Charm++ parallel programming system

[Evan Ramos <evan AT hpccharm.com>]

It looks like you're trying to compile NAMD. A more recent version of NAMD should work with Charm++ 6.9.0.

--

Evan A. Ramos

Software Engineer

Charmworks, Inc.

On Tue, Mar 12, 2019 at 5:35 PM Joshua Studen <joshua.studen AT gmail.com> wrote:

Here is an example of the compilation error I get when compiling against charm 6.9.0:

icpc -I/home/jstuden/charm-6.9.0/mpi-linux-x86_64-smp///include -DCMK_OPTIMIZE=1 -Isrc -Iinc    -Iplugins/include -DSTATIC_PLUGIN -I/usr//include -DNAMD_TCL -I/home/jstuden/fftw_install/include -DNAMD_FFTW     -DNAMD_VERSION=\"2.11\" -DNAMD_PLATFORM=\"Linux-x86_64\"  -DREMOVE_PROXYRESULTMSG_EXTRACOPY -DNODEAWARE_PROXY_SPANNINGTREE -DUSE_NODEPATCHMGR    -DUSE_CKLOOP=1 -O3 -xCORE-AVX512 -mcmodel medium -shared-intel -DNAMD_KNL -std=c++11  -o obj/Compute.o -c src/Compute.C

src/Compute.C(38): error: _expression_ must have class type

    ldObjHandle.id.id[0] = -1;

I wonder if this would be an easy change I could make in the application directly?

On Tue, Mar 12, 2019 at 2:13 PM Joshua Studen <joshua.studen AT gmail.com> wrote:

Okay, I get a little further, but I get a segfault now:

 ./charmrun ./hello ++autoProvision

Running on 1 processors:  ./hello ++autoProvision

charmrun>  /usr/bin/setarch x86_64 -R  mpirun -np 1  ./hello ++autoProvision

Charm++> Running on MPI version: 3.1

Charm++> level of thread support used: MPI_THREAD_FUNNELED (desired: MPI_THREAD_FUNNELED)

Charm++> Running in SMP mode: numNodes 1,  1 worker threads per process

Charm++> The comm. thread both sends and receives messages

Converse/Charm++ Commit ID: v6.7.0-0-g46f867c-namd-charm-6.7.0-build-2015-Dec-21-45876

CharmLB> Load balancer assumes all CPUs are same.

Charm++> Running on 1 unique compute nodes (72-way SMP).

Charm++> cpu topology info is gathered in 0.002 seconds.

WARNING: ++autoProvision is a command line argument beginning with a '+' but was not parsed by the RTS.

If any of the above arguments were intended for the RTS you may need to recompile Charm++ with different options.

Running Hello on 1 processors for 0 elements

[0] Assertion "bound > 0" failed in file ../bin/../include/ckarrayindex.h line 295.

------------- Processor 0 Exiting: Called CmiAbort ------------

Reason:

[0] Stack Traceback:

  [0:0] _ZN8CkLocMgrC1E14CkArrayOptions+0x6ce  [0x4cc9fe]

  [0:1] _ZN16CkIndex_CkLocMgr24_call_CkLocMgr_marshall1EPvS0_+0x31d  [0x4cc00d]

  [0:2] CkDeliverMessageFree+0x95  [0x4b0965]

  [0:3] CkCreateLocalGroup+0xab2  [0x49da42]

  [0:4] _Z12_createGroup10_ckGroupIDP8envelope+0x25c  [0x49cf6c]

  [0:5] CkCreateGroup+0x148  [0x49ecc8]

  [0:6] _ZN15CProxy_CkLocMgr5ckNewERK14CkArrayOptionsPK14CkEntryOptions+0x151  [0x4d4df1]

  [0:7]   [0x50275d]

  [0:8] _ZN16CProxy_ArrayBase13ckCreateArrayEP14CkArrayMessageiRK14CkArrayOptions+0x356  [0x508fc6]

  [0:9] _ZN12CkIndex_Main19_call_Main_CkArgMsgEPvS0_+0x17f  [0x48cf2f]

  [0:10] _Z10_initCharmiPPc+0x1eac  [0x493c4c]

  [0:11]   [0x616502]

  [0:12]   [0x613aeb]

  [0:13] +0x8744  [0x7ffff7bc7744]

  [0:14] clone+0x6d  [0x7ffff3402aad]

Abort(1) on node 0 (rank 0 in comm 496): application called MPI_Abort(comm=0x84000002, 1) - process 0

------------- Processor 1 Exiting: Caught Signal ------------

Reason: Segmentation fault

On Tue, Mar 12, 2019 at 2:07 PM Evan Ramos <evan AT hpccharm.com> wrote:

Hi Joshua,

The ++autoProvision parameter was introduced in Charm++ 6.9.0, so it will not be possible to use it with 6.7.0. Additionally, it is only implemented for the netlrts and verbs machine layers and is not available with MPI.

What kinds of compilation errors do you get with 6.9.0?

Regards,

Evan A. Ramos

Software Engineer

Charmworks, Inc.

On Tue, Mar 12, 2019 at 3:55 PM Joshua Studen <joshua.studen AT gmail.com> wrote:

Okay, I have gotten this to build. The app that I am using right now uses charm 6.7.0. If I compile against 6.9.0, I get compilation errors, so I am sticking with 6.7.0 for now. 

When I try something with ppn, I get mpirun related errors. Here's an example using the simplearrayhello test:

./charmrun ++autoProvision ./hello

Running on 1 processors:  ++autoProvision ./hello

charmrun>  /usr/bin/setarch x86_64 -R  mpirun -np 1  ++autoProvision ./hello

[skynet] HYD_spawn (../../../../../src/pm/i_hydra/libhydra/spawn/hydra_spawn.c:113): execvp error on file ++autoProvision (No such file or directory)

Any advice?

On Thu, Mar 7, 2019 at 1:41 PM Kanakagiri, Raghavendra <raghaven AT illinois.edu> wrote:

Hi Joshua Studen,

There seems to be no issue with the steps you are following.

Is it a fresh build? In other words, are you running the build command on a previous installation?

Can you try a fresh smp build, and also run the test program without specifying the hosts parameter (just trying to get the basic setup running first).

Also, is there any particular reason to use the 6.7.0 version of Charm++? (The last update to this version was in 2015). We recommend you to try version 6.9.0.

If it doesn't work, please don't hesitate to email us back.

 

 

Thank you,

Raghavendra

 

From: Joshua Studen <joshua.studen AT gmail.com>
Reply-To: Joshua Studen <joshua.studen AT gmail.com>
Date: Thursday, 7 March 2019 at 10:49 AM
To: <charm AT lists.cs.illinois.edu>
Subject: [charm] Can't get seem to build an SMP version of charm++

 

Hi All,

I am trying to compile an SMP version of charm++, but I can't seem to get it to work. I always get an error:

 

Reason: +ppn cannot be used in non SMP version!

 

Here is how I compile:

 

./build charm++ mpi-linux-x86_64 smp --with-production -j8 --destination=$HOME/charm_install

 

I then try to sample test program:

 

~/charm_install/tests/charm++/simplearrayhello> ./charmrun -hosts c002n0008 ./hello +p 2 +ppn 2

 

charmrun>  /usr/bin/setarch x86_64 -R  mpirun -np 1  -hosts c002n0008 ./hello +ppn 2

------- Partition 0 Processor 0 Exiting: Called CmiAbort ------

Reason: +ppn cannot be used in non SMP version!

 

The charm version is 6.7.0. Any advice on what I'm doing wrong?

 

Regards,

Josh