OpenCL SNP Simulation

An OpenCL implementation of a simulator for Spiking Neural Systems.

The project uses OpenCL 1.2 libraries on top of C++14.

Limitations

• Non-determenism is not yet supported
• Built-in Environment not yet supported
  • This means you have still have to create an "Environment neuron" to simulate an environment
  • As an effect, input and output spike train not yet supported
• Tested upto input with 2752 neurons and 4032 rules

Dependencies

This project is hardware dependent so machines with different GPUs have different dependencies.

GENERAL DEPENDENCIES

• Git
• OpenCL ICD Loader
• OpenCL headers
• C++ Boost library
• GNU Make
• GNU g++
• Google re2

MACHINES WITH NVIDIA CARD

• nVidia drivers for the card in your machine
• CUDA SDK
• opencl-nvidia (Needed for execution) - cuda (Note: Nvidia implementation only currently supports OpenCL 1.1)

MACHINES WITH ATI CARD

• ATI drivers for the card in your machine

• AMD App SDK

• opencl-catalyst or opencl-mesa

• amdapp-sdk

Software Versions

Linux Distributions

• git 2.7.0
• g++ 5.3.0
• boost 1.6.0
• make 4.1
• cuda 8.0

Windows

• git 2.7.0
• g++ 4.8.1
• boost 1.55.0
• cuda 8.0

Installing Dependencies

Arch-Linux and Arch-based Linux Distributions

1. Update repositories

   $ sudo pacman -Syu

2. Install packer

   $ sudo pacman -S wget expac jshon
   $ mkdir packer
   $ cd packer
   $ sudo wget https://aur.archlinux.org/cgit/aur.git/plain/PKGBUILD?h=packer
   $ mv PKGBUILD?h=packer PKGBUILD
   $ makepkg
   $ sudo pacman -U packer-*********-*-any.pkg.tar.xz
   $ cd ..
   $ sudo rm -dR packer 
   

3. Install gcc

   $ sudo pacman -S gcc

4. Install and configure git

   $ sudo pacman -S git
   $ git config --global user.name "your name"
   $ git config --global user.email your@email.com

5. Install OpenCL headers and ICD Loaders

   $ sudo pacman -S ocl-icd opencl-headers

6. Install GPU dependent dependencies

   AMD

   ```bash
   $ sudo packer -S opencl-catalyst amdapp-sdk
   ```
   

   NVIDIA

   ```bash
   $ sudo pacman -S opencl-nvidia cuda
   ```
   

7. Install boost

   $ sudo pacman -S boost

8. Install re2

   $ sudo pacman -S re2-git

Ubuntu and Ubuntu based systems

1. Update apt repositories

   $ sudo apt-get update
   

2. Install git

   $ sudo apt-get install git
   

3. Install boost

   $ sudo apt-get install libboost-all-dev
   

4. Install boost compute: https://boostorg.github.io/compute/boost_compute/getting_started.html

5. Install google re2 https://github.com/google/re2/wiki/Install

6. Install opencl icd headers

   $ sudo apt install ocl-icd-opencl-dev
   

7. Install AMD App SDK or CUDA SDK

Windows

1. Install the latest driver for your GPU

2. Install and configure git

   1. Open git bash

   2. Set global variables

      > git config --global user.name "your name"
      > git config --global user.email your@email.com

3. Install OpenCL SDK

   1. For NVIDIA install CUDA SDK
   2. For AMD install AMD App SDK

4. Install MinGW

   1. Set destination folder as "C:\MinGW" (default)
   2. On installer select
      • mingw-developer-toolkit
      • mingw32-base
      • mingw32-gcc-g++
      • msys-base
      • msys-make (optional)
   3. Add C:\MinGW\bin to PATH environment variable

5. Download Boost 1.60 for windows then extract to C:\

   1. Rename folder to boost
   2. cd into boost folder

    ```
    > cd C:\boost
    ```
   

   3. Build boost

    ```
    > bootsrap mingw
    > b2 toolset=gcc
    ````
   

Building

On Linux Distros

1. CD into project root

2. Build the project using make

   $ make

On Windows

1. CD into project root

2. CD into scripts

   > cd scripts
   

3. Run compile.bat

   > compile
   

Running

On Linux Distros

1. cd into bin from project root

   $ cd bin

2. For Parallel

   $ ./oclsnp <input_binary> [--o output_file] [--txt | --silent]

3. For Linear

   $ ./linsnp <input_binary> [--o output_file] [--txt | --silent]

On Windows

1. cd into bin from project root

   > cd bin
   

2. For Parallel

   > oclsnp <input_binary>
   

3. For Linear

   > linsnp <input_binary>