Memory Layouts for GPU-Data Transfer Buffering in SPH

Abouzied M.A.Nasar; Benedict D. Rogers; Georgios Fourtakas; Matthieu Schaller; Mladen Ivkovic; Scott T. Kay; Tobias Weinzierl

arxiv: 2606.23891 · v1 · pith:EKG3MK27new · submitted 2026-06-22 · 💻 cs.DC · cs.MS· cs.PF

Memory Layouts for GPU-Data Transfer Buffering in SPH

Mladen Ivkovic , Abouzied M.A.Nasar , Tobias Weinzierl , Matthieu Schaller , Benedict D. Rogers , Georgios Fourtakas , Scott T. Kay This is my paper

Pith reviewed 2026-06-26 06:47 UTC · model grok-4.3

classification 💻 cs.DC cs.MScs.PF

keywords Smoothed Particle HydrodynamicsGPU offloadingmemory layoutdata transfer bufferingarray of structureshost-device transferparticle simulation

0 comments

The pith

Splitting monolithic particle structs into access-pattern sub-structs cuts GPU buffer packing time by 20-40% in SPH.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper studies memory layouts for particle data in a Smoothed Particle Hydrodynamics solver when offloading work to GPUs. GPU arithmetic has advanced faster than host-device transfers, so the fraction of time spent moving data now grows and developers must pack data into compact buffers before each offload. The authors replace a single array-of-struct layout with a split array-of-struct layout in which each particle record is broken into several smaller sub-structs chosen according to the read and write patterns of the SPH kernels and the types of the stored attributes. Measurements on the resulting code show that the time to assemble and disassemble the transfer buffers falls by roughly 20 to 40 percent, which in turn reduces the entire GPU-offloading phase by 12 to 25 percent.

Core claim

Splitting classic array-of-struct data structures into a split array-of-struct arrangement, in which each logical struct decomposes into substructs determined by kernel read/write access patterns and attribute types, can reduce the time required to pack data to and from buffers by ~20% - 40%, lowering total time spent on GPU-offloading by ~12% - 25%.

What carries the argument

Split array-of-struct arrangement that decomposes each particle struct into multiple finer-grained sub-structs matched to kernel access patterns.

Load-bearing premise

The read and write patterns of the SPH kernels allow the particle struct to be split into sub-structs without adding offsetting overhead during kernel execution or data access.

What would settle it

Measure packing and total offloading times on the same SPH run before and after applying the split layout; the reported 20-40% and 12-25% reductions must appear in those timings.

Figures

Figures reproduced from arXiv: 2606.23891 by Abouzied M.A.Nasar, Benedict D. Rogers, Georgios Fourtakas, Matthieu Schaller, Mladen Ivkovic, Scott T. Kay, Tobias Weinzierl.

**Figure 1.** Figure 1: Relative fraction of time spent in offloading cycles for the interaction loops density, gradient, and force in each of the three offloading steps (packing, launch, unpacking) Swift employs for 16 simulation steps of the Gresho256 test. Timings have been obtained on two architectures: A node comprising two Intel Xeon Gold 6430 CPUs connected via PCIe 4 to a NVDIA A30 GPU (blue bars) and an NVIDIA Grace Hopp… view at source ↗

**Figure 2.** Figure 2: Time to complete all packing and unpacking operations for varying particle memory layout variants (x-axis) and accessor methods (line colours) over all compute kernel types. Results obtained from Intel+A30 (left) and Grace Hopper (right) normalised against the part-struct accessor and AoS baseline [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 3.** Figure 3: Normalised time to complete all packing and unpacking operations for varying particle memory layout variants (x-axis) and accessor methods (line colours) on Intel+A30 for different compute kernels. We observe that particular particle sublayouts have a pronounced effect on different kernels: the total impact of a given layout optimisation is therefore a weighted combination of the runtime improvement due t… view at source ↗

