August 2017 - discuss - mailman3.cels.anl.gov

need help to run hybrid code
by Pasha Pashaei 14 Aug '17

14 Aug '17

Dear Colleague Hi I am working on hybrid MPI -OpenMp code and face with a problem. I can not run the code as the way that I want. I can run pure mpi code and also pure openmp code, but not hybrid one. I used "export OMP_NUM_THREADS=6" and then "mpirun -np 4 ./hybrid.exe > output.txt" It run but I noticed that It used only virtual threads of CPUs not physical threads. I have both OpenMpi and Mpich on my computer. My problem is I want to use mpi for communication between nodes and sockets and Openmp for cores which are inside of sockets and nodes, I do not know how to pin cpu to each other in each socket and nodes. I could find some material on internet such as https://aciref.org/how-to-gain-hybrid-mpi-openmp-code-performance-without-c… How to gain hybrid MPI-OpenMP code performance without ...<https://aciref.org/how-to-gain-hybrid-mpi-openmp-code-performance-without-c…> aciref.org How to gain hybrid MPI-OpenMP code performance without changing a line of code a.k.a. dealing with task affinity But it did not work, I think I need a PBS file to assign those commands on my grid with qsub command. If you need more information please let me know. I am looking forward to hearing from you, Best regards. Pasha Pashaei _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Need help for running hybrid code
by Pasha Pashaei 14 Aug '17

14 Aug '17

Dear Colleague Hi I am working on hybrid MPI -OpenMp code and face with a problem. I can not run the code as the way that I want. I can run pure mpi code and also pure openmp code, but not hybrid one. I used "export OMP_NUM_THREADS=6" and then "mpirun -np 4 ./hybrid.exe > output.txt" It run but I noticed that It used only virtual threads of CPUs not physical threads. I have both OpenMpi and Mpich on my computer. My problem is I want to use mpi for communication between nodes and sockets and Openmp for cores which are inside of sockets and nodes, I do not know how to pin cpu to each other in each socket and nodes. I could find some material on internet such as https://aciref.org/how-to-gain-hybrid-mpi-openmp-code-performance-without-c… How to gain hybrid MPI-OpenMP code performance without ...<https://aciref.org/how-to-gain-hybrid-mpi-openmp-code-performance-without-c…> aciref.org How to gain hybrid MPI-OpenMP code performance without changing a line of code a.k.a. dealing with task affinity But it did not work, I think I need a PBS file to assign those commands on my grid with qsub command. If you need more information please let me know. I am looking forward to hearing from you, Best regards. Pasha ________________________________ Pasha Pashaei _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] mpi_comm_spawn - process not destroyed
by Stanislav Simko 13 Aug '17

13 Aug '17

Hi Min,attached are configure and make logs for the build that I did for myself on our local cluster with Intel compilers. I.e., build is definitely not guaranteed to be perfect/optimal. But please keep in mind that I get the same behaviour with standard MPICH package in Fedora distribution - I think that configuration options should be available online, maybe prepared by someone from MPICH? Also, I tested with simple c++ hello world like program and it's the same. Thank you.best,stanislav. On Sun, 2017-08-13 at 19:58 +0100, Min Si wrote: > Hi Stanislav, > > > > This seems interesting. Could you please also attach the MPICH > config.log ? You can find under the directory where you build > MPICH. > I will look into this problem then and keep you updated. > > > > Min > > > > On 8/11/17 1:31 PM, Stanislav Simko > wrote: > > > > > > > Dear all, > > > > > > > > I'm just trying some very basic stuff with MPI_COMM_SPAWN in > > python (i.e. I use mpi4py package), but I see behaviour > > that I > > do not understand - the child process gets spawned, does > > its > > stuff and then "should" finish. I see though, that the > > process > > created for the child stays alive. I see this only with the > > MPICH, OPENMPI does what I would (naively) expect. In this > > way I > > can end up with N "ghost" process, after calling SPAWN N- > > times. > > My minimal working example is following: > > > > > > > > > > > > > > ______________________________________________ > > parent.py > > > > > > > > from __future__ import print_function > > from mpi4py import MPI > > comm = MPI.COMM_WORLD > > spawned = > > MPI.COMM_SELF.Spawn(sys.executable,args=['child.py'],maxpro > > cs=1) > > print("parent process is waiting for child") > > spawned.Barrier() > > > > > > > > > > > > > > ______________________________________________ > > child.py > > > > > > > > from __future__ import print_function > > from mpi4py import MPI > > parent = MPI.Comm.Get_parent() > > # just do some stupid stuff that takes a bit of time > > for i in range(5000000): > > a = i*i+1-(i*10) + math.sin(math.pow(i,i%8)) > > parent.Barrier() > > > > > > > > ______________________________________________ > > > > > > > > I run with e.g.: > > mpirun -n 1 python mpi.py > > > > > > > > Do I miss something with SPAWN method? > > (I tested on two independent systems, our local cluster with > > mpich v3.0.4, and my laptop - fedora 26, mpich v3.2.8 from > > repositories) > > > > > > > > thank you very much for suggestions. > > > > > > > > Regards, > > stanislav. > > > > > > > > > > > > > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > > > > > > _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] mpi_comm_spawn - process not destroyed
by Min Si 13 Aug '17

