Ella Wu
Ella Wu
TODOs: - [ ] fix the complex build - [ ] finalize implementation (based on the pyspline PR) - [ ] add printing of the version in the header printed...
Also, `petsc4py` is specified as a dependency but actually only PETSc is needed. Same thing happens in IDwarp.
Can confirm that it is taking a very long time to compile ADflow now, but I'm not sure if it's related to PETSc or if it's something else.
To add on to this, it appears to only affect the intel/impi build when testflo is oversubscribing. I cannot replicate this locally with my gcc build.
@ArshSaja @gang525 what's the status on this?
Is this ready @eirikurj?
Oh I didn't realize we are doing `-ffast-math` by default, that doesn't seem right. I don't think we're doing that in other packages right?
> I agree the cluster builds should include optimizations. Do we still use gfortran on Great Lakes? If so, maybe compiling `myisnan` without the flag is better than removing it...
Ideally, we generate training data without any IEEE-unsafe optimizations, and we run tests with whatever flags we would be running on production cases. I'm not sure how feasible that is...
> With a cheeky little `-fallow-argument-mismatch` I'm able to run it with gcc 11 I would like to eventually fix these, so that we do not have to conditionally add...