1 |
209 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
2 |
204 |
dgisselq |
//
|
3 |
|
|
// Filename: mpy_tb.cpp
|
4 |
|
|
//
|
5 |
|
|
// Project: Zip CPU -- a small, lightweight, RISC CPU soft core
|
6 |
|
|
//
|
7 |
|
|
// Purpose: Bench testing for the multiply ALU instructions used within the
|
8 |
|
|
// Zip CPU. This depends upon the cpuops.v module, but should be
|
9 |
|
|
// independent of the internal settings within the module.
|
10 |
|
|
//
|
11 |
|
|
//
|
12 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
13 |
|
|
// Gisselquist Technology, LLC
|
14 |
|
|
//
|
15 |
209 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
16 |
204 |
dgisselq |
//
|
17 |
|
|
// Copyright (C) 2015-2016, Gisselquist Technology, LLC
|
18 |
|
|
//
|
19 |
|
|
// This program is free software (firmware): you can redistribute it and/or
|
20 |
|
|
// modify it under the terms of the GNU General Public License as published
|
21 |
|
|
// by the Free Software Foundation, either version 3 of the License, or (at
|
22 |
|
|
// your option) any later version.
|
23 |
|
|
//
|
24 |
|
|
// This program is distributed in the hope that it will be useful, but WITHOUT
|
25 |
|
|
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
|
26 |
|
|
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
27 |
|
|
// for more details.
|
28 |
|
|
//
|
29 |
209 |
dgisselq |
// You should have received a copy of the GNU General Public License along
|
30 |
|
|
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
|
31 |
|
|
// target there if the PDF file isn't present.) If not, see
|
32 |
|
|
// <http://www.gnu.org/licenses/> for a copy.
|
33 |
|
|
//
|
34 |
204 |
dgisselq |
// License: GPL, v3, as defined and found on www.gnu.org,
|
35 |
|
|
// http://www.gnu.org/licenses/gpl.html
|
36 |
|
|
//
|
37 |
|
|
//
|
38 |
209 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
39 |
204 |
dgisselq |
//
|
40 |
|
|
//
|
41 |
|
|
#include <signal.h>
|
42 |
|
|
#include <time.h>
|
43 |
|
|
#include <unistd.h>
|
44 |
|
|
#include <assert.h>
|
45 |
209 |
dgisselq |
#include <stdint.h>
|
46 |
204 |
dgisselq |
|
47 |
|
|
#include <stdlib.h>
|
48 |
|
|
#include <ctype.h>
|
49 |
|
|
|
50 |
|
|
#include "verilated.h"
|
51 |
|
|
#include "Vcpuops.h"
|
52 |
|
|
|
53 |
209 |
dgisselq |
#ifdef NEW_VERILATOR
|
54 |
|
|
#define VVAR(A) cpuops__DOT_ ## A
|
55 |
|
|
#else
|
56 |
|
|
#define VVAR(A) v__DOT_ ## A
|
57 |
|
|
#endif
|
58 |
|
|
|
59 |
|
|
|
60 |
204 |
dgisselq |
#include "testb.h"
|
61 |
|
|
#include "cpudefs.h"
|
62 |
|
|
// #include "twoc.h"
|
63 |
|
|
|
64 |
|
|
class CPUOPS_TB : public TESTB<Vcpuops> {
|
65 |
|
|
public:
|
66 |
|
|
// Nothing special to do in a startup.
|
67 |
|
|
CPUOPS_TB(void) {}
|
68 |
|
|
|
69 |
|
|
// ~CPUOPS_TB(void) {}
|
70 |
|
|
|
71 |
|
|
//
|
72 |
209 |
dgisselq |
// Calls TESTB<>::reset to reset the core. Makes sure the i_stb line
|
73 |
204 |
dgisselq |
// is low during this reset.
