Stage architecture

Large projects benefit from explicit checker stages. Stage boundaries make ownership, performance, and debug responsibilities clear.

[ENV][UVM][CHECK] pipeline stages

stage 0: ingest
stage 1: normalize
stage 2: correlate
stage 3: compare
stage 4: classify + report

for each stage:
  input counter
  output counter
  drop/reject reason counter

1
2
3
4
5
6
7

function txn pipeline_normalize(txn in);
  txn out = in.clone();
  out.addr = out.addr & 'hFFFF;
  out.byte_en = sanitize_be(out.byte_en);
  norm_hits++;
  return out;
endfunction

• Stage counters localize failures immediately.

• Normalization logic should be pure and deterministic.

• Stage ownership can be split across team members safely.

Key takeaways

• Pipelines scale checker complexity without losing debug clarity.

• Countered stage boundaries make regressions observable.

Common pitfalls

• Monolithic compare functions with hidden preprocessing.

• No stage-level counters, forcing wave-only debug.

Applied Patterns

Capture stage transition metadata so each mismatch includes full provenance.

[CHECK] provenance fields

txn.debug.stage0_time
txn.debug.stage1_time
txn.debug.stage2_time
txn.debug.stage3_time
txn.debug.source_stream
txn.debug.correlation_id

diagnostic policy:
  include provenance in all mismatch reports

1
2
3
4

function void stamp_stage(txn t, string stage_name);
  t.debug.last_stage = stage_name;
  t.debug.last_time  = $time;
endfunction

• Provenance data converts vague mismatches into actionable reports.

• Keep provenance lightweight to avoid throughput impact.

• Standardize stage names across scoreboards.

Integration Drilldown

Run stage-isolation tests that bypass one stage at a time and verify expected failure signatures.

[CHECK] stage isolation

test A: bypass normalize
  expect format mismatches

test B: bypass correlate
  expect missing-pair errors

test C: bypass compare
  expect suspiciously zero mismatches warning

Key takeaways

• Stage isolation proves each pipeline block is genuinely active.

Common pitfalls

• Assuming stage enable flags work without targeted tests.