sky130hd dont_use cells are inferred

[aviadb]

Hello.
Doing some exploration trials and I want to avoid specific cells.

when running the ORFS, I set the "DONT_USE_CELLS" variable in the config.mk file
export DONT_USE_CELLS += sky130_fd_sc_hd__fa_*

I see in the 1_2_yosys.log file that abc has received the correct -dont_use flag
abc -script /home/aviadb/asic/OpenROAD-flow-scripts/flow/scripts/abc_speed.script -liberty ./objects/sky130hd/adder_8bit_structural/base/lib/sky130_fd_sc_hd__tt_025C_1v80.lib -constr ./objects/sky130hd/adder_8bit_structural/base/abc.constr -dont_use sky130_fd_sc_hd__fa_*

However, in the final netlist I still see such cells

also checking in the synth_stat.txt report file

Any other config that I am missing?

[maliberty]

They are specifically used in the ADDER_MAP_FILE. If you want to avoid them you'll need to unset that for your design.

[aviadb]

Thanks @maliberty
Well, that is tricky.
The flow/platforms/sky130hd/cells_adders_hd.v contains both half-adders and full_adders.

Also, I assume the resizer will next can change the inferred fa cell despite included in the DONT_USE_CELLS.

Wouldn't it be more correct to overrule the ADDER_MAP_FILE with the DONT_USE_CELLS configuration?

[maliberty]

What would it mean to overrule it? It is a user given configuration.

Btw, why do you want to eliminate these cells?

[aviadb]

I mean that despite having the the fa cell explicitly listed as a dont_use it was inferred due to the mapper that contains both fa and ha cells. My intent for now is to allow ha cells but avoid fa.
The only option I see available is to edit the cells_adders_hd.v file.

So my question is, shouldn't DONT_USE_CELLS have higher priority over any other map file?

and to your question,
I want to avoid fa cells because I am exploring the usage of my own extension of std cells.

[maliberty]

No as the map file makes no sense in that case. I think the path forward is clear in any case.

[rovinski]

@aviadb you will likely have to create your own mapping as ABC does not map to multi-output gates, hence why the extract_fa pass exists.