timofeymukha
commented
4 months ago You might have caught something here, because I don't think we run the checks on the device.
Closed timfelle closed 3 months ago
timofeymukha
commented
4 months ago You might have caught something here, because I don't think we run the checks on the device.
timfelle
commented
4 months ago Hmm alright interesting. I'll investigate a bit more.
timfelle
commented
4 months ago It does complete fine on my tablet, which does not have a GPU.
njansson
commented
4 months ago It does complete fine on my tablet, which does not have a GPU.
Looking at the failing tests, it seems like that one is missing a call to device_init
njansson
commented
4 months ago The best way to avoid this is to add generic setup and teardown routines as in device_math
timfelle
commented
4 months ago Yep that did it. I have added a PR with the corrections. It might still be useful to revisit the tests and make them behave a bit closer to what you describes the math one does, but so far it is a quickfix.
Lets keep this issue open so we remember to do it more thoroughly.
njansson
commented
3 months ago Yep that did it. I have added a PR with the corrections. It might still be useful to revisit the tests and make them behave a bit closer to what you describes the math one does, but so far it is a quickfix.
Lets keep this issue open so we remember to do it more thoroughly.
Should we keep this for v0.8, or postpone the fixes until 0.9?
timfelle
commented
3 months ago Yep that did it. I have added a PR with the corrections. It might still be useful to revisit the tests and make them behave a bit closer to what you describes the math one does, but so far it is a quickfix. Lets keep this issue open so we remember to do it more thoroughly.
Should we keep this for v0.8, or postpone the fixes until 0.9?
When do we plan to do the 0.8 release?
njansson
commented
3 months ago Yep that did it. I have added a PR with the corrections. It might still be useful to revisit the tests and make them behave a bit closer to what you describes the math one does, but so far it is a quickfix. Lets keep this issue open so we remember to do it more thoroughly.
Should we keep this for v0.8, or postpone the fixes until 0.9?
When do we plan to do the 0.8 release?
Current target is feature freeze end of April, and release mid May
timfelle
commented
3 months ago Yep that did it. I have added a PR with the corrections. It might still be useful to revisit the tests and make them behave a bit closer to what you describes the math one does, but so far it is a quickfix. Lets keep this issue open so we remember to do it more thoroughly.
Should we keep this for v0.8, or postpone the fixes until 0.9?
When do we plan to do the 0.8 release?
Current target is feature freeze end of April, and release mid May
I'll look through it before then.
I have managed to setup pFUnit on my local PC and are attempting to get the unit tests up and running. However, there seem to be one that keeps failing on me.
The
point_interpolationcheck keeps failing. It seems to be written as a MPI based test, but it is only executed on a single processor. There seem to be an issue with the underlying device allocation in the space_t. However, the space and device unit tests does not fail.Any ideas what might be happening?
Error message
``` 1/1 Test #26: neko_point_interpolation_parallel ...***Failed 2.79 sec [ctest] Start:System information
``` Windows: Edition Windows 11 Enterprise Version 23H2 Installed on 25/09/2023 OS build 22631.3155 Experience Windows Feature Experience Pack 1000.22684.1000.0 WSL: Distributor ID: Ubuntu Description: Ubuntu 22.04.4 LTS Release: 22.04 Codename: jammy Compilers : gfortran : GNU Fortran (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 nvcc : NVIDIA (R) Cuda compiler driver, release 12.3, V12.3.103 Hardware: CPU type : 11th Gen Intel(R) Core(TM) i9-11950H @ 2.60GHz Bcknd type: Accelerator (CUDA) Dev. name : NVIDIA RTX A3000 Laptop GPU ```