**Figure 4.** Figure 4: Time required to complete all packing and unpacking operations for varying particle memory layout variants (x-axis) and accessor methods (line styles) for different loop fission methods (colours) relative to the by-particle baseline. Top: results obtained on Intel+A30. Bottom: results obtained on Grace Hopper. Left: results for the Gresho256 experiment. Right: results for the Eagle25 experiment. Recommend… view at source ↗

read the original abstract

The rise in GPU compute speed has outpaced improvements in host-to-device memory transfer speeds, despite the advent of shared-memory superchips. Consequently, memory transfer times now constitute an increasingly large fraction of total time-to-solution, compelling developers to compress GPU kernel input and output data into compact, minimal formats prior to GPU-offloading. This complements existing work on GPU- and compute-friendly data arrangements. We study a Smoothed Particle Hydrodynamics solver and propose memory layout strategies for host-side particle data that are particularly well-suited to GPU-offloading. Specifically, we advocate splitting classic array-of-struct data structures into a split array-of-struct arrangement, in which each logical struct decomposes into substructs determined by kernel read/write access patterns and attribute types. Splitting a monolithic particle struct into several bespoke, finer-grained structs can reduce the time required to pack data to and from buffers by ~20% - 40%, lowering total time spent on GPU-offloading by ~12% - 25%.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper proposes splitting monolithic particle structs into access-pattern-driven sub-structs to cut SPH GPU packing time, but the quantitative claims rest on unshown experiments.

read the letter

The core idea is to decompose the usual array-of-struct particle layout into several narrower arrays, each containing only the fields that a given SPH kernel phase actually touches. The authors argue this reduces the work of packing and unpacking host buffers before and after GPU transfer, yielding 20-40% faster packing and 12-25% lower total offload time.

This is a straightforward engineering step rather than a new theory. It extends standard memory-layout advice to the specific constraints of SPH data movement on current GPUs, where transfer bandwidth lags compute. The choice of splits is guided by actual read/write sets and attribute types, which is the concrete contribution.

The abstract states the gains clearly, yet supplies no hardware description, no error bars, no kernel-only timings, and no end-to-end numbers that include the kernels after the layout change. The stress-test concern is therefore live: if the split layout increases cache misses or register pressure inside the kernels, the net saving could be smaller than claimed. Without those measurements the central performance assertion cannot be checked.

The paper targets practitioners already optimizing SPH or similar particle codes for GPU offload. A reader working on data-layout tuning for transfer-bound workloads will find the concrete split strategy useful to try. It is not aimed at readers seeking first-principles memory models or broad performance portability results.

I would bring the full paper to a reading group as maybe, mainly to see whether the experiments close the gap the abstract leaves open. I would not cite it until the numbers are verified. The work is coherent enough on its own terms to merit peer review rather than desk rejection, provided the full manuscript contains reproducible benchmarks that address kernel runtime impact.

Referee Report

2 major / 0 minor

Summary. The manuscript studies memory layouts for host-side particle data in a Smoothed Particle Hydrodynamics (SPH) solver to reduce the cost of packing/unpacking data for GPU offloading. It proposes decomposing a monolithic array-of-structs into a split array-of-structs whose sub-structs are chosen according to kernel read/write patterns and attribute types. The central quantitative claim is that this decomposition reduces packing time by ~20–40 % and total GPU-offloading time by ~12–25 %.

Significance. If the claimed reductions are shown to be net gains after accounting for any changes in kernel execution time, the work would provide a practical, access-pattern-driven technique for mitigating the growing fraction of time spent on host–device transfers in GPU-accelerated particle codes. The approach complements existing GPU-friendly data arrangements and could be relevant to other particle-based or irregular-access simulations.

major comments (2)

[Abstract] Abstract: the performance claims rest on the assumption that the chosen split produces no material increase in SPH kernel runtime or cache behavior that would offset the reported packing-time savings. The text gives no indication that end-to-end timings (packing + transfer + kernel execution + unpacking) or isolated kernel timings before/after the layout change were measured; without such data the net 12–25 % improvement cannot be verified.
[Abstract] The manuscript supplies no experimental setup, hardware description, compiler flags, error bars, or discussion of potential confounding factors (cache effects, register pressure, indirection overhead). Consequently the quantitative claims cannot be assessed from the given text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We agree that additional experimental details and validation measurements are needed to fully support the quantitative claims. We will revise the manuscript to address these points.

