fasm.cpp: return from visit if pb is open

[acomodi]

This change avoids the net visitor to write fasm outputs of pb_graph_nodes which do not have valid corresponding PBs and valid parent PBs (this could suggest that the PB is actually open).

These PBs are skipped and the visit goes on to the next PB in the net.

Signed-off-by: Alessandro Comodi [email protected]

[acomodi]

I am running all_xc7 and all_ice40 with this version on f4pga/f4pga-arch-defs#779.

[mithro]

@litghost

[litghost]

Let's merge this, but can we add some tests for this corner? Something boiled down from the test case that exposed this behavior?

Something I've been thinking about is whether it is worth walking the pb tree explicitly and building the prefixes in a forward way, rather than search the tree like this. Thoughts?

[acomodi]

@litghost well, IMO the backward approach is quite confusing, mostly from a debugging point of view. I think that exploring the pb tree from root to leaves could be a good improvement.

From a brief analysis, I thing that this change would require the implementation of visit_clb only, I guess, from which we can expand accordingly through all the correct PBs downwards (basically we should implement our version of walk_blocks).

Anyways, I will wait for the successful completion of the tests and then I'll merge, only to avoid a possible regression.

[acomodi]

@litghost I need also to verify how to scale-down the test and add it to the fasm tests, in order to easily reproduce it. I will open an issue on that to keep track of this.

BTW local builds of all_ice40 and all_xc7 passed, merging.