pmi config #2557
pmi config #2557
Conversation
|
test:jenkins/ch4/ofi |
raffenet
force-pushed
the
pmi-config
branch
5 times, most recently
from
c6553fd to
61b91bd
Compare
|
The first few commits are fine with me. The last commit to use PMI2, I am not sure. Changing CH4 to use PMI2 is a larger effort I thought Intel MPI was doing. They may have a patch for it. |
|
Okay. The reason I did that was because the PMI calls in shm init did not work when configuring with Cray PMI. If we go PMI2 across the board, everything just worked. Maybe @ddurnov can comment on any PMI2 patch for CH4 from Intel? |
|
It also occurs to me that in CH4, the netmods are calling |
src/mpid/ch4/netmod/ofi/ofi_init.h
Outdated
| @@ -755,7 +755,8 @@ static inline int MPIDI_NM_mpi_init_hook(int rank, | |||
| sprintf(keyS, "OFI-%d", i); | |||
| #ifdef USE_CRAYPMI_API | |||
| MPIDI_OFI_PMI_CALL_POP(PMI2_KVS_Get | |||
| (NULL, -1, keyS, valS, MPIDI_KVSAPPSTRLEN, &len), pmi); | |||
There was a problem hiding this comment.
the int len couple lines above should also be removed?
|
test:jenkins/ch4/ofi |
If we are building with an external pmi library, we should be using the external version of the pmi.h/pmi2.h. Move the MPICH copies into subdirs that will be conditionally added to CPPFLAGS based on configure options. Signed-off-by: Gengbin Zheng <[email protected]>
Signed-off-by: Gengbin Zheng <[email protected]>
Signed-off-by: Gengbin Zheng <[email protected]>
Fixes a compile error for undeclared variable. Signed-off-by: Gengbin Zheng <[email protected]>
Signed-off-by: Dmitry Durnov <[email protected]>
No description provided.