- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I am trying to compile Intel(R) Xeon(R) E5-2670 v2 to verify the benchmarks , published here .I first want to verify the performance benchmark on my host processor , though later i will compile it for the Intel MIC. Presently i am am stuck at an error (attached: make_logs_unseting4.txt) regarding mic routines even though i have not enabled any offload flag( unset USE_OFFLOAD
after sourcing
nwchem_make_env.sh.txt
).
Variables on my path:(attached: nwchem_make_env.sh.txt)
Please help me regarding finding out the root cause of this problem.
Eagerly awaiting your reply,
Regards,
Puneet
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Dear Puneet,
Can you please tell me which version of Intel Composer you are using? Do you have access to a later version of the compiler suite? The problem should be gone with versions 15.x of Intel Composer.
Please let me know if this resolves your problem.
Cheers,
-michael
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
UPDATE: The multiple definition error creeps in only if i set the USE_OFFLOAD=1. If i start afresh without setting USE_OFFLOAD in my environment , the compilation succeeds and binary is created, but the on running the binary ,hovever, the process gets killed with error(will post error in separate thread)
Dear Michael,
I am using evaluation version of composer_xe_2015.0.090 (which i guess is latest!) & in the make_logs_unseting4.txt file you can find the compilation log.I posted the question at NWChem Forum, there is no reply from their side yet, so now, I guess intel forum is the suitable platform for the resolution of this issue.
I went through previous Xeon Phi threads and was unable to find issue similar to mine.If you came across this issue then please help me with pointers to resources also, If you can then please try compiling NWChem for Xeon Phi Cards and suggest correct compiler/flag settings.
Eagerly awaiting your reply,
Regards,
Puneet
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Dear Puneet,
Hm... This is really strange. Can you please checkout the latest version of NWChem from the Subversion repository (you can find it on the NWChem website). I have seen a few commits that also touch that code, so maybe the developers have done some more work on making this issue go away for more cases.
Cheers,
-michael
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I download the 6.5 version and tried to compile it myself using icc+ifort 15; I even applied two ifort15 patches from the website. I get the exact same error that Puneet is seeing.
Further investigation shows that the ifort linking stage does not appreciate it if a (static) library is listed twice, i.e.
ifort -o newchem .... -lnwcutil .... -lnwcutil ....
fails. The order of the libraries is also quite important. Eventually I ended up with the following commandline to successfully link 'nwchem':
ifort -Ltools/install/lib -L../lib/LINUX64 -o ../bin/LINUX64/nwchem nwchem.o stubs.o -lnwctask -lccsd -lmcscf -lselci -lmp2 -lmoints -lstepper -ldriver -loptim -lnwdft -lgradients -lcphf -lesp -lddscf -ldangchang -lguess -lhessian -lvib -lrimp2 -lproperty -lsolvation -lnwints -lprepar -lnwmd -lnwpw -lofpw -lpaw -lpspw -lband -lnwpwlib -lnwxc -lcafe -lspace -lanalyze -lqhop -lpfft -ldplot -lnwpython -ldrdy -lvscf -lqmmm -lqmd -letrans -lpspw -ltce -lbq -lcons -lperfm -ldntmc -lccca -lnwcutil -lga -larmci -lpeigs -lcons -lbq -lperfm /usr/lib64/python2.6/config/libpython2.6.so -mkl -openmp -lmkl_scalapack_ilp64 -lmkl_blacs_intelmpi_ilp64 -lpthread -lm -mkl -openmp -lpthread -lm -L/opt/intel/impi/5.0.1.035/intel64/lib -lmpi -lmpigf -lmpigi -lrt -lpthread -loffload -lpthread -lutil -ldl -lz
Now I have a working executable but when I launch it on my host with a single Xeon Phi it always returns
offload error: cannot offload to MIC - device is not available
I guess this is an MPI setup issue on my end.
However, as for the 'ifort' link stage failing: could this be a bug in the ifort 15 linker?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Puneet
I have posted a reply to your question on NWChem forum
http://www.nwchem-sw.org/index.php/Special:AWCforum/st/id1614
Cheers, Edo
puneet s. wrote:
I am trying to compile Intel(R) Xeon(R) E5-2670 v2 to verify the benchmarks , published here .I first want to verify the performance benchmark on my host processor , though later i will compile it for the Intel MIC. Presently i am am stuck at an error (attached: make_logs_unseting4.txt) regarding mic routines even though i have not enabled any offload flag( unset USE_OFFLOAD after sourcing nwchem_make_env.sh.txt ).
Variables on my path:(attached: nwchem_make_env.sh.txt)
Please help me regarding finding out the root cause of this problem.Eagerly awaiting your reply,
Regards,
Puneet
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
How do i convert a normal nwchem input.nw file running on CPU to nwchem cuda input file.?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page