Update CITING.md with new papers #1507
Conversation
pratikvn
added
reg:documentation
There was a problem hiding this comment.
I think it's missing clear statements on which papers to actually cite. The TOMS paper should probably have something like this at the top:
Please cite this if you are using Ginkgo in your research.
And for the other features there should also be a single paper highlighted, which best represents the feature. (But I get that this might not make sense everywhere.)
|
Thanks Pratik for taking this up! I agree with Marcel, we should not clutter this file too much. The main point from my PoV is to concentrate citations on e.g. journal papers rather than have a few citations spread over too many papers. It's up to everyone to pick which papers they prefer is cited for their research, but I do think we should keep this list as short as possible. |
|
I think that is quite difficult to do. In many cases, papers are mostly intertwined with applications, so specifying one paper for the feature is difficult. I mostly added a few main papers for each, and let the user decide which they want to use. Maybe we can make a rule that we add two papers per section to keep this file limited. I think maybe one approach would be to have a recommendation only for the main Ginkgo paper, but list maybe two/three papers for each major feature, and let the user decide if they want to cite something additional to the main Ginkgo paper. |
| eprint={2006.14290}, | ||
| archivePrefix={arXiv}, | ||
| primaryClass={cs.MS} | ||
| @inproceedings{tsaiPreparingGinkgoAMD2021, |
There was a problem hiding this comment.
I think the main paper which summarizes these efforts would be:
title = {{Ginkgo—A math library designed for platform portability}},
journal = {{Parallel Computing}},
volume = {111},
pages = {102902},
year = {2022},
issn = {0167-8191},
doi = {https://doi.org/10.1016/j.parco.2022.102902},
url = {https://www.sciencedirect.com/science/article/pii/S0167819122000096},
author = {Terry Cojean and Yu-Hsiang Mike Tsai and Hartwig Anzt},
keywords = {Porting to GPU accelerators, Platform Portability, Performance portability, AMD, NVIDIA, Intel}
}
I generally think it's better to cite journals first or important conferences (e.g., IPDPS).
| @@ -119,15 +148,217 @@ articleno = {2}, | |||
| numpages = {26}, | |||
| keywords = {irregular matrices, GPUs, Sparse Matrix Vector Product (SpMV)} | |||
| } | |||
|
|
|||
| @inproceedings{anztEvaluatingPerformanceNVIDIA2020, | |||
There was a problem hiding this comment.
This paper is hard to put in one place, it's more than sparse to begin with, and also has solvers performance. It's essentially an A100 benchmarking paper, and comparing to V100 performance, I don't know if it fits here.
| ### Preconditioners | ||
|
|
||
| ```bibtex | ||
| @inproceedings{gobelMixedPrecisionIncomplete2021, |
There was a problem hiding this comment.
We could also consider Hartwig and Tobias' ParILUT paper in this section?
https://scholar.google.com/citations?view_op=view_citation&citation_for_view=ctNCF8YAAAAJ:RGFaLdJalmkC
| keywords = {Batched preconditioners,Batched solvers,GINKGO,GPU,Sparse linear systems} | ||
| } | ||
|
|
||
| @inproceedings{kashiBatchedSparseIterative2022, |
There was a problem hiding this comment.
I would put IPDPS first I think.
Following the suggestion from @tcojean, this PR adds new papers and sections to the CITING.md file