|
74 |
|
|
//
|
75 |
|
|
void reset(void) {
|
76 |
|
|
// m_flash.debug(false);
|
77 |
209 |
dgisselq |
m_core->i_stb = 0;
|
78 |
204 |
dgisselq |
|
79 |
|
|
TESTB<Vcpuops>::reset();
|
80 |
|
|
}
|
81 |
|
|
|
82 |
|
|
//
|
83 |
|
|
// dbgdump();
|
84 |
|
|
//
|
85 |
|
|
// Just before the positive edge of every clock, we call this function
|
86 |
|
|
// (if the debug flag is set). This prints out a line of information
|
87 |
|
|
// telling us what is going on within the logic, allowing us access
|
88 |
|
|
// for debugging purposes to inspect things.
|
89 |
|
|
//
|
90 |
|
|
// Other than debugging, this isn't necessary for the functioning of the
|
91 |
|
|
// test bench. At the same time, what are you using a test bench for if
|
92 |
|
|
// not for debugging?
|
93 |
|
|
//
|
94 |
|
|
void dbgdump(void) {
|
95 |
|
|
char outstr[2048], *s;
|
96 |
209 |
dgisselq |
sprintf(outstr, "Tick %4lld %s%s ",
|
97 |
|
|
(unsigned long long)m_tickcount,
|
98 |
|
|
(m_core->i_reset)?"R":" ",
|
99 |
|
|
(m_core->i_stb)?"CE":" ");
|
100 |
204 |
dgisselq |
switch(m_core->i_op) {
|
101 |
|
|
case 0: strcat(outstr, " SUB"); break;
|
102 |
|
|
case 1: strcat(outstr, " AND"); break;
|
103 |
|
|
case 2: strcat(outstr, " ADD"); break;
|
104 |
|
|
case 3: strcat(outstr, " OR"); break;
|
105 |
|
|
case 4: strcat(outstr, " XOR"); break;
|
106 |
|
|
case 5: strcat(outstr, " LSR"); break;
|
107 |
|
|
case 6: strcat(outstr, " LSL"); break;
|
108 |
|
|
case 7: strcat(outstr, " ASR"); break;
|
109 |
|
|
case 8: strcat(outstr, " MPY"); break;
|
110 |
|
|
case 9: strcat(outstr, "LODILO"); break;
|
111 |
|
|
case 10: strcat(outstr, "MPYUHI"); break;
|
112 |
|
|
case 11: strcat(outstr, "MPYSHI"); break;
|
113 |
|
|
case 12: strcat(outstr, " BREV"); break;
|
114 |
|
|
case 13: strcat(outstr, " POPC"); break;
|
115 |
|
|
case 14: strcat(outstr, " ROL"); break;
|
116 |
|
|
case 15: strcat(outstr, " MOV"); break;
|
117 |
|
|
default: strcat(outstr, "UNKWN!"); break;
|
118 |
|
|
} s = &outstr[strlen(outstr)];
|
119 |
|
|
sprintf(s, "(%x) 0x%08x 0x%08x -> 0x%08x [%x] %s%s",
|
120 |
|
|
m_core->i_op,
|
121 |
|
|
m_core->i_a, m_core->i_b,
|
122 |
|
|
m_core->o_c, m_core->o_f,
|
123 |
|
|
(m_core->o_valid)?"V":" ",
|
124 |
|
|
(m_core->o_busy)?"B":" ");
|
125 |
|
|
s = &outstr[strlen(outstr)];
|
126 |
|
|
|
127 |
|
|
#if(OPT_MULTIPLY==1)
|
128 |
209 |
dgisselq |
#define mpy_result VVAR(_mpy_result)
|
129 |
204 |
dgisselq |
sprintf(s, "1,MPY[][][%016lx]",
|
130 |
209 |
dgisselq |
(unsigned long)m_core->mpy_result);
|
131 |
204 |
dgisselq |
s = &outstr[strlen(outstr)];
|
132 |
|
|
#elif(OPT_MULTIPLY==2)
|
133 |
|
|
sprintf(s, "2,MPY[%016lx][%016lx][%016lx]",
