OSHMEM/MCA/SPML/UCX: added support for team management functions #13177
Conversation
|
Hello! The Git Commit Checker CI bot found a few problems with this PR: 26401af: OSHMEM/MCA/SPML/UCX: removed unnecessary team_type...
Please fix these problems and, if necessary, force-push new commits back up to the PR branch. Thanks! |
roiedanino
force-pushed
the
shmem/1.5-support-ucx
branch
2 times, most recently
from
41068b3 to
cff7c93
Compare
|
Hello! The Git Commit Checker CI bot found a few problems with this PR: 450086a: OSHMEM/MCA/SPML/UCX: WIP
Please fix these problems and, if necessary, force-push new commits back up to the PR branch. Thanks! |
roiedanino
force-pushed
the
shmem/1.5-support-ucx
branch
from
450086a to
ba0d28e
Compare
|
Hi @MamziB can you please take a look? |
roiedanino
force-pushed
the
shmem/1.5-support-ucx
branch
2 times, most recently
from
a024387 to
a6babaf
Compare
oshmem/mca/spml/ucx/spml_ucx.c
Outdated
| ucx_new_team->config = calloc(1, sizeof(mca_spml_ucx_team_config_t)); | ||
|
|
||
| if (config != NULL) { | ||
| memcpy(&ucx_new_team->config->super, config, sizeof(shmem_team_config_t)); | ||
| } | ||
|
|
||
| ucx_new_team->config = (mca_spml_ucx_team_config_t*)config; |
There was a problem hiding this comment.
ucx_new_team->config is getting leaked here.
oshmem/mca/spml/ucx/spml_ucx.c
Outdated
| /* In order to simplify pe translations start and stride are calculated with respect to | ||
| * world_team */ | ||
| ucx_new_team = (mca_spml_ucx_team_t *)malloc(sizeof(mca_spml_ucx_team_t)); | ||
| ucx_new_team->start = parent_start + start; |
There was a problem hiding this comment.
this is incorrect, we should consider the parent's stride as well when we want to calculate the new start based on the comm world:
ucx_new_team->start = parent_start + (start * parent_stride)
oshmem/mca/spml/ucx/spml_ucx.c
Outdated
| int parent_stride; | ||
| int my_pe; | ||
|
|
||
| SPML_UCX_ASSERT(((start + size * stride) <= oshmem_num_procs()) && (start < size) && (stride > 0) && (size > 0)); |
There was a problem hiding this comment.
size can be any positive value , and start is PE index in the parent team, so start < size is not required to be true.
|
looks good now |
Signed-off-by: Roie Danino <[email protected]>
roiedanino
force-pushed
the
shmem/1.5-support-ucx
branch
from
c8cce70 to
d0a4e07
Compare
3 participants
Implemented team-management functions according to OpenSHMEM 1.5 spec
Tests PR: https://github.com/openshmem-org/tests-mellanox/pull/59/files