Prof. Christian Bischof

In high-performance computing, the so-called software gap, i.e. the discrepancy between the theoretical power of parallel computers, and the ability of computational practitioners to realize that power with a reasonable time investment for software development, is slowing down the potential of HPC technology in computational engineering. To address this issue, the research group focusses on parallel programming and software development, and automated software development techniques such as automatic differentiation. In particular, we work on performance models of parallel programs, best-practice methods for code tuning, and code instrumentation techniques in order to assess or improve a parallel programs performance. Automatic differentiation (AD), on the other hand, is a methodology for augmenting computer codes with derivatives. Thus, in effect, AD turns a simulation code into a design tool as accurate and efficient derivatives are a key ingredient in numerical optimization approaches. The research group works, in particular, on AD for Matlab, which is a challenge due to the dynamic nature of this language. A particular intriguing feature of AD is that the chain rule of differential calculus allows the exploitation of parallelism in places that in the original code were inherently sequential, thus enabling both additional vector- and task-based parallelism. As such, AD is an example of a domain-specific semantic transformation, where knowledge about the particular problem domain is used to automate the code generation process, taking into account the possibilities of parallel computers.

Prof. Dr. C. Bischof

Contact

Technische Universität Darmstadt

Graduate School CE
Dolivostraße 15
D-64293 Darmstadt

Phone+49 6151/16-24401    or
-24402
Fax+49 6151/16-24404
OfficeS4|10-322

to assistants' office

Open BSc/MSc Theses

Show a list of open BSc/MSc topics at GSC CE.

 Print |  Impressum |  Sitemap |  Search |  Contact |  Privacy Policy
zum Seitenanfangzum Seitenanfang