|
134 |
209 |
dgisselq |
#define MPY2VAR(A) VVAR(_thempy__DOT__IMPY__DOT__MPN1__DOT__MPY2CK__DOT_ ## A)
|
135 |
|
|
#define r_mpy_a_input MPY2VAR(_r_mpy_a_input)
|
136 |
|
|
#define r_mpy_b_input MPY2VAR(_r_mpy_b_input)
|
137 |
|
|
#define mpy_result VVAR(_mpy_result)
|
138 |
|
|
m_core->r_mpy_a_input,
|
139 |
|
|
m_core->r_mpy_b_input,
|
140 |
|
|
m_core->mpy_result);
|
141 |
204 |
dgisselq |
s = &outstr[strlen(outstr)];
|
142 |
|
|
#elif(OPT_MULTIPLY==3)
|
143 |
209 |
dgisselq |
#define MPY3VAR(A) VVAR(_thempy__DOT__IMPY__DOT__MPN1__DOT__MPN2__DOT__MPY3CK__DOT_ ## A)
|
144 |
|
|
#define r_mpy_a_input MPY3VAR(_r_mpy_a_input)
|
145 |
|
|
#define r_mpy_b_input MPY3VAR(_r_mpy_b_input)
|
146 |
|
|
#define r_smpy_result MPY3VAR(_r_smpy_result)
|
147 |
|
|
#define mpypipe MPY3VAR(_mpypipe)
|
148 |
|
|
sprintf(s, "3,MPY[%08x][%08x][%016llx], P[%d]",
|
149 |
|
|
m_core->r_mpy_a_input,
|
150 |
|
|
m_core->r_mpy_b_input,
|
151 |
|
|
(long long)m_core->r_smpy_result,
|
152 |
|
|
m_core->mpypipe);
|
153 |
204 |
dgisselq |
|
154 |
|
|
#endif
|
155 |
|
|
|
156 |
|
|
#if(OPT_MULTIPLY != 1)
|
157 |
209 |
dgisselq |
#define this_is_a_multiply_op ((m_core->i_stb)&&(((m_core->i_op&0xe) == 5)||((m_core->i_op&0x0f)==0xc))) // VVAR(_this_is_a_multiply_op)
|
158 |
|
|
if (this_is_a_multiply_op)
|
159 |
204 |
dgisselq |
strcat(s, " MPY-OP");
|
160 |
|
|
#endif
|
161 |
|
|
puts(outstr);
|
162 |
|
|
}
|
163 |
|
|
|
164 |
|
|
//
|
165 |
|
|
// tick()
|
166 |
|
|
//
|
167 |
|
|
// Call this to step the processor.
|
168 |
|
|
//
|
169 |
|
|
// This is a bit unusual compared to other tick() functions I have in
|
170 |
|
|
// my simulators in that there are a lot of calls to eval() with clk==0.
|
171 |
|
|
// This is because the multiply logic for OPT_MULTIPLY < 3 depends upon
|
172 |
|
|
// it to be valid. I assume any true Xilinx, or even higher level,
|
173 |
|
|
// implementation wouldn't have this problem.
|
174 |
|
|
//
|
175 |
|
|
void tick(void) {
|
176 |
|
|
bool debug = false;
|
177 |
|
|
|
178 |
|
|
// Insist that we are never both busy and producing a valid
|
179 |
|
|
// result at the same time. One or the other may be true,
|
180 |
|
|
// but never both.
|
181 |
|
|
assert((!m_core->o_busy)||(!m_core->o_valid));
|
182 |
|
|
//
|
183 |
|
|
TESTB<Vcpuops>::tick();
|
184 |
|
|
|
185 |
|
|
if (debug)
|
186 |
|
|
dbgdump();
|
187 |
|
|
}
|
188 |
|
|
|
189 |
|
|
//
|
190 |
|
|
// clear_ops
|
191 |
|
|
//
|
192 |
|
|
// Runs enough clocks through the device until it is neither busy nor
|
193 |
|
|
// valid. At this point, the ALU should be thoroughly clear. Then
|
194 |
|
|
// we tick things once more.
