Hardware interview practice
Regress cross-transaction histogram leakage
Build a focused regression for an earliest-unique-event DUT suspected of retaining histogram counts between requests. Prove the history dependence, then generalize the check into a fresh-per-request oracle.
Reviewed example
Work through one case
Input
back-to-back requests without reset: first events=[5,5]; second events=[5]Expected output
first result not_found; second result found at index0A fresh histogram makes code 5 unique in the second request; leaked counts would incorrectly retain its prior frequency of two.
What to cover
Requirements
- Issue legal requests back to back without reset and allocate a fresh 16-entry count table for every prediction.
- Return the lowest active index whose final count is one; empty and all-repeated frames return not found.
- Treat event_count above 32 as an error with all result fields zero.
- Include a minimal two-request sequence in which a repeated code becomes unique in the second request.
- Require exactly one done pulse per accepted start and identify the repeated-to-unique transition as the key follow-up coverage target.