1 |
3 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
2 |
|
|
//
|
3 |
|
|
// Filename: fastcache.v
|
4 |
|
|
//
|
5 |
|
|
// Project: Zip CPU -- a small, lightweight, RISC CPU soft core
|
6 |
|
|
//
|
7 |
|
|
// Purpose: Keeping our CPU fed with instructions, at one per clock and
|
8 |
|
|
// with no stalls. An unusual feature of this cache is the
|
9 |
|
|
// requirement that the entire cache may be cleared (if necessary).
|
10 |
|
|
//
|
11 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
12 |
|
|
// Gisselquist Technology, LLC
|
13 |
|
|
//
|
14 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
15 |
|
|
//
|
16 |
|
|
// Copyright (C) 2015, Gisselquist Technology, LLC
|
17 |
|
|
//
|
18 |
|
|
// This program is free software (firmware): you can redistribute it and/or
|
19 |
|
|
// modify it under the terms of the GNU General Public License as published
|
20 |
|
|
// by the Free Software Foundation, either version 3 of the License, or (at
|
21 |
|
|
// your option) any later version.
|
22 |
|
|
//
|
23 |
|
|
// This program is distributed in the hope that it will be useful, but WITHOUT
|
24 |
|
|
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
|
25 |
|
|
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
26 |
|
|
// for more details.
|
27 |
|
|
//
|
28 |
|
|
// License: GPL, v3, as defined and found on www.gnu.org,
|
29 |
|
|
// http://www.gnu.org/licenses/gpl.html
|
30 |
|
|
//
|
31 |
|
|
//
|
32 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
33 |
|
|
//
|
34 |
|
|
module fastcache(i_clk, i_rst, i_new_pc, i_clear_cache,
|
35 |
|
|
// i_early_branch, i_from_addr,
|
36 |
|
|
i_stall_n, i_pc, o_i, o_pc, o_v,
|
37 |
|
|
o_wb_cyc, o_wb_stb, o_wb_we, o_wb_addr, o_wb_data,
|
38 |
|
|
i_wb_ack, i_wb_stall, i_wb_err, i_wb_data,
|
39 |
|
|
o_illegal);
|
40 |
|
|
parameter LGCACHELEN = 8, ADDRESS_WIDTH=24,
|
41 |
|
|
CACHELEN=(1<<LGCACHELEN), BUSW=32, AW=ADDRESS_WIDTH,
|
42 |
|
|
CW=LGCACHELEN, PW=LGCACHELEN-5;
|
43 |
|
|
input i_clk, i_rst, i_new_pc;
|
44 |
|
|
input i_clear_cache;
|
45 |
|
|
input i_stall_n;
|
46 |
|
|
input [(AW-1):0] i_pc;
|
47 |
|
|
output wire [(BUSW-1):0] o_i;
|
48 |
|
|
output wire [(AW-1):0] o_pc;
|
49 |
|
|
output wire o_v;
|
50 |
|
|
//
|
51 |
|
|
output reg o_wb_cyc, o_wb_stb;
|
52 |
|
|
output wire o_wb_we;
|
53 |
|
|
output reg [(AW-1):0] o_wb_addr;
|
54 |
|
|
output wire [(BUSW-1):0] o_wb_data;
|
55 |
|
|
//
|
56 |
|
|
input i_wb_ack, i_wb_stall, i_wb_err;
|
57 |
|
|
input [(BUSW-1):0] i_wb_data;
|
58 |
|
|
//
|
59 |
|
|
output reg o_illegal;
|
60 |
|
|
|
61 |
|
|
// Fixed bus outputs: we read from the bus only, never write.
|
62 |
|
|
// Thus the output data is ... irrelevant and don't care. We set it
|
63 |
|
|
// to zero just to set it to something.
