Interpretation of dont_use

[povik]

Should OpenROAD try to remove all occurrences of the blacklisted cells from the netlist, or should the list be taken as a restriction on the currently running commands not to introduce new cells of this type into the netlist?

[maliberty]

I take it as the later. If the incoming data has such cells it isn't clear they want them removed. Even if we did, what would it mean? Leave a hole, resynthesize the logic, size swap (if possible)?

[QuantamHD]

I would also concur. Imagine the common use case of a power switch or tri-state buffer. You may not want OpenROAD to use these like normal buffers and would DONT_USE them, but you also wouldn't want OpenROAD to remove them either.

I would say in general DONT_USE probably needs context. If you're designing the PDN and it's supposed to touch power switches then the fact that it's in the DONT_USE list probably doesn't matter, but in a context-less environment like optimization I would say a cell being present in the DONT_USE list that also appears in the netlist is an implicit dont_touch