13 Aug '17

Hi Stanislav, This seems interesting. Could you please also attach the MPICH config.log ? You can find under the directory where you build MPICH. I will look into this problem then and keep you updated. Min On 8/11/17 1:31 PM, Stanislav Simko wrote: > Dear all, > > I'm just trying some very basic stuff with MPI_COMM_SPAWN in python > (i.e. I use mpi4py package), but I see behaviour that I do not > understand - the child process gets spawned, does its stuff and then > "should" finish. I see though, that the process created for the child > stays alive. I see this only with the MPICH, OPENMPI does what I would > (naively) expect. In this way I can end up with N "ghost" process, > after calling SPAWN N-times. My minimal working example is following: > > > ______________________________________________ > parent.py > > from __future__ import print_function > from mpi4py import MPI > comm = MPI.COMM_WORLD > spawned = MPI.COMM_SELF.Spawn(sys.executable,args=['child.py'],maxprocs=1) > print("parent process is waiting for child") > spawned.Barrier() > > > ______________________________________________ > child.py > > from __future__ import print_function > from mpi4py import MPI > parent = MPI.Comm.Get_parent() > # just do some stupid stuff that takes a bit of time > for i in range(5000000): > a = i*i+1-(i*10) + math.sin(math.pow(i,i%8)) > parent.Barrier() > > ______________________________________________ > > I run with e.g.: > mpirun -n 1 python mpi.py > > Do I miss something with SPAWN method? > (I tested on two independent systems, our local cluster with mpich > v3.0.4, and my laptop - fedora 26, mpich v3.2.8 from repositories) > > thank you very much for suggestions. > > Regards, > stanislav. > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] test failures on armhf architecture on Debian
by Jeff Hammond 13 Aug '17

13 Aug '17

Those failures are in OpenPA, which is a separate component you can test by itself for convenience. http://git.mpich.org/openpa.git/ is the official project source. You should try different implementations. The GCC intrinsics should be portable but if they don't work, that's a bug either in the compiler or OpenPA. The most portable OpenPA port uses Pthread locks. That's quite slow but should not fail. You may need to force that if the ARM assembly port and the GCC __sync intrinsics ones break. I wrote a C11 port with https://github.com/jeffhammond/OpenPA that has not been tested anywhere besides x86 but could be better if your compiler does a good job with C11. It's trivial to make a __atomic or __c11 intrinsics port for GCC or Clang if that is necessary. Those should be more efficient on ARM than __sync intrinsics, if used properly. My only programmable ARM system is an Adapteva Parallella board that hasn't been powered on in 3 years, so I'm not in a good position to test anything, but I'll hack on OpenPA if necessary. Jeff Sent from my iPhone > On Aug 13, 2017, at 4:23 AM, Alastair McKinstry <mckinstry(a)debian.org> wrote: > > Hi, > > I've built 3.3a2 on Debian on all supported architectures. Failures were > seen on the armhf architecture (Arm with hardware float support, > 32-bit v7 architecture). > > Testing integer swap with 4 threads *FAILED* > at test_primitives.c:3399 in test_threaded_swap_int()... > Value 0 was encountered 0 times. Expected: 1 > Testing pointer swap with 4 threads *FAILED* > at test_primitives.c:3554 in test_threaded_swap_ptr()... > Value 3 was encountered 2 times. Expected: 1 > > Details: https://buildd.debian.org/status/fetch.php?pkg=mpich&arch=armhf&ver=3.3%7Ea… > > Is anyone working on this? If necessary I can do testing on armhf systems > > Best regards > Alastair > > -- > Alastair McKinstry, <alastair(a)sceal.ie>, <mckinstry(a)debian.org>, https://diaspora.sceal.ie/u/amckinstry > Misentropy: doubting that the Universe is becoming more disordered. H > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