|
64 |
|
|
assign o_wb_we = 1'b0;
|
65 |
|
|
assign o_wb_data = 0;
|
66 |
|
|
|
67 |
|
|
wire r_v;
|
68 |
|
|
reg [(BUSW-1):0] cache [0:((1<<CW)-1)];
|
69 |
|
|
reg [(AW-CW-1):0] tags [0:((1<<(CW-PW))-1)];
|
70 |
|
|
reg [((1<<(CW-PW))-1):0] vmask;
|
71 |
|
|
|
72 |
|
|
reg [(AW-1):0] lastpc;
|
73 |
|
|
reg [(CW-1):0] rdaddr;
|
74 |
|
|
reg [(AW-1):CW] tagvalipc, tagvallst;
|
75 |
|
|
wire [(AW-1):CW] tagval;
|
76 |
|
|
wire [(AW-1):PW] lasttag;
|
77 |
|
|
reg illegal_valid;
|
78 |
|
|
reg [(AW-1):PW] illegal_cache;
|
79 |
|
|
|
80 |
|
|
// initial o_i = 32'h76_00_00_00; // A NOOP instruction
|
81 |
|
|
// initial o_pc = 0;
|
82 |
|
|
reg [(BUSW-1):0] r_pc_cache, r_last_cache;
|
83 |
|
|
reg [(AW-1):0] r_pc, r_lastpc;
|
84 |
|
|
reg isrc;
|
85 |
|
|
always @(posedge i_clk)
|
86 |
|
|
if (~r_v)
|
87 |
|
|
isrc <= 1'b0;
|
88 |
|
|
else if ((i_stall_n)||(i_new_pc))
|
89 |
|
|
isrc <= 1'b1;
|
90 |
|
|
always @(posedge i_clk)
|
91 |
|
|
r_pc_cache <= cache[i_pc[(CW-1):0]];
|
92 |
|
|
always @(posedge i_clk)
|
93 |
|
|
r_last_cache <= cache[lastpc[(CW-1):0]];
|
94 |
|
|
always @(posedge i_clk)
|
95 |
|
|
r_pc <= i_pc;
|
96 |
|
|
always @(posedge i_clk)
|
97 |
|
|
r_lastpc <= lastpc;
|
98 |
|
|
assign o_pc = (isrc) ? r_pc : r_lastpc;
|
99 |
|
|
assign o_i = (isrc) ? r_pc_cache : r_last_cache;
|
100 |
|
|
|
101 |
|
|
reg tagsrc;
|
102 |
|
|
always @(posedge i_clk)
|
103 |
|
|
// It may be possible to recover a clock once the cache line
|
104 |
|
|
// has been filled, but our prior attempt to do so has lead
|
105 |
|
|
// to a race condition, so we keep this logic simple.
|
106 |
|
|
if (((r_v)&&(i_stall_n))||(i_clear_cache)||(i_new_pc))
|
107 |
|
|
tagsrc <= 1'b1;
|
108 |
|
|
else
|
109 |
|
|
tagsrc <= 1'b0;
|
110 |
|
|
initial tagvalipc = 0;
|
111 |
|
|
always @(posedge i_clk)
|
112 |
|
|
tagvalipc <= tags[i_pc[(CW-1):PW]];
|
113 |
|
|
initial tagvallst = 0;
|
114 |
|
|
always @(posedge i_clk)
|
115 |
|
|
tagvallst <= tags[lastpc[(CW-1):PW]];
|
116 |
|
|
assign tagval = (tagsrc)?tagvalipc : tagvallst;
|
117 |
|
|
|
118 |
|
|
// i_pc will only increment when everything else isn't stalled, thus
|
119 |
|
|
// we can set it without worrying about that. Doing this enables
|
120 |
|
|
// us to work in spite of stalls. For example, if the next address
|
121 |
|
|
// isn't valid, but the decoder is stalled, get the next address
|
122 |
|
|
// anyway.