|
195 |
|
|
//
|
196 |
|
|
void clear_ops(void) {
|
197 |
209 |
dgisselq |
m_core->i_stb = 0;
|
198 |
204 |
dgisselq |
m_core->i_op = 0;
|
199 |
|
|
|
200 |
|
|
do {
|
201 |
|
|
tick();
|
202 |
|
|
} while((m_core->o_busy)||(m_core->o_valid));
|
203 |
|
|
tick();
|
204 |
|
|
}
|
205 |
|
|
|
206 |
|
|
//
|
207 |
|
|
// This is a fairly generic CPU operation call. What makes it less
|
208 |
|
|
// than generic are two things: 1) the ALU is cleared before any
|
209 |
|
|
// new instruction, and 2) the tick count at the end is compared
|
210 |
|
|
// against the tick count OPT_MULTIPLY says we should be getting.
|
211 |
|
|
// A third difference between this call in simulation and a real
|
212 |
|
|
// call within the CPU is that we never set the reset mid-call, whereas
|
213 |
|
|
// the CPU may need to do that if a jump is made and the pipeline needs
|
214 |
|
|
// to be cleared.
|
215 |
|
|
//
|
216 |
209 |
dgisselq |
uint32_t op(int op, int a, int b) {
|
217 |
204 |
dgisselq |
// Make sure we start witht he core idle
|
218 |
|
|
if (m_core->o_valid)
|
219 |
|
|
clear_ops();
|
220 |
|
|
|
221 |
|
|
// Set the arguments to the CPUOPS core to get a multiple
|
222 |
|
|
// started
|
223 |
209 |
dgisselq |
m_core->i_stb = 1;
|
224 |
204 |
dgisselq |
m_core->i_op = op;
|
225 |
|
|
m_core->i_a = a;
|
226 |
|
|
m_core->i_b = b;
|
227 |
|
|
|
228 |
209 |
dgisselq |
uint64_t now = m_tickcount;
|
229 |
204 |
dgisselq |
|
230 |
|
|
// Tick once to get it going
|
231 |
|
|
tick();
|
232 |
|
|
|
233 |
|
|
// Clear the input arguments to the multiply
|
234 |
209 |
dgisselq |
m_core->i_stb = 0;
|
235 |
204 |
dgisselq |
m_core->i_a = 0;
|
236 |
|
|
m_core->i_b = 0;
|
237 |
|
|
|
238 |
|
|
// Wait for the result to be valid
|
239 |
|
|
while(!m_core->o_valid)
|
240 |
|
|
tick();
|
241 |
|
|
|
242 |
|
|
// Check that we used the number of clock ticks we said we'd
|
243 |
|
|
// be using. OPT_MULTIPLY is *supposed* to be equal to this
|
244 |
|
|
// number.
|
245 |
|
|
if((m_tickcount - now)!=OPT_MULTIPLY) {
|
246 |
209 |
dgisselq |
printf("%lld ticks seen, %d ticks expected\n",
|
247 |
|
|
(unsigned long long)(m_tickcount-now), OPT_MULTIPLY);
|
248 |
204 |
dgisselq |
dbgdump();
|
249 |
|
|
printf("TEST-FAILURE!\n");
|
250 |
|
|
closetrace();
|
251 |
|
|
exit(EXIT_FAILURE);
|
252 |
|
|
}
|
253 |
|
|
|
254 |
|
|
return m_core->o_c;
|
255 |
|
|
}
|
256 |
|
|
|
257 |
|
|
//
|
258 |
|
|
// Here's our testing function. Pardon the verbosity of the error
|
259 |
|
|
// messages within it, but ... well, hopefully you won't ever encounter
|
260 |
|
|
// any of those errors. ;)
|
261 |
|
|
//
|
262 |
|
|
// The function works by applying the two inputs to all three of the
|
263 |
|
|
// multiply functions, MPY, MPSHI, and MPYUHI. Results are compared
|
264 |
|
|
// against a local multiply on the local (host) machine. If there's
|
265 |
|
|
// any mismatch, an error message is printed and the test fails.
