| 1 |
51 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
| 2 |
|
|
//
|
| 3 |
|
|
// Filename: dblfetch.v
|
| 4 |
|
|
//
|
| 5 |
|
|
// Project: Zip CPU -- a small, lightweight, RISC CPU soft core
|
| 6 |
|
|
//
|
| 7 |
|
|
// Purpose: This is one step beyond the simplest instruction fetch,
|
| 8 |
|
|
// prefetch.v. dblfetch.v uses memory pipelining to fetch two
|
| 9 |
|
|
// instruction words in one cycle, figuring that the unpipelined CPU can't
|
| 10 |
|
|
// go through both at once, but yet recycles itself fast enough for the
|
| 11 |
|
|
// next instruction that would follow. It is designed to be a touch
|
| 12 |
|
|
// faster than the single instruction prefetch, although not as fast as
|
| 13 |
|
|
// the prefetch and cache found elsewhere.
|
| 14 |
|
|
//
|
| 15 |
|
|
// There are some gotcha's in this logic, however. For example, it's
|
| 16 |
|
|
// illegal to switch devices mid-transaction, since the second device
|
| 17 |
|
|
// might have different timing. I.e. the first device might take 8
|
| 18 |
|
|
// clocks to create an ACK, and the second device might take 2 clocks, the
|
| 19 |
|
|
// acks might therefore come on top of each other, or even out of order.
|
| 20 |
|
|
// But ... in order to keep logic down, we keep track of the PC in the
|
| 21 |
|
|
// o_wb_addr register. Hence, this register gets changed on any i_new_pc.
|
| 22 |
|
|
// The i_pc value associated with i_new_pc will only be valid for one
|
| 23 |
|
|
// clock, hence we can't wait to change. To keep from violating the WB
|
| 24 |
|
|
// rule, therefore, we *must* immediately stop requesting any transaction,
|
| 25 |
|
|
// and then terminate the bus request as soon as possible.
|
| 26 |
|
|
//
|
| 27 |
|
|
// This has consequences in terms of logic used, leaving this routine
|
| 28 |
|
|
// anything but simple--even though the number of wires affected by
|
| 29 |
|
|
// this is small (o_wb_cyc, o_wb_stb, and last_ack).
|
| 30 |
|
|
//
|
| 31 |
|
|
//
|
| 32 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
| 33 |
|
|
// Gisselquist Technology, LLC
|
| 34 |
|
|
//
|
| 35 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
| 36 |
|
|
//
|
| 37 |
|
|
// Copyright (C) 2017, Gisselquist Technology, LLC
|
| 38 |
|
|
//
|
| 39 |
|
|
// This program is free software (firmware): you can redistribute it and/or
|
| 40 |
|
|
// modify it under the terms of the GNU General Public License as published
|
| 41 |
|
|
// by the Free Software Foundation, either version 3 of the License, or (at
|
| 42 |
|
|
// your option) any later version.
|
| 43 |
|
|
//
|
| 44 |
|
|
// This program is distributed in the hope that it will be useful, but WITHOUT
|
| 45 |
|
|
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
|
| 46 |
|
|
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
| 47 |
|
|
// for more details.
|
| 48 |
|
|
//
|
| 49 |
|
|
// You should have received a copy of the GNU General Public License along
|
| 50 |
|
|
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
|
| 51 |
|
|
// target there if the PDF file isn't present.) If not, see
|
| 52 |
|
|
// <http://www.gnu.org/licenses/> for a copy.