|
123 |
|
|
initial lastpc = 0;
|
124 |
|
|
always @(posedge i_clk)
|
125 |
|
|
if (((r_v)&&(i_stall_n))||(i_clear_cache)||(i_new_pc))
|
126 |
|
|
lastpc <= i_pc;
|
127 |
|
|
|
128 |
|
|
assign lasttag = lastpc[(AW-1):PW];
|
129 |
|
|
// initial lasttag = 0;
|
130 |
|
|
// always @(posedge i_clk)
|
131 |
|
|
// if (((r_v)&&(i_stall_n))||(i_clear_cache)||(i_new_pc))
|
132 |
|
|
// lasttag <= i_pc[(AW-1):PW];
|
133 |
|
|
|
134 |
|
|
wire w_v_from_pc, w_v_from_last;
|
135 |
|
|
assign w_v_from_pc = ((i_pc[(AW-1):PW] == lasttag)
|
136 |
|
|
&&(tagvalipc == i_pc[(AW-1):CW])
|
137 |
|
|
&&(vmask[i_pc[(CW-1):PW]]));
|
138 |
|
|
assign w_v_from_last = (
|
139 |
|
|
//(lastpc[(AW-1):PW] == lasttag)&&
|
140 |
|
|
(tagvallst == lastpc[(AW-1):CW])
|
141 |
|
|
&&(vmask[lastpc[(CW-1):PW]]));
|
142 |
|
|
|
143 |
|
|
reg [1:0] delay;
|
144 |
|
|
|
145 |
|
|
initial delay = 2'h3;
|
146 |
|
|
reg rvsrc;
|
147 |
|
|
always @(posedge i_clk)
|
148 |
|
|
if ((i_rst)||(i_clear_cache)||(i_new_pc)||((r_v)&&(i_stall_n)))
|
149 |
|
|
begin
|
150 |
|
|
// r_v <= r_v_from_pc;
|
151 |
|
|
rvsrc <= 1'b1;
|
152 |
|
|
delay <= 2'h2;
|
153 |
|
|
end else if (~r_v) begin // Otherwise, r_v was true and we were
|
154 |
|
|
// stalled, hence only if ~r_v
|
155 |
|
|
rvsrc <= 1'b0;
|
156 |
|
|
if (o_wb_cyc)
|
157 |
|
|
delay <= 2'h2;
|
158 |
|
|
else if (delay != 0)
|
159 |
|
|
delay <= delay + 2'b11; // i.e. delay -= 1;
|
160 |
|
|
end
|
161 |
|
|
reg r_v_from_pc, r_v_from_last;
|
162 |
|
|
always @(posedge i_clk)
|
163 |
|
|
r_v_from_pc <= w_v_from_pc;
|
164 |
|
|
always @(posedge i_clk)
|
165 |
|
|
r_v_from_last <= w_v_from_last;
|
166 |
|
|
|
167 |
|
|
assign r_v = ((rvsrc)?(r_v_from_pc):(r_v_from_last));
|
168 |
|
|
assign o_v = ((rvsrc)?(r_v_from_pc):(r_v_from_last))&&(~i_new_pc);
|
169 |
|
|
|
170 |
|
|
reg last_ack;
|
171 |
|
|
initial last_ack <= 1'b0;
|
172 |
|
|
always @(posedge i_clk)
|
173 |
|
|
last_ack <= (o_wb_cyc)&&(
|
174 |
|
|
(rdaddr[(PW-1):1]=={(PW){1'b1}})
|
175 |
|
|
&&((rdaddr[0])||(i_wb_ack)));
|
176 |
|
|
|
177 |
|
|
reg needload;
|
178 |
|
|
initial needload = 1'b0;
|
179 |
|
|
always @(posedge i_clk)
|
180 |
|
|
needload <= ((~r_v)&&(delay==0)
|
181 |
|
|
&&((tagvallst != lastpc[(AW-1):CW])
|
182 |
|
|
||(~vmask[lastpc[(CW-1):PW]]))
|
183 |
|
|
&&((~illegal_valid)
|
184 |
|
|
||(lastpc[(AW-1):PW] != illegal_cache)));
|
185 |
|
|
|
186 |
|
|
reg last_addr;
|
187 |
|
|
initial last_addr = 1'b0;
|
188 |
|
|
always @(posedge i_clk)
|
189 |
|
|
last_addr <= (o_wb_cyc)&&(o_wb_addr[(PW-2):1] == {(PW-1){1'b1}})
|
190 |
|
|
&&((~i_wb_stall)|(o_wb_addr[0]));
|
191 |
|
|
|
192 |
|
|
initial o_wb_cyc = 1'b0;
|
193 |
|
|
initial o_wb_stb = 1'b0;
|
194 |
|
|
initial o_wb_addr = {(AW){1'b0}};
|
195 |
|
|
initial rdaddr = 0;
|
196 |
|
|
always @(posedge i_clk)
|
197 |
|
|
if ((i_rst)||(i_clear_cache))
|
198 |
|
|
begin
|
199 |
