Charm++ parallel programming system

[Jozsef Bakosi <jbakosi AT gmail.com>]

Hi folks,

I'm trying to use MSAs and running into a runtime error producing the following trace:

===

------------- Processor 7 Exiting: Called CmiAbort ------------

Reason: Error! This group proxy has not been initialized!

[5] Stack Traceback:

  [5:0] CmiAbortHelper+0xc1  [0x8284a1]

  [5:1] CmiAbort+0x2b  [0x825ebb]

  [5:2] _ZNK12CProxy_Group7ckCheckEv+0x31  [0x5a4671]

  [5:3] _ZNK28CProxyElement_MSA_CacheGroupIm12DefaultEntryImLb0EELj1024EE7ckCheckEv+0x15  [0x5a44f5]

  [5:4] _ZN28CProxyElement_MSA_CacheGroupIm12DefaultEntryImLb0EELj1024EE7AckPageEjPK14CkEntryOptions+0x29  [0x5ab799]

  [5:5] _ZN13MSA_PageArrayIm12DefaultEntryImLb0EELj1024EE16PAReceiveRLEPageEPK15MSA_WriteSpan_tjPKmji16MSA_Page_Fault_t+0x122  [0x5ab632]

  [5:6] _ZN21CkIndex_MSA_PageArrayIm12DefaultEntryImLb0EELj1024EE32_call_PAReceiveRLEPage_marshall5EPvS3_+0x13f  [0x5ac04f]

  [5:7] CkDeliverMessageFree+0x41  [0x706d21]

  [5:8] _ZN14CkLocRec_local11invokeEntryEP12CkMigratablePvib+0x1ec  [0x72a16c]

  [5:9] _ZN14CkLocRec_local7deliverEP14CkArrayMessage11CkDeliver_ti+0x218  [0x72a578]

  [5:10] _ZN8CkLocMgr7deliverEP9CkMessage11CkDeliver_ti+0x31d  [0x72d2bd]

  [5:11] _ZN8CkLocMgr13deliverInlineEP9CkMessage+0x24  [0x70e854]

  [5:12]   [0x708adc]

  [5:13] _Z15_processHandlerPvP11CkCoreState+0x193  [0x708633]

  [5:14] CmiHandleMessage+0x48  [0x82b378]

  [5:15] CsdScheduleForever+0xc6  [0x82b6a6]

  [5:16] CsdScheduler+0x1a  [0x82b3ca]

  [5:17] StartInteropScheduler+0x12  [0x84d952]

  [5:18] CharmLibInit+0x88  [0x825b58]

  [5:19] main+0x4ab  [0x545acb]

  [5:20] __libc_start_main+0xf5  [0x7ffff50b6b45]

  [5:21]   [0x528e24]

===

Following the example in examples/multiphaseSharedArrays/histogram, I create a 1D MSA, successfully fill it up, and try to switch to reading mode by calling syncToRead() which is when the problem occurs and only on certain number of CPUs and with certain array size, e.g., 1-7 all fine, 8 error, 9-15 fine, 16 error with array size 34168; all fine on all CPUs with array size 10000. Here is some code describing what I'm doing, all this is called from the main chare on a single CPU:

using MSA1D = MSA::MSA1D< std::size_t, DefaultEntry< std::size_t >, MSA_DEFAULT_ENTRIES_PER_PAGE >;

MSA1D msa_elements( 34168, 1, MSA_DEFAULT_MAX_BYTES );  // 31468 entries, 1 worker

msa_elements.enroll( 1 );

// write

MSA1D::Write msa_writer = msa_elements.getInitialWrite();

for (int i=0; i<34168; ++i) msa_writer.set(i) = i;

// switch to read

MSA1D::Read msa_reader = msa_writer.syncToRead();   // error if entries = 31468, no error if entries = 10000

Does anyone have an idea what I'm missing?

Thanks,

Jozsef