|
| 53 |
|
|
//
|
| 54 |
|
|
// License: GPL, v3, as defined and found on www.gnu.org,
|
| 55 |
|
|
// http://www.gnu.org/licenses/gpl.html
|
| 56 |
|
|
//
|
| 57 |
|
|
//
|
| 58 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
| 59 |
|
|
//
|
| 60 |
|
|
//
|
| 61 |
|
|
module dblfetch(i_clk, i_rst, i_new_pc, i_clear_cache,
|
| 62 |
|
|
i_stall_n, i_pc, o_i, o_pc, o_v,
|
| 63 |
|
|
o_wb_cyc, o_wb_stb, o_wb_we, o_wb_addr, o_wb_data,
|
| 64 |
|
|
i_wb_ack, i_wb_stall, i_wb_err, i_wb_data,
|
| 65 |
|
|
o_illegal);
|
| 66 |
|
|
parameter ADDRESS_WIDTH=32, AUX_WIDTH = 1;
|
| 67 |
|
|
localparam AW=ADDRESS_WIDTH;
|
| 68 |
|
|
input i_clk, i_rst, i_new_pc, i_clear_cache,
|
| 69 |
|
|
i_stall_n;
|
| 70 |
|
|
input [(AW-1):0] i_pc;
|
| 71 |
|
|
output reg [31:0] o_i;
|
| 72 |
|
|
output reg [(AW-1):0] o_pc;
|
| 73 |
|
|
output wire o_v;
|
| 74 |
|
|
// Wishbone outputs
|
| 75 |
|
|
output reg o_wb_cyc, o_wb_stb;
|
| 76 |
|
|
output wire o_wb_we;
|
| 77 |
|
|
output reg [(AW-1):0] o_wb_addr;
|
| 78 |
|
|
output wire [31:0] o_wb_data;
|
| 79 |
|
|
// And return inputs
|
| 80 |
|
|
input i_wb_ack, i_wb_stall, i_wb_err;
|
| 81 |
|
|
input [31:0] i_wb_data;
|
| 82 |
|
|
// And ... the result if we got an error
|
| 83 |
|
|
output reg o_illegal;
|
| 84 |
|
|
|
| 85 |
|
|
assign o_wb_we = 1'b0;
|
| 86 |
|
|
assign o_wb_data = 32'h0000;
|
| 87 |
|
|
|
| 88 |
|
|
reg last_ack, last_stb, invalid_bus_cycle;
|
| 89 |
|
|
|
| 90 |
|
|
reg [31:0] cache [0:1];
|
| 91 |
|
|
reg cache_read_addr, cache_write_addr;
|
| 92 |
|
|
reg [1:0] cache_valid;
|
| 93 |
|
|
|
| 94 |
|
|
initial o_wb_cyc = 1'b0;
|
| 95 |
|
|
initial o_wb_stb = 1'b0;
|
| 96 |
|
|
always @(posedge i_clk)
|
| 97 |
|
|
if ((i_rst)||(i_wb_err))
|
| 98 |
|
|
begin
|
| 99 |
|
|
o_wb_cyc <= 1'b0;
|
| 100 |
|
|
o_wb_stb <= 1'b0;
|
| 101 |
|
|
// last_stb <= 1'b0;
|
| 102 |
|
|
// last_ack <= 1'b0;
|
| 103 |
|
|
end else if (o_wb_cyc)
|
| 104 |
|
|
begin
|
| 105 |
|
|
if ((o_wb_stb)&&(!i_wb_stall))
|
| 106 |
|
|
begin
|
| 107 |
|
|
// last_stb <= 1'b1;
|
| 108 |
|
|
o_wb_stb <= !last_stb;
|
| 109 |
|
|
end
|
| 110 |
|
|
// if (i_wb_ack)
|
| 111 |
|
|
// last_ack <= 1'b1;
|
| 112 |
|
|
if ((i_new_pc)||(invalid_bus_cycle))
|
| 113 |