|
|
o_wb_cyc <= 1'b0;
|
200 |
|
|
o_wb_stb <= 1'b0;
|
201 |
|
|
end else if (o_wb_cyc)
|
202 |
|
|
begin
|
203 |
|
|
if (i_wb_err)
|
204 |
|
|
o_wb_stb <= 1'b0;
|
205 |
|
|
else if ((o_wb_stb)&&(~i_wb_stall)&&(last_addr))
|
206 |
|
|
o_wb_stb <= 1'b0;
|
207 |
|
|
|
208 |
|
|
if (((i_wb_ack)&&(last_ack))||(i_wb_err))
|
209 |
|
|
o_wb_cyc <= 1'b0;
|
210 |
|
|
|
211 |
|
|
// else if (rdaddr[(PW-1):1] == {(PW-1){1'b1}})
|
212 |
|
|
// tags[lastpc[(CW-1):PW]] <= lastpc[(AW-1):CW];
|
213 |
|
|
|
214 |
|
|
end else if (needload)
|
215 |
|
|
begin
|
216 |
|
|
o_wb_cyc <= 1'b1;
|
217 |
|
|
o_wb_stb <= 1'b1;
|
218 |
|
|
end
|
219 |
|
|
|
220 |
|
|
always @(posedge i_clk)
|
221 |
|
|
if (o_wb_cyc) // &&(i_wb_ack)
|
222 |
|
|
tags[o_wb_addr[(CW-1):PW]] <= o_wb_addr[(AW-1):CW];
|
223 |
|
|
always @(posedge i_clk)
|
224 |
|
|
if ((o_wb_cyc)&&(i_wb_ack))
|
225 |
|
|
rdaddr <= rdaddr + 1;
|
226 |
|
|
else if (~o_wb_cyc)
|
227 |
|
|
rdaddr <= { lastpc[(CW-1):PW], {(PW){1'b0}} };
|
228 |
|
|
|
229 |
|
|
always @(posedge i_clk)
|
230 |
|
|
if ((o_wb_stb)&&(~i_wb_stall)&&(~last_addr))
|
231 |
|
|
o_wb_addr[(PW-1):0] <= o_wb_addr[(PW-1):0]+1;
|
232 |
|
|
else if (~o_wb_cyc)
|
233 |
|
|
o_wb_addr <= { lastpc[(AW-1):PW], {(PW){1'b0}} };
|
234 |
|
|
|
235 |
|
|
// Can't initialize an array, so leave cache uninitialized
|
236 |
|
|
// We'll also never get an ack without sys being active, so skip
|
237 |
|
|
// that check. Or rather, let's just use o_wb_cyc instead. This
|
238 |
|
|
// will work because multiple writes to the same address, ending with
|
239 |
|
|
// a valid write, aren't a problem.
|
240 |
|
|
always @(posedge i_clk)
|
241 |
|
|
if (o_wb_cyc) // &&(i_wb_ack)
|
242 |
|
|
cache[rdaddr] <= i_wb_data;
|
243 |
|
|
|
244 |
|
|
// VMask ... is a section loaded?
|
245 |
|
|
initial vmask = 0;
|
246 |
|
|
always @(posedge i_clk)
|
247 |
|
|
if ((i_rst)||(i_clear_cache))
|
248 |
|
|
vmask <= 0;
|
249 |
|
|
else begin
|
250 |
|
|
if ((o_wb_cyc)&&(i_wb_ack)&&(last_ack))
|
251 |
|
|
vmask[rdaddr[(CW-1):PW]] <= 1'b1;
|
252 |
|
|
if ((~o_wb_cyc)&&(needload))
|
253 |
|
|
vmask[lastpc[(CW-1):PW]] <= 1'b0;
|
254 |
|
|
end
|
255 |
|
|
|
256 |
|
|
initial illegal_cache = 0;
|
257 |
|
|
initial illegal_valid = 0;
|
258 |
|
|
always @(posedge i_clk)
|
259 |
|
|
if ((i_rst)||(i_clear_cache))
|
260 |
|
|
begin
|
261 |
|
|
illegal_cache <= 0;
|
262 |
|
|
illegal_valid <= 0;
|
263 |
|
|
end else if ((o_wb_cyc)&&(i_wb_err))
|
264 |
|
|
begin
|
265 |
|
|
illegal_cache <= o_wb_addr[(AW-1):PW];
|
266 |
|
|
illegal_valid <= 1'b1;
|
267 |
|
|
end
|
268 |
|
|
|
269 |
|
|
initial o_illegal = 1'b0;
|
270 |
|
|
always @(posedge i_clk)
|
271 |
|
|
if ((i_rst)||(i_clear_cache))
|
272 |
|
|
o_illegal <= 1'b0;
|
273 |
|
|
else
|
274 |
|
|
o_illegal <= (illegal_valid)
|
275 |
|
|
&&(illegal_cache == i_pc[(AW-1):PW]);
|
276 |
|
|
|
277 |
|
|
endmodule
|