Overview

We present Whippletree, a novel approach to scheduling dynamic, irregular workloads on the GPU. We introduce a new programming model which offers the simplicity and expressiveness of task-based parallelism while retaining all aspects of the multi-level execution hierarchy essential to unlocking the full potential of a modern GPU. At the same time, our programming model lends itself to efficient implementation on the SIMD-based architecture typical of a current GPU. We demonstrate the practical utility of our model by providing a reference implementation on top of current CUDA hardware. Furthermore, we show that our model compares favorably to traditional approaches in terms of both performance as well as the range of applications that can be covered. We demonstrate the benefits of our model for recursive Reyes rendering, procedural geometry generation and volume rendering with concurrent irradiance caching.

Paper reference

Markus Steinberger, Michael Kenzel, Pedro Boechat, Bernhard Kerbl, Mark Dokter, and Dieter Schmalstieg. Whippletree: Task-based Scheduling of Dynamic Workloads on the GPU. ACM Transactions on Graphics (Proc. SIGGRAPH Asia 2014), December 2014. To appear. [ .pdf ]

Getting started

Building on Linux

On Linux the CUDA compiler with C++11 support (CUDA 6.5RC or later) and cmake are required.

Clone the source tree and build basic examples:

$ git clone https://github.com/apc-llc/whippletree.git
$ cd whippletree/examples/queuing
$ mkdir build
$ cd build
$ cmake ..
$ make

Building on Windows

On Windows Microsoft Visual Studio with C++11 support (e.g. version 13), cmake and CUDA Toolkit are required.

Clone the source tree and build basic examples:

> git clone https://github.com/apc-llc/whippletree.git
> cd whippletree\examples\queuing
> mkdir build
> cd build
> cmake -DCUDA_TOOLKIT_ROOT_DIR="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v6.5" ..

Then open Example VC++ project in Visual Studio and build Example target.

Running

Three different procedures are defined using proc0.cuh, proc1.cuh and proc2.cuh. The host control logic is found in test.cu and could be executed via queuing binary:

$ ./queuing

If launch fails, you may need to add code generation for Compute Capability of your GPU and recompile:

$ cmake -DCUDA_BUILD_CC30=TRUE ..
$ make