Xilinx test bench and user group

[Joseph A. Zammit]

Is there any other way to specify test benches in Xilinx WebPack other
than doing it graphicaly or using the pattern generator?

For example i am testing a full adder through all cominbations and
that requires quite a large testbench. Is there any other way to do
it?

Is there any user group/forum that is on Xilinx Coolrunner devices
only?

Thanks for your help

Joseph Zammit

Malta

 

[valentin tihomirov]

Typically, special architecture is used that instantiates your design,
generates stimuli and analyzes responses.

architecture TB
constant WIDTH : integer := A_WIDTH + B_WIDTH;
signal CNT_INT: integer range 0 to 2**WIDTH - 1;
signal CNT_VEC: std_logic_vector(WIDTH- 1 downto 0);
alias A is CNT_VEC(A_WIDTH-1 downto 0);
alias B is CNT_VEC(WIDTH-1 downto A_WIDTH);
signal SUM: std_logic_vector(O_WIDTH-1 downto 0);
begin

CNT_VEC <= conv_std_logic_vector(CNT, WIDTH);

UUT: entity ADDER(RTL)
port map (A=>A, B=> B, SUM => SUM);

process
begin
for I in CNT_INT'range loop
null;
end loop;
wait; -- wait forever
end;


CLOCK : process
begin
CLK <= '0'; wait for PERIOD/2;
CLK <= '1'; wait for PERIOD/2;
end process;

end architecture TB;

 

[Ralf Hildebrandt]

Is there any other way to specify test benches in Xilinx WebPack other
than doing it graphicaly or using the pattern generator?

A Testbench is nothing more than a VHDL component, than instatiates you
design als sub-component. The testbench generates some stimuli and may
react on the output of the sub-component.

You are free to use the complete language, because the testbench hasn't
to be synthesized.

A graphical tool is nothing more than a GUI, that outputs such a plain
text file.

For example i am testing a full adder through all cominbations and
that requires quite a large testbench.

If you want to test your design, you should think about functional tests
and special test in "corners" (e.g. addition with overflow). Otherwise
the number of test vectors becomes very huge. The probability of a fault
in a fulladder should be _very_ low. Much higher is the probability,
that the whole adder does not work, because of a different error.

Furthermore you should think about fault injection: Insert a
stuck-at-zero or stuck-at-one fault at every point, you think an error
could occur. Make a test vector, that stimulates the fault and makes the
faul visible (not masked). Do it for all faults and you will get a much
lower amout of nessecary test vectors.
AFAIK some synthesis tools support test vector generation.


Ralf

 

[TDB]

Yes, just 'add new source' and pick 'test bench'