hmm this is not right, the decission how many nodes work together in the distributed kernel should be made elsewhere. Perhaps a DistributedAccess instance should be given to a DataFactory Definition at line 288 of file DataFactory.h. References mDataSet, mDistributedAccess, mKernelMatrixName, and Impala::Core::Matrix::DistributedAccess::Subscribe(). Referenced by Impala::Application::mainCrossValidate(), Impala::Application::mainTrainModel(), Impala::Core::Training::Factory::MakeDataSrc(), and ServeDistributedAccess(). 00289 { 00290 if(mDistributedAccess == 0) 00291 { 00292 if(Link::Mpi::NrProcs() == 1) 00293 mDistributedAccess = new Matrix::DistributedAccess 00294 (mKernelMatrixName, mDataSet, 0, 0, 1); 00295 else 00296 { 00297 mDistributedAccess = new Matrix::DistributedAccess 00298 (mKernelMatrixName, mDataSet, 0, 1, Link::Mpi::NrProcs() - 1); 00299 if(Link::Mpi::MyId() == 0) 00300 mDistributedAccess->Subscribe(); 00301 } 00302 } 00303 return mDistributedAccess; 00304 }
Here is the call graph for this function:
|