< Back

Datasets for Benchmarking Floating-Point Compressors

Datasets are little-endian single- and double-precision IEEE 754 arrays, with the first dimension of the extent being the slowest dimension in the multidimensional array. Files are compressed using Zstandard.

astro_mhd and astro_pt simulations were performed separately with single- and double-precision arithmetic; the remaining single-precision datasets are simply truncated from their double-precision counterparts.

dataset	description	extent	download
`rsim`	Radiosity field from room response simulation for time-of-flight imaging	2,048 × 11,509	single	double
`astro_mhd`	Temperature component of a magnetohydrodynamic simulation of solar wind interactions	(*)	single	double
`astro_pt`	Velocity vector component of a particle transport simulation in the LS 5039 system	512 × 256 × 640	single	double
`wave`	Time steps from a wave propagation simulation on a two-dimensional surface	512 × 512 × 512	single	double

(*) 128 × 512 × 1024 for the single-precision dataset, 130 × 514 × 1026 for double precision

Acknowledgement

If you are using the entire dataset collection as part of a benchmark suite, please refer to:

Fabian Knorr, Peter Thoman, and Thomas Fahringer: "Datasets for benchmarking floating-point compressors", arXiv.org, 2020.

If you intend to refer to individual datasets in detail, please acknowledge:

For rsim:
Thoman, Peter, et al. "RTX-RSim: Accelerated Vulkan room response simulation for time-of-flight imaging." Proceedings of the International Workshop on OpenCL. 2020.
For astro_mhd:
Kissmann, R., et al. "Colliding-wind binaries with strong magnetic fields." The Astrophysical Journal 831.2 (2016): 121.
For astro_pt:
Huber, David, Ralf Kissmann, and Olaf Reimer. "Relativistic fluid modelling of gamma-ray binaries. II. Application to LS 5039." arXiv preprint arXiv:2103.00995 (2021).
For wave:
Thoman, Peter, et al. "Celerity: High-level C++ for accelerator clusters." European Conference on Parallel Processing. Springer, Cham, 2019.