| 1 |
3 |
dgisselq |
///////////////////////////////////////////////////////////////////////////////
|
| 2 |
|
|
//
|
| 3 |
|
|
// Filename: div.v
|
| 4 |
|
|
//
|
| 5 |
|
|
// Project: Zip CPU -- a small, lightweight, RISC CPU soft core
|
| 6 |
|
|
//
|
| 7 |
|
|
// Purpose: Provide an Integer divide capability to the Zip CPU.
|
| 8 |
|
|
//
|
| 9 |
|
|
//
|
| 10 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
| 11 |
|
|
// Gisselquist Technology, LLC
|
| 12 |
|
|
//
|
| 13 |
|
|
///////////////////////////////////////////////////////////////////////////////
|
| 14 |
|
|
//
|
| 15 |
|
|
// Copyright (C) 2015-2016, Gisselquist Technology, LLC
|
| 16 |
|
|
//
|
| 17 |
|
|
// This program is free software (firmware): you can redistribute it and/or
|
| 18 |
|
|
// modify it under the terms of the GNU General Public License as published
|
| 19 |
|
|
// by the Free Software Foundation, either version 3 of the License, or (at
|
| 20 |
|
|
// your option) any later version.
|
| 21 |
|
|
//
|
| 22 |
|
|
// This program is distributed in the hope that it will be useful, but WITHOUT
|
| 23 |
|
|
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
|
| 24 |
|
|
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
| 25 |
|
|
// for more details.
|
| 26 |
|
|
//
|
| 27 |
|
|
// License: GPL, v3, as defined and found on www.gnu.org,
|
| 28 |
|
|
// http://www.gnu.org/licenses/gpl.html
|
| 29 |
|
|
//
|
| 30 |
|
|
//
|
| 31 |
|
|
///////////////////////////////////////////////////////////////////////////////
|
| 32 |
|
|
//
|
| 33 |
|
|
// `include "cpudefs.v"
|
| 34 |
|
|
//
|
| 35 |
|
|
module div(i_clk, i_rst, i_wr, i_signed, i_numerator, i_denominator,
|
| 36 |
|
|
o_busy, o_valid, o_err, o_quotient, o_flags);
|
| 37 |
25 |
dgisselq |
parameter BW=32, LGBW = 5;
|
| 38 |
|
|
input i_clk, i_rst;
|
| 39 |
3 |
dgisselq |
// Input parameters
|
| 40 |
|
|
input i_wr, i_signed;
|
| 41 |
|
|
input [(BW-1):0] i_numerator, i_denominator;
|
| 42 |
|
|
// Output parameters
|
| 43 |
|
|
output reg o_busy, o_valid, o_err;
|
| 44 |
|
|
output reg [(BW-1):0] o_quotient;
|
| 45 |
|
|
output wire [3:0] o_flags;
|
| 46 |
|
|
|
| 47 |
|
|
// r_busy is an internal busy register. It will clear one clock
|
| 48 |
|
|
// before we are valid, so it can't be o_busy ...
