has anyone had the openlane flow fail at this step...
# openlane
has anyone had the openlane flow fail at this step?
Copy code
16. Executing ABC pass (technology mapping using ABC).

16.1. Extracting gate netlist of module `\mac_cluster' to `/tmp/yosys-abc-ScuQlb/input.blif'..
Extracted 11382 gates and 11705 wires to a netlist network with 323 inputs and 384 outputs.

16.1.1. Executing ABC.
ERROR: Can't open ABC output file `/tmp/yosys-abc-ScuQlb/output.blif'.
It's happening on rc3, but on two completely different machines and environments. Same design though. Maybe it's huge and doesn't get written?
ah, nevermind. in the yosys.log i found this:
Copy code
983 ABC: Error: The network is combinational.
984 ABC: + map -p -B 0.2 -A 0.9 -M 0 
985 ABC: The cell areas are multiplied by the factor: <num_fanins> ^ (0.90).
986 ABC: The cell delays are multiplied by the factor: <num_fanins> ^ (0.20).
987 ABC: + retime -D -D 30000 
988 ABC: + buffer -N 5 -S 3000.0 
989 ABC: ** cmd error: aborting 'source /tmp/yosys-abc-uhjRHz/abc.script'
990 ABC: node 9893 has no fanout
991 ABC: Error: The command has failed.
this might have had something to do with low-ish disk space on my PC, though none of the errors suggest as much
i get the same error now on another machine with no space issues
ok. i don't think it's space related. it's definitely flakiness. sometimes it works, sometimes it doesn't.
here's the outputs of two runs with the same config, one working, one not. diff of the yosys.log. not clear why there's a divergence
OOOOH is the difference the synth strategy?
omg i think that's it
also, confusingly, i don't think SYNTH_STRATEGY was getting reset unless i did ./clean_runs.tcl
You were probably using the same run tag over and over without -overwrite?
yeah - i didn't think i needed it because it blows away all the old results in most of the subdirs that i've seen
in any case the only real takeaway here is that abc failing with such an obtuse error might be better furnished with a warning to try a different strategy