[petsc-users] Problem building petsc +rocm variant

[Satish Balay]

Looks like an issue with --with-batch=1 [where the MPI checks are skipped]

Can you try the build without: '+batch'?

Satish


On Fri, 11 Mar 2022, Roskop, Luke B wrote:

> As requested, I attached the configure.log and make.log files
> Thanks!
> Luke
> 
> 
> From: Barry Smith <bsmith at petsc.dev>
> Date: Friday, March 11, 2022 at 2:47 PM
> To: "Roskop, Luke B" <luke.roskop at hpe.com>
> Cc: "petsc-users at mcs.anl.gov" <petsc-users at mcs.anl.gov>
> Subject: Re: [petsc-users] Problem building petsc +rocm variant
> 
> 
>   Please send configure.log and make.log to petsc-maint at mcs.anl.gov<mailto:petsc-maint at mcs.anl.gov>  For some reason PETSc's configure does not detect that the MPI does provide MPI_Type_dup() even though it is prototyped in the MPI include file.
> 
> 
> 
> On Mar 11, 2022, at 3:34 PM, Roskop, Luke B <luke.roskop at hpe.com<mailto:luke.roskop at hpe.com>> wrote:
> 
> Hi, I’m hoping you can help me figure out how to build PETSc targeting AMDGPUs (gfx90a GPU).
> 
> I’m attempting to build on the CrayEx ORNL system called crusher, using ROCmCC (AMD’s compiler), and cray-mpich. In case it helps, I’m using spack to build petsc with the “petsc at main%rocmcc+batch+rocm amdgpu_target=gfx90a” spec.
> 
> Spack ends up invoking the following configure for PETSc:
> '/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/python-3.9.10-7y7mxajn5rywz5xdnba4azphcdodxiub/bin/python3.9' 'configure' '--prefix=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/petsc-main-mccbycx66la7rlx6jv44f6zd63cmdzm7' '--with-ssl=0' '--download-c2html=0' '--download-sowing=0' '--download-hwloc=0' 'CFLAGS=' 'FFLAGS=-fPIC' 'CXXFLAGS=' 'LDFLAGS=-Wl,-z,notext' '--with-cc=/opt/cray/pe/mpich/8.1.12/ofi/amd/4.4/bin/mpicc' '--with-cxx=/opt/cray/pe/mpich/8.1.12/ofi/amd/4.4/bin/mpicxx' '--with-fc=/opt/cray/pe/mpich/8.1.12/ofi/amd/4.4/bin/mpif90' '--with-precision=double' '--with-scalar-type=real' '--with-shared-libraries=1' '--with-debugging=0' '--with-openmp=0' '--with-64-bit-indices=0' 'COPTFLAGS=' 'FOPTFLAGS=' 'CXXOPTFLAGS=' '--with-blas-lapack-lib=/opt/cray/pe/libsci/21.08.1.2/AMD/4.0/x86_64/lib/libsci_amd.so' '--with-batch=1' '--with-x=0' '--with-clanguage=C' '--with-cuda=0' '--
 with-hip
 =1' '--with-hip-dir=/opt/rocm-4.5.0/hip' '--with-metis=1' '--with-metis-include=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/metis-5.1.0-zn5rn5srr7qzxyo5tq36d46adcsyc5a7/include' '--with-metis-lib=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/metis-5.1.0-zn5rn5srr7qzxyo5tq36d46adcsyc5a7/lib/libmetis.so' '--with-hypre=1' '--with-hypre-include=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/hypre-develop-vgfx3lhhloq4cnethsrrpz7iez7x6wad/include' '--with-hypre-lib=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/hypre-develop-vgfx3lhhloq4cnethsrrpz7iez7x6wad/lib/libHYPRE.so' '--with-parmetis=1' '--with-parmetis-include=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/parmetis-4.0.3-6jqxqmt7qqq73rxmx3beu5ba4vj3r253/include' '--with-parmetis-lib=
