Chair News

Published on

graphic with text '10 software campus' and a tree

The CC Chair, together with Prof. Christian Pilato and Dr. Christoph Hagleitner, organised a workshop at the DATE Conference 2022 entitled "Data-driven applications for industrial and societal challenges: Problems, methods, and computing platforms" (DATA-DREAM'22). The workshop provided an international forum for researchers in the field of large-scale data analysis technologies to discuss challenges and ideas for the future of big data analytics. Participants shared their insights into current political, economic and technological developments, and placed special focus on their approaches to overcoming the hurdles towards wide adoption of reconfigurable computing platforms. This event is an initiative of the EVEREST project, which we hope will trigger collaborations and awareness in the community.

Published on

graphic with text '10 software campus' and a tree

TU Dresden has been one of the 21 partners of Software Campus since 2017. Here, master's and doctoral students of computer science receive further training and are prepared for leadership positions. In the process, participants in the program lead their own IT research project and manage the entire process independently. They receive support from experienced leaders. Each project is funded by the BMBF with up to 100,000 euros over a period of two years.

Published on

The CC Chair was present this year at Embedded Systems Week, one of the top venues for hardware and software design. Robert Khasanov presented "Domain-specific Hybrid Mapping for Energy-efficient Baseband Processing in Wireless Networks", joint work with Julian, Christian, and Andres. This work proposes a hybrid design-time/runtime mapping methodology, which exploits the application domain knowledge and refines the scheduling algorithm (see video). Christian Menard and Clément Fournier, a diploma student at the CC Chair, presented their work on the coordination language Lingua Franca jointly with collaborators from UC Berkeley, UT Dallas, Hanyang University and Kiel University in the tutorial "Deterministic Reactive Programming for Cyber-Physical Systems Using Lingua Franca" (see video).

Published on

The CC chair was present at this year's International Conference on Parallel Architectures and Compilation Techniques (PACT), one of the top venues for compilers for parallel architectures. At PACT, Andres presented PolyGym: Polyhedral Optimizations as an Environment for Reinforcement Learning, a joint work with Alex that defines loop optimizations in the polyhedral framework as a problem that can be solved with machine learning methods. The work also highlights the challenges of the current heuristics and presents a search space that contains significantly better-performing loop variants. We are looking forward to improved polyhedral schedule heuristics in follow-up work!

 

Published on

The FPGA for HPC workshop (HPCFPGA'21) targets FPGA technologies for High Performance Computing, and focuses on topics including hardware synthesis, high bandwidth memory, design optimization, and interconnected clusters of FPGA devices. At the workshop, Karl Friebel presented the paper "From Domain-Specific Languages to Memory-Optimized Accelerators for Fluid Dynamics". This marks the first joint publication of researchers from Politecnico di Milano, and in particular Prof. Christian Pilato's group, and our team in the context of the EU H2020 EVEREST Project. Karl described a prototype implementation of an end-to-end DSL-to-FPGA compilation flow for scientific computing.

 

Published on

On July 9, Prof. Castrillon gave the closing keynote address at the Platform for Advanced Scientific Computing (PASC) Conference, co-sponsored by the Association for Computing Machinery (ACM) and the Swiss National Supercomputing Centre (CSCS). PASC is a thriving conference that brings together researchers from the scientific computing, the computational science, and the computer science communities. Prof. Castrillon talked about how domain specific languages and compilers are badly needed to democratise high-performance computing, especially in the advent of emerging technologies and new computing non Von Neumann paradigms. The slides of the talk are available and a recording of the virtual keynote will be made available later in the year (here). 

Published on

Hauke Mewes, a former Masters's thesis student at the chair for Compiler Construction, was awarded the N. J. Lehmann Price award 2021 for his work "Towards Optimizing Compilers for Systems with Racetrack Memories". The N.J.Lehmann award, named after the renowned TU Dresden alumni Nikolaus Joachim Lehmann (1921-1998), is awarded every two years to two outstanding theses at TU Dresden in the fields of Computer Science and Mathematics. The award particularly encourages research at the intersection of mathematics and real-world systems. Asif Ali Khan supervised Hauke's work at the chair, which turned into an excellent publication "Polyhedral Compilation for Racetrack Memories" at the peer-reviewed IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (TCAD). Hauke developed a prototype compiler to optimize and generate efficient codes for systems with racetrack memories. The award was handed over at this year's OUTPUT.DD science event, where Hauke also briefly talked about his work (see video). We are proud of Hauke's achievement and wish him all the best in his current and future endeavors.

Published on

Andrés, who recently finished his PhD at the CC chair to then join the Barkhausen Institute, presented his work on “Embeddings of Task Mappings to Multicore Systems” at this year’s SAMOS Conference. What a pity that he couldn’t enjoy a trip to the beautiful island of Samos in Greece and profit from the the great atmosphere that characterises the conference. Andrés talked about an innovative way to think about the optimization space that arises when mapping applications to multicores. His approach makes it possible to transfer geometry-based optimization approaches to an otherwise complex and unstructured discrete space. 

 

Published on

Alex Brauckmann, a member of the CC Chair from 2020, has been awarded the “Zeiss Diplompreis für Informatik” for his thesis “Investigating Input Representations and Representation Models of Source Code for Machine Learning”, supervised by Andres Goens. Alex’s thesis was the basis for a paper on “Compiler-Based Graph Representations for Deep Learning Models of Code” at the renown ACM SIGPLAN International Conference on Compiler Construction (CC’20), featuring a novel deep learning method that improves predictive compiler heuristics. The Zeiss prize is awarded to outstanding thesis work at TU Dresden and the HTW Dresden in the areas of Scientific Computing, Machine Learning, Image Processing and Framework Design. Due to the pandemic, Alex couldn't receive the honour during this year's OUTPUT.DD science event, so we had to improvise a picture. We are happy for Alex and proud to have him in our team! 

Published on

The CC Chair was present at the 21st International Conference on Computational Science (ICCS), which was held online during the week of June 14 2021. ICCS is an A-ranked conference bringing together researchers and scientists from different disciplines to highlight the role of recent developments and challenges in computational science. Nesrine presented our work entitled “The OpenPME Problem Solving Environment for Numerical Simulations”. In this work, Nesrine introduces the OpenPME domain-specific language (DSL) for particle-mesh simulations that is built atop a domain metamodel general enough to cover the main types of numerical simulations: simulations using particles, meshes, and hybrid combinations of particles and meshes. This work is a collaboration with professor Ivo Sbalzarini and his group at the Chair of Scientific Computing for Systems Biology.