OpenSTA/test/write_path_spice_arc_sense....

7 lines
392 B
Plaintext
Raw Permalink Normal View History

write_path_spice: match side input values to the path arc's transitions (#475) * write_path_spice: match side input values to the path arc's transitions gatePortValues() chose side input values from the first CUDD cube of the Boolean difference d(f)/d(input), which sensitizes the gate but ignores the transition directions of the arc the path used: - For non-unate gates whose Boolean difference is a tautology (xor2, xnor2) every side variable came back don't-care, and the unknown value fell through to tie-low in writeSubcktInstVoltSrcs() -- wrong whenever the path used the when-condition requiring the side high. - For mux select arcs the cube was an arbitrary data assignment, unrelated to the output edge the path reported. Either way the written deck's gate drives the opposite direction from the reported path: the simulated chain switches with inverted polarity from that gate onward, edge-qualified arrival measurements fail, and the deck sums delays from the wrong rise/fall tables. Constrain the side input condition to the cofactor pair matching this arc -- f1 & !f0 when the input and driver edges agree (non-inverting), f0 & !f1 when they differ (inverting) -- threading the gate input RiseFall from the path stage into gatePortValues(). Also release the CUDD nodes that were previously leaked (the old code Cudd_Ref'd the Boolean difference after the generator was freed and never deref'd it). Fixes #474 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * test: write_path_spice arc-sense regression for #474, and outlined in #475 --------- Co-authored-by: Brian Degnan <bpdegnan@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 19:23:15 +02:00
* Placeholder SPICE model file for the write_path_spice_arc_sense regression.
*
* write_path_spice requires a -model_file and emits it as a ".include" line in
* the generated deck, but does not read its contents. The regression only
* diffs the written deck, so no transistor models are needed here. Supply the
* real SKY130 sky130_fd_pr models if you want to simulate the deck in ngspice.