Unrelated cells sharing the same name can cause conflicting drivers in an amended partition (triggerd by the default ghdl-yosys-plugin output)

[fayalalebrun]

I am trying to use eqy in conjuction with ghdl-yosys-plugin. However, I get the following error:

ERROR: conflicting matches for gold bit $auto$ghdl.cc:806:import_module$28: $auto$ghdl.cc:806:import_module$100 vs $auto$ghdl.cc:806:import_module$42

There are many such auto-generated names in the netlist, so the next thing I tried was to add the following:

[match *]
nodefault
gold-match /^((?!import_module).)*$/

However, this has no effect on the error.

I am not sure if preventing those signals from being matched would fix the issue, or what the implications would be.

Therefore, I am wondering whether it is possible to use eqy with ghdl-yosys-plugin? If not, is this a limitation of eqy, or of ghdl-yosys-plugin?

[jix]

Can you share the complete .eqy file and design you are using?

[fayalalebrun]

I was using it for a cache controller which is rather complex. However, I transcribed one of the examples to vhdl and it seems to fail with a similar error:
https://gist.github.com/fayalalebrun/055de56382081b62b47fe2639ae8e000

[jix]

Just for reference, since you mentioned "a similar error" and I'm not sure I'm getting the same error, with EQY, yosys and the ghdl-plugin from the latest nightly oss-cad-suite build I get:

EQY 14:26:23 [ex_amend] partition: Warning: multiple conflicting drivers for gate.top.Y.\__cp_X [1]:
EQY 14:26:23 [ex_amend] partition: module input __cp_X[1]
EQY 14:26:23 [ex_amend] partition: module input __cp_auto$ghdl.cc:806:import_module$1[0]
EQY 14:26:23 [ex_amend] partition: ERROR: Found 1 problems in 'check -assert'.

While looking into the cause, I noticed that the ghdl plugin uses public RTLIL names for auto-generated sequential cell names. This can result in EQY trying to align the design along completely unrelated cells that happen to receive the same sequential name. Ideally the ghdl plugin would generate private RTLIL names, but a workaround is to manually hide the public names for those cells, which can be done by running rename -hide */[0123456789]* after the ghdl plugin command. This hides all names starting with a digit.

With that done, the example doesn't reproduce the issue and this might also avoid the issue in your larger use case. It's still an actual bug that this can confuse the partitioning into generating multiple drivers for the same signal, but I haven't tracked down the exact reason for that happening yet.

[jix]

I reduced it further to the following eqy file directly containing two smaller netlists that reproduce this without needing to use the ghdl plugin: https://gist.github.com/jix/ff72778c03cd6769a277b3c98658441e

gold

gate

The cells named "13" and "2" have the same name although they were unrelated even in the ghdl example.

@clairexen Any idea how that gets the partitioning code to produce conflicting drivers?

[clairexen]

First attempt at a fix for this issue:

• Use https://github.com/YosysHQ/yosys/tree/claire/bufnorm
• Replace insbuf -chain in eqy with bufnorm -bits -chain