Description
One long property does not prove at depth 2. Split it into short ones that do, without losing any of its meaning.
A whole-path property needs induction depth 4, and the budget is depth 2. Replace it with short properties, one per arc of the transaction. Each must prove at depth 2, and together they must pin down every arc on a model with free outputs.
Pinned interface
| Port | Direction | Bits |
|---|---|---|
| clk | input | 1 |
| rst | input | 1 |
| start | input | 1 |
| retry | input | 1 |
| busy | output | 1 |
| s1 | output | 1 |
| s2 | output | 1 |
| fin | output | 1 |
| code | output | 2 |
Full specification
The block runs one transaction at a time. A start or a retry in an idle cycle sends an item through two stages, s1 and s2. Then fin rises for one cycle with code equal to 2. busy is high while an item is in either stage.
The team wrote one property for the whole path: (start || retry) && !busy |-> ##3 fin && code == 2. An induction of depth 2 cannot close it. At depth 4 it proves, but the regression budget is depth 2.
Submit one property file with one short property per arc of the transaction. Every property must prove on the block at depth 2. Together they must pin down each arc on a model with free outputs, so that the whole-path property follows. Three faulty blocks must still fail.
The judge first searches the block for a counterexample to depth 10, so a false arc fails with a trace. A true property that a depth-2 induction cannot close is rejected as over budget.
module top(input clk, input rst, input start, input retry, output reg busy=0, output reg s1=0, output reg s2=0, output reg fin=0, output reg [1:0] code=0);
always @* busy = s1 || s2;
always @(posedge clk)
if (rst) begin s1 <= 0; s2 <= 0; fin <= 0; code <= 0; end
else begin
s1 <= (start || retry) && !busy;
s2 <= s1;
fin <= s2;
code <= s2 ? 2'd2 : 2'd0;
end
endmodule
| Execution rule | Meaning |
|---|---|
| Interface | Use the pinned port names, directions and bit widths shown above. Observed state and tracker outputs are not environment inputs. |
| Sampling | One positive edge of clk; binary values. Non-overlapping implication observes the next sample. Reset and initialization follow the source and contract above. |
| Submission | Submit assertion properties using the starter’s format. The checks apply to this stated model and scope. |
| How this is graded | The reference obligations, independent reachability checks and fault checks must all succeed. An unresolved proof or resource limit is inconclusive and is never a rated miss. |
Measured Counterexample
FAIL · bmc · depth 10 · smtbmc yices · 1236 ms
This measured failing run uses the published interface. Follow the sampled violation back to the preceding inputs and state. The worked solution below explains the required behaviour and the failure mechanisms.
Samples from the measured run. Free input lanes are shaded; the terminal VCD record is omitted.
Sample 2: s1=0, s2=0, fin=0, code=0. Previous sample: start=0, retry=1, busy=0, s1=0, s2=0. In the previous sample retry was high and the block was idle, so s1 must be high here. This faulty block ignores retry. The arc property reports it after one cycle. The whole-path property would need three.
Examples
One Arc
Whole path: a |-> ##2 c, through a register b.
a |=> b and b |=> c.
Each arc looks back one cycle, so a short induction closes it. The two arcs together give the two-cycle path.
Requirements
- Use the published ports and widths. The checker and the DUT variants are owned by the judge.
- Do not add assumptions unless the statement explicitly asks for one. A passing assertion without a reachable trigger is insufficient.
Hints
Hint 1
Draw the timing diagram of one transaction and name each edge between two neighbouring signals.
Hint 2
An OR on the left of an implication and an AND on the right are both places to split.
Hint 3
Use the split table of the "Simplify the property" section in F4.1.
Single clock, binary values, synchronous reset. The judge checks the pinned interface, reachability and the declared proof or bound. UNKNOWN, resource limits and timeouts remain inconclusive.
INCONCLUSIVE never counts as acceptance or a rated miss. Run previews the same checks that Submit uses.
More problems on this topic