IDEFIX: A versatile performance-portable Godunov code for astrophysical flows

Lesur, Geoffroy; Baghdadi, Siwar; Wafflard-Fernandez, G.; J., Mauxion,; Robert, Carmelle; Bossche, Matthias Van den

doi:10.1051/0004-6361/202346005

Cited by 7 publications

(1 citation statement)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is true for many different research fields where fluid and plasma turbulence play a relevant role, from engineering applications of standard hydrodynamic flows [35][36][37][38][39][40], to relativistic MHD and GRMHD [41][42][43][44]. Unfortunately, the process of porting an original code previously engineered to work on CPU-based systems can be very long and it usually requires a complete rewriting of the code in the CUDA language (created by NVIDIA, available for both C and Fortran HPC languages) or using meta-programming libraries like KOKKOS (e.g., [45]) or SYCL/Data Parallel C++ (recently successfully applied to a reduced and independent version of our code, see https://www.intel.com/content/www/us/en/developer/articles/news/ parallel-universe-magazine-issue-51-january-2023.html, accessed on 27 November 2023).…”

Section: Introductionmentioning

confidence: 99%

A GPU-Accelerated Modern Fortran Version of the ECHO Code for Relativistic Magnetohydrodynamics

Del Zanna,

Landi,

Serafini

et al. 2024

Fluids

View full text Add to dashboard Cite

The numerical study of relativistic magnetohydrodynamics (MHD) plays a crucial role in high-energy astrophysics but unfortunately is computationally demanding, given the complex physics involved (high Lorentz factor flows, extreme magnetization, and curved spacetimes near compact objects) and the large variety of spatial scales needed to resolve turbulent motions. A great benefit comes from the porting of existing codes running on standard processors to GPU-based platforms. However, this usually requires a drastic rewriting of the original code, the use of specific languages like CUDA, and a complex analysis of data management and optimization of parallel processes. Here, we describe the porting of the ECHO code for special and general relativistic MHD to accelerated devices, simply based on native Fortran language built-in constructs, especially do concurrent loops, few OpenACC directives, and straightforward data management provided by the Unified Memory option of NVIDIA compilers. Thanks to these very minor modifications to the original code, the new version of ECHO runs at least 16 times faster on GPU platforms as compared to CPU-based ones. The chosen benchmark is the 3D propagation of a relativistic MHD Alfvén wave, for which strong and weak scaling tests performed on the LEONARDO pre-exascale supercomputer at CINECA are provided (using up to 256 nodes corresponding to 1024 GPUs, and over 14 billion cells). Finally, an example of high-resolution relativistic MHD Alfvénic turbulence simulation is shown, demonstrating the potential for astrophysical plasmas of the new GPU-based version of ECHO.

show abstract

Section: Introductionmentioning

confidence: 99%

A GPU-Accelerated Modern Fortran Version of the ECHO Code for Relativistic Magnetohydrodynamics

Del Zanna,

Landi,

Serafini

et al. 2024

Fluids

View full text Add to dashboard Cite

show abstract

Toward an efficient second‐order method for computing the surface gravitational potential on spherical‐polar meshes

Gressel,

Ziegler

2024

Astron Nachr

View full text Add to dashboard Cite

Astrophysical accretion discs that carry a significant mass compared with their central object are subject to the effect of self‐gravity. In the context of circumstellar discs, this can, for instance, cause fragmentation of the disc gas, and—under suitable conditions—lead to the direct formation of gas‐giant planets. If one wants to study these phenomena, the disc's gravitational potential needs to be obtained by solving the Poisson equation. This requires to specify suitable boundary conditions. In the case of a spherical‐polar computational mesh, a standard multipole expansion for obtaining boundary values is not practicable. We hence compare two alternative methods for overcoming this limitation. The first method is based on a known Green's function expansion (termed “CCGF”) of the potential, while the second (termed “James' method”) uses a surface screening mass approach with a suitable discrete Green's function. We demonstrate second‐order convergence for both methods and test the weak scaling behavior when using thousands of computational cores. Overall, James' method is found superior owing to its favorable algorithmic complexity of compared with the scaling of the CCGF method.

show abstract

Planet-disk-wind interaction: The magnetized fate of protoplanets

Wafflard-Fernandez¹,

Lesur²

2023

A&A

View full text Add to dashboard Cite

Context. Models of a planet-disk interaction are mainly based on 2D and 3D viscous hydrodynamic simulations. In such models, accretion is classically prescribed by an αv parameter which characterizes the turbulent radial transport of angular momentum in the disk. This accretion scenario has been questioned for a few years and an alternative paradigm has been proposed that involves the vertical transport of angular momentum by magneto-hydrodynamic (MHD) winds. Aims. We revisit planet–disk interactions in the context of MHD wind-launching protoplanetary disks. In particular, we focus on the planet’s ability to open a gap and produce meridional flows. The accretion, magnetic field, and wind torque in the gap are also explored, as well as the evaluation of the gravitational torque exerted by the disk onto the planet. Methods. We carried out high-resolution 3D global nonideal MHD simulations of a gaseous disk threaded by a large-scale vertical magnetic field harboring a planet in a fixed circular orbit using the code IDEFIX, which is accelerated with graphics processing units. We considered various planet masses (10 Earth masses, 1 Saturn mass, 1 Jupiter mass, and 3 Jupiter masses for a solar-mass star) and disk magnetizations (104 and 103 for the β-plasma parameter, defined as the ratio of the thermal pressure over the magnetic pressure). Results. We find that a gap opening always occurs for sufficiently massive planets, typically on the order of a few Saturn masses for β0 = 103, with deeper gaps when the planet mass increases and when the initial magnetization decreases. We propose an expression for the gap-opening criterion when accretion is dominated by MHD winds. We show that accretion is unsteady and comes from surface layers in the outer disk, bringing material directly toward the planet poles. A planet gap is a privileged region for the accumulation of a large-scale magnetic field, preferentially at the gap center or at the gap edges in some cases. This results in a fast accretion stream through the gap, which can become sonic at high magnetizations. The torque due to the MHD wind responds to the planet presence in a way that leads to a more intense wind in the outer gap compared to the inner gap. More precisely, for massive planets, the wind torque is enhanced as it is fed by the planet torque above the gap’s outer edge, whereas the wind torque is seemingly diminished above the gap’s inner edge due to the planet-induced deflection of magnetic field lines at the disk surface. This induces an asymmetric gap, both in depth and in width, that progressively erodes the outer gap edge, reducing the outer Lindblad torque and potentially reversing the migration direction of Jovian planets in magnetized disks after a few hundreds of orbits. For low-mass planets, we find strongly fluctuating gravitational torques that are mostly positive on average, indicating a stochastic outward migration. Conclusions. The presence of MHD winds strongly affects planet-disk interaction, both in terms of flow kinematics and protoplanet migration. This work illustrates the tight dependence between the planet torque, the wind torque, and magnetic field transport that is required to get the correct dynamics of such systems. In particular, many of the predictions from “effective” models that use parameterized wind torques are not recovered (such as gap formation criteria, the migration direction, and speed) in our simulations.

show abstract

IDEFIX: A versatile performance-portable Godunov code for astrophysical flows

Cited by 7 publications

References 56 publications

A GPU-Accelerated Modern Fortran Version of the ECHO Code for Relativistic Magnetohydrodynamics

A GPU-Accelerated Modern Fortran Version of the ECHO Code for Relativistic Magnetohydrodynamics

Toward an efficient second‐order method for computing the surface gravitational potential on spherical‐polar meshes

Planet-disk-wind interaction: The magnetized fate of protoplanets

Contact Info

Product

Resources

About