Regression Merge, Closure, and report_phase

Why merge is required

A single simulation seed rarely hits every bin and cross. Different seeds exercise different scenarios — random address ranges, varied burst lengths, different FSM paths. Regression merge unions bin hits across all seeds to produce the closure number for sign-off.

Legend: [COVER]

  REGRESSION MERGE WORKFLOW

  Seed 1 sim  ──► cov_seed1.ucdb  ──┐
  Seed 2 sim  ──► cov_seed2.ucdb  ──┤
  Seed 3 sim  ──► cov_seed3.ucdb  ──├──► MERGE (union of bins)
  ...                               ──┤         │
  Seed N sim  ──► cov_seedN.ucdb  ──┘         ▼
                                        merged.ucdb
                                             │
                                             ▼
                                    Map to plan rows
                                             │
                                             ▼
                                    Closure report
                                    F-040 ✓  F-042 ✓  F-055 ✗ (open)
                                             │
                                             ▼
                                    Sign-off or add tests

report_phase and merge commands

function void report_phase(uvm_phase phase);
  super.report_phase(phase);
  `uvm_info("COV", $sformatf("inst coverage %0.1f%%",
    cg.get_inst_coverage()), UVM_LOW)
  `uvm_info("COV", $sformatf("group coverage %0.1f%%",
    cg.get_coverage()), UVM_LOW)
endfunction

# Run each seed with coverage enabled
simv +UVM_TESTNAME=bus_test +ntb_random_seed=1 -coverage -covfile bus.cf
simv +UVM_TESTNAME=bus_test +ntb_random_seed=2 -coverage -covfile bus.cf

# Merge all seed databases (union of bin hits)
vcover merge -out merged.ucdb cov_seed1.ucdb cov_seed2.ucdb cov_seed3.ucdb

# Report merged coverage
vcover report merged.ucdb -detail -binratio

# Questa equivalent:
# vcover merge -stats=none -out merged.ucdb cov_seed*.ucdb

Closure sign-off workflow

Step-by-step

1. Run regression with coverage enabled — each seed writes a .ucdb or .vdb file.

2. Merge all seed databases — union operation, not average.

3. Generate merged report — map bin names to plan feature IDs.

4. Identify open bins — plan rows with zero hits after merge.

5. Add directed tests or constraint tweaks for open bins.

6. Re-run regression and re-merge until all plan rows hit.

7. Document sign-off — merged report + plan traceability matrix.

CLOSURE DECISION [COVER]

  Merged bin hit > 0 for plan row?
       │
       ├─ YES → row closed ✓
       │
       └─ NO
             ├─ Reachable with current constraints?
             │     ├─ YES → add directed test or adjust weights
             │     └─ NO  → set_goal(0) or ignore_bins (document why)
             │
             └─ illegal_bins hit in any seed?
                   └─ YES → protocol bug, not coverage gap

• Single seed rarely closes all crosses — always merge before assessing closure.

• ignore_bins or set_goal(0) for provably unreachable features — document in plan.

• option.per_instance = 1 for multi-agent — merge instances separately or together per plan.

• illegal_bins hit in merge report — investigate protocol violation, not coverage gap.

Key takeaways

• Closure = merged coverage mapped to verification plan, not single-run %.

• report_phase coverage summary is per-seed; sign-off uses merged database.

• Open bins after merge drive directed test creation — that is the closure loop.

Common pitfalls

• Declaring closure on single seed without merge — false sign-off.

• Averaging coverage % across seeds — wrong; merge unions bin hits.

• Ignoring per_instance coverage — one agent at 100%, other at 0%.