 /gpfs/al
 pine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/parmetis-4.0.3-6jqxqmt7qqq73rxmx3beu5ba4vj3r253/lib/libparmetis.so' '--with-superlu_dist=1' '--with-superlu_dist-include=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/superlu-dist-develop-mpiyhomp4k72bilqn6xk7uol36ulsdve/include' '--with-superlu_dist-lib=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/superlu-dist-develop-mpiyhomp4k72bilqn6xk7uol36ulsdve/lib/libsuperlu_dist.so' '--with-ptscotch=0' '--with-suitesparse=0' '--with-hdf5=1' '--with-hdf5-include=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/hdf5-1.12.1-dp5vqo4tjh6oi7szpcsqkdlifgjxknzf/include' '--with-hdf5-lib=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/hdf5-1.12.1-dp5vqo4tjh6oi7szpcsqkdlifgjxknzf/lib/libhdf5.so' '--with-zlib=1' '--with
 -zlib-in
 clude=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/zlib-1.2.11-2ciasfxwyxanyohroisdpvidg4gs2fdy/include' '--with-zlib-lib=/gpfs/alpine/ven114/scratch/lukebr/confidence/spack/install_tree/cray-sles15-zen3/rocmcc-4.5.0/zlib-1.2.11-2ciasfxwyxanyohroisdpvidg4gs2fdy/lib/libz.so' '--with-mumps=0' '--with-trilinos=0' '--with-fftw=0' '--with-valgrind=0' '--with-gmp=0' '--with-libpng=0' '--with-giflib=0' '--with-mpfr=0' '--with-netcdf=0' '--with-pnetcdf=0' '--with-moab=0' '--with-random123=0' '--with-exodusii=0' '--with-cgns=0' '--with-memkind=0' '--with-p4est=0' '--with-saws=0' '--with-yaml=0' '--with-hwloc=0' '--with-libjpeg=0' '--with-scalapack=1' '--with-scalapack-lib=/opt/cray/pe/libsci/21.08.1.2/AMD/4.0/x86_64/lib/libsci_amd.so' '--with-strumpack=0' '--with-mmg=0' '--with-parmmg=0' '--with-tetgen=0' '--with-cxx-dialect=C++11'
> 
> Using spack, I see this error at compile time:
> 
> /tmp/lukebr/spack-stage/spack-stage-petsc-main-5jlv6jcfdaa37iy5zm77umvb6uvgwdo7/spack-src/src/vec/is/sf/impls/basic/sfpack.c:463:19: error: static declaration of 'MPI_Type_dup' follows non-static declaration
> static inline int MPI_Type_dup(MPI_Datatype datatype,MPI_Datatype *newtype)
>                   ^
> /opt/cray/pe/mpich/8.1.12/ofi/cray/10.0/include/mpi.h:1291:5: note: previous declaration is here
> int MPI_Type_dup(MPI_Datatype oldtype, MPI_Datatype *newtype) MPICH_API_PUBLIC;
>     ^
> 1 error generated.
> 
> To get around this error, I pass “-DPETSC_HAVE_MPI_TYPE_DUP” but then I see the following lining error:
> 
> CLINKER arch-linux-c-opt/lib/libpetsc.so.3.016.5
> ld.lld: error: undefined hidden symbol: PetscSFCreate_Window
> >>> referenced by sfregi.c
> >>>               arch-linux-c-opt/obj/vec/is/sf/interface/sfregi.o:(PetscSFRegisterAll)
> clang-13: error: linker command failed with exit code 1 (use -v to see invocation)
> gmake[3]: *** [gmakefile:113: arch-linux-c-opt/lib/libpetsc.so.3.016.5] Error 1
> gmake[2]: *** [/tmp/lukebr/spack-stage/spack-stage-petsc-main-5jlv6jcfdaa37iy5zm77umvb6uvgwdo7/spack-src/lib/petsc/conf/rules:56: libs] Error 2
> 
> 
> Before I continue, is there a preferred way to build PETSc on an AMDGPU system? Could you share this?
> 
> Thanks,
> Luke
> 
>