|
266 |
|
|
void mpy_test(int a, int b) {
|
267 |
209 |
dgisselq |
const int OP_MPY = 0x0c, OP_MPYSHI=0xb, OP_MPYUHI=0x0a;
|
268 |
|
|
const bool debug = false;
|
269 |
|
|
int64_t ia, ib, sv;
|
270 |
|
|
uint64_t ua, ub, uv;
|
271 |
204 |
dgisselq |
unsigned r, s, u;
|
272 |
|
|
|
273 |
|
|
clear_ops();
|
274 |
|
|
|
275 |
209 |
dgisselq |
if (debug)
|
276 |
204 |
dgisselq |
printf("MPY-TEST: 0x%08x x 0x%08x\n", a, b);
|
277 |
|
|
|
278 |
|
|
ia = (long)a; ib = (long)b; sv = ia * ib;
|
279 |
209 |
dgisselq |
ua = ((uint64_t)a)&0x0ffffffffu;
|
280 |
|
|
ub = ((uint64_t)b)&0x0ffffffffu;
|
281 |
204 |
dgisselq |
uv = ua * ub;
|
282 |
|
|
|
283 |
|
|
r = op(OP_MPY, a, b);
|
284 |
|
|
s = op(OP_MPYSHI, a, b);
|
285 |
|
|
u = op(OP_MPYUHI, a, b);
|
286 |
|
|
tick();
|
287 |
|
|
|
288 |
|
|
// Let's check our answers, and see if we got the right results
|
289 |
|
|
if ((r ^ sv)&0x0ffffffffu) {
|
290 |
|
|
printf("TEST FAILURE(MPY), MPY #1\n");
|
291 |
209 |
dgisselq |
printf("Comparing 0x%08x to 0x%016llx\n", r, (long long)sv);
|
292 |
204 |
dgisselq |
printf("TEST-FAILURE!\n");
|
293 |
|
|
closetrace();
|
294 |
|
|
exit(EXIT_FAILURE);
|
295 |
|
|
} if ((r ^ uv)&0x0ffffffffu) {
|
296 |
|
|
printf("TEST FAILURE(MPY), MPY #2\n");
|
297 |
209 |
dgisselq |
printf("Comparing 0x%08x to 0x%016llx\n", r, (unsigned long long)uv);
|
298 |
204 |
dgisselq |
printf("TEST-FAILURE!\n");
|
299 |
|
|
closetrace();
|
300 |
|
|
exit(EXIT_FAILURE);
|
301 |
|
|
}
|
302 |
|
|
|
303 |
|
|
if ((s^(sv>>32))&0x0ffffffffu) {
|
304 |
|
|
printf("TEST FAILURE(MPYSHI), MPY #3\n");
|
305 |
209 |
dgisselq |
printf("Comparing 0x%08x to 0x%016llx\n", s, (long long)sv);
|
306 |
204 |
dgisselq |
printf("TEST-FAILURE!\n");
|
307 |
|
|
closetrace();
|
308 |
|
|
exit(EXIT_FAILURE);
|
309 |
|
|
} if ((u^(uv>>32))&0x0ffffffffu) {
|
310 |
|
|
printf("TEST FAILURE(MPYUHI), MPY #4\n");
|
311 |
209 |
dgisselq |
printf("Comparing 0x%08x to 0x%016llx\n", u, (unsigned long long)uv);
|
312 |
204 |
dgisselq |
printf("TEST-FAILURE!\n");
|
313 |
|
|
closetrace();
|
314 |
|
|
exit(EXIT_FAILURE);
|
315 |
|
|
}
|
316 |
|
|
}
|
317 |
|
|
};
|
318 |
|
|
|
319 |
|
|
void usage(void) {
|
320 |
|
|
printf("USAGE: mpy_tb [a b]\n");
|
321 |
|
|
printf("\n");
|
322 |
|
|
printf(
|
323 |
|
|
"The test is intended to be run with no arguments. When run in this fashion,\n"
|
324 |
|
|
"a series of multiplcation tests will be conducted using all three multiply\n"
|
325 |
|
|
"instructions. Any test failure will terminate the program with an exit\n"
