Stop Thinking, Just Do!

Sungsoo Kim's Blog

Aparapi for Unconventional Cores

tagsTags

12 May 2015



Aparapi-Ucores - Aparapi for Unconventional Cores

This project is a fork of AMD’s Aparapi framework meant to allow using simple Java code to target any type of heterogeneous device that supports OpenCL.

Why we created Aparapi-Ucores

The original Aparapi framework was built with GPUs/APUs in mind. We believe that heterogeneous computing has a lot more to offer than just combining CPUs and GPUs together. In the past few years we have started to see OpenCL support for FPGAs (Altera/Xilinx) and DSPs(TI) and more types computing devices will soon follow suite.

Combining several types of devices and architectures is at the heart of heterogeneous computing’s power efficiency advantage. We believe that in order to optimize performance per watt and accommodate the needs of today’s high level programming requirements we need to be able to work with all these types of devices using a single high level code base.

The framework has been tested and known to work on the following device types/OpenCL SDKs*:

  1. CPUs (AMD/Intel)
  2. GPUs (AMD/NVidia)
  3. APUs (AMD)
  4. FPGAs(Altera - Nallatech/Terasic boards)
  • Tested operating systems - CentOS 6.3/4/5/6

Changes from original work

  1. Support for FPGA devices (right now Altera OpenCL is supported)
  2. Support for multiple platforms (platform selection)
  3. Support for accelerators and OpenCL binary file format and flow
  4. Built against OpenJDK (to allow more control/flexibility in future Java/OpenCL integration)
  5. Misc small changes - Improve profiling usability etc.

Project History

To the best of my knowledge the original idea of trying to target FPGAs from Java using Aparapi was the brain child of Sai Rahul Chalamalasetti and Mitch Wright from HP Servers.

I started writing the first implementation of the project (called APARAPI-FPGA’s) while working as an intern at HP Servers in 2013. It started as a proof of concept but showed great promise.

During that work I started envisioning a more general solution using Aparapi that will allow us to target any OpenCL device. i.e. not only GPUs or FPGAs.

The project culminated with two publications in Sept 2014 (FPL 2014 and FSP 2014) in which we described our findings and promised to release our initial work back to the open source community.

Since then we received a lot of interest and requests to release the source and we have been planning to release our implementation for some time now.

This project was left on the back burner for a long time, but a couple of months back I started to dedicate some time to rewrite the framework to what I originally envisioned it to be i.e. a general, programmer friendly, high level framework for heterogeneous devices.

Two things were important to me working towards this public release:

  1. I wanted to make sure it will be something that can be usable by regular Java programmers and not just release another difficult to understand research project that would require significant effort to make it work.
  2. I wanted to implement the more general idea (not just an FPGA frame work or GPU framework)

The initial release is still far from being programmer “Friendly”, but it is a step forward and I hope it will encourage high level programmers to experiment with “exotic” heterogeneous architectures.

I welcome any one that wants to be involved in improving this framework.

If you use this work for academic purposes please reference the original FPL/FSP papers describing our initial Aprapi work(see below).

References

The modified Aparapi framework that supports FPGAs was first introduced in the following papers:

Oren Segal, Sai Rahul Chalamalasetti, Mitch Wright and Martin Margala. “High Level Programming Framework for FPGAs in the Data Center”, Field Programmable Logic and Applications (FPL), 2014 24th International Conference on. IEEE, 2014. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6927442

Segal, Oren, Martin Margala, Sai Rahul Chalamalasetti, and Mitch Wright. “High Level Programming for Heterogeneous Architectures.” arXiv preprint arXiv:1408.4964 (2014). http://arxiv.org/abs/1408.4964

The original Aparapi project can be found at: https://code.google.com/p/aparapi/

This work is part of the Heterogeneous computing research that is conducted by the Mora research group at UML: http://mora.uml.edu/

Documentation

For information on how to setup the environment and run some tests please check the [project root]/docs folder: https://gitlab.com/mora/aparapi-ucores/tree/master/docs

Wiki: https://gitlab.com/mora/aparapi-ucores/wikis/home

Screenshots: https://gitlab.com/mora/aparapi-ucores/wikis/Screenshots

Acknowledgments

  1. AMD for doing such a great job with Aparapi without which this project would be an order of magnitude more complex.

  2. HP Servers which believed back in 2012 that Java-OpenCL on FPGAs is something worth looking at.

  3. Altera, Nallatech and Terassic for contributing FPGA hardware for us to develop and test on.


comments powered by Disqus