Joe Wallwork
Joe Wallwork
Alternatively, could we set the whole test suite up so that it's executed using `ctest` rather than with the `run_test_suite` script? That way we wouldn't need to worry about the...
@TomMelt found https://epubs.stfc.ac.uk/manifestation/4395/RAL_TR_2009_019.pdf
And/or use formatting for Fortran `write` statements.
Seems to all be set up although the CI hits a linking error for pFUnit ``` -- Building with Fortran PyTorch coupling -- Configuring done (1.8s) -- Generating done (0.1s)...
Also note that the Docker image is fairly chunky. This is mainly due to `intel-openapi-mkl` getting pulled in as a dependency for Spack's `py-torch` package. I haven't figured out if...
> Also note that the Docker image is fairly chunky. This is mainly due to `intel-openapi-mkl` getting pulled in as a dependency for Spack's `py-torch` package. I haven't figured out...
> Hi @jwallwork23, I am going to take a look at this PR and likely will atomize it the same way I approached the Intel CI in a previous PR....
Added a load of testing. There are some cases that seem to be inconsistent with what I'd expect from PyTorch. (See the `NOTE`s in the unit tests.)
> Though I'm surprised its showing up unless you are building on an obscure system or compiler? Ah it's because I hadn't accounted for the `UNIX` macro. Works if I...
> But yes, try a version of this rebased on #416 and it should hopefully go away (just don't forget to remove the `UNIX` macro if you added it ;)...