Current Theses

Title

Student

Supervisor(s)

Description

Finite element method on GPU using OpenCL

OpenCL: Matrix Library

• Becoming familiar with the cloud infrastructure and GPU instances
• Write some benchmarks for GPUs
• Perform benchmarks and compare the cloud instance with stand-alone hardware
• Write code to simplify and automate creation and configuration of cloud instances
• Write library/broker to distribute a computation over n instances with m GPUs
• Evaluate and benchmark using an existing deep learning algorithm provided by IIS/LFU
• Evaluate and benchmark using another algorithm provided by JAIST, optimize and verify with results from the deep learning algorithm

• Interests in parallelizable algorithms
• Distributed Systems
• Parallel Programming (OpenCL)

• Good C and OpenCL knowledge

• Familiarize with ESS
• Find appropriate architecture
• Prepare the code for division
• Implement server backend
• Bring the server to a cluster
• Speedup data-processing by caching
• Adopt current native client
• Implement a mobile clien

• Java
• Javascript

•   Learn and familiarize with OpenCL and WebCL
•   Test current available WebCL applications.
•   Implement a web infrastructure which schedules OpenCL tasks over distributed web-based clients
•   Test the final infrastructure with at least two different test codes

• Javascript
• OpenCL (attending\have attended Einführung in das Parallelrechnen und parallele Algorithmen is a plus)

An interesting rendering technique is Instant Radiosity (Keller 1997). IR is a method that approximates the indirect lighting, as part of global illumination, by creating additional light sources (called Virtual Point Lights). The algorithm is fairly simple: each light source it casts N photon rays into the scene. At each intersection the photon either bounces and another ray is cast or, through Russian roulette, is killed. At each of the intersections it creates a Virtual Point Light (VPL) that has the same radiance value as the photon.

Although IR can produce extremely realistic results, they can only be achieved with exponentially increasing the number of VPLs. A solution to scale the computation needed for the huge number of VPLs is provided by Lightcuts (Walter et al. 2005). The idea is to group lights in a binary light tree, then to use a perceptual metric to adaptively partition the lights into groups which can be rendered at the desired quality level. This algorithm offers a user-controllable tradeoff between the desired quality and the computational cost (i.e. execution time).

The goal of this thesis is to develop a parallel version of Lightcut and to evaluate its performance on different scenes.

The project should use an internal optimized implementation of the ray tracing algorithm.

• Familiarize yourself with the provided ray tracing implementation
• Study and implementation of Instant Radiosity
• Study and implementation of Lightcuts
  • Light tree building
  • Perceptual cuts
  • Reconstruction of cuts
• Test the system with different configuration on at least three scenes.

• Rendering algorithms
• Concurrent Programming

• C/C++
• OpenGL (basic knowledge)
• OpenCL (basic knowledge)

Global Illumination with the Instant Radiosity technique.
Instant Radiosity, by Alexander Keller
Lightcuts, by Walter et al.

• Spezifikation der Pseudocode-Grammatik
• Implementierung eines Scanners und Parsers
• Entwurf und Erzeugung eines abstrakten Syntaxbaumes und einer Symboltabelle
• Implementierung einer semantischen Analyse
• Kleine Programmtransformationen und Optimierungen
• Code Generierung in der C-Programmiersprache

• C oder C++

• study state-of-the-art research in the fields of program code instrumentation and energy efficiency;
• familiarize yourself with the power measuring device and its data collection interfaces;
• implement a utility library for connecting to the power measuring device;
• design and implement the program code instrumentation library for power measurements;
• implement a test suite for accurate evaluation of power consumption for small program code regions.

The goal of these thesis is develop a framework for building scientific applications in Cloud computing infrastructures based on Apache Hadoop technology.

• Einarbeiten in OpenCL
• Einarbeiten in Grundlagen der FEM
• paralleles Assemblieren der Steifigkeitsmatrix
• Performace-Analyse
• Implementierung verschiedener Elementstypen in OpenCL

• C++
• OpenCL
• Grundlagen Matrizen

• Read into OpenCL
• Read into matrix problems and Gnu MP
• Write some test programs using Gnu MP
• Model the workflow data structures
• Build “simple” implementation
• Benchmark
• Optimize workflow on memory layers (data transfers)
• (Optimize/Parallelize with known algorithms)
• Benchmark
• Unit-Test library extensively

• Interests in optimized algorithms
• Interests in workflow based systems
• (Grid Workflows)
• (Parallel Systems)

• Good C/C++
• (OpenCL)

The objective of this thesis is to investigate the potential of using hybrid multicore architectures consisting of CPUs, GPUs, and Cell processors for parallelising list sorting algorithms such as bubble sort, quick sort, rank sort, bucket sort, selection sort, merge sort, etc.

• Implementation of several (3-4) parallel sorting algorithms in OpenCL
• Scheduling and optimisation of the parallel sorting algorithms on hybrid multiprocessor platforms
• Speedup and efficiency analysis
• Overhead analysis