|
|
o_wb_stb <= 1'b0;
|
| 114 |
|
|
|
| 115 |
|
|
if ((i_wb_ack)&&(
|
| 116 |
|
|
// Relase the bus on the second ack
|
| 117 |
|
|
(last_ack)
|
| 118 |
|
|
// Or on the first ACK, if we've been told
|
| 119 |
|
|
// we have an invalid bus cycle
|
| 120 |
|
|
||((o_wb_stb)&&(i_wb_stall)&&(last_stb)&&(
|
| 121 |
|
|
(i_new_pc)||(invalid_bus_cycle)))
|
| 122 |
|
|
))
|
| 123 |
|
|
begin
|
| 124 |
|
|
o_wb_cyc <= 1'b0;
|
| 125 |
|
|
o_wb_stb <= 1'b0;
|
| 126 |
|
|
end
|
| 127 |
|
|
|
| 128 |
|
|
if ((!last_stb)&&(i_wb_stall)&&((i_new_pc)||(invalid_bus_cycle)))
|
| 129 |
|
|
// Also release the bus with no acks, if we
|
| 130 |
|
|
// haven't made any requests
|
| 131 |
|
|
begin
|
| 132 |
|
|
o_wb_cyc <= 1'b0;
|
| 133 |
|
|
o_wb_stb <= 1'b0;
|
| 134 |
|
|
end
|
| 135 |
|
|
end else if ((invalid_bus_cycle)
|
| 136 |
|
|
||((o_v)&&(i_stall_n)&&(cache_read_addr))) // Initiate a bus cycle
|
| 137 |
|
|
begin
|
| 138 |
|
|
o_wb_cyc <= 1'b1;
|
| 139 |
|
|
o_wb_stb <= 1'b1;
|
| 140 |
|
|
// last_stb <= 1'b0;
|
| 141 |
|
|
// last_ack <= 1'b0;
|
| 142 |
|
|
end
|
| 143 |
|
|
|
| 144 |
|
|
initial last_stb = 1'b0;
|
| 145 |
|
|
always @(posedge i_clk)
|
| 146 |
|
|
if ((o_wb_cyc)&&(o_wb_stb)&&(!i_wb_stall))
|
| 147 |
|
|
last_stb <= 1'b1;
|
| 148 |
|
|
else if (!o_wb_cyc)
|
| 149 |
|
|
last_stb <= 1'b0;
|
| 150 |
|
|
|
| 151 |
|
|
initial last_ack = 1'b0;
|
| 152 |
|
|
always @(posedge i_clk)
|
| 153 |
|
|
if ((o_wb_cyc)&&(i_wb_ack))
|
| 154 |
|
|
last_ack <= 1'b1;
|
| 155 |
|
|
else if ((o_wb_cyc)&&(o_wb_stb)&&(i_wb_stall)&&(
|
| 156 |
|
|
(i_new_pc)||(invalid_bus_cycle)))
|
| 157 |
|
|
last_ack <= 1'b1;
|
| 158 |
|
|
else if ((o_wb_cyc)&&(o_wb_stb)&&(!i_wb_stall)&&(!last_stb)&&(
|
| 159 |
|
|
(i_new_pc)||(invalid_bus_cycle)))
|
| 160 |
|
|
last_ack <= 1'b1;
|
| 161 |
|
|
else if (!o_wb_cyc)
|
| 162 |
|
|
last_ack <= 1'b0;
|
| 163 |
|
|
|
| 164 |
|
|
initial invalid_bus_cycle = 1'b0;
|
| 165 |
|
|
always @(posedge i_clk)
|
| 166 |
|
|
if (i_rst)
|
| 167 |
|
|
invalid_bus_cycle <= 1'b0;
|
| 168 |
|
|
else if ((i_new_pc)||(i_clear_cache))
|
| 169 |
|
|
invalid_bus_cycle <= 1'b1;
|
| 170 |
|
|
else if (!o_wb_cyc)
|
| 171 |
|
|
invalid_bus_cycle <= 1'b0;
|
| 172 |
|
|
|
| 173 |
|
|
