CMK_CXI not defined on non-Slingshot OFI build

[heatherkellyucl]

Charm++ 8.0.0.

This is a Spack build, requesting backend=ofi pmi=pmix, build command it creates is below.

The cluster is older and has Intel OmniPath, no Slingshot.

The build fails because #if CMK_CXI statements are being reached when CMK_CXI is not defined.

==> Error: ProcessError: Command exited with status 2:                                                                
    './build' 'LIBS' 'ofi-linux-x86_64' 'gcc' 'gfortran' '-j6' '--destination=/lustre/apps/spack/0.23/deploy/2025-02/s
pack/opt/spack/linux-rhel9-broadwell/gcc-12.3.0/charmpp-8.0.0-4hct6qiefz6ssrb34tfuxlkbddapfhsr' 'ompipmix' '--basedir=
/lustre/apps/spack/0.23/deploy/2025-02/spack/opt/spack/linux-rhel9-broadwell/gcc-12.3.0/openmpi-4.1.6-qbo4kz6ww5zqnhja
bsnojud5h74eqqea' 'smp' '--build-shared' '--with-production'                                                          
                                                                                                                      
33 errors found in build log:                                                                                         
     1019    [ 61%] Building CXX object src/ck-core/CMakeFiles/ck.dir/ck.C.o                                          
     1020    [ 61%] Building CXX object src/ck-core/CMakeFiles/ck.dir/ckarray.C.o                                     
     1021      CC       mpi-io/get_atom.lo                                                                            
     1022    [ 61%] Building CXX object CMakeFiles/converse.dir/src/arch/ofi/machine.C.o                              
     1023      CC       mpi-io/get_bytoff.lo                                                                          
     1024      CC       mpi-io/get_errh.lo                                                                            
  >> 1025    /lustre/apps/spack/0.23/deploy/2025-02/build_stage/ccspapp/spack-stage-charmpp-8.0.0-4hct6qiefz6ssrb34tf 
             uxlkbddapfhsr/spack-src/src/arch/ofi/machine.C:170:12: error: #if with no expression                     
     1026      170 | #if CMK_CXI                                                                                      
     1027          |            ^ 

I'm assuming that it needs defining as 0 somewhere where it currently isn't.

Does it just need an else in here?

charm/CMakeLists.txt

Line 1038 in 1a877e8

if(CMK_BUILD_ON_CXI)

[ericjbohm]

The non-CXI OFI implementation is probably sub-optimal. I'll work on a fix for this issue so that it at least compiles and runs, but in the meanwhile you will probably get better performance using the verbs or ucx builds.

[heatherkellyucl]

@ericjbohm Ok - I was previously using ucx which compiled and ran, but we've come across a RHEL9 issue where running namd with charm on kernels newer than 9.0 is hard locking up and then kernel panicking the nodes we run the job on, if the hfi1 module is loaded . So I wanted to check if the same thing happened with this build.

(We have a case open with RedHat).

ETA: actually, newer than kernel 9.1, identified as specific patch in 9.2.