Rtl Model Consistency

<Purpose> Verify that all three models of the design — reference C model, SystemC TLM BFM, and RTL — produce bitexact identical outputs on a shared test vector set. Outputs: sim/consistency/consistency_report.md with per-vector comparison matrix. </Purpose>

<Use_When>

• All three models exist (refc/, bfm/, rtl/)
• Checking for drift between models after independent updates
• Pre-regression gate to ensure baseline consistency
• Debugging discrepancy between two models (need third to arbitrate) </Use_When>

<Do_Not_Use_When>

• Only two models exist (use rtl-p5s-func-verify for RTL vs ref, or rtl-p5s-perf-verify for RTL vs BFM)
• Models are known to be out of sync (fix the diverging model first)
• Only functional verification of RTL needed (use rtl-p5s-func-verify) </Do_Not_Use_When>

<Why_This_Exists> When ref model, BFM, and RTL are developed independently, silent divergence accumulates. 3-way comparison isolates which model is wrong: if ref == BFM != RTL, RTL has a bug; if ref != BFM == RTL, ref model diverged; if ref == RTL != BFM, BFM has an issue. </Why_This_Exists>

<Execution_Policy>

• Use skills/rtl-model-consistency/scripts/compare_3way.py for automated pairwise comparison
• Use as report scaffold

<Steps> 1. Select shared test vector set (use sim/consistency/test_vectors.bin or generate 50 vectors) 2. Run all three models in parallel on identical input via Bash CLI: a. ref-model-dev: `./refc/build/ref_model < sim/consistency/test_vectors.bin > sim/consistency/ref_output.bin` b. bfm-dev: `./bfm/build/bfm_smoke < sim/consistency/test_vectors.bin > sim/consistency/bfm_output.bin` c. func-verifier: simulate RTL with vectors using iverilog/cocotb, capture output to sim/consistency/rtl_output.bin - RTL ports use project convention: i_/o_ prefixes, {domain}_clk, {domain}_rst_n 3. Compare outputs pairwise via Bash CLI (diff, cmp, or Python script): ref==BFM? ref==RTL? BFM==RTL? 4. Write sim/consistency/consistency_report.md: - Per-vector comparison matrix (PASS/FAIL per pair) - First divergence for each mismatch (byte offset, value expected vs actual) - Diagnosis: which model is the likely source of error 5. Report overall consistency status to user </Steps> <Examples> <Good> All 3 models run on 50 vectors; ref==BFM on 50/50; RTL diverges on vector 23; consistency report shows RTL is the outlier; first divergence at byte 142;