Continuous integration testing of GPUSolver

mit-crpg / OpenMOC

A method of characteristics code for nuclear reactor physics calculations.

https://mit-crpg.github.io/OpenMOC/

Other

150 stars 84 forks source link

Continuous integration testing of GPUSolver #445

Open GiudGiud opened 5 years ago

GiudGiud commented 5 years ago

Now that the GPUSolver is back online (thanks to @gridley), it'd be great to make sure new PRs do not break it. To do that, it should be part of our test suite, with a few limitations.

Travis CI doesn't provide a GPU in its free version to my knowledge, so this test would only be run locally
Most users did not install CUDA or many do not have a nVidia GPU, so it should be an option to the general test suite. Like : python run_tests.py --with-cuda

lishunde commented 5 years ago

I tried to use the GPU to solve the c5g7-3D algorithm, but always reported an error (2d can run normally), have you tested it in 3dMOC?

gridley commented 5 years ago

Hey @lishunde, it currently is not set up for 3D. Sorry about that.

gridley commented 5 years ago

@GiudGiud did you ever do this? Because it'll probably take like 10 minutes to write this.

GiudGiud commented 5 years ago

No I haven't. It's all yours if you want

gridley commented 4 years ago

Hey, so, this is something I'd like to get done. If we want to be testing on the latest CUDA version, it seems to require gcc 7. However, the Travis CI seems to currently be done with gcc-6. @GiudGiud you cool to switch to gcc 7?

GiudGiud commented 4 years ago

That's fine with me

GiudGiud commented 4 years ago

@gridley any progress on this ? I'm adding CMFD to the GPUSolver hence the renewed interest in proper testing of the GPUSolver.

gridley commented 4 years ago

Whoa, really!?? That’s awesome! I’ll definitely set this up then.