test-design-policy

RTL Test Case Design Policy

Purpose

Codifies systematic test case design techniques for deriving test vectors from specifications. Referenced by testbench-dev and verification orchestrators.

Equivalence Class Partitioning (ECP)

For each input signal:

• Identify valid equivalence classes from uarch spec (e.g., opcode ranges, mode encodings)
• Identify invalid equivalence classes (undefined encodings, reserved ranges)
• Select at least one representative value from each class
• Document: {input_name: [{class: "valid_normal", range: "0x00-0x0F", representative: "0x05"}, ...]}

Boundary Value Analysis (BVA)

For each integer input of width W:

• Unsigned: test 0, 1, 2^(W-1)-1, 2^(W-1), 2^W-2, 2^W-1
• Signed: test -(2^(W-1)), -(2^(W-1))+1, -1, 0, +1, 2^(W-1)-1
• For address inputs: test base, base+1, top-1, top (alignment boundaries)
• For counter/depth inputs: test 0, 1, depth-1, depth (full/empty conditions)

State Transition Testing

For each FSM defined in uarch spec:

• Extract state list and valid transitions from spec
• Build state transition matrix (source_state x event -> target_state)
• Mandatory tests: all valid transitions (0-switch coverage)
• Recommended: all 1-switch transition pairs (N-1 switch coverage)
• At least one illegal transition attempt per state (verify no state corruption)
• Reset recovery: verify return to spec-defined reset state from every reachable state

Decision Table Testing

For modules with >=3 boolean control inputs:

• Enumerate condition combinations (2^N for N conditions, or reduced set for don't-cares)
• Map each combination to expected action/output
• Generate one test per row (or per distinct action group if rows collapse)

Corner Case Categories (by module type)

Datapath modules

• All-zeros, all-ones input
• Sign extension boundary (MSB toggle)
• Overflow/underflow at arithmetic boundaries
• Back-to-back operations (no idle between)

FSM/Control modules

• Immediate re-trigger after completion
• Simultaneous conflicting requests
• Stall/backpressure at every state boundary
• Timeout/watchdog expiry

Interface modules (valid/ready)

• valid without ready (backpressure, hold >N cycles)
• ready without valid (idle consumption)
• Simultaneous assert/deassert
• Back-to-back transfers (zero-gap throughput)
• Single-beat and max-burst-length transfers

Test Vector Derivation Workflow

1. Read uarch spec + io_definition.json
2. For each input: apply ECP -> list equivalence classes
3. For each integer input: apply BVA -> list boundary values
4. For each FSM: extract state transition matrix
5. For modules with >=3 boolean controls: build decision table
6. Merge results into unified test plan with coverage mapping
7. Prioritize: boundary values + state transitions first, then ECP representatives

Anti-Patterns

• "100 random vectors" without coverage goal
• Testing only happy path (no error injection)
• Testing only reset -> single operation -> check (no sustained/stress scenarios)
• Copy-paste test vectors without traceability to spec requirement