|
| 49 |
|
|
//
|
| 50 |
|
|
reg r_busy;
|
| 51 |
|
|
reg [(2*BW-2):0] r_divisor;
|
| 52 |
|
|
reg [(BW-1):0] r_dividend;
|
| 53 |
|
|
wire [(BW):0] diff; // , xdiff[(BW-1):0];
|
| 54 |
|
|
assign diff = r_dividend - r_divisor[(BW-1):0];
|
| 55 |
|
|
// assign xdiff= r_dividend - { 1'b0, r_divisor[(BW-1):1] };
|
| 56 |
|
|
|
| 57 |
|
|
reg r_sign, pre_sign, r_z, r_c, last_bit;
|
| 58 |
|
|
reg [(LGBW-1):0] r_bit;
|
| 59 |
|
|
|
| 60 |
|
|
reg zero_divisor;
|
| 61 |
|
|
initial zero_divisor = 1'b0;
|
| 62 |
|
|
always @(posedge i_clk)
|
| 63 |
|
|
zero_divisor <= (r_divisor == 0)&&(r_busy);
|
| 64 |
|
|
|
| 65 |
|
|
initial r_busy = 1'b0;
|
| 66 |
|
|
always @(posedge i_clk)
|
| 67 |
|
|
if (i_rst)
|
| 68 |
|
|
r_busy <= 1'b0;
|
| 69 |
|
|
else if (i_wr)
|
| 70 |
|
|
r_busy <= 1'b1;
|
| 71 |
|
|
else if ((last_bit)||(zero_divisor))
|
| 72 |
|
|
r_busy <= 1'b0;
|
| 73 |
|
|
|
| 74 |
|
|
initial o_busy = 1'b0;
|
| 75 |
|
|
always @(posedge i_clk)
|
| 76 |
|
|
if (i_rst)
|
| 77 |
|
|
o_busy <= 1'b0;
|
| 78 |
|
|
else if (i_wr)
|
| 79 |
|
|
o_busy <= 1'b1;
|
| 80 |
|
|
else if (((last_bit)&&(~r_sign))||(zero_divisor))
|
| 81 |
|
|
o_busy <= 1'b0;
|
| 82 |
|
|
else if (~r_busy)
|
| 83 |
|
|
o_busy <= 1'b0;
|
| 84 |
|
|
|
| 85 |
|
|
always @(posedge i_clk)
|
| 86 |
|
|
if ((i_rst)||(i_wr))
|
| 87 |
|
|
o_valid <= 1'b0;
|
| 88 |
|
|
else if (r_busy)
|
| 89 |
|
|
begin
|
| 90 |
|
|
if ((last_bit)||(zero_divisor))
|
| 91 |
|
|
o_valid <= (zero_divisor)||(~r_sign);
|
| 92 |
|
|
end else if (r_sign)
|
| 93 |
|
|
begin
|
| 94 |
|
|
o_valid <= (~zero_divisor); // 1'b1;
|
| 95 |
|
|
end else
|
| 96 |
|
|
o_valid <= 1'b0;
|
| 97 |
|
|
|
| 98 |
|
|
initial o_err = 1'b0;
|
| 99 |
|
|
always @(posedge i_clk)
|
| 100 |
|
|
if((i_rst)||(o_valid))
|
| 101 |
|
|
o_err <= 1'b0;
|
| 102 |
|
|
else if (((r_busy)||(r_sign))&&(zero_divisor))
|
| 103 |
|
|
o_err <= 1'b1;
|
| 104 |
|
|
else
|
| 105 |
|
|
o_err <= 1'b0;
|
| 106 |
|
|
|
| 107 |
|
|
initial last_bit = 1'b0;
|
| 108 |
|
|
always @(posedge i_clk)
|
| 109 |
|
|
if ((i_wr)||(pre_sign)||(i_rst))
|
| 110 |
|
|
last_bit <= 1'b0;
|
| 111 |
|
|
else if (r_busy)
|
| 112 |
|
|
last_bit <= (r_bit == {{(LGBW-1){1'b0}},1'b1});
|
| 113 |
|
|
|
| 114 |
|
|
always @(posedge i_clk)
|
| 115 |
|
|
// if (i_rst) r_busy <= 1'b0;
|
| 116 |
|
|
// else
|
| 117 |
|
|
if (i_wr)
|
| 118 |
|
|
begin
|
| 119 |
|
|
//
|
| 120 |
|
|
// Set our values upon an initial command. Here's
|
| 121 |
|
|
// where we come in and start.
|
| 122 |
|
|
//
|
| 123 |
|
|
// r_busy <= 1'b1;
|
| 124 |
|
|
//
|
| 125 |
|
|
o_quotient <= 0;
|
| 126 |
|
|
r_bit <= {(LGBW){1'b1}};
|
| 127 |
|
|
r_divisor <= { i_denominator, {(BW-1){1'b0}} };
|
| 128 |
|
|
r_dividend <= i_numerator;
|
| 129 |
|
|
r_sign <= 1'b0;
|
| 130 |
|
|
pre_sign <= i_signed;
|
| 131 |
|
|
r_z <= 1'b1;
|
| 132 |
|
|
end else if (pre_sign)
|
| 133 |
|
|
begin
|
| 134 |
|
|
//
|
| 135 |
|
|
// Note that we only come in here, for one clock, if
|
| 136 |
|
|
// our initial value may have been signed. If we are
|
| 137 |
|
|
// doing an unsigned divide, we then skip this step.
