Datapath Inference: Debug Playbook

Debug playbook

Debug Playbook for Datapath Inference focuses on inferred arithmetic structure count, depth reduction, datapath area delta. The goal is to connect observed QoR movement to mechanism, ownership, and regression risk.

Synthesis debug finds the first causal drift, not the loudest downstream symptom.

Root-cause tree

ROOT-CAUSE TREE — Datapath Inference

inferred arithmetic structure count, depth reduction, datapath area delta regressed
        |
   same RTL/constraints tag?
     /              \
   no                yes
   |                  |
input drift      transform side-effect
 /    \             /         \
SDC    libs      mapping      physical estimate
diff   diff      choice       mismatch

1. Freeze baseline and regressed run manifests.

2. Verify RTL/SDC/library deltas before transform tuning.

3. Classify issue: constraints, mapping choice, physical estimate, or ECO side effect.

4. Pick one minimal reversible change.

5. Re-run full timing/area/power checks with ownership signoff.

Review memo template

STAFF SYNTHESIS REVIEW MEMO — Datapath & Retiming / Datapath Inference

1. Symptom
   - Watched metric: inferred arithmetic structure count, depth reduction, datapath area delta
   - Affected compile run: <tag/build ID>
   - Impacted path/class: <critical group / power lane / area lane>
   - Database tags: <RTL, SDC, libs, switches>

2. Mechanism hypothesis
   - Primary mechanism: Datapath-aware synthesis recognizes arithmetic patterns and maps them to optimized structures; RTL shape determines whether inference succeeds.
   - Competing hypothesis: <constraint drift, mapping choice, physical estimate mismatch>
   - Missing evidence: <report diff, dashboard trend, ownership board>

3. Proposed action
   - Minimal reversible change: <constraint patch, compile knob, ECO cell move>
   - Expected movement: <timing / area / power delta>
   - Regression risk: hold, power spike, leakage drift, formal mismatch

4. Signoff
   - Re-run artifact: datapath extraction report, inferred operator map, QoR comparison
   - Required owners: synthesis owner, RTL owner, library owner
   - Final decision: merge, rollback, or escalate

Synthesis deep dive

Datapath and retiming gains are only real if formal and timing remain clean.

Concept diagram

DATAPATH + RETIME

operator inference -> stage balancing -> formal proof -> QoR validation

Metric graph

FREQUENCY VS LATENCY

Fmax gain   ████████
latency cost ████

Reports and artifacts

• retiming report

• datapath inference log

• equivalence status

• latency impact sheet

Mini case study

Retiming met frequency target, but missing formal hooks delayed closure by two days.

Debug branches

• Check retime blockers

• Formal first for aggressive moves

• Compare with pipeline option

Senior review question

Ask: what evidence proves this QoR move is real and stable?

Key takeaways

• State exact run context (RTL, SDC, libs, switches) with every QoR claim.

• Re-run timing, area, and power regressions after each synthesis ECO.

Common pitfalls

• Comparing runs with mismatched constraints or library views.

• Timing-only fixes that violate power or area budgets.

• Skipping equivalence checks after structural changes.

Principal synthesis review addendum

Datapath-aware synthesis recognizes arithmetic patterns and maps them to optimized structures; RTL shape determines whether inference succeeds.

Metric: inferred arithmetic structure count, depth reduction, datapath area delta