test failures on armhf architecture on Debian
by Alastair McKinstry 13 Aug '17

13 Aug '17

Hi, I've built 3.3a2 on Debian on all supported architectures. Failures were seen on the armhf architecture (Arm with hardware float support, 32-bit v7 architecture). Testing integer swap with 4 threads *FAILED* at test_primitives.c:3399 in test_threaded_swap_int()... Value 0 was encountered 0 times. Expected: 1 Testing pointer swap with 4 threads *FAILED* at test_primitives.c:3554 in test_threaded_swap_ptr()... Value 3 was encountered 2 times. Expected: 1 Details: https://buildd.debian.org/status/fetch.php?pkg=mpich&arch=armhf&ver=3.3%7Ea… Is anyone working on this? If necessary I can do testing on armhf systems Best regards Alastair -- Alastair McKinstry, <alastair(a)sceal.ie>, <mckinstry(a)debian.org>, https://diaspora.sceal.ie/u/amckinstry Misentropy: doubting that the Universe is becoming more disordered. H _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

mpi_comm_spawn - process not destroyed
by Stanislav Simko 11 Aug '17

11 Aug '17

Dear all, I'm just trying some very basic stuff with MPI_COMM_SPAWN in python (i.e. I use mpi4py package), but I see behaviour that I do not understand - the child process gets spawned, does its stuff and then "should" finish. I see though, that the process created for the child stays alive. I see this only with the MPICH, OPENMPI does what I would (naively) expect. In this way I can end up with N "ghost" process, after calling SPAWN N-times. My minimal working example is following: ______________________________________________ parent.py from __future__ import print_function from mpi4py import MPI comm = MPI.COMM_WORLD spawned = MPI.COMM_SELF.Spawn(sys.executable,args=['child.py'],maxprocs=1) print("parent process is waiting for child") spawned.Barrier() ______________________________________________ child.py from __future__ import print_function from mpi4py import MPI parent = MPI.Comm.Get_parent() # just do some stupid stuff that takes a bit of time for i in range(5000000): a = i*i+1-(i*10) + math.sin(math.pow(i,i%8)) parent.Barrier() ______________________________________________ I run with e.g.: mpirun -n 1 python mpi.py Do I miss something with SPAWN method? (I tested on two independent systems, our local cluster with mpich v3.0.4, and my laptop - fedora 26, mpich v3.2.8 from repositories) thank you very much for suggestions. Regards, stanislav. _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] mpich2-1.3.1 make error
by Halim Amer 07 Aug '17

07 Aug '17

First, you are using a very old version of MPICH. Second, we no longer support MPICH on Windows; try Microsoft MPI. Halim www.mcs.anl.gov/~aamer On 8/7/17 12:40 AM, 163邮箱团队 wrote: > Hi: > My mpich2-1.3.2 make is failing (configure seems to complete > successfully). I am testing this on Windows 7. Attached is the c.txt and > m.txt files. I am configuring this using: configure > --prefix=/public/home/szl1978/programfiles/mpich2 2>&1 | tee c.txt, and > make VERBOSE=1 2>&1 | tee m.txt > Any help to resolve this error is appreciated! > Thanks. > > > > > > > > > > > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

mpich2-1.3.1 make error
by 163邮箱团队 07 Aug '17

07 Aug '17

Hi: My mpich2-1.3.2 make is failing (configure seems to complete successfully). I am testing this on Windows 7. Attached is the c.txt and m.txt files. I am configuring this using: configure --prefix=/public/home/szl1978/programfiles/mpich2 2>&1 | tee c.txt, and make VERBOSE=1 2>&1 | tee m.txt Any help to resolve this error is appreciated!Thanks.

1 0