|
| 138 |
|
|
//
|
| 139 |
|
|
r_sign <= ((r_divisor[(2*BW-2)])^(r_dividend[(BW-1)]));
|
| 140 |
|
|
// Negate our dividend if necessary so that it becomes
|
| 141 |
|
|
// a magnitude only value
|
| 142 |
|
|
if (r_dividend[BW-1])
|
| 143 |
|
|
r_dividend <= -r_dividend;
|
| 144 |
|
|
// Do the same with the divisor--rendering it into
|
| 145 |
|
|
// a magnitude only.
|
| 146 |
|
|
if (r_divisor[(2*BW-2)])
|
| 147 |
|
|
r_divisor[(2*BW-2):(BW-1)] <= -r_divisor[(2*BW-2):(BW-1)];
|
| 148 |
|
|
//
|
| 149 |
|
|
// We only do this stage for a single clock, so go on
|
| 150 |
|
|
// with the rest of the divide otherwise.
|
| 151 |
|
|
pre_sign <= 1'b0;
|
| 152 |
|
|
end else if (r_busy)
|
| 153 |
|
|
begin
|
| 154 |
|
|
// While the divide is taking place, we examine each bit
|
| 155 |
|
|
// in turn here.
|
| 156 |
|
|
//
|
| 157 |
|
|
r_bit <= r_bit + {(LGBW){1'b1}}; // r_bit = r_bit - 1;
|
| 158 |
|
|
r_divisor <= { 1'b0, r_divisor[(2*BW-2):1] };
|
| 159 |
|
|
if (|r_divisor[(2*BW-2):(BW)])
|
| 160 |
|
|
begin
|
| 161 |
|
|
end else if (diff[BW])
|
| 162 |
|
|
begin
|
| 163 |
|
|
//
|
| 164 |
|
|
// diff = r_dividend - r_divisor[(BW-1):0];
|
| 165 |
|
|
//
|
| 166 |
|
|
// If this value was negative, there wasn't
|
| 167 |
|
|
// enough value in the dividend to support
|
| 168 |
|
|
// pulling off a bit. We'll move down a bit
|
| 169 |
|
|
// therefore and try again.
|
| 170 |
|
|
//
|
| 171 |
|
|
end else begin
|
| 172 |
|
|
//
|
| 173 |
|
|
// Put a '1' into our output accumulator.
|
| 174 |
|
|
// Subtract the divisor from the dividend,
|
| 175 |
|
|
// and then move on to the next bit
|
| 176 |
|
|
//
|
| 177 |
|
|
r_dividend <= diff[(BW-1):0];
|
| 178 |
|
|
o_quotient[r_bit[(LGBW-1):0]] <= 1'b1;
|
| 179 |
|
|
r_z <= 1'b0;
|
| 180 |
|
|
end
|
| 181 |
|
|
r_sign <= (r_sign)&&(~zero_divisor);
|
| 182 |
|
|
end else if (r_sign)
|
| 183 |
|
|
begin
|
| 184 |
|
|
r_sign <= 1'b0;
|
| 185 |
|
|
o_quotient <= -o_quotient;
|
| 186 |
|
|
end
|
| 187 |
|
|
|
| 188 |
|
|
// Set Carry on an exact divide
|
| 189 |
|
|
wire w_n;
|
| 190 |
|
|
always @(posedge i_clk)
|
| 191 |
|
|
r_c <= (r_busy)&&((diff == 0)||(r_dividend == 0));
|
| 192 |
|
|
assign w_n = o_quotient[(BW-1)];
|
| 193 |
|
|
|
| 194 |
|
|
assign o_flags = { 1'b0, w_n, r_c, r_z };
|
| 195 |
|
|
endmodule
|
