One rule written as restrict and as assume, run in simulation and in formal. Say what each engine does with it.
Hardware context: An arbiter whose formal run is restricted to single requests. Know which engine checks a restriction and which one only obeys it.
The rule "never both requests" appears as a restrict and as an assume. A simulation breaks it. Three formal runs use the restriction, one of them on an arbiter with a bug.
Write answer(q). Questions 3 and 4 ask about the simulation. Questions 5 to 7 ask for formal results: 1 for PASS or cover reached, 0 for FAIL or unreached.
Questions 0 to 2 are worked examples with assume. Questions 3 to 7 are graded only by the full check.
Signals in the Timeline
| Signal | Width | Role |
|---|---|---|
question | 4 | input of each check |
Full specification
module tb;
logic clk = 0;
always #5 clk = ~clk; // rising edges at 5, 15, ..., 65
logic req0 = 0, req1 = 0;
int a_fails = 0, a_first = 0;
only_one_r: restrict property (@(posedge clk) !(req0 && req1));
only_one_a: assume property (@(posedge clk) !(req0 && req1))
else begin a_fails++; if (a_first == 0) a_first = $time; end
initial begin
#12 req0 = 1; // t=12
#5 req1 = 1; // t=17
#15 req1 = 0; // t=32
#10 req1 = 1; // t=42
#20 req0 = 0; // t=62
#6 $display("a_fails=%0d a_first=%0d", a_fails, a_first);
$finish;
end
endmodule
module top(input clk, input req0, input req1, output reg gnt0 = 0, output reg gnt1 = 0);
always @(posedge clk) begin
gnt0 <= req0;
gnt1 <= req1;
end
`ifdef FORMAL
always @(posedge clk) restrict(!(req0 && req1)); // formal only
`endif
endmodule
Simulator note: The formal engine takes no restrict property. The runs use its immediate form restrict(expr) under ifdef FORMAL, which constrains the run in the same way. That form is the tool's, not IEEE 1800 syntax: the standard writes restrict property (@(posedge clk) !(req0 && req1)); (16.14.4).
c_gnt1: cover property (@(posedge clk) gnt1);
both: cover property (@(posedge clk) req0 && req1);
one_grant: assert property (@(posedge clk) !(gnt0 && gnt1));
| q | Return |
|---|---|
| 0 | worked example: the failures of only_one_a if req0 dropped at 52 instead of 62 |
| 1, 2 | worked examples: the two formal runs below, on the arbiter without the restriction |
| 3 | a_fails |
| 4 | a_first |
| 5, 6, 7 | the three formal runs above |
one_req: assume property (@(posedge clk) !(req0 && req1));
one_grant: assert property (@(posedge clk) !(gnt0 && gnt1));
one_req: assume property (@(posedge clk) !(req0 && req1));
single: cover property (@(posedge clk) req0 && !req1);
Simulator note: Simulation keys were measured on Verilator 5.052, formal keys with the formal engine v1. Your answer function is checked by a small bench on Icarus 12.
Examples
Example 1
An Assumption Is Checked in Simulation- Input
answer(0)
- Output
answer = 2
Each edge that breaks the rule fails the assumption once.
Example 2
An Assumption in Formal- Input
answer(1)
- Output
answer = 1
With both requests removed, two grants are impossible: PASS.
Example 3
A Cover Keeps the Assumption Honest- Input
answer(2)
- Output
answer = 1
A single request is still allowed, so the cover is reached.
Constraints and Contract
- Return one value per question from answer(q).
- Keep the function header.
Hints
Hint 1
List the sampled requests at each edge: a change between two edges shows at the later one.
Hint 2
A restriction removes input sequences, as an assumption does. It removes no behaviour of the design itself.
Hint 3
A cover on an output asks whether some allowed input sequence reaches it.
What the grader checks: Eight constant answers: three worked examples and five graded questions. They are measured on Verilator 5.052 and the formal engine v1, and checked against IEEE 1800-2023 16.14.4.
8 checks on Icarus · browser samples / server checks. Run the examples first, then check the full regression. Full checks use the server judge. Downloads contain public samples only. A passing result establishes the tested cases, not a proof for all possible inputs.
See all 3 steps and the final capstone