`timescale 1ns/1ps

// =============================================================================
// Multiply a value by a constant
// Use a shift-add algorithm instead of a multiplier
// parameters:
//      -- W : data width
//      -- C : constant
// inout:
//      -- x : input of width W
//      -- y : output of widht 2W
// =============================================================================
module mul_const_shiftadd#(
    parameter integer W   = 16,
    parameter integer C   = 16'sh7fff
)(
    input  wire signed [W-1:0] x,
    output wire signed [2*W-1:0] y
);
    // Absolute value and sign of C
    localparam integer C_NEG   = (C < 0) ? 1 : 0;
    localparam integer C_ABS   = (C < 0) ? -C : C;

    // Find MSB index of C_ABS to size the network
    function integer msb_index;
        input integer v;
        integer i;
        begin
            msb_index = -1;
            for (i = 0; i < 32; i = i + 1)
                if (v >> i) msb_index = i;
        end
    endfunction
    localparam integer I_MAX = (C_ABS == 0) ? 0 : msb_index(C_ABS);

    // Partial products
    wire signed [W+I_MAX:0] part [0:I_MAX];
    genvar i;
    generate
        for (i = 0; i <= I_MAX; i = i + 1) begin : GEN_PARTS
            assign part[i] = (C_ABS[i]) ? ($signed(x) <<< i) : { (W+I_MAX+1){1'b0} };
        end
    endgenerate

    // Adder chain (simple; replace with tree if you want higher performance)
    wire signed [W+I_MAX:0] sum [0:I_MAX];
    generate
        if (I_MAX == 0) begin
            assign sum[0] = part[0];
        end else begin
            assign sum[0] = part[0];
            for (i = 1; i <= I_MAX; i = i + 1) begin : GEN_SUM
                assign sum[i] = sum[i-1] + part[i];
            end
        end
    endgenerate

    // Apply sign of C
    wire signed [W+I_MAX:0] mag = (I_MAX==0) ? part[0] : sum[I_MAX];
    wire signed [W+I_MAX:0] prod = C_NEG ? -mag : mag;

    // Stretch to fixed y width (truncate/extend as you wish outside)
    assign y = prod;

endmodule