Centos 7.3 - do I really need OFED from openfabric.org ?

[lejeczek]

hi everyone

One quick question - do I really really need OFED from openfabrics.org and all the packages from centos/rhel + epel repos will NOT do?

Reason I'm asking is - compilation/installation of OFED 3.18.x which MPSS manuals says to use fails. Compiler fails somewhere in rdma-compat I think.

[Rob_J_]

In the MPSS Users Guide there are four distributions mentioned. See section 3.7.1

[lejeczek]

And both 3.18 and newer which is not metioned 4.8.1 fail to build on my vanilla Centos:

$ perl install.pl --with-xeon-phi --basic

...

Running rpmbuild --rebuild --define 'PSM_HAVE_SCIF 1' --define '_topdir /var/tmp//OFED_topdir' --nodeps --define '_dist .unsupported' --define 'configure_options   --with-core-mod --with-user_mad-mod --with-user_access-mod --with-addr_trans-mod --with-mlx4-mod --with-mlx4_en-mod --with-mlx5-mod --with-cxgb3-mod --with-cxgb4-mod --with-nes-mod --with-qib-mod --with-ipoib-mod --with-qed-mod --with-firmware-mod --with-qede-mod --with-qedr-mod --with-bnxt_en-mod --with-bnxt_re-mod' --define 'KVERSION 3.10.0-693.21.1.el7.x86_64' --define 'K_SRC /lib/modules/3.10.0-693.21.1.el7.x86_64/source' --define 'K_SRC_OBJ /lib/modules/3.10.0-693.21.1.el7.x86_64/build' --define '_release 1.1.gd5db855.3.10.0_693.21.1.el7.x86_64' --define 'network_dir /etc/sysconfig/network-scripts' --define '_prefix /usr' --define '__arch_install_post %{nil}' /root/rpmbuild/tars/OFED-4.8-1/SRPMS/compat-rdma-4.8-1.1.gd5db855.src.rpm
Failed to build compat-rdma RPM
See /tmp/OFED.16178.logs/compat-rdma.rpmbuild.log

I see that many of those rpm packeges OFED from openfabrics.org is trying to install are already there in default/epel repos so I am hoping maybe rather those than adding extra weight, if it can work?

With OFED installation problem - where to look for help, here or seek some at openfabric.org?

[Rob_J_]

Like I said, there are three other distros available: TSR 7.4.2, OFED 3.12-1 and Mellanox OFED 2.4

The build line you used also varies from the instructions. There's no --basic appended.

[Rocha__João]

Dear all,

I'm trying to run an offload .c code but i'm having trouble getting some libs in the right place, although i've put (so i suppose) the correct path's on my bashrc. I get this line when running the executable:

error while loading shared libraries: liboffload.so.5: cannot open shared object file: No such file or directory

Location for the lib:

/opt/intel/composer_xe_2015.1.133/compiler/lib/intel64/liboffload.so.5
/opt/intel/composer_xe_2015.1.133/compiler/lib/mic/liboffload.so.5

which i've exported in bashrc as:

export SINK_LD_LIBRARY_PATH="/opt/intel/composer_xe_2015.1.133/compiler/lib/intel64:$SINK_LD_LIBRARY_PATH"
export SINK_LD_LIBRARY_PATH="/opt/intel/composer_xe_2015.1.133/compiler/lib/mic:$SINK_LD_LIBRARY_PATH"

I've tried to -rpath and -rpath-link but still doesn't work. Is my line wrongly written?
(e.g.): icc -fopenmp -rpath=/opt/intel/composer_xe_2015.1.133/compiler/lib/intel64 flops3_offload.c -o flops3_offload

Can someone give a clue on this?
Thanks a lot,

João Rocha

[Rob_J_]

Don't know the answer to that. You probably should start a new thread since it's not related to OFED.