1 |
330 |
jeremybenn |
/* frv simulator fr550 dependent profiling code.
|
2 |
|
|
|
3 |
|
|
Copyright (C) 2003, 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
|
4 |
|
|
Contributed by Red Hat
|
5 |
|
|
|
6 |
|
|
This file is part of the GNU simulators.
|
7 |
|
|
|
8 |
|
|
This program is free software; you can redistribute it and/or modify
|
9 |
|
|
it under the terms of the GNU General Public License as published by
|
10 |
|
|
the Free Software Foundation; either version 3 of the License, or
|
11 |
|
|
(at your option) any later version.
|
12 |
|
|
|
13 |
|
|
This program is distributed in the hope that it will be useful,
|
14 |
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
15 |
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
16 |
|
|
GNU General Public License for more details.
|
17 |
|
|
|
18 |
|
|
You should have received a copy of the GNU General Public License
|
19 |
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
20 |
|
|
|
21 |
|
|
*/
|
22 |
|
|
#define WANT_CPU
|
23 |
|
|
#define WANT_CPU_FRVBF
|
24 |
|
|
|
25 |
|
|
#include "sim-main.h"
|
26 |
|
|
#include "bfd.h"
|
27 |
|
|
|
28 |
|
|
#if WITH_PROFILE_MODEL_P
|
29 |
|
|
|
30 |
|
|
#include "profile.h"
|
31 |
|
|
#include "profile-fr550.h"
|
32 |
|
|
|
33 |
|
|
/* Initialize cycle counting for an insn.
|
34 |
|
|
FIRST_P is non-zero if this is the first insn in a set of parallel
|
35 |
|
|
insns. */
|
36 |
|
|
void
|
37 |
|
|
fr550_model_insn_before (SIM_CPU *cpu, int first_p)
|
38 |
|
|
{
|
39 |
|
|
if (first_p)
|
40 |
|
|
{
|
41 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
42 |
|
|
d->cur_fr_load = d->prev_fr_load;
|
43 |
|
|
d->cur_fr_complex_1 = d->prev_fr_complex_1;
|
44 |
|
|
d->cur_fr_complex_2 = d->prev_fr_complex_2;
|
45 |
|
|
d->cur_ccr_complex = d->prev_ccr_complex;
|
46 |
|
|
d->cur_acc_mmac = d->prev_acc_mmac;
|
47 |
|
|
}
|
48 |
|
|
}
|
49 |
|
|
|
50 |
|
|
/* Record the cycles computed for an insn.
|
51 |
|
|
LAST_P is non-zero if this is the last insn in a set of parallel insns,
|
52 |
|
|
and we update the total cycle count.
|
53 |
|
|
CYCLES is the cycle count of the insn. */
|
54 |
|
|
void
|
55 |
|
|
fr550_model_insn_after (SIM_CPU *cpu, int last_p, int cycles)
|
56 |
|
|
{
|
57 |
|
|
if (last_p)
|
58 |
|
|
{
|
59 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
60 |
|
|
d->prev_fr_load = d->cur_fr_load;
|
61 |
|
|
d->prev_fr_complex_1 = d->cur_fr_complex_1;
|
62 |
|
|
d->prev_fr_complex_2 = d->cur_fr_complex_2;
|
63 |
|
|
d->prev_ccr_complex = d->cur_ccr_complex;
|
64 |
|
|
d->prev_acc_mmac = d->cur_acc_mmac;
|
65 |
|
|
}
|
66 |
|
|
}
|
67 |
|
|
|
68 |
|
|
static void fr550_reset_fr_flags (SIM_CPU *cpu, INT fr);
|
69 |
|
|
static void fr550_reset_ccr_flags (SIM_CPU *cpu, INT ccr);
|
70 |
|
|
static void fr550_reset_acc_flags (SIM_CPU *cpu, INT acc);
|
71 |
|
|
|
72 |
|
|
static void
|
73 |
|
|
set_use_is_fr_load (SIM_CPU *cpu, INT fr)
|
74 |
|
|
{
|
75 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
76 |
|
|
fr550_reset_fr_flags (cpu, (fr));
|
77 |
|
|
d->cur_fr_load |= (((DI)1) << (fr));
|
78 |
|
|
}
|
79 |
|
|
|
80 |
|
|
static void
|
81 |
|
|
set_use_not_fr_load (SIM_CPU *cpu, INT fr)
|
82 |
|
|
{
|
83 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
84 |
|
|
d->cur_fr_load &= ~(((DI)1) << (fr));
|
85 |
|
|
}
|
86 |
|
|
|
87 |
|
|
static int
|
88 |
|
|
use_is_fr_load (SIM_CPU *cpu, INT fr)
|
89 |
|
|
{
|
90 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
91 |
|
|
