Source

Address Translator (Static)

This is the fixed-range version of the Arithmetic Address Translator.

Translates a fixed, arbitrary, unaligned range of N locations into an aligned range (starting at zero, up to N-1) so the translated address can be used to sequentially index into other addressed components (multiplexers, address decoders, RAMs, etc...).

When memory-mapping a small memory or a number of control registers to a base address that isn't a power-of-2, the Least Significant Bits (LSBs) which vary over the address range will not address the mapped entries in order. This addressing offset scrambles the order of the control registers and of the memory locations so that the mapped order no longer matches the physical order of the actual hardware, which makes design and debugging harder.

If the address range does not start at a power-of-2 boundary, the LSBs will not count in strictly increasing order and will not start at zero: the order of the numbers the LSBs represent is rotated by the offset to the nearest power-of-2 boundary. Also, if the address range does not span a power-of-2 sized block, then not all possible LSB values will be present.

However, we can describe a translation table as a small asynchronous read-only memory which translates only the raw LSBs into consecutive LSBs to then directly address the memory or control registers. A separate Address Decoder signals when the translation is valid.

For example, take 4 locations at addresses 7 to 10, which we would like to map to a range of 0 to 3. We want address 7 to access the zeroth location, and so on. To map 4 locations, the address' two LSBs must be translated as follows:

0111 (7)  --> 0100 (4)
1000 (8)  --> 1001 (5)
1001 (9)  --> 1010 (6)
1010 (10) --> 1011 (7)

You can see the raw two LSBs are in the sequence 3,0,1,2, which we must map to 0,1,2,3. We can pre-fill a table with the right values to do that, where address 3 will contain value 0, address 0 will contain value 1, etc... Translating the LSBs like so re-aligns the addresses to the nearest power-of-2 boundary, leaving the LSBs in the order we need.

Typically, you'll need this Address Translator alongside an Address Decoder module which decodes a fixed address range: either Static, or Behavioural or Arithmetic with constant base and bound addresses.

Since we only translate the LSBs, the higher address bits are unused here, and would trigger CAD tool warnings about module inputs not driving anything. Thus, we don't select the LSBs from the whole address here, and it is up to the enclosing module to feed the Address Translator only the necessary LSBs, which is fine since the enclosing module must already know the number of LSBs to translate. The output is the translated LSBs in consecutive order.

`default_nettype none

module Address_Translator_Static
#(
    parameter       INPUT_ADDR_BASE     = 0,
    parameter       OUTPUT_ADDR_WIDTH   = 0
)
(
    input   wire    [OUTPUT_ADDR_WIDTH-1:0] input_addr,
    output  reg     [OUTPUT_ADDR_WIDTH-1:0] output_addr
);

Let's create the translation table and specify its implementation as LUT logic, otherwise it might end up as a Block RAM or asynchronous LUT RAM at random, and then the table cannot get optimized into other logic, and will likely be too slow. We make the table deep enough to hold all possible LSB values, which enables better logic optimization (no special cases for unused addresses).

    localparam OUTPUT_ADDR_DEPTH = 2**OUTPUT_ADDR_WIDTH;

    (* ramstyle = "logic" *)        // Quartus
    (* ram_style = "distributed" *) // Vivado
    reg [OUTPUT_ADDR_WIDTH-1:0] translation_table [OUTPUT_ADDR_DEPTH-1:0];

Now lets construct the translation table entry index j. When initializing j, we must zero-pad the narrow input address up to the width of a Verilog integer, else the width mismatch raises warnings at its assigment.

We also need a simple loop counter i to iterate over the number of addresses to translate, which is the entire range described by the LSBs.

    integer i;
    integer j;
    localparam INTEGER_WIDTH    = 32;
    localparam PADSIZE          = INTEGER_WIDTH - OUTPUT_ADDR_WIDTH;
    localparam PADDING          = {PADSIZE{1'b0}};

We then initialize j by slicing out the LSBs from the integer base address, and then padding it back to an integer. We then store the (strictly incrementing) corresponding LSBs from the i counter into each translation table entry indexed by j, wrapping around to the start of the table if necessary.

    initial begin
        j = {PADDING, INPUT_ADDR_BASE[OUTPUT_ADDR_WIDTH-1:0]};
        for(i = 0; i < OUTPUT_ADDR_DEPTH; i = i + 1) begin
            translation_table[j] = i[OUTPUT_ADDR_WIDTH-1:0];
            j = (j + 1) % OUTPUT_ADDR_DEPTH;
        end
    end

Finally, after all these details, the translation itself is trivial.

    always @(*) begin
        output_addr = translation_table[input_addr];
    end

endmodule

back to FPGA Design Elements

fpgacpu.ca