analogdevicesinc/hdl

[BUG] Possible bug in request_arb.v of axi_dmac core

catkira opened this issue · 2 comments

catkira commented

I am looking at this function and cant make sense of it:

function compare_id;
  input [ID_WIDTH-1:0] a;
  input [ID_WIDTH-1:0] b;
  begin
    compare_id = a[ID_WIDTH-1] == b[ID_WIDTH-1];  // compare MSB
    if (ID_WIDTH >= 2) begin
      if (a[ID_WIDTH-2] == b[ID_WIDTH-2]) begin //  compare 2nd MSB
        compare_id = 1'b1;
      end
    end
    if (ID_WIDTH >= 3) begin
      if (a[ID_WIDTH-3:0] != b[ID_WIDTH-3:0]) begin      // compare LSBs except 2 MSBs
        compare_id = 1'b1;
      end
    end
  end
endfunction

(code comments are added by me)

Why is it comparing in some cases "==" and in another case "!="?
That numbers that are compared are gray coded, maybe that helps to figure out the meaning of this function...
The code is taken from here.

@larsclausen I think You now the code best (judging by the commit history of the file)

larsclausen commented

It's standard logic for comparing gray code FIFO addresses.

To understand this maybe going through a simple example will help. Imagine a FIFO with 4 entries. The FIFO can be in 5 different states (0, 1, 2, 3, 4 elements used). So you need 3 bits. Draw the table of all 3 bit gray codes in the right order and then look at it to see the pattern when FIFO should be considered full.

catkira commented

thanks, that helped a lot!
I would maybe name the function something like "test_not_full" or so :)

Its the inverted version of this code from zipcpu

	assign	o_wfull  = (wgray[AW:AW-1] == ~wq2_rgray[AW:AW-1])
				&& (wgray[AW-2:0]==wq2_rgray[AW-2:0]);