Michael Lange

A second instance of this bug, which might not be immediately obvious, arises when there are no loop-carried dependencies in the time loop and the DLE decides to parallelise across...

Well, what conversation? "first touch is broken" is not longer true, and since this is now hopefully tested sufficiently, we can close this issue. There are a bunch of follow-on...

> 30 seconds to initialise an array should be a cause for immediate concern Actually, that also depends on context, since 30s can be very little or a lot. What's...

Not my point. What I mean is, what percentage of overall runtime of my application/PDE operator does the allocation take? A 10x speedup is still useless if the component I...

Ok, let's start with a simple forward-reverse; what's the cost of each operator and the cost of the sequential/parallel initialisation for a given (large) grid size/number of timesteps?

Just a quick update, after offline discussion: Due to the inherent circular dependencies between subroutine/module, the IR nodes/visitors and the scope/type classes, we will probably move scope/type and the program_units...

Hi @ecossevin , apologies for the delay. I started to look into this, and then realised where the hidden complexities come from, and had to drop it due to time...

Same. Nothing to add to @reuterbal 's suggestions; GTG from me once addressed.

@reuterbal Ok, I've rebased this over recent master and renamed the outline method, flags and pragma names. The rebase was quite tricky (esp. splitting the new `inline` sub-package), so I'd...

Quick note: This is ready for review, but currently still fails ec-phys regression locally and will need some small fixes.