Uncovering the Unimaginable Pace of Raspberry Pi

From Computer Laboratory Group Design Projects
Jump to navigationJump to search

Client: MathWorks

Contact: Milos Puzovic Milos.Puzovic@mathworks.co.uk

modified proposal

Unlocking the graphics power of the Raspberry Pi

The Raspberry Pi has become incredibly popular, and it's great for hobby applications, but its appeal to children is reduced by the fact that it's a little slow (for example, the Pi Edition of Minecraft doesn't support mobile characters). However, most applications are using only a fraction of its computational power. The good news is that the Raspberry Pi's System-On-Chip BCM2835 has a hidden gem: a Graphics Processing Unit (GPU) that can significantly improve performance of applications that use large blocks of data. At the moment, the GPU on Raspberry Pi is heavily underutilised. Your task is to enable applications written in the MATLAB language to exploit the Raspberry Pi GPU. As a demonstration of what can be achieved, it should be possible to implement high performance game physics such as real time cloth dynamics on a human character with moving clothes and hair (created using MATLAB Simulink toolboxes). It's your choice whether this is a mobile character in a sandbox game, a narrative cut-scene, or even an intelligent avatar such as the Zoe talking head.

added detail

Simulink has the built-in support for Raspberry Pi to build standalone applications (http://www.mathworks.com/hardware-support/raspberry-pi.html). This support has enable access to various audio and video algorithms through toolboxes such as DSP System Toolbox (http://www.mathworks.com/products/dsp-system/) and Computer Vision Toolbox (http://www.mathworks.com/products/computer-vision/). Unfortunately, the code that is generated to run on the Raspberry Pi only targets CPU and it does not make any use of GPU that is available on the chip.

In order to be able to generate code that targets the GPU it would be necessary to customize build process (http://www.mathworks.com/help/rtw/ug/customizing-the-target-build-process-with-the-stf-make-rtw-hook-file.html). Using the hooks provided by the Simulink Coder you would be able to analyze the generated code and replace parts of the code with code that can target VideoCore ISA (https://github.com/hermanhermitage/videocoreiv/wiki/VideoCore-IV-Programmers-Manual). There are various toolchains that can target VideoCore and students will have an opportunity to evaluate them and decide which one would be the best suitable for integration. The toolchains available are:

To demonstrate the performance improvement the students would start with one of simpler algorithms that can be found in image processing or computer vision that are already implemented and available in Simulink. Once they have demonstrated the improvement one part of the group could implement an application in Simulink that adds real physics to the gaming application such as cloth motion.

original proposal

Raspberry Pi, the credit-card size computer, has unleashed an army of children, hobbyists and professionals who have toyed with the computer to make many interesting uses for it, from monitoring weather outside of a shed to replacing old satellite navigation systems in the car. As many more people are joining this new wave they will want to have access to faster hardware. The good news is that Raspberry Piís System-On-Chip BCM2835 has a secret gem in the form of a Graphics Processing Unit (GPU), which can significantly improve performance of applications that use large blocks of data. At the moment, the GPU on Raspberry Pi is heavily underutilised. Your task is to enable applications written in the MATLAB language to exploit Raspberry Piís GPU in order to maximise the performance of the application. At the end of the project, you will be able to demonstrate how applications that rely on computer graphics execute faster with your project, opening up new possibilities for users of Raspberry Pi.