For verification, what's the best way to introduce delay off

[py]

In a testbench, let say we have a simple loop that assigns data and valid signals to a DUT's size 2 input array for NUM_CLK cycles, ex:


for ii in 0 to NUM_CLK-1 loop
for jj in 0 to 1 loop
valid(jj) <= '1';
data(jj) <= data_source(jj)(ii);
end loop;
wait for rising_edge(clk);
end loop;

My question is, what's the easiest way to introduce a delay offset between the valid and data stream in this format? To clarify, let say data_source for data(0) and data(1) are identical, a delay offset of 2 would make the waveform look like:

data(0) 0 0 F D F E F 0 0
valid(0) 0 0 1 1 1 1 1 0 0
data(1) 0 0 0 0 F D F E F
valid(1) 0 0 0 0 1 1 1 1 1


One way that I used, and kind of work is to factor in the offset into the loop structure, ex:

for ii in 0 to NUM_CLK-1+delay loop
if jj=1 then
--only assign signal when ii = delay, otherwise idle
end if;
....


But I want to know if there is a better solution out there. Ideally, I would like to not change the loop structure. So I was thinking about piping the input signal through some kind of intermediate function call or construct that would take care of the delay, like

delay(0) := 0;
delay(1) := 2;
for ii in 0 to NUM_CLK-1 loop
for jj in 0 to 1 loop
valid(jj) <= '1';
data(jj) <= magical_delay_function( data_source(jj)(ii), delay(jj));
end loop;
wait for rising_edge(clk);
end loop;

So the function would acts as a barrel shifter. Not sure if this is the right way to go.

Any thoughts into this will be appreciated. Thanks!

 

Am Mittwoch, 20. Februar 2013 05:48:12 UTC+1 schrieb py:

In a testbench, let say we have a simple loop that assigns data and valid signals to a DUT's size 2 input array for NUM_CLK cycles, ex:





for ii in 0 to NUM_CLK-1 loop

for jj in 0 to 1 loop

valid(jj) <= '1';

data(jj) <= data_source(jj)(ii);

end loop;

wait for rising_edge(clk);

end loop;



My question is, what's the easiest way to introduce a delay offset between the valid and data stream in this format? To clarify, let say data_source for data(0) and data(1) are identical, a delay offset of 2 would make the waveform look like:



data(0) 0 0 F D F E F 0 0

valid(0) 0 0 1 1 1 1 1 0 0

data(1) 0 0 0 0 F D F E F

valid(1) 0 0 0 0 1 1 1 1 1





One way that I used, and kind of work is to factor in the offset into the loop structure, ex:



for ii in 0 to NUM_CLK-1+delay loop

if jj=1 then

--only assign signal when ii = delay, otherwise idle

end if;

...





But I want to know if there is a better solution out there. Ideally, I would like to not change the loop structure. So I was thinking about piping the input signal through some kind of intermediate function call or construct that would take care of the delay, like



delay(0) := 0;

delay(1) := 2;

for ii in 0 to NUM_CLK-1 loop

for jj in 0 to 1 loop

valid(jj) <= '1';

data(jj) <= magical_delay_function( data_source(jj)(ii), delay(jj));

end loop;

wait for rising_edge(clk);

end loop;



So the function would acts as a barrel shifter. Not sure if this is the right way to go.



Any thoughts into this will be appreciated. Thanks!

Hi,
why not using the simple after statement.
e.g. like this:

data(jj) <= transport data_source(jj)(ii) after ((NUM_CLK*CLK_PERIOD) + 1 ps) ;

The +1ps is to make sure that the data definitely changes after the clock edge.
Otherwise there may occur delta cycle issues.

Have a nice simulation
Eilert