Difference between revisions of "SC²S Colloquium - August 31, 2017"
Line 12: | Line 12: | ||
In this thesis the relevant code for neighborhood lists in the ls1 mardyn project is vectorized with SIMD-instructions for Intel’s Xeon Phi architecture. The performance improvements are measured and compared to the existing vectorized Linked Cells approach. We will see that efficient memory usage is crucial for high performance with neighborhood lists. Also, we evaluate different optimizations for the vectorized neighborhood lists and try to outperform the existing Linked Cells. Some other minor improvements for ls1 mardyn in general are examined as well. | In this thesis the relevant code for neighborhood lists in the ls1 mardyn project is vectorized with SIMD-instructions for Intel’s Xeon Phi architecture. The performance improvements are measured and compared to the existing vectorized Linked Cells approach. We will see that efficient memory usage is crucial for high performance with neighborhood lists. Also, we evaluate different optimizations for the vectorized neighborhood lists and try to outperform the existing Linked Cells. Some other minor improvements for ls1 mardyn in general are examined as well. | ||
− | == Thomas Schilling: | + | == Thomas Schilling: Neighbor Lists in Molecular Dynamics Simulations: Implementation and Parallelization == |
− | + | The molecular dynamics code ls1 mardyn is analysed with the aim to enhance its source code. Only taking linked-cell algorithm the code is to receive an upgrade using neighbor lists. The goal to outperform the old linked-cell implementation with the combination of neighbor-lists and linked-cell method was not achieved even tough the new implementation gets really close to the old one. | |
[[Category:ShowComingUp]] | [[Category:ShowComingUp]] | ||
[[Category:news]] | [[Category:news]] |
Latest revision as of 12:11, 30 August 2017
Date: | August 31, 2017 |
Room: | 02.07.023 |
Time: | 16:00 pm, s.t. |
Micha Müller: Neighbourhood Lists in Molecular Dynamics Simulations: SIMD-Vectorizierung
In this thesis the relevant code for neighborhood lists in the ls1 mardyn project is vectorized with SIMD-instructions for Intel’s Xeon Phi architecture. The performance improvements are measured and compared to the existing vectorized Linked Cells approach. We will see that efficient memory usage is crucial for high performance with neighborhood lists. Also, we evaluate different optimizations for the vectorized neighborhood lists and try to outperform the existing Linked Cells. Some other minor improvements for ls1 mardyn in general are examined as well.
Thomas Schilling: Neighbor Lists in Molecular Dynamics Simulations: Implementation and Parallelization
The molecular dynamics code ls1 mardyn is analysed with the aim to enhance its source code. Only taking linked-cell algorithm the code is to receive an upgrade using neighbor lists. The goal to outperform the old linked-cell implementation with the combination of neighbor-lists and linked-cell method was not achieved even tough the new implementation gets really close to the old one.