Update pynucastro networks

[yut23]

Regenerated from the current pynucastro main branch: pynucastro/pynucastro@556584d.

Also add a python script to do this with a single command.

[yut23]

Going to wait for pynucastro/pynucastro#580 to be merged

[zingale]

do we want to wait until we do the PR that changes the interpolation for tables to work on logs?

[yut23]

Not sure. What do the tests look like now?

[zingale]

I'm not sure I understand.

We haven't implemented that yet. So once we do implement the proper interpolation, we'd want to regenerate these again.

[yut23]

Oh, I see. I was wondering if that would give bigger diffs on the regression tests. I think the Suzuki table updates are the only thing that would change the results in this PR right now.

[zhichen3]

I was trying to build subchandra with CUDA on perlmutter with updated network, but it seems like it's complaining that std::upper_bound doesn't work with cuda.

error: calling a __host__ function("T1  ::std::upper_bound<const double *, double> (T1, T1, const T2 &)") from a __host__ __device__ function("part_fun::interpolate_pf<(int)72> ") is not allowed

I also tried on groot, getting the same error, but I do see the cuda test passed here...

[yut23]

That's unfortunate, we can probably just write a manual binary search instead (or maybe reuse vector_index_lu() from the tabular rates). My guess for why the test passes is that interpolate_pf() is a templated function and so it doesn't get compiled unless it's used somewhere. The CUDA test doesn't build with any networks that use partition functions; maybe we should add that?

[zhichen3]

we do compile test_react with subch_simple in the cuda test though?

[yut23]

Yes, but DO_DERIVED_RATES is false.

Microphysics/networks/subch_simple/partition_functions.H

Lines 66 to 81 in eb5189f

	void get_partition_function(const int inuc, [[maybe_unused]] const tf_t& tfactors,
	Real& pf, Real& dpf_dT) {
	// inuc is the 1-based index for the species
	switch (inuc) {
	default:
	pf = 1.0_rt;
	dpf_dT = 0.0_rt;
	}
	}

[zhichen3]

ah okay. We could build withase network, it has DO_DERIVED_RATES=True.

[yut23]

Yeah, that would be good since it exercises approximate rates too.

[zingale]

with the release of pynucastro 2.1.0, I think we are ready for this now. We should update to get the latest table changes.

[zingale]

I think we need to update some benchmarks now