neuromapp

The Neuronm(ini)app(lication)library reproduces the algorithms of the main software of the BBP as a collection of mini-apps For its first release, the Neuromapp framework focuses on CoreNeuron application.

The framework is organised under the form

neuronmapp/app

This directory neuromapp contains the main program (C++) that organises all the mini applications. A specific README is added, it gives directives for the plug of new applications.

neuronmapp/coreneuron_1.0

This directory contains miniapps (C) about the main behaviour of first release of coreneuron, 3 miniapps are available: kernel and solver

• kernel: This miniapp extracts the computation kernel of CoreNeuron, the miniapp focuses on the vectorization by the compiler

• solver: This miniapp extracts the linear algbra solver of CoreNeuron, the Hines solver (similar to Thomas Solver for 3 bands matrix)

• cstep: It combines the kernel and the solver to mimic a step time of neuron simulator

• queue: This miniapp simulates the queueing system of CoreNeuron.

neuromapp/hello

This directory provides an example of how could be design a miniapp (C++)

You can disable the compilation of this mini-app by using the following variable in cmake command: '-DNEUROMAPP_DISABLE_HELLO=TRUE'

neuromapp/iobench

This directory contains a miniapp that creates a fake report in a key/value fashion and stores and reads it from a key/value backend. Several parameters can be tuned, like the size of keys and values, the amount of k/v pairs, parallelism (MPI/OMP), randomness of reads and writes, etc. This miniapp is similar to the keyvalue miniapp, but in this case the parameter configuration is more flexible.

For more information, run the mini-app with the --help argument.

You can disable the compilation of this mini-app by using the following variable in cmake command: '-DNEUROMAPP_DISABLE_IOBENCH=TRUE'

neuromapp/keyvalue

This directory contains a miniapp that mimics CoreNeuron simulation loop but outputs the results in a key/value fashion. This miniapp is similar to the iobench miniapp, but in this case the options are more restrict to a BG/Q environment: three different use cases are defined in order to fill 25%, 50% or 75% of the main memory of a BG/Q node.

For more information, run the mini-app with the --help argument.

You can disable the compilation of this mini-app by using the following variable in cmake command: '-DNEUROMAPP_DISABLE_KEYVALUE=TRUE'

neuromapp/readi

This directory contains a miniapp implementing a reaction-diffusion simulator based on operator splitting, as in TetOpSplit solver of STEPS software package. To run the simulation on a simple model with 10 species and 8 reactions, go to the build directory and execute the mini-app with the following arguments:

./neuromapp/app/app readi --filename_mesh ../neuromapp/readi/dataset/data_mesh --filename_model ../neuromapp/readi/dataset/data_model

For more information, run the mini-app with the --help argument.

You can disable the compilation of this mini-app by setting -DNEUROMAPP_DISABLE_READI=TRUE when running cmake.

neuromapp/replib

This directory contains a miniapp that mimics the behavior of Neuron's ReportingLib. It uses MPI I/O collective calls to write a fake report to a shared file. The miniapp provides several options to distribute data across ranks in different ways.

For more information, run the mini-app with the --help argument.

You can disable the compilation of this mini-app by using the following variable in cmake command: '-DNEUROMAPP_DISABLE_REPLIB=TRUE'

neuromapp/nest/synapse

This directory contains a miniapp that simulates synapse models from NEST.

You can disable the compilation of this mini-app by using the following variable in cmake command: '-DNEUROMAPP_DISABLE_NEST=TRUE'

neuromapp/compression

This directory contains a miniapp that simulates in memory compression for neuro science.

You can disable the compilation of this mini-app by using the following variable in cmake command: '-DNEUROMAPP_COMPRESSION=FALSE'

Installation Instructions

1. Create a build directory, for example inside the source directory (NEUROMAPP_SRC_DIR):

$ mkdir $NEUROMAPP_SRC_DIR/build

2. Configure, compile and install the framework. Optionally, specify the desired installation directory (NEUROMAPP_INST_DIR):

$ cd $NEUROMAPP_SRC_DIR/build
$ cmake .. -DCMAKE_INSTALL_PREFIX=$NEUROMAPP_INST_DIR
$ make
$ make install

Note : On OS X with Clang compiler we need to disable compression miniapp using '-DNEUROMAPP_COMPRESSION=FALSE'

If you're using GCC 5.1 and above, you may have issues if your full stack (e.g. BOOST) has not been compiled with C++11 due to the std::string implementation that changes with GCC 4.9. In this case, please add the following flag to your cmake command to tell GCC to not use the C++11 ABI:

-DCMAKE_CXX_FLAGS=-D_GLIBCXX_USE_CXX11_ABI=0

Dependency Handling

Mandatory dependencies are:

• BOOST >= 1.44
• HDF5
• SLURM
• MPI

The following variables can be set in the cmake command to disable the compilation of certain parts of the framework or ignore certain external libraries that would be used otherwise:

'-DNEUROMAPP_DISABLE_HDF5MAPP=TRUE': Ignore the code under neuromapp/hdf5

'-DNEUROMAPP_DISABLE_HELLO=TRUE': Ignore the code under neuromapp/hello

'-DNEUROMAPP_DISABLE_IOBENCH=TRUE': Ignore the code under neuromapp/iobench

'-DNEUROMAPP_DISABLE_KEYVALUE=TRUE': Ignore the code under neuromapp/keyvalue

'-DNEUROMAPP_DISABLE_NEST=TRUE': Ignore the code under neuromapp/nest

'-DNEUROMAPP_DISABLE_REPLIB=TRUE': Ignore the code under neuromapp/replib

'-DNEUROMAPP_DISABLE_CASSANDRA=TRUE': Do not try to find the Cassandra installation

'-DNEUROMAPP_DISABLE_CEPH=TRUE': Do not try to find the CEPH installation

'-DNEUROMAPP_DISABLE_LEVELDB=TRUE': Do not try to find the LevelDB installation

'-DNEUROMAPP_DISABLE_SKV=TRUE': Do not try to find the IBM SKV installation