read point-by-point responses

Referee: [Abstract] Abstract: the performance claims rest on the assumption that the chosen split produces no material increase in SPH kernel runtime or cache behavior that would offset the reported packing-time savings. The text gives no indication that end-to-end timings (packing + transfer + kernel execution + unpacking) or isolated kernel timings before/after the layout change were measured; without such data the net 12–25 % improvement cannot be verified.

Authors: We agree that net end-to-end gains require confirmation that kernel execution time is not materially increased. The reported 12–25 % reduction in total GPU-offloading time refers specifically to the combined cost of packing, host-to-device transfer, and unpacking; the GPU-side data layout and kernel code remain unchanged by the host-side split. Nevertheless, to address the concern, the revised manuscript will include isolated kernel execution timings (before and after the layout change) along with a brief discussion of cache and indirection effects. This will allow readers to verify that any overhead is negligible relative to the packing savings. revision: yes
Referee: [Abstract] The manuscript supplies no experimental setup, hardware description, compiler flags, error bars, or discussion of potential confounding factors (cache effects, register pressure, indirection overhead). Consequently the quantitative claims cannot be assessed from the given text.

Authors: We acknowledge that the current manuscript lacks a dedicated experimental-methods section. The revision will add a new subsection describing the test hardware (CPU, GPU, interconnect), compiler and optimization flags, number of runs used for timing, error-bar methodology, and a short discussion of potential confounding factors including cache behavior, register pressure, and any indirection cost introduced by the split array-of-structs representation. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical timing measurements with no derivations or fitted predictions

full rationale

The paper presents an empirical study of memory layout changes for SPH particle data, reporting measured reductions in packing and GPU-offloading times (20-40% and 12-25% respectively) based on access-pattern-driven struct splitting. No equations, first-principles derivations, parameter fits, or predictions appear in the abstract or described content; the central claims rest on direct timing benchmarks rather than any self-referential construction, self-citation chain, or renamed known result. The derivation chain is therefore self-contained as a set of performance observations.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no equations, parameters, or background assumptions can be extracted.

pith-pipeline@v0.9.1-grok · 5733 in / 1088 out tokens · 24261 ms · 2026-06-26T06:47:55.953369+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 7 canonical work pages

[1]

In: 13th SPHERIC International Workshop

Borrow, J., Bower, R.G., Draper, P.W., Gonnet, P., Schaller, M.: SWIFT: Main- taining weak-scalability with a dynamic range of 10^4 in time-step size to harness extreme adaptivity. In: 13th SPHERIC International Workshop. pp. 44–51. Gal- way, Ireland (Jul 2018)

2018
[2]

, keywords =

Borrow, J., Schaller, M., Bower, R.G., Schaye, J.:Sphenix: Smoothed particle hydrodynamics for the next generation of galaxy formation simulations. Monthly Notices of the Royal Astronomical Society511(2), 2367–2389 (Feb 2022). https: //doi.org/10.1093/mnras/stab3166

work page doi:10.1093/mnras/stab3166 2022
[3]

Journal of Par- allel and Distributed Computing pp

Carter Edwards, H., Trott, C.R., Sunderland, D.: Kokkos: Enabling manycore performance portability through polymorphic memory access patterns. Journal of Parallel and Distributed Computing74(12), 3202–3216 (Dec 2014). https: //doi.org/10.1016/j.jpdc.2014.07.003

work page doi:10.1016/j.jpdc.2014.07.003 2014
[4]

Monthly Notices of the Royal Astronomical Society 539(1), 1–33 (2025)

David-Cléris, T., Laibe, G., Lapeyre, Y.: The shamrock code: I–smoothed parti- cle hydrodynamics on gpus. Monthly Notices of the Royal Astronomical Society 539(1), 1–33 (2025)

