MPIDI_OFI_handle_cq_error(1042): OFI poll failed

[Lumos]

Hi，

I use the latest mpiicx2024.0.1 to compile hdf5.14.3, when make will have:

icx: warning: argument unused during compilation: '-fno-alias' [-Wunused-command-line-argument]

How to remove this warning. With mpiicx -v, there will be a similar warning: icx: warning: argument unused during compilation: '-i/home/Compiler/intel2024 / oneapi/intelpython3 / include' [- Wunused - the command - line - argument].

and make hdf5.14.3 error: 

make[1]: *** [Makefile:1132: t_bigio.o] Error 1
make[1]: Leaving directory '/home/LIBRARIES/intel2024/hdf5-1.14.3/testpar'
make: *** [Makefile:730: all-recursive] Error 1

My system is CENTOS7. Do you have any suggestions?

My steps:

export CC=mpiicx
export CXX=mpiicpx
export FC=mpiifx
export F90=mpiifx
export F77=mpiifx
export CPPFLAGS=-I/home/Compiler/intel2024/oneapi/mpi/2021.11/include
export LDFLAGS="-L/home/Compiler/intel2024/oneapi/mpi/2021.11/lib"
export CFLAGS="-O3 -fPIC"
export FCFLAGS="-O3 -fPIC"
export CXXFLAGS="-O3 -fPIC"

./configure --prefix=$DIR/hdf51.14.3 --enable-fortran --enable-shared --enable-parallel --with-pic CC=mpiicx FC=mpiifx CXX=mpiicpx CFLAGS="-fPIC -O3 -xHost -fno-alias -align" FFLAGS="-fPIC -O3 -xHost -fno-alias -align" CXXFLAGS="-fPIC -O3 -xHost -fno-alias -align" FFLAGS="-I/home/Compiler/intel2024/oneapi/mpi/2021.11/include -L/home/Compiler/intel2024/oneapi/mpi/2021.11/lib" --with-szlib=$DIR/szip2.1.1 --with-zlib=$DIR/zlib1.3

Moreover, I found that the inability to use oneapi2024.0 in the virtual machine Rocky9 system would also cause the system to not boot up and run.

[TobiasK]

@Lumos please open a new thread for a new issue.

There seems to be a problem compiling t_bigio.o with 2024.0.2 icx at optimization level -O3. However, the next release fixes that bug.

Please consider using -O2 for this release.

"-fno-alias" is still not supported, unfortunately.

Also your configure and export defines / overwrites a couple of flags. For your reference, I successfully build hdf5 with those flags:

./configure CFLAGS="-fPIC -O2 -xHost -fno-alias -align" FFLAGS="-fPIC -O3 -xHost -fno-alias -align" CXXFLAGS="-fPIC -O3 -xHost -fno-alias -align" CC=mpiicx CXX=mpiicpx F90=mpiifx FC=mpiifx --enable-shared --enable-parallel --enable-fortran --with-zlib=yes --with-szlib=no

make -j

For the issue with Rocky Linux 9.3 in a virtual machine, please post this in the proper forum here:

oneAPI Registration, Download, Licensing and Installation

[Lumos]

Thank you very much for your timely reply. I will make it according to your flags, but there is a new problem:

ptest.F90(19): error #7002: Error in opening the compiled module file. Check INCLUDE paths. [MPI]
USE MPI
------^
ptest.F90(42): warning #8889: Explicit interface or EXTERNAL declaration is required. [MPI_INIT]
CALL mpi_init(mpierror)
-------^
ptest.F90(43): error #6404: This name does not have a type, and must have an explicit type. [MPI_SUCCESS]
IF (mpierror .NE. MPI_SUCCESS) THEN
--------------------^
ptest.F90(46): warning #8889: Explicit interface or EXTERNAL declaration is required. [MPI_COMM_RANK]
CALL mpi_comm_rank( MPI_COMM_WORLD, mpi_rank, mpierror )
-------^
ptest.F90(46): error #6404: This name does not have a type, and must have an explicit type. [MPI_COMM_WORLD]
CALL mpi_comm_rank( MPI_COMM_WORLD, mpi_rank, mpierror )
----------------------^
ptest.F90(50): warning #8889: Explicit interface or EXTERNAL declaration is required. [MPI_COMM_SIZE]
CALL mpi_comm_size( MPI_COMM_WORLD, mpi_size, mpierror )
-------^
ptest.F90(67): warning #8889: Explicit interface or EXTERNAL declaration is required. [HYPER]
CALL hyper(length, do_collective(j), do_chunk(i), mpi_size, mpi_rank, ret_total_error)
-------------^
ptest.F90(78): warning #8889: Explicit interface or EXTERNAL declaration is required. [MULTIPLE_DSET_WRITE]
CALL multiple_dset_write(length, do_collective(1), do_chunk(1), mpi_size, mpi_rank, ret_total_error)
-------^
ptest.F90(87): warning #8889: Explicit interface or EXTERNAL declaration is required. [PMULTIPLE_DSET_HYPER_RW]
CALL pmultiple_dset_hyper_rw(do_collective(j), do_chunk(i), mpi_size, mpi_rank, ret_total_error)
-------------^
ptest.F90(98): warning #8889: Explicit interface or EXTERNAL declaration is required. [MPI_ALLREDUCE]
CALL MPI_ALLREDUCE(total_error, sum, 1, MPI_INTEGER, MPI_SUM, MPI_COMM_WORLD, mpierror)
-------^
ptest.F90(98): error #6404: This name does not have a type, and must have an explicit type. [MPI_INTEGER]
CALL MPI_ALLREDUCE(total_error, sum, 1, MPI_INTEGER, MPI_SUM, MPI_COMM_WORLD, mpierror)
------------------------------------------^
ptest.F90(98): error #6404: This name does not have a type, and must have an explicit type. [MPI_SUM]
CALL MPI_ALLREDUCE(total_error, sum, 1, MPI_INTEGER, MPI_SUM, MPI_COMM_WORLD, mpierror)
-------------------------------------------------------^
ptest.F90(106): warning #8889: Explicit interface or EXTERNAL declaration is required. [MPI_FINALIZE]
CALL mpi_finalize(mpierror)
----------^
ptest.F90(112): warning #8889: Explicit interface or EXTERNAL declaration is required. [MPI_ABORT]
CALL mpi_abort(MPI_COMM_WORLD, 1, mpierror)
----------^
compilation aborted for ptest.F90 (code 1)
make[2]: *** [Makefile:1322: ptest.o] Error 1
make[2]: Leaving directory '/home/LIBRARIES/intel2024/hdf5-1.14.3/fortran/testpar'
make[1]: *** [Makefile:902: all-recursive] Error 1
make[1]: Leaving directory '/home/LIBRARIES/intel2024/hdf5-1.14.3/fortran'
make: *** [Makefile:730: all-recursive] Error 1

Do you have any suggestions?

Best,

Lumos

[TobiasK]

Please do not set the variables via export.

Only use one way, not both. Probably the path to your MPI include directory is wrong, the mpi* compiler wrappers will set them automatically.

[Lumos]

Actually, I didn't set the variable by export.

My flags:

./configure --prefix=$DIR/hdf51.14.3 CFLAGS="-fPIC -O2 -xHost -fno-alias -align" FFLAGS="-fPIC -O3 -xHost -fno-alias -align" CXXFLAGS="-fPIC -O3 -xHost -fno-alias -align" CC=mpiicx CXX=mpiicpx F90=mpiifx FC=mpiifx --enable-shared --enable-parallel --enable-fortran --with-zlib=yes --with-szlib=no

make

I get the same error, what's going on?

[TobiasK]

@Lumos

I just realized that you are using CentOS 7.9 which is not supported anymore.

https://www.intel.com/content/www/us/en/developer/articles/system-requirements/mpi-library-system-requirements.html

The last thing I can advise you to do is to start from scratch, delete the entire folder, extract it again in a clean environment.

[Lumos]

Thank you for your help, I do use CentOS7, but I am only a user of the machine and cannot change the system. And I have tried your suggestion, which does not solve my problem.

[Lumos]

That's great! I found the problem: I installed intelpython, and I can successfully pass make by removing intelpython from my environment. But in fact, make check made a new mistake:

*** UNEXPECTED RETURN from H5Oget_native_info is -1 at line 5438 in tfile.c
HDF5-DIAG: Error detected in HDF5 (1.14.3) thread 0:
  #000: H5O.c line 1325 in H5Oget_native_info(): invalid location identifier
    major: Invalid arguments to routine
    minor: Inappropriate type
  #001: H5VLint.c line 1741 in H5VL_vol_object(): invalid identifier type to function
    major: Invalid arguments to routine
    minor: Inappropriate type
*** UNEXPECTED VALUE from H5Oget_native_info should be 2, but is 6520800 at line 5439 in tfile.c
HDF5-DIAG: Error detected in HDF5 (1.14.3) thread 0:
  #000: H5O.c line 1325 in H5Oget_native_info(): invalid location identifier
    major: Invalid arguments to ro0.07user 0.09system 0:00.68elapsed 23%CPU (0avgtext+0avgdata 14112maxresident)k
1000inputs+3072outputs (0major+7824minor)pagefaults 0swaps
make[4]: *** [Makefile:3982: testhdf5.chkexe_] Error 1
make[4]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/test'
make[3]: *** [Makefile:3968: build-check-s] Error 2
make[3]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/test'
make[2]: *** [Makefile:3962: test] Error 2
make[2]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/test'
make[1]: *** [Makefile:3401: check-am] Error 2
make[1]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/test'
make: *** [Makefile:730: check-recursive] Error 1

[TobiasK]

@Lumos

I did not check make check

It seems to be related to the "-align" compiler flag that you use. Can you please try again with:

./configure CFLAGS="-fPIC -O3 -xHost" FFLAGS="-fPIC -O3 -xHost" CXXFLAGS="-fPIC -O3 -xHost" CC=mpiicx CXX=mpiicpx F90=mpiifx FC=mpiifx --enable-shared --enable-parallel --enable-fortran --with-zlib=yes --with-szlib=no

Where did you get the compiler flags from?

[Lumos]

I tried your suggestion and it seems to solve my problem. My compiler flag is the one I used when I used OneAPI 2022.

[Lumos]

What a bummer. make check gets a new error:

Testing  -- Collective I/O with Independent metadata writes (COLLIO_INDMDWR)
===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 0 PID 30288 RUNNING AT pam2
=   KILLED BY SIGNAL: 14 (Alarm clock)
===================================================================================

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 1 PID 30289 RUNNING AT pam2
=   KILLED BY SIGNAL: 14 (Alarm clock)
===================================================================================

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 2 PID 30290 RUNNING AT pam2
=   KILLED BY SIGNAL: 14 (Alarm clock)
===================================================================================

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 3 PID 30291 RUNNING AT pam2
=   KILLED BY SIGNAL: 14 (Alarm clock)
===================================================================================

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 4 PID 30292 RUNNING AT pam2
=   KILLED BY SIGNAL: 14 (Alarm clock)
===================================================================================

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 5 PID 30293 RUNNING AT pam2
=   KILLED BY SIGNAL: 14 (Alarm clock)
===================================================================================
64.74user 26.77system 21:38.86elapsed 7%CPU (0avgtext+0avgdata 511028maxresident)k
7143448inputs+1667208outputs (0major+599930minor)pagefaults 0swaps
make[4]: *** [Makefile:1671: testphdf5.chkexe_] Error 1
make[4]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/testpar'
make[3]: *** [Makefile:1804: build-check-p] Error 1
make[3]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/testpar'
make[2]: *** [Makefile:1652: test] Error 2
make[2]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/testpar'
make[1]: *** [Makefile:1393: check-am] Error 2
make[1]: Leaving directory '/LIBRARIES/intel2024/hdf5-1.14.3/testpar'
make: *** [Makefile:730: check-recursive] Error 1

[TobiasK]

@Lumos

that seems to be a MPI problem, however, since you are running on an unsupported OS I cannot help you much here.

Maybe the HDF5 group can provide more insights here.

[Lumos]

Thank you for your reply.

In fact, I have encountered similar problems when using intel oneapi2022. I have asked about it in hdf5 group, but I have not received any reply.

make[4]： *** [t_2Gio.chkexe_] Error 1 - HDF5 Library - HDF 论坛 --- Make[4]: *** [t_2Gio.chkexe_] Error 1 - HDF5 Library - HDF Forum (hdfgroup.org)

In addition, I have suggested that the administrator of the machine upgrade the system, but this can not be implemented right away, do you have any recommended open source system?

[TobiasK]

@Lumos

sorry to hear that. As for the OS I would of course try to stick to one of the OS we list in our support matrix.

https://www.intel.com/content/www/us/en/developer/articles/system-requirements/mpi-library-system-requirements.html

[Lumos]

Thank you very much for your help and advice. I'll keep trying. In fact, when I used oneapi2022.3 to compile hdf5, although make check did not pass, I successfully make install hdf5, I do not know if there will be any impact after use. In addition, I will tell your suggestion to the administrator of our machine for reference.

Thanks again!