Debugging a Half-Adder Testbench: A Journey in Four Lessons

When I set out expecting a quick smoke test for my pure combinational half-adder in SystemVerilog, I expected it to be trivial. Instead, I learned more about assertions, build scripts, and wildcard patterns than I ever thought I would. What began as a syntax error turned into a four-step masterclass in testbench design.

Lesson 1: The SVA “Watch” Myth

I was sure I could “watch” my 2-bit input vector directly:

@(ab_combined) …

But SystemVerilog properties demand explicit edges (posedge or negedge). The parser’s complaint—“expecting edge or negedge or posedge”—taught me that there’s no built‑in change‑detector syntax in SVA. You either:

• Drive checks on a clock edge, or

• Use immediate assertions in an always_comb block

Lesson 2: Combinational Tests vs. Clocked Assertions

My half-adder doesn’t need a clock—but SVA properties do. I discovered two clean patterns:

1. Immediate assertions

   always_comb begin
     assert(sum === a^b)   else $error("…");
     assert(carry === a&b) else $error("…");
   end

   No clock required; checks fire as soon as inputs or outputs change.

2. Dedicated TB clock
   Declare

   input logic clk;

   in the module header and drive it from the C++ harness. That clock only paces the assertions—your DUT remains purely combinational.

Along the way I learned that declaring logic clk; inside the module body doesn’t expose it to Verilator’s C++ driver—only ports become public members.

Lesson 3: Makefile Wildcard Woes

My Makefile used this pattern:

modules/*/tb/half_adder_tb.sv

…but my testbench lived at:

modules/arithmetic/half_adder/tb/half_adder_tb.sv

A missing * cost me countless rebuilds. Changing to:

modules/*/*/tb/half_adder_tb.sv

instantly fixed Verilator’s “Cannot find file containing module” error.

Lesson 4: Sequential vs. Combinational Dispatch

The build script classifies any TB containing the string posedge clk as sequential, pulling in sim_main.cpp that tries to toggle a clk port. Without that port, compilation fails. Removing the stray posedge clk (or properly making clk a port) let the combinational rule run with sim_main_minimal.cpp, and suddenly everything passed.

Takeaway:

• SVA assertions aren’t magical “watches”—they need edges or immediate checks.

• Combinational DUTs can use either always_comb asserts or a separate TB clock.

• Wildcard patterns in Makefiles must match directory depth exactly.

• Build rules can hinge on simple text matches, so stray keywords change behavior.

With these insights, my half-adder now simulates cleanly every time.