CSE Seminar High-Performance Libraries in Computational Science and Engineering - Winter 08
- Winter 08
- Dr. Michael Bader
- Time and Place
- block seminar (see below for schedule)
- Computational Science and Engineering, 3rd semester
- Semesterwochenstunden / ECTS Credits
- 2 SWS (2S) / 4 Credits
Tuesday, July 15, 2008 (15:00) in room 02.07.023. The seminar session will be in three block sessions between Dec 8 and Dec 12 (see schedule below).
Optimised libraries for specific tasks in computational science and engineering are important building blocks for CSE applications. The seminar will generate an overview of the algorithmic and implementational approaches adopted in such libraries to achieve high performance on modern architectures.
The following libraries are intended for presentation:
- FFTW - the fastest Fourier Transformation in the West ...
- GotoBLAS - Anatomy of a fast matrix multiplication
- METIS/PARMETIS - partitioning algorithms for discretisation meshes (1 or 2 talks)
- LAPACK/ScaLAPACK - libraries for numerical algorithms (1 or 2 talks)
- FLAME: Formal Linear Algebra Method Environment
- OSKI - Optimised kernels for sparse matrix computations
- PetSc - Portable, Extensible Toolkit for Scientific Computation (1 or 2 talks)
- ARMCI and Global Arrays
- Mondriaan - Partitioning of Sparse Matrices
In all presentations, the focus should not be on the library features, but on the numerical, algorithmic, and implementational approaches!
Each student will give one presentation (approx. 45 minutes) in the seminar. In addition to her or his talk, each student will have to prepare a short paper (8-10 pages), which will be reviewed by two other seminar participants. The paper and the two reviews will, in addition to the talk, be considered to determine the final grade.
- deadline for paper: Mon, Nov 10, 2008.
- deadline for review: Mon, Dec 1, 2008.
- deadline for final paper: Mon, Dec 15, 2008
The seminar talks will be given in presumably 3 block sessions in the week of December 8-12.
The following suggested session schedule is preliminary: sessions might start one week (or even a couple of weeks) earlier or later, and sessions might be swapped.
|Tue, Dec 9 (16:00)
|FFTW - the Fastest Fourier Transform in the West||Kaveh Rahnema|
|GotoBLAS - Anatomy of a fast matrix multiplication||Alexander Krivutsenko|
|OSKI - optimised kernels for sparse matrix computations||Dheevatsa Mudigere|
|Thu, Dec 11 (9:00)
|LAPACK/ScaLAPACK||Md Golam Suman|
|FLAME: Formal Linear Algebra Method Environment||Daniel Butnaru|
|Tue, Dec 16 (16:00)
|PetSc - Portable, Extensible Toolkit for Scientific Computation||Viet Pham Anh|
|ARMCI and Global Arrays||Ventsislav Petkov|