Sergi Siso

The initial implementation of InlineTransformation is now on master, but some remaining TODOs need to be fixed to close this issue.

My comments regarding the initial "towards" PR of this issue: - Currently resulting datatype of accesses to UknownTypes can become DeferredType, my worry with deferredtype is that it could imply...

> Is there a way to go from the CodedKern object to the actually kernel itself, or are the kernels just treated as black boxes? Yes.. the get_kernel_schedule() method >...

It also looks like you have a very thin transformation and all the logic is in the node. Could the logic be brought to the transformation. I found it is...

> Ok, i thought i'd tried this and it not worked but I'll test again! You may need the "-d" in the psyclone command to point to the directory where...

Ah you are right. This is a good reason!

> Without proper dependency analysis, I don't think we can do much better with the placement of the update directive. OpenACC compute constructs and subroutine calls should be treated differently...

> I've also just realised that I'm the original author of this PR so GitHub isn't going to let me approve it. I suggest responding to all of my current...

@arporter This is ready for review. It is a relatively small PR to close #1815 . I could also use the PR to move the OMPTargetTrans to src/psyclone/psyir/transformations if you...

@arporter I reduced the scope of this PR to just remove codeblocks (original issue) and move the transformation locations. The UBOUND and LBOUND issues are a compiler bug that only...