Are you saying that for every occurrence of “a” then variable b occurs after 50 clock cycles?
No, we are only concerned about the last occurrence of “a”.I mean after last occurrence of “a”
Or, are you saying that if I have something like:
a[*3] ##1 !a then b after the last b? Not this
Pehaps something like: $rose(a) ##1 !a[->1] |-> ##50 b;
but something like
a[*3] ##1 !a then b after the last a? this but here a is getting repeated n number of times at any interval as it is randomised so
Pehaps something like: $rose(a) ##1 !a[->1] |-> ##50 b;// donot know this will catch the last occurance of a
$rose(a) ##1 !a[->1]
Catches the last occurrence of a before another rose of a.
No assertion will predict another repeats of a.
You may need some support logic or other info to limit what you are searching for. Again, you have to define what “Last” means.
Ben
In reply to ben@SystemVerilog.us:
Hi Ben ,
I mean to say
Are you saying that for every occurrence of “a” then variable b occurs after 50 clock cycles?
No, we are only concerned about the last occurrence of “a”.I mean after last occurrence of “a”
Or, are you saying that if I have something like:
a[*3] ##1 !a then b after the last b? Not this
Pehaps something like: $rose(a) ##1 !a[->1] |-> ##50 b;
but something like
a[*3] ##1 !a then b after the last a? this but here a is getting repeated n number of times at any interval as it is randomised so
Pehaps something like: $rose(a) ##1 !a[->1] |-> ##50 b;// donot know this will catch the last occurance of a
a will be asserted within 50 cycles. if a is not asserted within 50 cycles , then b should be asserted. Should we consider that a will be the last occurrence?
always@(posedge clk) begin
if(a) count <= 0 ;
else count <= count + 1 ;
end
assert property(@ posedge clk count >= 50 |-> b)
is there case there in signal a can be asserted after 50 clock cycles?
In reply to kddholak:
yes there can be case where a can be asserted after 50 clock edge
How would design know its last a, and b should be asserted after 50 clock cycles ? here its not predictable behaviour.
if the design can predict the last a , and asserts b after 50 clock cycles. there should be some deterministic behavior