dfe-snippets

dfe-snippets is a collection of building blocks for projects using Maxeler Dataflow Engines (DFEs). It contains

• library code for CPU and DFE designs (include, src)
• system benchmarks for DRAM & interconnect throughput and resource usage (test/Benchmarks)
• application benchmarks (applications)

The structure is:

scripts/          ---> Useful scripts
include/          ---> Header only C++ libraries
src/              ---> MaxJ libraries

test/             ---> Test projects, benchmarks, etc.
  Benchmarks/     ---> System benchmarks
  concept1/
    implementation1/
       src/          ---> source code
       build         ---> build directory
  ...

applications      ---> Application Benchmarks

Citing

If you find this work useful, please consider reading and citing our paper on dfe-snippets:

@inproceedings{grigoras2017dfesnippets,
	author = {Grigoras, P. and Burovskiy, P. and Arram, J. and Xinyu Niu and Cheung, K. and Xie, J. and Luk, W.},
	booktitle = {13th International Symposium on Applied Reconfigurable Computing},
	title = {{dfesnippets:An Open-Source Library for Dataflow Acceleration on FPGAs}},
	year = {2017}
}

Requirements

1. MaxCompiler 2013+
2. Compatible Maxeler hardware
3. python 2.7 (for running scripts)
4. a C++ compiler with C++11 support

Using

DFE Libraries

To use the DFE libraries you only need to add the src directory to your classpath when running maxjc. For example: maxjc -cp <path/to/snippets>/src/:<other/class/paths> ....

If you use the Makefile.common provided with dfe-snippets you can just add this to the JFLAGS variable. For example: JFLAGS=-cp $(MAXCOMPILERDIR)/lib/MaxCompiler.jar:$(DFESNIPPETS)/src -1.6 -d .

You should then be able to import any class using the fully qualified name. For example

// A Maxeler DFE kernel

import com.custom_computing_ic.dfe_snippets.blas.DenseBlasLib;

class MyKernel extends Kernel {...}

Note Use a full, absolute path for dfe-snippets to avoid compilation issues.

CPU Libraries

We also provide a number of utilities for CPU code. These are header only libraries, so no compilation is required. To use them simply add the include/ directory to your include path.

If you use the Makefile.common provided with dfe-snippets you can just add this path to the CFLAGS variable. For example: CFLAGS=(....) -I<path/to/dfesnippets>/include. You should then be able to use functions, for example:

#include <dfesnippets/blas/Blas.hpp>

int main() {
  dfesnippets::blas::Matrix m(10);
  // ....
  return 0;
}

Note You may also have to add the -std=c++11 flag to CFLAGS to ensure C++11 support is enabled.

Creating new projects

Use create.py -h for a list of available options.

To create and run a new dfe-snippet from the snippets directory run:

python scripts/create.py <ProjectName> -c <ProjectConcept>

Test this project with:

cd <ProjectConcept>/<ProjectName>/build && make runsim

To create an external, stand-alone project run:

python scripts/create.py <ProjectName>

Test this project with:

cd <ProjectName>/build && make runsim

Note If you select c++ as the project language, the generated Makefile expects a compiler which supports C++ 11 (i.e the --std=c++11 command line argument).