Quick Start Guide

Installing C++ Compiler and CMake

Building RAD-Sim requires a C++ compiler and CMake (3.19 or higher), so make sure these are installed on your system or install them using the following command:

$ sudo apt install build-essential cmake cmake-curses-gui

Installing SystemC

You have to also install SystemC libraries on your system following these steps:

1. Download the SystemC 2.3.4 (Core SystemC Language and Examples tar.gz) from this link.

2. Unzip the tar archive in your directory of choice (I usually prefer to create a “tools” directory in my home for all the tools I am installing on my system):

   $ tar -xzvf systemc-2.3.4.tar.gz systemc-2.3.4/
   

3. In the systemc-2.3.4 directory, build the SystemC libraries using CMake by executing the following steps:

   • Create a new build directory:

     $ pwd
     <path_to_systemc>/systemc-2.3.4
     $ mkdir build
     $ cd build
     $ ccmake ..
     

   • Press c to configure the CMake build, then press t to toggle into the advanced configuration mode. Make sure to change CMAKE_CXX_STANDARD to 17 and DISABLE_COPYRIGHT_MESSAGE to ON. You can change this entry by pressing enter to type in the new value or toggle between ON/OFF. You can also change any of the other parameters as needed for your system.

   • After changing all needed configuration parameters, press c again to configure and then g to generate the configuration file.

   • Build and install the SystemC libraries by executing the following commands:

     $ make
     ....
     [100%] Linking CXX shared library libsystemc.so
     [100%] Built target systemc
     
     $ sudo make install
     

Installing Python Dependencies

It is preferable to install Anaconda and create a Python virtual environment for the RAD flow by following these steps:

1. Download Anaconda and install it following these instructions

2. Create a new conda environment with all the required packages for the RAD flow using this command:

   $ conda env create -f <rad_flow_root_dir>/rad-flow-env.yml
   Collecting package metadata (repodata.json): /
   ....
   #
   # To activate this environment, use
   #
   #     $ conda activate radflow
   #
   # To deactivate an active environment, use
   #
   #     $ conda deactivate
   

Building RAD-Sim

You can configure RAD-Sim for your example design simulation using the following commands executed at the rad-sim root directory (the commands use the mlp example design which can be replaced by your own design under the rad-flow/rad-sim/example-designs directory):

$ cd <rad_flow_root_dir>/rad-sim
$ python config.py mlp
$ cd build
$ cmake ..
$ make
....
[100%] Linking CXX executable build/system
[100%] Built target system

Running RAD-Sim

To test that RAD-Sim is working as intended, the multi-layer perceptron (MLP) example design can be used as a test. You can simulate this example design following these steps:

1. Activate the RAD flow conda environment:

   $ conda activate radflow
   

2. Generate an MLP test case using the provided compiler:

   $ cd <rad_flow_root_dir>/rad-sim/example-designs/mlp/compiler
   # python gen_testcase.py <num_layers> <input_size> {<hidden_dims>} {<mvms_per_layer>}
   $ python gen_testcase.py 4 512 512 512 256 128 4 3 2 2
   # Number of layers = 4
   # Input dimension = 512
   # Hidden dimensions = [ 512 512 256 128 ]
   # Number of MVMs = [ 4 3 2 2 ]
   

3. Run RAD-Sim simulation:

   $ cd <rad_flow_root_dir>/rad-sim/build
   $ make run
   # ....
   # Inputs per dispatcher = 256
   # Outputs = 256
   # placement_filepath: /home/andrew/rad-flow/rad-sim/example-designs/mlp/mlp.place
   # SUCCESS - All outputs are matching!
   # Simulation Cycles = 12280
   # Aggregate NoC BW = 211.071 Gbps
   
   # Info: /OSCI/SystemC: Simulation stopped by user.
   # Simulation Cycles from main.cpp = 12282
   # [100%] Built target run