Andy Porter
Andy Porter
e.g. if we have some code: my_array(:,:) = cp_air(array1(:,:)) * array2(:,:) + array3(:,:) where `cp_air` is an array-valued function that performs compute then this is hard to parallelise/ensure it remains...
Yes, I said as much to Chris when we were discussing this. Hopefully there's some reuse possible.
Obviously, a step 0 for this work is to be able to identify array-valued functions. Currently, all RoutineSymbols that represent Fortran functions are given DeferredType (#1294).
This is going quite well. I need to handle the case where there's a name clash for a `ContainerSymbol` as we can't simply rename it in that case.
Note to self: I need to identify when a routine accesses variables made available from some outer scoping region as this prevents inlining.
The first review and subsequent work on `validate` has thrown up a number of things that I'm adding TODOs for this time around: - [ ] `validate` will be much...
The problem is that the current PSyclone support for intergrid kernels was motivated by the multi-grid solver. As such it assumes that each grid in a hierarchy increases (or decreases)...
I'd argue that the workaround is just that - it means the metadata is no longer correct and, once we get around to implementing checks that supplied kernel arguments are...
I realise that in our meeting today we didn't discuss the problem with configurable nesting of meshes and the fact that the halos then become too deep. That feels like...
Created branch `1542_cont_write` to alter the rules for fields with `GH_WRITE` access.