 without actually creating a wrap around. The reason this patch still works in this case is because Hydra creates a new node list starting from node ID 0 for user-specified nodes during MPI_Comm_spawn{_multiple}. If a different process manager searches for allocated nodes in the user-specified list, this patch will break. Fixes #2007. Signed-off-by: Ken Raffenetti <[email protected]> diff --git a/src/mpid/ch3/src/mpid_vc.c b/src/mpid/ch3/src/mpid_vc.c index fc2c970..b9289fe 100644 --- a/src/mpid/ch3/src/mpid_vc.c +++ b/src/mpid/ch3/src/mpid_vc.c @@ -1085,6 +1085,7 @@ static int populate_ids_from_mapping(char *mapping, int *num_nodes, MPIDI_PG_t * int rank; int block, block_node, node_proc; int *tmp_rank_list, i; + int found_wrap; MPIU_CHKLMEM_DECL(1); *did_map = 1; /* reset upon failure */ @@ -1096,8 +1097,25 @@ static int populate_ids_from_mapping(char *mapping, int *num_nodes, MPIDI_PG_t * MPIU_ERR_CHKINTERNAL(mt != VECTOR_MAPPING, mpi_errno, "unsupported mapping type"); /* allocate nodes to ranks */ + found_wrap = 0; for (rank = 0;;) { - for (block = 0; block < nblocks; block++) { + /* FIXME: The patch is hacky because it assumes that seeing a + * start node ID of 0 means a wrap around. This is not + * necessarily true. A user-defined node list can, in theory, + * use the node ID 0 without actually creating a wrap around. + * The reason this patch still works in this case is because + * Hydra creates a new node list starting from node ID 0 for + * user-specified nodes during MPI_Comm_spawn{_multiple}. If + * a different process manager searches for allocated nodes in + * the user-specified list, this patch will break. */ + + /* If we found that the blocks wrap around, repeat loops + * should only start at node id 0 */ + for (block = 0; found_wrap && mb[block].start_id; block++); + + for (; block < nblocks; block++) { + if (mb[block].start_id == 0) + found_wrap = 1; for (block_node = 0; block_node < mb[block].count; block_node++) { for (node_proc = 0; node_proc < mb[block].size; node_proc++) { pg->vct[rank].node_id = mb[block].start_id + block_node; http://git.mpich.org/mpich.git/commitdiff/7a9210fff1f9545215bea4a2a67161e88c... commit 7a9210fff1f9545215bea4a2a67161e88c8dcab9 Author: Ken Raffenetti <[email protected]> Date: Wed Jan 29 17:19:16 2014 -0600 handle when qsort is not available in vc setup When qsort is not available, don't define comparision function and fallback to simple insertion sort implementation. In the future, a more general function with fallback should be added in MPL so it can be used in other cases like comm_split. Refs #2007 Signed-off-by: Pavan Balaji <[email protected]> diff --git a/src/mpid/ch3/src/mpid_vc.c b/src/mpid/ch3/src/mpid_vc.c index acec520..fc2c970 100644 --- a/src/mpid/ch3/src/mpid_vc.c +++ b/src/mpid/ch3/src/mpid_vc.c @@ -1056,6 +1056,7 @@ done: #endif +#if defined HAVE_QSORT static int compare_ints(const void *orig_x, const void *orig_y) { int x = *((int *) orig_x); @@ -1068,6 +1069,7 @@ static int compare_ints(const void *orig_x, const void *orig_y) else return 1; } +#endif #undef FUNCNAME #define FUNCNAME populate_ids_from_mapping @@ -1120,7 +1122,27 @@ break_out: #if defined HAVE_QSORT qsort(tmp_rank_list, pg->size, sizeof(int), compare_ints); #else -#error "qsort not available" + /* fall through to insertion sort if qsort is unavailable/disabled */ + { + int j, tmp; + + for (i = 1; i < pg->size; ++i) { + tmp = tmp_rank_list[i]; + j = i - 1; + while (1) { + if (tmp_rank_list[j] > tmp) { + tmp_rank_list[j+1] = tmp_rank_list[j]; + j = j - 1; + if (j < 0) + break; + } + else { + break; + } + } + tmp_rank_list[j+1] = tmp; + } + } #endif *num_nodes = 1; http://git.mpich.org/mpich.git/commitdiff/687dd1dc8d9d64cdda9926d48d75ebfc70... commit 687dd1dc8d9d64cdda9926d48d75ebfc70b81301 Author: Pavan Balaji <[email protected]> Date: Wed Jan 29 13:20:41 2014 -0600 Improve PMI_process_mapping parsing. The original PMI process mapping parsing code had a number of assumptions that would allow it to only work on COMM_WORLD. This patch corrects these to work for dynamic processes as well. It also corrects the evaluation of the number of nodes used to be correct in the general case. Refs #2007. Signed-off-by: Ken Raffenetti <[email protected]> diff --git