Projects:Ganga

From Epl

1 Ganga

Ganga

Overview

The goal of this project is to create new parallel programming paradigms and build systems that help developers of Vision and other compute intensive applications to target a wide range of hardware platforms from Graphics Processing Units (GPUs), the Cell Broadband Engine, to other Multi-Core processors.

[edit]

Capsules

Designing high-level parallel programming models and languages have led to the granularity problem where the execution of parallel task instances that are too fine-grain incur large overheads in the parallel runtime and amortize the speed-up gain achieved by parallel execution. In this work we attempt to address this issue with a unifying concept of expressing composable computation in a parallel programming model.

We show that this concept not only allows the user to express the decision on granularity of execution, but also the granularity of the garbage collection mechanism, and other features that may also be supported by the programming model.

We argue that this adaptability of execution granularity leads to efficient parallel execution by matching the available application concurrency to the available hardware concurrency. In the heterogeneous computing world that is increasing becoming common-place even in HPC application targets, such adaptability is crucial to decrease parallel runtime overheads and increase effective utilization of the hardware concurrency. Our results using the Cascade Face Detector show that adjusting the execution granularity through profiling helps determine the serial execution granularity of parallel tasks to yield optimal performance.

[edit]