Re: Quartus Prime complete lack of real data type support?

Altera_Forum · ‎08-09-2017

Hi,

I have a function (basicaly implements a LUT) something like the following:

function integer gen_lut;
    input integer addr;
    
    localparam SOME_CONSTANT = 418.39; // Gets calculated at runtime.
    localparam D0 = 0.785398163 * SOME_CONSTANT;
    localparam D1 = 0.463647609 * SOME_CONSTANT;
    ...
    begin
        case (addr)
            0: gen_lut = D0;
            1: gen_lut = D1;
            ...
        endcase
    end
endfunction

The thing is I need to convert d0, d1, ... from real type to integer at elaboration time to use it later in design.

I've tried using system function $rtoi, $floor but Quartus complains: error (10174): verilog hdl unsupported feature error at top.v(54): system function "$floor" is not supported for synthesis

Then I've tried implementing a constant function, since I already have log2 function implemented and it works, floor that would take real number and output a integer like so:

function integer floor;    
    input real x;
    
    integer int_part;
    
    begin
        int_part = x;
        
        if (int_part > x) begin floor = int_part - 1; end
        else begin floor = int_part; end
    end
endfunction

And the tool complains again: error (10172): verilog hdl unsupported feature error at const_functions.sv(21): real variable data type values are not supported.

IMHO, this is just ridiculous. I am not even trying to synthesise real values I just want to convert them to integer at elaboration time. Does Quartus even have any kind of support for real data types? I mean even Xillinx ISE and Icarus Verilog have these features supported and have no problem synthesizing.

Is there a workaround this problem? I would still like to keep real numbers in the code for easier parametrization.

P.S.

I saw a similar thread that was posted in 2008, so these features havent been implemented for almost 10 years??? Come on, its not like the whole IDE needs to be re-implemented...

Altera_Forum · ‎08-09-2017

I've found one possible workaround, how to round real data type down to nearest integer:

1) Create a constant function that will convert real to nearest integer (up or down):

function integer rtoi;
    input integer x;
    
    
    begin
        rtoi = x;
    end
endfunction

2) Create a macro that will round real down to integer:

`define FLOOR(x) ((rtoi(x) > x) ? rtoi(x) - 1 : rtoi(x))

3) Use the macro:

localparam real R_TEST = 0.785398163 * SOME_CONSTANT;
integer floor_int = `FLOOR(R_TEST);

Best regards!