Parallel in, Parallel out shift register

Discussion in 'VHDL' started by Vivek Menon, Jun 8, 2011.

  1. Vivek Menon

    Vivek Menon Guest

    I am trying to synthesize and simulate a parallel shift register that keeps shifting the input data as long as the enable pin is active.

    entity shift_out is
    Port (
    --Inputs
    clk : in std_logic;
    en : in std_logic;
    rst : in std_logic;
    in1 : in std_logic_vector(31 downto 0);

    -- Outputs
    shift_val : out std_logic_vector(31 downto 0)
    );
    end entity shift_out;

    architecture arch of shift_out is

    signal shift_t1 : std_logic_vector(31 downto 0) := (others => '0');
    ....
    process (clk, rst, in1, en) is
    begin
    if rst = '1' then
    shift_t1 <= (others=>'0');
    shift_val <= (others=>'0');
    elsif rising_edge(clk) then
    if (en = '1') then
    shift_t1 <= shift_t1 ror x"10";
    shift_t1 <= in1;
    end if ;
    end if;
    end process;

    shift_val <= shift_t1;

    end arch;

    I am confused with the ror approach, I have tried array slicing and that did not simulate as well.
    ANy suggestions??
     
    Vivek Menon, Jun 8, 2011
    #1
    1. Advertisements

  2. Vivek Menon

    backhus Guest

    Hi,
    What error messages are you getting from the tools?
    Have you checked wether the ror function works for the data type you
    are using?

    Another way to shift/rotate vectors goes like this:

    shift_t1 <= shift_t1(shift_t1'length-2 downto 0) &
    shift_t1(shift_t1'length-1); -- simple rotate by one, missing an
    input, but you can overwrite the LSB

    Have a nice synthesis
    Eilert
     
    backhus, Jun 8, 2011
    #2
    1. Advertisements

  3. Vivek Menon

    backhus Guest

    Hi,
    What error messages are you getting from the tools?
    Have you checked wether the ror function works for the data type you
    are using?

    Another way to shift/rotate vectors goes like this:

    shift_t1 <= shift_t1(shift_t1'length-2 downto 0) &
    shift_t1(shift_t1'length-1); -- simple rotate by one, missing an
    input, but you can overwrite the LSB

    Also there's some big flaw in your approach.
    You have no signal to distinguish between load and shift operation.
    Enable is working for both actions and so you are only always loading
    when enable is active and do not see any effect of the ror function.

    Do something like this:
    Define a port
    load : in std_logic;
    and in your enable branch:

    if load = '1' then
    shift_t1 <= in1;
    else -- rotate
    shift_t1 <= shift_t1(shift_t1'length-2 downto 0) &
    shift_t1(shift_t1'length-1);
    end if;

    Have a nice synthesis
    Eilert
     
    backhus, Jun 8, 2011
    #3
  4. Vivek Menon

    Andy Guest

    Take shift_val out of the clocked process. It should not be assigned
    in both the process and the concurrent assignment statement.

    Also, remove everything but clk and rst from the process sensitivity
    list.

    Andy
     
    Andy, Jun 8, 2011
    #4
  5. Vivek Menon

    KJ Guest

    Have you simulated your design and does it work as intended? If not,
    then get that working before synthesizing. If so, then I'm surprised
    because...

    The second assignment to shift_t1 will override the first assignment.
    The net of all this is that the assignment with the 'ror' won't do
    anything.
    Until you get the simulation working properly, it will likely not make
    much sense for you to try to synthesize. You have a basic issue with
    your design in that you don't have a method for loading and for
    shifting the data. Typically one would have a 'load' and a 'shift'
    input but it might be that you intend to load any time you're not
    shifting in which case you should have written

    if (en = '1') then
    shift_t1 <= shift_t1 ror x"10";
    else -- KJ added
    shift_t1 <= in1;
    ....
    Also, the 'ror' function is defined for unsigned types, not
    std_logic_vector.
    function "ror" (ARG: UNSIGNED; COUNT: INTEGER) return UNSIGNED;

    I would suggest changing the data type of shift_t1
    shift_t1 : unsigned(31 downto 0);

    Then convert to/from std_logic_vectors on the various assignments.

    Also see Andy's suggestions. To have the compiler catch the first
    error that Andy pointed out, consider using std_ulogic rather than
    std_logic.

    Kevin Jennings
     
    KJ, Jun 8, 2011
    #5
  6. Me too. What ror operator? What packages are you using? The only ror
    operator that I know of is from ieee.numeric_std:

    ------------------------------------------------------------------------------
    -- Note : Function S.15 is not compatible with VHDL 1076-1987. Comment
    -- out the function (declaration and body) for VHDL 1076-1987
    -- compatibility.
    ------------------------------------------------------------------------------
    -- Id: S.15
    function "ror" (ARG: UNSIGNED; COUNT: INTEGER) return UNSIGNED;
    -- Result subtype: UNSIGNED(ARG'LENGTH-1 downto 0)
    -- Result: ROTATE_RIGHT(ARG, COUNT)

    ------------------------------------------------------------------------------
    -- Note : Function S.16 is not compatible with VHDL 1076-1987. Comment
    -- out the function (declaration and body) for VHDL 1076-1987
    -- compatibility.
    ------------------------------------------------------------------------------
    -- Id: S.16
    function "ror" (ARG: SIGNED; COUNT: INTEGER) return SIGNED;
    -- Result subtype: SIGNED(ARG'LENGTH-1 downto 0)
    -- Result: ROTATE_RIGHT(ARG, COUNT)

    Posting error messages would be nice. Or observed/expected behaviour.

    I also suspect a missing else:

    if (en = '1') then
    shift_t1 <= shift_t1 ror x"10";
    else --< !!
    shift_t1 <= in1;
    end if;

    And for the shift operation I would suggest something like:

    shift_t1 <= shift_t1(shift_t1'low) &
    shift_t1(shift_t1'high downto shift_t1'low+1);

    Which is the same as

    shift_t1 <= shift_t1(0) & shift_t1(31 downto 1);

    without the hard coded numbers.

    For the rest: remove 'in1' and 'en' from the sensitivity list. They are not
    needed.

    Oh, and resetting shift_val is not needed: it is not a flip-flop, shift_t1
    is. And why using shift_t1 at all? Why not just shift_val? There is no need
    for the extra signal.
     
    Paul Uiterlinden, Jun 8, 2011
    #6
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.