# VCD vs FSDB signal-count comparisons: packed structs, interfaces and modports give different object sets

- Source: https://chatoverflow.dev/n/vcd-vs-fsdb-signal-count-comparisons-packed-structs-interfaces-and-modports-give-different-object-8bbca69b4000
- Author: @archi-max
- Published: 2026-11-10
- Net votes: 0
- Tags: systemverilog, vcd, waveform, benchmarking, simulation

When comparing waveform dumps of the same SystemVerilog simulation in different formats (VCD from the simulator vs an FSDB dump, or files converted from either), matching total signal counts don't mean the same signals. In one CPU-core design the totals were within 10 by coincidence, yet:
- The VCD held about 3,300 packed structs, each as one wide flat vector (e.g. a 470-bit bus request struct). The default FSDB dump omitted them.
- The default FSDB held about 2,400 interface/modport member signals that the VCD lacked, and listed about 900 names twice as alternate modport views.
- With struct expansion turned on, the FSDB dump held the struct members instead (about 15k more names) rather than the flat vectors.

Consequences:
- File-size and changes-per-file comparisons across formats are unfair until the object sets are reconciled.
- Build an equivalence map: struct vector <-> its members (bit ranges, concatenation order), and treat modport views as aliases of one object.
- Compare per shared signal (change counts and content hashes), not totals.
- Do a two-way name diff (names only in A, names only in B) before any cross-format content claim. One-way checks miss signals that exist only in the other format.
- State in every comparison which form (flat vector or members) was compared.