`hiop~mpi` includes MPI headers

[cameronrutherford]

https://github.com/LLNL/hiop/blob/develop/src/Interface/hiopInterface.hpp#L60

https://github.com/pnnl/ExaGO/actions/runs/6304772661/job/17116842345?pr=15

This is a really weird bug, as even when building petsc~mpi in the exago package here, petsc insists on having an mpi.h lying around that is also picked up...

I am still trying to figure out who to blame here, but this seemed like the right place to start.

[cnpetra]

are you somehow building hiop without MPI? or different mpi headers are with hiop and petsc

[cameronrutherford]

are you somehow building hiop without MPI? or different mpi headers are with hiop and petsc

From the ExaGO pipeline, we are building:

exago@develop+hiop~ipopt~mpi~python+raja+tests arch=None-None-x86_64
 -   tx7nd5d  exago@develop%gcc@9.4.0~cuda+hiop~ipo~ipopt+logging~mpi~python+raja~rocm+tests build_system=cmake build_type=RelWithDebInfo dev_path=/__w/ExaGO/ExaGO arch=linux-ubuntu20.04-x86_64
 -   ybikngp      ^camp@0.2.3%gcc@9.4.0~cuda~ipo+openmp~rocm~tests build_system=cmake build_type=RelWithDebInfo arch=linux-ubuntu20.04-x86_64
 -   kl43gwj          ^blt@0.4.1%gcc@9.4.0 build_system=generic arch=linux-ubuntu20.04-x86_64
 -   7bzaewm      ^cmake@3.25.2%gcc@9.4.0~doc+ncurses+ownlibs~qt build_system=generic build_type=Release arch=linux-ubuntu20.04-x86_64
 -   3bxcabf          ^ncurses@6.4%gcc@9.4.0~symlinks+termlib abi=none build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   yekhgie          ^openssl@1.1.1t%gcc@9.4.0~docs~shared build_system=generic certs=mozilla arch=linux-ubuntu20.04-x86_64
 -   djeruao              ^ca-certificates-mozilla@2023-01-10%gcc@9.4.0 build_system=generic arch=linux-ubuntu20.04-x86_64
 -   jymwj6w              ^zlib@1.2.13%gcc@9.4.0+optimize+pic+shared build_system=makefile arch=linux-ubuntu20.04-x86_64
 -   vcsqn5o      ^hiop@0.7.1%gcc@9.4.0~cuda+deepchecking~ginkgo~ipo~jsrun~kron~mpi+raja~rocm~shared~sparse build_system=cmake build_type=RelWithDebInfo arch=linux-ubuntu20.04-x86_64
 -   wtvhbiz      ^openblas@0.3.21%gcc@9.4.0~bignuma~consistent_fpcsr+fortran~ilp64+locking+pic+shared build_system=makefile patches=114f95f,a4c642f,c20f518,d3d9b15 symbol_suffix=none threads=none arch=linux-ubuntu20.04-x86_64
 -   5qydzbx          ^perl@5.36.0%gcc@9.4.0+cpanm+open+shared+threads build_system=generic arch=linux-ubuntu20.04-x86_64
 -   e5g7oef              ^berkeley-db@18.1.40%gcc@9.4.0+cxx~docs+stl build_system=autotools patches=26090f4,b231fcc arch=linux-ubuntu20.04-x86_64
 -   gs4r33x              ^bzip2@1.0.8%gcc@9.4.0~debug~pic+shared build_system=generic arch=linux-ubuntu20.04-x86_64
 -   7wdyruu              ^gdbm@1.23%gcc@9.4.0 build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   wslvyrk      ^petsc@3.18.3%gcc@9.4.0~X~batch~cgns~complex~cuda~debug+double~exodusii~fftw+fortran~giflib~hdf5~hpddm~hwloc~hypre~int64~jpeg~knl~kokkos~libpng~libyaml~memkind+metis~mkl-pardiso~mmg~moab~mpfr~mpi~mumps~openmp~p4est~parmmg~ptscotch~random123~rocm~saws~scalapack+shared~strumpack~suite-sparse~superlu-dist~tetgen~trilinos~valgrind build_system=generic clanguage=C arch=linux-ubuntu20.04-x86_64
 -   kwz7ftm          ^diffutils@3.8%gcc@9.4.0 build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   y4xrp3s              ^libiconv@1.17%gcc@9.4.0 build_system=autotools libs=shared,static arch=linux-ubuntu20.04-x86_64
 -   wnqabk7          ^metis@5.1.0%gcc@9.4.0~gdb~int64~ipo~real64+shared build_system=cmake build_type=RelWithDebInfo patches=4991da9,93a7903,b1225da arch=linux-ubuntu20.04-x86_64
 -   quyjgw3          ^python@3.10.8%gcc@9.4.0+bz2+crypt+ctypes+dbm~debug+libxml2+lzma~nis~optimizations+pic+pyexpat+pythoncmd+readline+shared+sqlite3+ssl~tkinter+uuid+zlib build_system=generic patches=0d98e93,7d40923,f2fd060 arch=linux-ubuntu20.04-x86_64
 -   pgvwni4              ^expat@2.5.0%gcc@9.4.0+libbsd build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   en3zuay                  ^libbsd@0.11.7%gcc@9.4.0 build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   ps7sxlx                      ^libmd@1.0.4%gcc@9.4.0 build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   wlq5rko              ^gettext@0.21.1%gcc@9.4.0+bzip2+curses+git~libunistring+libxml2+tar+xz build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   j6aqcps                  ^libxml2@2.10.3%gcc@9.4.0~python build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   zt4ocio                  ^tar@1.34%gcc@9.4.0 build_system=autotools zip=pigz arch=linux-ubuntu20.04-x86_64
 -   xoxeujp                      ^pigz@2.7%gcc@9.4.0 build_system=makefile arch=linux-ubuntu20.04-x86_64
 -   3vtuapf                      ^zstd@1.5.2%gcc@9.4.0+programs build_system=makefile compression=none libs=shared,static arch=linux-ubuntu20.04-x86_64
 -   6sswith              ^libffi@3.4.4%gcc@9.4.0 build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   2evlwmd              ^libxcrypt@4.4.33%gcc@9.4.0~obsolete_api build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   iuswzm4              ^readline@8.2%gcc@9.4.0 build_system=autotools patches=bbf97f1 arch=linux-ubuntu20.04-x86_64
 -   ghcuaen              ^sqlite@3.40.1%gcc@9.4.0+column_metadata+dynamic_extensions+fts~functions+rtree build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   swhrnzy              ^util-linux-uuid@2.38.1%gcc@9.4.0 build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   qkxtzoa              ^xz@5.4.1%gcc@9.4.0~pic build_system=autotools libs=shared,static arch=linux-ubuntu20.04-x86_64
 -   w6opye6      ^pkgconf@1.8.0%gcc@9.4.0 build_system=autotools arch=linux-ubuntu20.04-x86_64
 -   7xqyl5b      ^raja@0.14.0%gcc@9.4.0~cuda+examples+exercises~ipo+openmp~rocm+shared~tests build_system=cmake build_type=RelWithDebInfo arch=linux-ubuntu20.04-x86_64
 -   4s36yj3      ^umpire@6.0.0%gcc@9.4.0+c~cuda~device_alloc~deviceconst+examples~fortran~ipo~numa~openmp~rocm+shared build_system=cmake build_type=RelWithDebInfo tests=none arch=linux-ubuntu20.04-x86_64

