exoclime/thor

Atmospheric fluid dynamics solver optimized for GPUs.

THOR

Flexible Global Circulation Model to Explore Planetary Atmospheres

THOR is a GCM that solves the three-dimensional non-hydrostatic Euler equations on an icosahedral grid. THOR was designed to run on Graphics Processing Units (GPUs).

If you use this code please cite: Mendonça, J.M., Grimm, S.L., Grosheintz, L., & Heng, K., ApJ, 829, 115, 2016

If using version 2.0 or later, please also cite: Deitrick, R., Mendonça, J.M., Schroffenegger, U., Grimm, S.L., Tsai, S.-M., & Heng, K., ApJS, 248, 2, 2020

Current code owners: Joao Mendonca: [email protected], Russell Deitrick: [email protected], Urs Schroffenegger: [email protected]

Copyright (C) 2017-2021 Exoclimes Simulation Platform

BUILD & RUN THOR (TL;DR instructions)

   $ sudo apt-get install git make gcc g++ cmake nvidia-cuda-toolkit nvidia-utils-390 libhdf5-dev libhdf5-100  libhdf5-serial-dev libhdf5-cpp-100
   $ git clone https://github.com/exoclime/THOR.git
   $ cd THOR
   $ cp Makefile.conf.template Makefile.conf

Find the SM value of your Nvidia GPU. Decide if you want to run without any physics module empty physics module, or the one with radiative transfer, the multi module. Then open Makefile.conf in a text editor and edit like so:

   MODULES_SRC := src/physics/managers/<module_type>/
   SM:=<SM value of your card> 

Set module_typeto empty (default) or multi.

Then head back to the command line and

   $ make -j8 release

Finally, run

   $ bin/esp ifile/<config file for your planet>

Furthur information

View our wiki pages here

Tutorial from ESP Summer School 2019

thor's Issues

Error in NAN check after PROFX:compute_pressure

After installing THOR (Ubuntu 18.04, compute power=3 NVIDIA GPU), most of the test cases in 'ifile' fail (after some successful initialization) with:

...

Physics module: empty

---

---

Simulation
Start from rest = true
Output directory = earth_hs
Start output numbering at 0.
Starting the model integration.

Time step number = 1/103680 || Time = 0.011574 days.
Elapsed 7s || Left: 8d 18h 37m 53s || Completion: 2020-05-09 11:30:19.

Error in NAN check after PROFX:compute_pressure!

However, the "earth_sync" case runs through at least the first several hundred steps... I'll keep looking into this, but it's odd that the 'out-of-the-box' examples should crash. Thanks for any insights!

Travis-CI -> Github Actions

Travis-CI has changed their pricing structure making unlimited free builds of the repo no longer available.

The best option may be to migrate the travis-ci testing tools here to GitHub Actions, which is an analogous service.

Would @deitrr or @nabajour be able to give me a quick verdict on whether or not it would be useful to update to GitHub Actions? Are the test results still useful?

Related/Links:

• Github Actions Migration guide
• astropy PR migrating from travis-ci to github actions

Cuda v 11.7 Not works

Energy budget

To control the convergence of a simulation, it would be interesting to print the time evolution of the usual variables of the energy budget: incident stellar radiation (ISR), absorbed stellar radiation (ASR), outgoing longwave radiation (OLR), downwelling longwave radiation (GLR), upwelling longwave radiation from the surface (LWU). These quantities correspond to spatially averaged fluxes over the sphere. High-frequency time oscillations of these quantities may indicate a too large time step for radiative transfer.

Porting plans? OpenCL or WebGPU?

Hi,

Have you considered porting THOR or any of the other exoclime projects to run on a more general parallel compute platform, e.g. OpenCL or WebGPU?

I'm on an Apple M1 and would love to try your code and potentially reference it from my web app (celestiary/web#8, https://celestiary.github.io/)

Cheers!
Pablo

Setting up a custom planet case

I am interested in setting up a case for Venus planet.

I have a background of Aerospace Engineering, therefore I am a newbie trying to crack meaning of various terms in available initial configurations (.thr) files.

My interest is into understanding super-rotation of planet Venus and use the wind profile data for aerial vehicle design. I am curious to know the extent to which THOR can model seasonal variation of winds in various parts of planet.

So far, I could build the THOR and run it on a local machine. I ran a few benchmarks provided with the tool. Please suggest on how to setup a case that closely resembles a "real" planet.

Thank you !