Floating point to fixed point coversion

Question

I'm creating a hardware module that is using fixed point for its computations. But the input is floating point, and I thus wish to convert the floating point input into fixed point (Q8.8).

I've been trying to use is David Bishops library (http://vhdl.org/fphdl/) for floating points and fixed point. It works well in simulation, but not when I synthesize it. E.g. in the following code the output y is routed to ground when synthesized.

library IEEE;
use IEEE.STD_LOGIC_1164.ALL;
use IEEE.NUMERIC_STD.ALL;
library ieee_proposed;
use ieee_proposed.fixed_float_types.all;
use ieee_proposed.fixed_pkg.all;
use ieee_proposed.float_pkg.all;

entity sigmoid is
    Port (
        clk     : in std_logic;
        x       : in  float32;
        y       : out sfixed(15 downto -16)
    );
end sigmoid;

architecture Behavioral of sigmoid is

    signal size : sfixed(15 downto -16);

begin


    set_c: process(clk)
    begin 
        if rising_edge(clk) then
           y <= to_sfixed(x, size);
        end if;
    end process;

end Behavioral;

From what I've been reading on various forums and in the documentation, this code should synthesize fine. Also, fixed to float conversion using to_float(fixed) works just fine. Have I missed something? Are there other simple ways of implementing float -> fixed conversion?

I assume your synthesis tool has ieee_proposed available, with the functionality you expect? Differences in support of libraries between simulation and synthesis tools are a frequent cause of problems. What tools are you using? Do they have VHDL-2008 support? — Josh
If the tools have good VHDL-2008 supprort, you don't need the ieee_proposed library, these packages were moved into library ieee. — user_1818839

ALPF ALPF · Accepted Answer · 2015-04-30T21:09:30

Whenever I use Bishop's library for synthesis, I just include the packages in the project and declare them like the following for your case:

use work.float_pkg.all;
use work.fixed_pkg.all;

I don't use the ieee_proposed library.

Regarding the conversion, I believe the easiest way is to really use the package functions (directly or with more than one type conversion).

Floating point to fixed point coversion

1 Answers