search

open-mpi / ompi

Open MPI main development repository
https://www.open-mpi.org
Other
2.12k stars 857 forks source link

error when install openmpi 4.0.1 #6581

Closed upuil closed 5 years ago

upuil commented 5 years ago

I try to install the new version of ompi from source code. CC=icc CXX=icpc FC=ifort F77=ifort ./configure --prefix=/home/opt/apps/openmpi/3.1.3/intel/2017a

In file included from common_ucx.c(12):
common_ucx.h(133): warning #266: function "ucp_request_free" declared implicitly
      MCA_COMMON_UCX_WAIT_LOOP(request, worker, msg, ucp_request_free(request));
      ^

In file included from common_ucx.c(18):
/opt/ucx/include/ucm/api/ucm.h(11): catastrophic error: cannot open source file "ucs/sys/compiler.h"
  #include <ucs/sys/compiler.h>
                               ^

compilation aborted for common_ucx.c (code 4)
make[2]: *** [libmca_common_ucx_la-common_ucx.lo] Error 1
make[2]: Leaving directory `/home/pu/openmpi-4.0.1/opal/mca/common/ucx'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/pu/openmpi-4.0.1/opal'
make: *** [all-recursive] Error 1
jsquyres commented 5 years ago

Are you compiling Open MPI v4.0.1 or 3.1.3? Both are listed in your output.

This looks like a bug in the UCX header file (i.e., not an Open MPI header file).

If the problem is really in /opt/ucx/include/ucm/api/ucm.h (i.e., is there a ucs/sys/compiler.h file on your machine anywhere?), this issue should probably be reported to the UCX community.

FYI @yosefe @jladd-mlnx

hppritcha commented 5 years ago

What version of UCX is installed on the system?

ggouaillardet commented 5 years ago

the root cause is a UCX header cannot find an other UCX header. This strongly suggests the issue is not in Open MPI but is either caused by an UCX bug or a local and busted UCX install.

dmitrygx commented 5 years ago

The similar problem was fixed in the https://github.com/openucx/ucx/pull/1336 The latest release 1.5.1 contains this fix

@upuil Could you check with UCX master or v1.5.1, please?

jsquyres commented 5 years ago

Let's go ahead and close the issue here -- sounds like followups with the UCX community are warranted if the problem persists.