2025
[5]

Computational Particle Mechanics9(5), 867–895 (2022)

Domínguez, J.M., Fourtakas, G., Altomare, C., Canelas, R.B., Tafuni, A., García- Feal, O., Martínez-Estévez, I., Mokos, A., Vacondio, R., Crespo, A.J., et al.: Du- alsphysics: from fluid dynamics to multiphysics problems. Computational Particle Mechanics9(5), 867–895 (2022)

2022
[6]

In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis

Frontiere, N., Emberson, J.D., Buehlmann, M., Rangel, E.M., Habib, S., Heitmann, K., Larsen, P., Morozov, V., Pope, A., Faucher-Giguère, C.A., et al.: Cosmological hydrodynamics at exascale: A trillion-particle leap in capability. In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis. pp. 25–35 (2025)

2025
[7]

Gingold, R.A., Monaghan, J.J.: Smoothed particle hydrodynamics: theory and applicationtonon-sphericalstars.Monthlynoticesoftheroyalastronomicalsociety 181(3), 375–389 (1977)

1977
[8]

Part 2: Implementation

Gresho, P.M., Chan, S.T.: On the theory of semi-implicit projection methods for viscous incompressible flow and its implementation via a finite element method that also introduces a nearly consistent mass matrix. Part 2: Implementation. International Journal for Numerical Methods in Fluids11(5), 621–659 (1990). https://doi.org/10.1002/fld.1650110510

work page doi:10.1002/fld.1650110510 1990
[9]

Nuncius1(aop), 1–20 (2026) 14 M

Jones, M.L.: From gaming to science: How the graphical processor unit became a supercomputer. Nuncius1(aop), 1–20 (2026) 14 M. Ivkovic et al

2026
[10]

In: Varbanescu, A.L., Bhatele, A., Luszczek, P., Marc, B

Li, B., Schulz, H., Weinzierl, T., Zhang, H.: Dynamic Task Fusion for a Block- Structured Finite Volume Solver over a Dynamically Adaptive Mesh with Local Time Stepping. In: Varbanescu, A.L., Bhatele, A., Luszczek, P., Marc, B. (eds.) High Performance Computing. pp. 153–173. Springer International Publishing, Cham (2022). https://doi.org/10.1007/978-3-031...

work page doi:10.1007/978-3-031-07312-0_8 2022
[11]

Astro- nomical Journal, vol

Lucy, L.B.: A numerical approach to the testing of the fission hypothesis. Astro- nomical Journal, vol. 82, Dec. 1977, p. 1013-1024.82, 1013–1024 (1977)

1977
[12]

RAS Techniques and Instruments5, rzag008 (Jan 2026)

Nasar, A.M.A., Rogers, B.D., Fourtakas, G., Ivkovic, M., Weinzierl, T., Kay, S.T., Schaller, M.: Task-parallelism in SWIFT for heterogeneous compute architectures. RAS Techniques and Instruments5, rzag008 (Jan 2026). https://doi.org/10.1093/ rasti/rzag008

2026
[13]

Jour- nal of Computational Physics231(3), 759–794 (Feb 2012)

Price, D.J.: Smoothed Particle Hydrodynamics and Magnetohydrodynamics. Jour- nal of Computational Physics231(3), 759–794 (Feb 2012). https://doi.org/10. 1016/j.jcp.2010.12.011

2012
[14]

arXiv e-prints arXiv:2502.16517 (Feb 2025)

Radtke, P.K., Weinzierl, T.: Annotation-guided AoS-to-SoA conversions and GPU offloading with data views in C++. arXiv e-prints arXiv:2502.16517 (Feb 2025). https://doi.org/10.48550/arXiv.2502.16517

work page doi:10.48550/arxiv.2502.16517 2025
[15]

https://doi.org/10.48550/ arXiv.2512.05516

Radtke, P.K., Weinzierl, T.: Compiler-supported reduced precision and AoS-SoA transformations for heterogeneous hardware (Dec 2025). https://doi.org/10.48550/ arXiv.2512.05516

arXiv 2025
[16]