And so we get the backtrace:

     459    In file included from /__w/ExaGO/ExaGO/tpl/spack/opt/spack/linux-ub
            untu20.04-x86_64/gcc-9.4.0/hiop-0.7.1-vcsqn5ocwcwfihlrbjqozv3ku2rkg
            zo7/include/hiopInterface.hpp:60,
     460                     from /__w/ExaGO/ExaGO/tpl/spack/opt/spack/linux-ub
            untu20.04-x86_64/gcc-9.4.0/hiop-0.7.1-vcsqn5ocwcwfihlrbjqozv3ku2rkg
            zo7/include/hiopNlpFormulation.hpp:59,
     461                     from /__w/ExaGO/ExaGO/tpl/spack/opt/spack/linux-ub
            untu20.04-x86_64/gcc-9.4.0/hiop-0.7.1-vcsqn5ocwcwfihlrbjqozv3ku2rkg
            zo7/include/hiopAlgFilterIPM.hpp:59,
     462                     from /__w/ExaGO/ExaGO/src/opflow/solver/hiop/opflo
            w_hiop.h:7,
     463                     from /__w/ExaGO/ExaGO/src/opflow/solver/hiop/opflo
            w_hiop.cpp:4:
  >> 464    /__w/ExaGO/ExaGO/tpl/spack/opt/spack/linux-ubuntu20.04-x86_64/gcc-9
            .4.0/petsc-3.18.3-wslvyrkkwofiig24a5rm7gctadb7g4fk/include/petsc/mp
            iuni/mpi.h:186:13: error: multiple types in one declaration
     465      186 | typedef int MPI_Comm;
     466          |             ^~~~~~~~
  >> 467    /__w/ExaGO/ExaGO/tpl/spack/opt/spack/linux-ubuntu20.04-x86_64/gcc-9
            .4.0/petsc-3.18.3-wslvyrkkwofiig24a5rm7gctadb7g4fk/include/petsc/mp
            iuni/mpi.h:186:13: error: declaration does not declare anything [-f
            permissive]
  >> 468    make[2]: *** [src/opflow/CMakeFiles/OPFLOW_obj_static.dir/build.mak
            e:261: src/opflow/CMakeFiles/OPFLOW_obj_static.dir/solver/hiop/opfl
            ow_hiop.cpp.o] Error 1

So the HiOp header hiopInterface.hpp on line 60 (linked in the issue description originally) is including hiopMPI.h, which is then including mpi.h. This looks for any header, and picks up a random PETSc one which errors out.

We are building PETSc and HiOp without MPI here, so I honestly think this could be a HiOp and a PETSc bug?

[nychiang]

@cnpetra @cameronrutherford
I can successfully build HiOp without MPI.
In hiopMPI.h, mpi.h is not included if we set HIOP_USE_MPI = OFF.

From your log file, I think the problems are:

1. When HIOP_USE_MPI = OFF, both HiOp and PETSc define their own MPI_Comm.
2. Not sure where mpi.h is included. Seems to be it is included via PETSc.

See here

[cameronrutherford]

@cnpetra @cameronrutherford

I can successfully build HiOp without MPI.

In hiopMPI.h, mpi.h is not included if we set HIOP_USE_MPI = OFF.

From your log file, I think the problems are:

1. When HIOP_USE_MPI = OFF, both HiOp and PETSc define their own MPI_Comm.

2. Not sure where mpi.h is included. Seems to be it is included via PETSc.

See here

I'm following, but some clarification. I am also able to build hiop~mpi, but issue only happens when exago~mpi tries to build with both petsc~mpi and hiop~mpi.

Why do HiOp and PETSc both need to define MPI_Comm in these non-mpi builds?

[cameronrutherford]

Again this might technically be an ExaGO (or PETSc or HiOp) issue, but trying to figure out who's to blame here

[cnpetra]

we had this issue before with mfem if I recall correctly. One the defines has to go. I think HiOp can take with however petsc defines MPI_Comm. So an easy fix would be for HiOp to check if already defined. This is for when HIOP_USE_MPI is off.