Understanding the world. Or at least trying to.
The number of people expecting Moore's law to fail doubles every 24 months.Anonymous
My name is Jens Breitbart and this web page is a basic overview of myself and my work over the last few years.
I have the same job as Batman (scientist), however, I am more than that: Among others, I am a software developer by passion, a teacher by importance, a biker for the fun and geek by history. And above all, I love quotes :-)
For one, if the search task is important, why haven't people been working on it previously?An anonymous reviewer
Isn't it easy to forget [...] that the answers which the Universe eventually elects to disclose appear to depend entirely on the Questions we pose?Juice Media
My research over the last years has been centered around easing parallel programming mostly with a focus on programming models and high performance computing. I have been working on existing programming models like OpenMP and OpenCL, as well as on designing new programming models/libraries.
My work on existing programming models mostly concentrated on OpenMP and OpenCL. Whereas with future hardware, memory accesses and data transfers seem to become more important than on current hardware, yet OpenMP does not take care of data locality. I have been working on adding the concept of tiling to OpenMP, to allow both better control over data locality and the possibility to effectively utilize tile based many-core chips. OpenCL already incorporates tiling has mechanisms for effective memory transfers, yet it does not effective data sharing between tasks. I have been working on proving ways to share data among task to eliminate off-chip memory accesses. My own programming model called Partitioned Single Assignment Memory (PSAM) is designed to deal with hybrid CPUs, missing cache coherency, and purely relies on RDMA memory accesses. Synchronization follows the dataflow semantic on a single assignment memory space. CuPP is one of my first research project and a library designed to allow easy integration of CUDA into C++ programs.