Patch from Nathan outlined with a crash the mishandling of the case where CMA is requested but not available.
Этот коммит содержится в:
родитель
d7c7bb3993
Коммит
55e49470de
@ -903,6 +903,16 @@ mca_btl_sm_component_init(int *num_btls,
|
|||||||
so no problems with accidentally overwriting this set earlier */
|
so no problems with accidentally overwriting this set earlier */
|
||||||
mca_btl_sm.super.btl_get = mca_btl_sm_get_sync;
|
mca_btl_sm.super.btl_get = mca_btl_sm_get_sync;
|
||||||
}
|
}
|
||||||
|
#else
|
||||||
|
/* If the user explicitly asked for CMA and we can't provide itm
|
||||||
|
* error */
|
||||||
|
if (mca_btl_sm_component.use_cma > 0) {
|
||||||
|
mca_btl_sm.super.btl_flags &= ~MCA_BTL_FLAGS_GET;
|
||||||
|
opal_show_help("help-mpi-btl-sm.txt",
|
||||||
|
"CMA requested but not available",
|
||||||
|
true, opal_process_info.nodename);
|
||||||
|
return NULL;
|
||||||
|
}
|
||||||
#endif /* OPAL_BTL_SM_HAVE_CMA */
|
#endif /* OPAL_BTL_SM_HAVE_CMA */
|
||||||
|
|
||||||
#if OPAL_CUDA_SUPPORT
|
#if OPAL_CUDA_SUPPORT
|
||||||
|
@ -90,6 +90,17 @@ mca_btl_sm_use_knem MCA parameter, but Knem support was either not
|
|||||||
compiled into this Open MPI installation, or Knem support was unable
|
compiled into this Open MPI installation, or Knem support was unable
|
||||||
to be activated in this process.
|
to be activated in this process.
|
||||||
|
|
||||||
|
The shared memory BTL will now deactivate itself, likely resulting in
|
||||||
|
lower performance for on-node communication.
|
||||||
|
|
||||||
|
Local host: %s
|
||||||
|
#
|
||||||
|
[CMA requested but not available]
|
||||||
|
WARNING: Linux kernel CMA support was requested via the
|
||||||
|
mca_btl_sm_use_cma MCA parameter, but CMA support was either not
|
||||||
|
compiled into this Open MPI installation, or CMA support was unable
|
||||||
|
to be activated in this process.
|
||||||
|
|
||||||
The shared memory BTL will now deactivate itself, likely resulting in
|
The shared memory BTL will now deactivate itself, likely resulting in
|
||||||
lower performance for on-node communication.
|
lower performance for on-node communication.
|
||||||
|
|
||||||
|
Загрузка…
x
Ссылка в новой задаче
Block a user