In: Proceedings of the 2026 SIAM Conference on Parallel Processing for Scientific Computing (PP)

Radtke, P.K., Weinzierl, T.: Compiler-supported reduced precision and aos-soa transformations for heterogeneous hardware. In: Proceedings of the 2026 SIAM Conference on Parallel Processing for Scientific Computing (PP). pp. 103–116. SIAM (2026)

2026
[17]

, keywords =

Schaller, M., Borrow, J., Draper, P.W., Ivkovic, M., McAlpine, S., Vandenbroucke, B., Bahé, Y., Chaikin, E., Chalk, A.B.G., Chan, T.K., Correa, C., van Daalen, M., Elbers, W., Gonnet, P., Hausammann, L., Helly, J., Huško, F., Kegerreis, J.A., Nobels, F.S.J., Ploeckinger, S., Revaz, Y., Roper, W.J., Ruiz-Bonilla, S., Sandnes, T.D., Uyttenhove, Y., Willis, ...

work page doi:10.1093/mnras/stae922 2024
[18]

Monthly Notices of the Royal Astronomical Society548(1), stag375 (2026)

Schaye, J., Chaikin, E., Schaller, M., Ploeckinger, S., Huško, F., McGibbon, R.J., Trayford, J.W., Benítez-Llambay, A., Correa, C., Frenk, C.S., et al.: The colibre project: cosmological hydrodynamical simulations of galaxy formation and evolu- tion. Monthly Notices of the Royal Astronomical Society548(1), stag375 (2026)

2026
[19]

A., Bower, R

Schaye, J., Crain, R.A., Bower, R.G., Furlong, M., Schaller, M., Theuns, T., Dalla Vecchia, C., Frenk, C.S., McCarthy, I.G., Helly, J.C., Jenkins, A., Rosas- Guevara, Y.M., White, S.D.M., Baes, M., Booth, C.M., Camps, P., Navarro, J.F., Qu, Y., Rahmati, A., Sawala, T., Thomas, P.A., Trayford, J.: The EA- GLE project: Simulating the evolution and assembly ...

work page doi:10.1093/mnras/stu2058 2015
[20]

Schaye, J., Kugel, R., Schaller, M., Helly, J.C., Braspenning, J., Elbers, W., McCarthy, I.G., Van Daalen, M.P., Vandenbroucke, B., Frenk, C.S., et al.: The flamingo project: cosmological hydrodynamical simulations for large-scale struc- tureandgalaxyclustersurveys.MonthlyNoticesoftheRoyalAstronomicalSociety 526(4), 4978–5020 (2023)

2023
[21]

Computer physics communications271, 108171 (2022)

Thompson, A.P., Aktulga, H.M., Berger, R., Bolintineanu, D.S., Brown, W.M., Crozier, P.S., In’t Veld, P.J., Kohlmeyer, A., Moore, S.G., Nguyen, T.D., et al.: Memory Layouts for GPU-Data Transfer Buffering in SPH 15 Lammps-a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales. Computer physics communicat...

2022

[1] [1]

In: 13th SPHERIC International Workshop

Borrow, J., Bower, R.G., Draper, P.W., Gonnet, P., Schaller, M.: SWIFT: Main- taining weak-scalability with a dynamic range of 10^4 in time-step size to harness extreme adaptivity. In: 13th SPHERIC International Workshop. pp. 44–51. Gal- way, Ireland (Jul 2018)

2018

[2] [2]

, keywords =

Borrow, J., Schaller, M., Bower, R.G., Schaye, J.:Sphenix: Smoothed particle hydrodynamics for the next generation of galaxy formation simulations. Monthly Notices of the Royal Astronomical Society511(2), 2367–2389 (Feb 2022). https: //doi.org/10.1093/mnras/stab3166

work page doi:10.1093/mnras/stab3166 2022

[3] [3]

