The FhSparseGen – Fraunhofer Sparse Matrix Layout Generator for Compound Entries ("FhSparseGen") is used to generate sparse matrix layouts and associated algorithms for varying parallel execution schedules.
FhSparseGen was developed at Fraunhofer IGD to enable joint optimization of sparse matrix layouts and schedules for parallel execution of the sparse matrix vector product (SpMV), primarily on massively parallel graphics processing units (GPUs).
The FhSparseGen code generator requires Python 3 and Jinja 2. To compile the resulting code you will also need a recent CUDA compiler and toolkit (9.2 or higher) and a compatible C++11 compiler.
Mueller-Roemer, J. S., A. Stork, and D. Fellner. “Analysis of Schedule and Layout Tuning for Sparse Matrices With Compound Entries on GPUs.” 2020. In: Computer Graphics Forum 39(6), pp. 133–143. DOI: 10.1111/cgf.13957.
@article{MUELLERROEMER2020,
author = {{Mueller-Roemer}, Johannes Sebastian and Stork, André and Fellner, Dieter},
title = {Analysis of Schedule and Layout Tuning for Sparse Matrices With Compound Entries on {GPUs}},
year = {2020},
journal = {Computer Graphics Forum},
volume = {39},
number = {6},
pages = {133--143},
doi = {10.1111/cgf.13957}
}Mueller-Roemer, J. S., A. Stork, and D. W. Fellner. “Joint Schedule and Layout Autotuning for Sparse Matrices with Compound Entries on GPUs.” 2019. In: Vision, Modeling and Visualization. VMV ’19. 2019, pp. 109–116. DOI: 10.2312/vmv.20191324.
@inproceedings{MUELLERROEMER2019,
author = {{Mueller-Roemer}, Johannes Sebastian and Stork, André and Fellner, Dieter W.},
title = {Joint Schedule and Layout Autotuning for Sparse Matrices with Compound Entries on {GPUs}},
year = {2019},
booktitle = {Vision, Modeling and Visualization},
editor = {Schulz, Hans-Jörg and Teschner, Matthias and Wimmer, Michael},
doi = {10.2312/vmv.20191324}
}FhSparseGen is licensed for non-commercial use under the terms found in LICENSE.md.