Andy Porter

4.4 neglects the need to create a mapping when creating a new script: 

There's another error in the same section:  We should do a `psyir.walk(Loop)` rather than look at children of subroutine.

> Difficulties validating the OpenACC Kernels version of the tra_adv miniapp on NVIDIA GPU - does this work on Glados? For the record, we had to apply the `-Mnofma` flag...

The one remaining piece is now to see whether we can run part 4 of the tutorial on GPU as part of the integration tests.

In order to use 'Unified Memory', the OpenMP standard mandates that a source file must contain `!$omp requires unified_shared_memory`. It turns out that the only compiler that doesn't require this...

Since most of the implementation is given in the issue description, this could be a good, self-contained issue for a new person to tackle.

Note, if the body of the loop doesn't contain any CodeBlocks then we could go one better and replace all References to the loop variable with the value supplied to...

> I believe we recently discussed to generalising the replace_induction_variables_trans? That would be another use case for a more generalised approach (the replace_induction... transformation just needs to restrict the constant...

#2707 is on now. Can this be closed @hiker?

Oh yes, I'd forgotten we'd been here before! I suspect that the `type` of r2d_field `extends` another type and we don't support that although I think @JulienRemy was looking at...