verdi-regression-diff
/verdi-regression-diff
Porównuje ten sam sygnał między dwoma przebiegami lub buildami w nWave i zapisuje różnicę jako obraz.
- Autor
- fayememe ↗
- Repozytorium
- verdimcp ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 29.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 29.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Porównuje ten sam sygnał między dwoma przebiegami lub buildami w nWave i zapisuje różnicę jako obraz.
Oryginalny opis (EN)
Use when the same signal behaved differently between two runs or two builds. Compares signals against another FSDB in nWave and captures the diff as an image; documents the verified limit that Verdi gives no textual comparison report.
Jak działa
- Scope
- Flow
- Limits (verified)
- Reporting
- Rules
Podgląd SKILL.md
---
name: verdi-regression-diff
description: Use when the same signal behaved differently between two runs or two builds. Compares signals against another FSDB in nWave and captures the diff as an image; documents the verified limit that Verdi gives no textual comparison report.
---
# Regression diff (two runs)
## Scope
**In:** "this worked in the previous run", "the transfer is one cycle later now", "did my fix change anything".
**Out:** formal equivalence checking, and any diff that needs a second pane's textual report (not
scriptable — see limits).
## Flow
1. Confirm the **two artifacts**: current session FSDB (`verdi_session_info` → `fsdb`) and the reference
FSDB path. Both must exist and be readable; the reference can be a copy of an older run.
2. **Do this before any cone capture** in the session (ordering hazard below).
3. `verdi_waveform_compare {signals, against_fsdb: [ref], capture: true}`.
- Signals are normalized to `/`-prefixed GUI names; if they are not yet in the window the tool adds them
once and reports `added_signals: true`.
- Returns `{window_id, signals, against_fsdb, image:{path,width,height,bytes,sha256}, text_available:false, warning}`.
4. Read the screenshot and state the difference shape per signal:
| Visual difference | Usual meaning |
|---|---|
| Same edges, offset by N cycles | pipeline/latency change or a new stall |
| Extra / missing transition | state machine path changed |
| Value differs on the same edge | data path or constant changed |
| Same picture, identical sha256 | no visible difference in this ROI |
5. If the diff must be quantified (e.g. "late by 2 cycles"), get the numbers with
`verdi_transitions_get` on each run separately and compare the times yourself. The compare tool gives
the picture; the evidence for "2 cycles" is the two transition lists.
## Limits (verified)
- `wvCompareResult -file ...` writes
`"You must open another waveform pane before running this option!"` — a second pane cannot be scripted,
so **no textual diff file exists**. The screenshot is the artifact. Never quote a diff report.
- Comparing a signal that exists only in one FSDB fails with `not_supported`; check the name in both runs
with `verdi_signals_list` first.
- **Ordering hazard:** comparing in a session that already opened nSchema windows killed Verdi W-2024.09
once. The tool returns `warning` when schematic windows are present; if so, prefer a fresh session.
## Reporting
```
run A: demo/sim/sync_fifo.fsdb (VERIFIED)
run B: /tmp/old_run/sync_fifo.fsdb (VERIFIED)
signal: /tb_sync_fifo/rd_data[7:0]
diff: first valid beat at 40 ns (A) vs 44 ns (B) (VERIFIED from two transitions_get calls)
visual: compare-<sid>-<ts>.png sha256 afb02699… (VERIFIED image, INFERRED about logic)
verdict: 1-cycle latency regression on the read path (INFERRED)
```
## Rules
- Never claim "no difference" from a screenshot alone; say "no visible difference in this ROI" and give
the ROI (`start`/`end`/zoom) and the sha256.
- Never claim a cycle count from pixels. Use transitions.
- Always name both FSDB paths — a diff without provenance is useless a week later.
Wersje
W bazie nie ma innych skilli o nazwie verdi-regression-diff.
Migawka bazy z 13.11.2026 · wersja paczki 29d65380e968 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100verdi
Kieruje sprawy z Synopsys Verdi, nWave i FSDB do właściwych narzędzi i przypomina zasady bezpiecznej pracy. - 100verdi-performance
Podpowiada, jak przyspieszyć zapytania i renderowanie w Verdi przy dużych plikach FSDB, wybierając GUI lub NPI. - 100x-npi
Pomaga pisać skrypty Pythona z użyciem pynpi do analizy plików FSDB, protokołów magistral i baz pokrycia.