a/src/mpid/ch3/src/mpid_vc.c b/src/mpid/ch3/src/mpid_vc.c index e35ed5d..acec520 100644 --- a/src/mpid/ch3/src/mpid_vc.c +++ b/src/mpid/ch3/src/mpid_vc.c @@ -1056,6 +1056,19 @@ done: #endif +static int compare_ints(const void *orig_x, const void *orig_y) +{ + int x = *((int *) orig_x); + int y = *((int *) orig_y); + + if (x == y) + return 0; + else if (x < y) + return -1; + else + return 1; +} + #undef FUNCNAME #define FUNCNAME populate_ids_from_mapping #undef FCNAME @@ -1069,6 +1082,8 @@ static int populate_ids_from_mapping(char *mapping, int *num_nodes, MPIDI_PG_t * int nblocks = 0; int rank; int block, block_node, node_proc; + int *tmp_rank_list, i; + MPIU_CHKLMEM_DECL(1); *did_map = 1; /* reset upon failure */ @@ -1078,31 +1093,43 @@ static int populate_ids_from_mapping(char *mapping, int *num_nodes, MPIDI_PG_t * if (NULL_MAPPING == mt) goto fn_fail; MPIU_ERR_CHKINTERNAL(mt != VECTOR_MAPPING, mpi_errno, "unsupported mapping type"); - rank = 0; - /* for a representation like (block,N,(1,1)) this while loop causes us to - * re-use that sole map block over and over until we have assigned node - * ids to every process */ - while (rank < pg->size) { - for (block = 0; block < nblocks; ++block) { - int node_id = mb[block].start_id; - for (block_node = 0; block_node < mb[block].count; ++block_node) { - if (node_id > *num_nodes) - *num_nodes = node_id; - - for (node_proc = 0; node_proc < mb[block].size; ++node_proc) { - pg->vct[rank].node_id = node_id; - ++rank; - if (rank == pg->size) - goto map_done; + /* allocate nodes to ranks */ + for (rank = 0;;) { + for (block = 0; block < nblocks; block++) { + for (block_node = 0; block_node < mb[block].count; block_node++) { + for (node_proc = 0; node_proc < mb[block].size; node_proc++) { + pg->vct[rank].node_id = mb[block].start_id + block_node; + if (++rank == pg->size) + goto break_out; } - ++node_id; } } } -map_done: - ++(*num_nodes); /* add one to get the num instead of the max */ +break_out: + /* Find the number of unique node ids. This is the classic + * element distinctness problem, for which the lower bound time + * complexity is O(N log N). Here we use a simple algorithm to + * sort the array and find the number of changes in the array + * through a linear search. There are certainly better algorithms + * available, which can be employed. */ + MPIU_CHKLMEM_MALLOC(tmp_rank_list, int *, pg->size * sizeof(int), mpi_errno, "tmp_rank_list"); + for (i = 0; i < pg->size; i++) + tmp_rank_list[i] = pg->vct[i].node_id; + +#if defined HAVE_QSORT + qsort(tmp_rank_list, pg->size, sizeof(int), compare_ints); +#else +#error "qsort not available" +#endif + + *num_nodes = 1; + for (i = 1; i < pg->size; i++) + if (tmp_rank_list[i] != tmp_rank_list[i-1]) + (*num_nodes)++; + fn_exit: + MPIU_CHKLMEM_FREEALL(); MPIU_Free(mb); return mpi_errno; fn_fail: @@ -1226,7 +1253,6 @@ int MPIDI_Populate_vc_node_ids(MPIDI_PG_t *pg, int our_pg_rank) /* See if process manager supports PMI_process_mapping keyval */ - /* FIXME 'PMI_process_mapping' only applies for the original PG (MPI_COMM_WORLD) */ if (pmi_version == 1 && pmi_subversion == 1) { pmi_errno = PMI_KVS_Get(kvs_name, "PMI_process_mapping", value, val_max_sz); if (pmi_errno == 0) { http://git.mpich.org/mpich.git/commitdiff/0f0f20a601889ae480958bc95112d0eee6... commit 0f0f20a601889ae480958bc95112d0eee686e822 Author: Pavan Balaji <[email protected]> Date: Wed Jan 29 13:20:35 2014 -0600 Remove unused flag. Signed-off-by: Ken Raffenetti <[email protected]> diff --git a/src/pm/hydra/pm/pmiserv/pmiserv_utils.c b/src/pm/hydra/pm/pmiserv/pmiserv_utils.c index 954a0fc..95abb81 100644 --- a/src/pm/hydra/pm/pmiserv/pmiserv_utils.c +++ b/src/pm/hydra/pm/pmiserv/pmiserv_utils.c @@ -201,7 +201,6 @@ static HYD_status pmi_process_mapping(struct HYD_pg *pg, char **process_mapping_ blocklist_tail = blocklist_head; } - create_mapping_key: /* Create the mapping out of the blocks */ HYD_STRING_STASH_INIT(stash); HYD_STRING_STASH(stash, HYDU_strdup("("), status); ----------------------------------------------------------------------- Summary of changes: src/mpid/ch3/src/mpid_vc.c | 104 +++++++++++++++++++++++++------ src/pm/hydra/pm/pmiserv/pmiserv_utils.c | 1 - 2 files changed, 85 insertions(+), 20 deletions(-) hooks/post-receive -- MPICH primary repository