How should we order indices in tracer_local?

[mnlevy1981]

Currently tracer_local is dimension(tracer_cnt, km). Is this what we want? In some subroutines, such as compute_autotroph_elemental_ratios() it would be advantageous to have have km be the inner-most dimension, while that might not be the case for other subroutines.

As I work through #53 and #316 I'll keep an eye out for examples of subroutines where the current ordering is preferable and update this issue accordingly.

[mnlevy1981]

@klindsay28 a starting point could be to keep both tracer_local and its transpose in memory and pass the optimal dimension ordering to each routine. This can help us track which routines are better-suited for each memory layout

@matt-long suggests this will be easier to test once we have a stand-alone driver testing interior_tendency_compute() because it will be faster to build / run

[mnlevy1981]

With the changes from #318 in place, we should swap the dimensions of all the PFT-specific variables as well. For example, compute_Pprime() would be more efficient with the following change:

     !  Compute Pprime for all autotrophs, used for loss terms
     do auto_ind = 1, autotroph_cnt
       where (temperature < autotroph_settings(auto_ind)%temp_thres)
         C_loss_thres = f_loss_thres * autotroph_settings(auto_ind)%loss_thres2
       else where
         C_loss_thres = f_loss_thres * autotroph_settings(auto_ind)%loss_thres
       end where
-      Pprime(auto_ind,:) = max(autotroph_local%C(auto_ind,:) - C_loss_thres, c0)
+      Pprime(:,auto_ind) = max(autotroph_local%C(:,auto_ind) - C_loss_thres, c0)
     end do

This specific example would require changing the constructor for both autotroph_local_type and autotroph_derived_terms_type.