Tomasz Bednarz "Heterogeneous Computing with OpenCL"
OpenCL (Open Computing Language) is an open, royalty-free, cross-platform standard specifically designed for general purpose parallel programming of heterogeneous systems, including modern desktop and workstation class multi-core processors (CPUs), graphics processing units (GPUs), and other accelerators such as Cell and digital signal processors (DSPs). The standard includes an API for coordinating execution between devices and a cross-platform parallel programming language. Based on ISO C99, the OpenCL-C language provides a well-defined execution environment with strict precision requirements, allowing device agnostic kernel methods to be written which can be run on any supported OpenCL device. The OpenCL standard supports task-based and data-based parallel computing, features consistent numerical requirements based on the IEEE standard for floating-point arithmetic (IEE 754), defines a configuration profile for handheld and embedded devices and efficiently interoperates with OpenGL, OpenGL ES, and other graphics APIs.
As requested, introduction to OpenCL will be presented.
Please also see some of our old slides from OpenCL Workshop at OzViz 2010
Suraj Pandey "Cloud Computing and Scientific Applications"
This talk briefly describes the underlying concept of computing Cloud. Using several real world applications, such as, image registrations for Functional Magnetic Resonance Imaging (fMRI), Gravitational wave search as workflow, distributed intrusion detection, and so forth, we describe our experiences when executing them using Amazon Cloud services.
Dadong Wang "Connected Components Labelling"
Connected component labelling is an algorithm to find a unique label for each of segmented regions in an image. The algorithm has been widely used in image analysis and computer vision. In this talk, we will briefly introduce the algorithm first, and then demonstrate how it is implemented in CUDA. The performance evaluation of the CUDA method together with the experimental results using various image datasets will also be presented.