|
326 |
|
|
"condition. Test success will terminate with a clear test condition. \n"
|
327 |
|
|
"During the test, you may expect a large amount of debug output to be\n"
|
328 |
|
|
"produced. This is a normal part of testing. For the meaning of the debug\n"
|
329 |
|
|
"output, please consider the source code. The last line of the debug output,\n"
|
330 |
|
|
"however, will always include either the word \"FAIL\" or \"SUCCESS\"\n"
|
331 |
|
|
"depending on whether the test succeeds or fails.\n\n"
|
332 |
|
|
"If the two arguments a and b are given, they will be interpreted according\n"
|
333 |
|
|
"to the form of strtol, and the test will only involve testing those two\n"
|
334 |
|
|
"parameters\n\n");
|
335 |
|
|
}
|
336 |
|
|
|
337 |
|
|
int main(int argc, char **argv) {
|
338 |
|
|
// Setup verilator
|
339 |
|
|
Verilated::commandArgs(argc, argv);
|
340 |
|
|
// Now, create a test bench.
|
341 |
|
|
CPUOPS_TB *tb = new CPUOPS_TB();
|
342 |
|
|
int rcode = EXIT_SUCCESS;
|
343 |
|
|
// tb->opentrace("mpy_tb.vcd");
|
344 |
|
|
|
345 |
|
|
// Get us started by a couple of clocks past reset. This isn't that
|
346 |
|
|
// unreasonable, since the CPU needs to load up the pipeline before
|
347 |
|
|
// any first instruction will be executed.
|
348 |
|
|
tb->reset();
|
349 |
|
|
tb->tick();
|
350 |
|
|
tb->tick();
|
351 |
|
|
tb->tick();
|
352 |
|
|
|
353 |
|
|
// Look for options, such as '-h'. Trap those here, and produce a usage
|
354 |
|
|
// statement.
|
355 |
|
|
if ((argc > 1)&&(argv[1][0]=='-')&&(isalpha(argv[1][1]))) {
|
356 |
|
|
usage();
|
357 |
|
|
exit(EXIT_SUCCESS);
|
358 |
|
|
}
|
359 |
|
|
|
360 |
|
|
if (argc == 3) {
|
361 |
|
|
// Were we given enough arguments to run a user-specified test?
|
362 |
|
|
tb->mpy_test(
|
363 |
|
|
strtol(argv[1], NULL, 0),
|
364 |
|
|
strtol(argv[2], NULL, 0));
|
365 |
|
|
} else {
|
366 |
|
|
// Otherwise we run through a canned set of tests.
|
367 |
|
|
tb->mpy_test(0,0);
|
368 |
|
|
tb->mpy_test(-1,0);
|
369 |
|
|
tb->mpy_test(-1,-1);
|
370 |
|
|
tb->mpy_test(1,-1);
|
371 |
|
|
tb->mpy_test(1,0);
|
372 |
|
|
tb->mpy_test(0,1);
|
373 |
|
|
tb->mpy_test(1,1);
|
374 |
|
|
|
375 |
|
|
for(int a=0; ((a&0xfff00000)==0); a+=137)
|
376 |
|
|
tb->mpy_test(139, a);
|
377 |
|
|
|
378 |
|
|
for(int a=0; ((a&0x80000000)==0); a+=0x197e2)
|
379 |
|
|
tb->mpy_test(0xf97e27ab, a);
|
380 |
|
|
}
|
381 |
|
|
|
382 |
|
|
printf("SUCCESS!\n");
|
383 |
|
|
exit(rcode);
|
384 |
|
|
}
|
385 |
|
|
|