This is the repository containing code and related material for the master's thesis project Effects of Design Space Discretization on Constraint Based Design Space Exploration. The following instructions are for the code contained in the experiments-subdirectory which is what is relevant for reproducing the results. The other directories in the repository are:
figures: Contains the figures used in the report.mra_example: Contains the code to generate the figure illustrating MRA included in the report.plan: The initial project plan.tests: Various solutions to constraint-programming problems.
Requirements:
- Java 11 or higher.
- Bash shell (tested in Cygwin and Ubuntu)
- Python 3.9 or higher
- Julia with packages: Printf, Plots, CSV, LinearAlgebra, Interpolations
Installation:
- Make sure that all requirements are met.
- Clone this repository.
- Download IDeSyDe 0.5.x, chose the correct version depending on your OS.
- Extract IDeSyDe into the
experiments-directory. This should add two directories: emodules and imodules; and the platform dependent IDeSyDe entry point. - (for Windows) change the name of the IDeSyDe entry point in
run.sh, i.e., change"idesyde"to"idesyde.exe"on line 16.
Each experiment is identified by a unique identifier. An experiment is defined by a directory named in_case_<identifier>. This directory should contain a configuration file called config with the following format:
tr1 tr2 tr3 ... trN
mr1 mr2 mr3 ... mrM
specifying the N time resolutions and M memory resolutions to test (both in ascending order) (the file should en with an empty line). The directory should also contain at least one .fiodl file specifying the set of application and platform models to investigate.
The following is a valid in-directory for an experiment called sobel8:
in_case_sobel8
├── a_sobel.hsdf.fiodl
├── bus_small_platform.fiodl
└── config
When a correct in-directory has been created, the experiment can be started by running the run.sh script. run.sh takes the following following format:
./run.sh [identifier] <cores <batches <halt-early>>>
where:
- identifier is the unique identifier for the experiment to run.
- cores are the number of cores to utilize, default is 1.
- batches is the number of times to run each case, default is 1.
- halt-early if set to 1 makes the script halt before generating plots.
run.sh reads the configuration and traverses the grid of configurations. Each case is ran and timed batches times. The resulting data is stored as a CSV-file and then plotted. As an example, the following runs the sobel8-case on 16 cores with a batch size of 4:
./run.sh sobel8 16 4