- DigiCortex reaches 300000 neurons in real-time on just a single GPU!
Presenting "DigiCortex - A Hybrid (CPU/GPU) Acceleration of Biological Spiking Neural Networks on Desktop Supercomputers"
at GPGPU Day 2013, StreamComputing, Amsterdam.
- PhD "Exploiting Multi-Level Parallelism in Streaming Applications for Heterogeneous Platforms with GPUs" - defence and party on 26th June 2013 in Leiden, The Netherlands.
Would you like to attend? Drop me a message! (ana.balevic _at_ gmail.com)
Short summary: Modern heterogeneous platforms provide the opportunity to exploit parallelism at different platform levels. However, the diversity of architectures and programming models makes tapping into the platform potential a challenging task. In this thesis, we present a novel methodology for compile-time generation of structured, multi-level programs (MLPs) from sequential applications. As fundamental contributions, we introduce a novel hierarchical intermediate program representation (HiPRDG) and a method to construct MLPs featuring multiple forms of parallelism (task, data, pipeline) that open the door towards efficient, tailor-made parallelization for heterogeneous platforms with accelerators.
Digital thesis and bibtex comping up soon!
Implemented CUDA acceleration of the DigiCortex engine for large-scale spiking neural network simulations! Checkout www.digicortex.net (twitter: #digicortex)
- Presenting a poster on "Multi-Level Parallelization in Polyhedral Model for Heterogeneous Platforms" at GTC2013.
Presented my work on multi-level parallelization at the thematic session on Hybrid Programming Models and Optimizations for Heterogeneous Manycores session
at HiPEAC Computing Systems Week, Paris, May 2013. Marisa, Eduard, and Chris - thank you for the invitation!
- Submitted my PhD thesis on multi-level parallelization techniques for heterogeneous platforms. PDF Coming soon!
- Created a Java-based KPN2GPU tool for automatic conversion of Polyhedral Process Network (PPN) specification
to CUDA as an extension to the Compaan compiler within the European project on Tera-Scale Multi-Core Architectures (TSAR).
While at University of Stuttgart I started and lead the developemnt of
the world's first open source CUDA JPEG2000 encoder entirely running on the GPU:
- Check-out my past CUDA projects on electromagnetic simulations and compression,
e.g. parallel variable-length encoding (Huffman compression).
CUDA/GPU Coding Projects