Issue with copy behavior in spiner EOS

[jhp-lanl]

In #375 , a test was added that looks like this:

  portableFor(
      "Positive temperature test", 0, n_eos, PORTABLE_LAMBDA(const int i) {
        const auto &this_eos = v_EOS[i];

        // Find the reference state (really we just want the density :shrug:)
        Real rho_ref;
        Real temp_ref;
        Real sie_ref;
        Real press_ref;
        Real cv_ref;
        Real bmod_ref;
        Real dpde_ref;
        Real dvdt_ref;
        this_eos.ValuesAtReferenceState(rho_ref, temp_ref, sie_ref, press_ref, cv_ref,
                                        bmod_ref, dpde_ref, dvdt_ref);

        // Sanity check on the density to make sure we don't screw this up
        PORTABLE_ALWAYS_REQUIRE(rho_ref > 0, "Zero reference density");

Before this, a host array of EOS objects was created and GetOnDevice() was called for each model (even if it technically wasn't necessary). Then the array was copied into the device array, v_EOS.

If instead I create a copy of the EOS in v_EOS by removing the auto & from the above code so that this_eos is now

        const auto this_eos = v_EOS[i];

then the PORTABLE_ALWAYS_REQUIRE statement above fails for the spiner EOS. It passes for all other EOS though.

This also happens when the const qualifier is removed:

       auto this_eos = v_EOS[i];

or when using the EOS type instead of auto.

I interpret this as the copied spiner EOS having an uninitialized state somehow, but I don't fully understand this issue.

[jhp-lanl]

Oh I should add that this only occurs on NVidia GPU systems.