Failing Tests on Latest Compilation

mumax / 3

GPU-accelerated micromagnetic simulator

Other

457 stars 151 forks source link

Failing Tests on Latest Compilation #127

Closed kahlos closed 6 years ago

kahlos commented 7 years ago

I've compiled mumax3 on an up to date Arch linux system running 2x 1080 Ti GPU's, driver 384.69 and cuda 8.0.61. Running the tests everything works fine except failure on the following:

fixedlayer //STtorque[0] : have: 0.00020790935377590358 want: 0.00020761338237207383 ± 1e-07

std5b //m[0] : have: -0.953765869140625 want: -0.9538384675979614 ± 1e-05

Anyone have any ideas? Is this anything to worry about?

barnex commented 7 years ago

This might have something to do with recent STT fixes by @godsic. The result is now possibly more precise, and the tests might have to be adjusted.

kahlos commented 7 years ago

Ah yes, that could be it. The other thing I forgot to mention, on an otherwise identical system, but with a Telsa K40c and Quadro K5200, I don't get these failures. I'd presume it's something to do with the latest compute version running on the 1080 Ti.

godsic commented 7 years ago

@kahlos @barnex I confirm STT-related test with non-zero field-like torque parameter are expected to fail when on the master branch. Hopefully I will find some time soon to make new (OOMMF-based) unit tests.

godsic commented 6 years ago

fixedlayer is fixed with a7384f54181f10f763baabc460cb9d59058c91e0

godsic commented 6 years ago

std5b is fixed with 0a27edafe1b57f9eb33a752d17aed705a777a0da