initial o_wb_addr = {(AW){1'b1}};
|
| 174 |
|
|
always @(posedge i_clk)
|
| 175 |
|
|
if (i_new_pc)
|
| 176 |
|
|
o_wb_addr <= i_pc;
|
| 177 |
|
|
else if ((o_wb_stb)&&(!i_wb_stall)&&(!invalid_bus_cycle))
|
| 178 |
|
|
o_wb_addr <= o_wb_addr + 1'b1;
|
| 179 |
|
|
|
| 180 |
|
|
initial cache_write_addr = 1'b0;
|
| 181 |
|
|
always @(posedge i_clk)
|
| 182 |
|
|
if (!o_wb_cyc)
|
| 183 |
|
|
cache_write_addr <= 1'b0;
|
| 184 |
|
|
else if ((o_wb_cyc)&&(i_wb_ack))
|
| 185 |
|
|
cache_write_addr <= cache_write_addr + 1'b1;
|
| 186 |
|
|
|
| 187 |
|
|
always @(posedge i_clk)
|
| 188 |
|
|
if ((o_wb_cyc)&&(i_wb_ack))
|
| 189 |
|
|
cache[cache_write_addr] <= i_wb_data;
|
| 190 |
|
|
|
| 191 |
|
|
initial cache_read_addr = 1'b0;
|
| 192 |
|
|
always @(posedge i_clk)
|
| 193 |
|
|
if ((i_new_pc)||(invalid_bus_cycle)
|
| 194 |
|
|
||((o_v)&&(cache_read_addr)&&(i_stall_n)))
|
| 195 |
|
|
cache_read_addr <= 1'b0;
|
| 196 |
|
|
else if ((o_v)&&(i_stall_n))
|
| 197 |
|
|
cache_read_addr <= 1'b1;
|
| 198 |
|
|
|
| 199 |
|
|
always @(posedge i_clk)
|
| 200 |
|
|
if ((i_new_pc)||(invalid_bus_cycle))
|
| 201 |
|
|
cache_valid <= 2'b00;
|
| 202 |
|
|
else begin
|
| 203 |
|
|
if ((o_v)&&(i_stall_n))
|
| 204 |
|
|
cache_valid[cache_read_addr] <= 1'b0;
|
| 205 |
|
|
if ((o_wb_cyc)&&(i_wb_ack))
|
| 206 |
|
|
cache_valid[cache_write_addr] <= 1'b1;
|
| 207 |
|
|
end
|
| 208 |
|
|
|
| 209 |
|
|
initial o_i = {(32){1'b1}};
|
| 210 |
|
|
always @(posedge i_clk)
|
| 211 |
|
|
if ((i_stall_n)&&(o_wb_cyc)&&(i_wb_ack))
|
| 212 |
|
|
o_i <= i_wb_data;
|
| 213 |
|
|
else
|
| 214 |
|
|
o_i <= cache[cache_read_addr];
|
| 215 |
|
|
|
| 216 |
|
|
initial o_pc = 0;
|
| 217 |
|
|
always @(posedge i_clk)
|
| 218 |
|
|
if (i_new_pc)
|
| 219 |
|
|
o_pc <= i_pc;
|
| 220 |
|
|
else if ((o_v)&&(i_stall_n))
|
| 221 |
|
|
o_pc <= o_pc + 1'b1;
|
| 222 |
|
|
|
| 223 |
|
|
assign o_v = cache_valid[cache_read_addr];
|
| 224 |
|
|
|
| 225 |
|
|
initial o_illegal = 1'b0;
|
| 226 |
|
|
always @(posedge i_clk)
|
| 227 |
|
|
if ((o_wb_cyc)&&(i_wb_err))
|
| 228 |
|
|
o_illegal <= 1'b1;
|
| 229 |
|
|
else if ((!o_wb_cyc)&&((i_new_pc)||(invalid_bus_cycle)))
|
| 230 |
|
|
o_illegal <= 1'b0;
|
| 231 |
|
|
|
| 232 |
|
|
endmodule
|
