Blocking TLM FIFO

Interview prompt

When use uvm_tlm_fifo instead of direct analysis connect?

WHITEBOARD CHAIN

1. DECLARE    interfaces / types / ports you need
2. SKELETON   class extends + utils macro + key methods
3. MECHANISM  fill one critical method while narrating
4. PITFALL    name one bug juniors make on this pattern
5. TEST       how you would smoke-test the component

Reference sketch (≤40 lines)

uvm_tlm_fifo #(txn) fifo;
// producer task
fifo.put(t);
// consumer task
fifo.get(got);

Use a FIFO when producer and consumer run in different tasks/phases and rates differ — predictor bursts while scoreboard compares slowly. Analysis push is fire-and-forget; FIFO adds blocking backpressure.

Mechanism to narrate

• blocking put waits if fifo full (bounded depth)

• blocking get waits if fifo empty

• Decouples threads without losing transactions

Smoke test (5 minutes)

• Producer faster than consumer — consumer still receives all items.

• Set depth=1 — producer stalls until consumer gets.

Common pitfalls

• Unbounded fifo on stress — memory growth.

• Using fifo for broadcast — use analysis fanout instead.