DO NOT MERGE: Experiment for TaskletFusion in Intra-Map Optimization#2454

Draft

philip-paul-mueller wants to merge 20 commits intoGridTools:mainfrom

philip-paul-mueller:better_in_map_dataflow_optimization_order

Commits on Jan 22, 2026

Let's try this order.
philip-paul-mueller
committed
Maybe this is better.
philip-paul-mueller
committed
Maybe the simplify call was unneeded.
philip-paul-mueller
committed
This is nearer at the empirical version, let's try it.
philip-paul-mueller
committed
This is a bit nicer than the previous version, i.e. it has an explanation.
philip-paul-mueller
committed
An experiment, let's see what happens.
philip-paul-mueller
committed
Revert "An experiment, let's see what happens."
philip-paul-mueller
committed
New experiment with a different inliner.
philip-paul-mueller
committed
Revert "New experiment with a different inliner."
philip-paul-mueller
committed
Now let's check what happens if we at the end of the fast wersion run Tasklet fusion again.
philip-paul-mueller
committed
Revert "Now let's check what happens if we at the end of the fast wersion run Tasklet fusion again."
philip-paul-mueller
committed
Essentially the same experiment as in `c40417421324dc274d3ce5fa01db21fa3b1b754e`.
philip-paul-mueller
committed
This version should be fast, but as soon as we know we will enabled the currently disacbled code and see if it is slow.
philip-paul-mueller
committed
Now let's activate the second FT pass.
philip-paul-mueller
committed
New experiment with running it twice.
philip-paul-mueller
committed

Commits on Jan 23, 2026