Journal of Par- allel and Distributed Computing pp

Carter Edwards, H., Trott, C.R., Sunderland, D.: Kokkos: Enabling manycore performance portability through polymorphic memory access patterns. Journal of Parallel and Distributed Computing74(12), 3202–3216 (Dec 2014). https: //doi.org/10.1016/j.jpdc.2014.07.003

work page doi:10.1016/j.jpdc.2014.07.003 2014

[4] [4]

Monthly Notices of the Royal Astronomical Society 539(1), 1–33 (2025)

David-Cléris, T., Laibe, G., Lapeyre, Y.: The shamrock code: I–smoothed parti- cle hydrodynamics on gpus. Monthly Notices of the Royal Astronomical Society 539(1), 1–33 (2025)

2025

[5] [5]

Computational Particle Mechanics9(5), 867–895 (2022)

Domínguez, J.M., Fourtakas, G., Altomare, C., Canelas, R.B., Tafuni, A., García- Feal, O., Martínez-Estévez, I., Mokos, A., Vacondio, R., Crespo, A.J., et al.: Du- alsphysics: from fluid dynamics to multiphysics problems. Computational Particle Mechanics9(5), 867–895 (2022)

2022

[6] [6]

In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis

Frontiere, N., Emberson, J.D., Buehlmann, M., Rangel, E.M., Habib, S., Heitmann, K., Larsen, P., Morozov, V., Pope, A., Faucher-Giguère, C.A., et al.: Cosmological hydrodynamics at exascale: A trillion-particle leap in capability. In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis. pp. 25–35 (2025)

2025

[7] [7]

Gingold, R.A., Monaghan, J.J.: Smoothed particle hydrodynamics: theory and applicationtonon-sphericalstars.Monthlynoticesoftheroyalastronomicalsociety 181(3), 375–389 (1977)

1977

[8] [8]

Part 2: Implementation

Gresho, P.M., Chan, S.T.: On the theory of semi-implicit projection methods for viscous incompressible flow and its implementation via a finite element method that also introduces a nearly consistent mass matrix. Part 2: Implementation. International Journal for Numerical Methods in Fluids11(5), 621–659 (1990). https://doi.org/10.1002/fld.1650110510

work page doi:10.1002/fld.1650110510 1990

[9] [9]

Nuncius1(aop), 1–20 (2026) 14 M

Jones, M.L.: From gaming to science: How the graphical processor unit became a supercomputer. Nuncius1(aop), 1–20 (2026) 14 M. Ivkovic et al

2026

[10] [10]

In: Varbanescu, A.L., Bhatele, A., Luszczek, P., Marc, B

Li, B., Schulz, H., Weinzierl, T., Zhang, H.: Dynamic Task Fusion for a Block- Structured Finite Volume Solver over a Dynamically Adaptive Mesh with Local Time Stepping. In: Varbanescu, A.L., Bhatele, A., Luszczek, P., Marc, B. (eds.) High Performance Computing. pp. 153–173. Springer International Publishing, Cham (2022). https://doi.org/10.1007/978-3-031...

work page doi:10.1007/978-3-031-07312-0_8 2022

[11] [11]

Astro- nomical Journal, vol

Lucy, L.B.: A numerical approach to the testing of the fission hypothesis. Astro- nomical Journal, vol. 82, Dec. 1977, p. 1013-1024.82, 1013–1024 (1977)

1977

[12] [12]

RAS Techniques and Instruments5, rzag008 (Jan 2026)

Nasar, A.M.A., Rogers, B.D., Fourtakas, G., Ivkovic, M., Weinzierl, T., Kay, S.T., Schaller, M.: Task-parallelism in SWIFT for heterogeneous compute architectures. RAS Techniques and Instruments5, rzag008 (Jan 2026). https://doi.org/10.1093/ rasti/rzag008

2026

[13] [13]

Jour- nal of Computational Physics231(3), 759–794 (Feb 2012)

Price, D.J.: Smoothed Particle Hydrodynamics and Magnetohydrodynamics. Jour- nal of Computational Physics231(3), 759–794 (Feb 2012). https://doi.org/10. 1016/j.jcp.2010.12.011

2012

[14] [14]

arXiv e-prints arXiv:2502.16517 (Feb 2025)

Radtke, P.K., Weinzierl, T.: Annotation-guided AoS-to-SoA conversions and GPU offloading with data views in C++. arXiv e-prints arXiv:2502.16517 (Feb 2025). https://doi.org/10.48550/arXiv.2502.16517

work page doi:10.48550/arxiv.2502.16517 2025

[15] [15]

https://doi.org/10.48550/ arXiv.2512.05516

Radtke, P.K., Weinzierl, T.: Compiler-supported reduced precision and AoS-SoA transformations for heterogeneous hardware (Dec 2025). https://doi.org/10.48550/ arXiv.2512.05516

arXiv 2025

[16] [16]

In: Proceedings of the 2026 SIAM Conference on Parallel Processing for Scientific Computing (PP)

Radtke, P.K., Weinzierl, T.: Compiler-supported reduced precision and aos-soa transformations for heterogeneous hardware. In: Proceedings of the 2026 SIAM Conference on Parallel Processing for Scientific Computing (PP). pp. 103–116. SIAM (2026)

2026

[17] [17]

, keywords =

Schaller, M., Borrow, J., Draper, P.W., Ivkovic, M., McAlpine, S., Vandenbroucke, B., Bahé, Y., Chaikin, E., Chalk, A.B.G., Chan, T.K., Correa, C., van Daalen, M., Elbers, W., Gonnet, P., Hausammann, L., Helly, J., Huško, F., Kegerreis, J.A., Nobels, F.S.J., Ploeckinger, S., Revaz, Y., Roper, W.J., Ruiz-Bonilla, S., Sandnes, T.D., Uyttenhove, Y., Willis, ...

work page doi:10.1093/mnras/stae922 2024

[18] [18]

Monthly Notices of the Royal Astronomical Society548(1), stag375 (2026)

Schaye, J., Chaikin, E., Schaller, M., Ploeckinger, S., Huško, F., McGibbon, R.J., Trayford, J.W., Benítez-Llambay, A., Correa, C., Frenk, C.S., et al.: The colibre project: cosmological hydrodynamical simulations of galaxy formation and evolu- tion. Monthly Notices of the Royal Astronomical Society548(1), stag375 (2026)

2026

[19] [19]

A., Bower, R

Schaye, J., Crain, R.A., Bower, R.G., Furlong, M., Schaller, M., Theuns, T., Dalla Vecchia, C., Frenk, C.S., McCarthy, I.G., Helly, J.C., Jenkins, A., Rosas- Guevara, Y.M., White, S.D.M., Baes, M., Booth, C.M., Camps, P., Navarro, J.F., Qu, Y., Rahmati, A., Sawala, T., Thomas, P.A., Trayford, J.: The EA- GLE project: Simulating the evolution and assembly ...

work page doi:10.1093/mnras/stu2058 2015

[20] [20]

Schaye, J., Kugel, R., Schaller, M., Helly, J.C., Braspenning, J., Elbers, W., McCarthy, I.G., Van Daalen, M.P., Vandenbroucke, B., Frenk, C.S., et al.: The flamingo project: cosmological hydrodynamical simulations for large-scale struc- tureandgalaxyclustersurveys.MonthlyNoticesoftheRoyalAstronomicalSociety 526(4), 4978–5020 (2023)

2023

[21] [21]

Computer physics communications271, 108171 (2022)

Thompson, A.P., Aktulga, H.M., Berger, R., Bolintineanu, D.S., Brown, W.M., Crozier, P.S., In’t Veld, P.J., Kohlmeyer, A., Moore, S.G., Nguyen, T.D., et al.: Memory Layouts for GPU-Data Transfer Buffering in SPH 15 Lammps-a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales. Computer physics communicat...

2022