return d->prev_fr_load & (((DI)1) << (fr));
|
92 |
|
|
}
|
93 |
|
|
|
94 |
|
|
static void
|
95 |
|
|
set_use_is_fr_complex_1 (SIM_CPU *cpu, INT fr)
|
96 |
|
|
{
|
97 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
98 |
|
|
fr550_reset_fr_flags (cpu, (fr));
|
99 |
|
|
d->cur_fr_complex_1 |= (((DI)1) << (fr));
|
100 |
|
|
}
|
101 |
|
|
|
102 |
|
|
static void
|
103 |
|
|
set_use_not_fr_complex_1 (SIM_CPU *cpu, INT fr)
|
104 |
|
|
{
|
105 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
106 |
|
|
d->cur_fr_complex_1 &= ~(((DI)1) << (fr));
|
107 |
|
|
}
|
108 |
|
|
|
109 |
|
|
static int
|
110 |
|
|
use_is_fr_complex_1 (SIM_CPU *cpu, INT fr)
|
111 |
|
|
{
|
112 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
113 |
|
|
return d->prev_fr_complex_1 & (((DI)1) << (fr));
|
114 |
|
|
}
|
115 |
|
|
|
116 |
|
|
static void
|
117 |
|
|
set_use_is_fr_complex_2 (SIM_CPU *cpu, INT fr)
|
118 |
|
|
{
|
119 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
120 |
|
|
fr550_reset_fr_flags (cpu, (fr));
|
121 |
|
|
d->cur_fr_complex_2 |= (((DI)1) << (fr));
|
122 |
|
|
}
|
123 |
|
|
|
124 |
|
|
static void
|
125 |
|
|
set_use_not_fr_complex_2 (SIM_CPU *cpu, INT fr)
|
126 |
|
|
{
|
127 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
128 |
|
|
d->cur_fr_complex_2 &= ~(((DI)1) << (fr));
|
129 |
|
|
}
|
130 |
|
|
|
131 |
|
|
static int
|
132 |
|
|
use_is_fr_complex_2 (SIM_CPU *cpu, INT fr)
|
133 |
|
|
{
|
134 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
135 |
|
|
return d->prev_fr_complex_2 & (((DI)1) << (fr));
|
136 |
|
|
}
|
137 |
|
|
|
138 |
|
|
static void
|
139 |
|
|
set_use_is_ccr_complex (SIM_CPU *cpu, INT ccr)
|
140 |
|
|
{
|
141 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
142 |
|
|
fr550_reset_ccr_flags (cpu, (ccr));
|
143 |
|
|
d->cur_ccr_complex |= (((SI)1) << (ccr));
|
144 |
|
|
}
|
145 |
|
|
|
146 |
|
|
static void
|
147 |
|
|
set_use_not_ccr_complex (SIM_CPU *cpu, INT ccr)
|
148 |
|
|
{
|
149 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
150 |
|
|
d->cur_ccr_complex &= ~(((SI)1) << (ccr));
|
151 |
|
|
}
|
152 |
|
|
|
153 |
|
|
static int
|
154 |
|
|
use_is_ccr_complex (SIM_CPU *cpu, INT ccr)
|
155 |
|
|
{
|
156 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
157 |
|
|
return d->prev_ccr_complex & (((SI)1) << (ccr));
|
158 |
|
|
}
|
159 |
|
|
|
160 |
|
|
static void
|
161 |
|
|
set_use_is_acc_mmac (SIM_CPU *cpu, INT acc)
|
162 |
|
|
{
|
163 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
164 |
|
|
fr550_reset_acc_flags (cpu, (acc));
|
165 |
|
|
d->cur_acc_mmac |= (((DI)1) << (acc));
|
166 |
|
|
}
|
167 |
|
|
|
168 |
|
|
static void
|
169 |
|
|
set_use_not_acc_mmac (SIM_CPU *cpu, INT acc)
|
170 |
|
|
{
|
171 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
172 |
|
|
d->cur_acc_mmac &= ~(((DI)1) << (acc));
|
173 |
|
|
}
|
174 |
|
|
|
175 |
|
|
static int
|
176 |
|
|
use_is_acc_mmac (SIM_CPU *cpu, INT acc)
|
177 |
|
|
{
|
178 |
|
|
MODEL_FR550_DATA *d = CPU_MODEL_DATA (cpu);
|
179 |
|
|
return d->prev_acc_mmac & (((DI)1) << (acc));
|
180 |
|
|
}
|
181 |
|
|
|
182 |
|
|
static void
|
183 |
|
|
fr550_reset_fr_flags (SIM_CPU *cpu, INT fr)
|
184 |
|
|
{
|
185 |
|
|
set_use_not_fr_load (cpu, fr);
|
186 |
|
|
set_use_not_fr_complex_1 (cpu, fr);
|
187 |
|
|
set_use_not_fr_complex_2 (cpu, fr);
|
188 |
|
|
}
|
189 |
|
|
|
190 |
|
|
static void
|
191 |
|
|
fr550_reset_ccr_flags (SIM_CPU *cpu, INT ccr)
|
192 |
|
|
{
|
193 |
|
|
set_use_not_ccr_complex (cpu, ccr);
|
194 |
|
|
}
|
195 |
|
|
|
196 |
|
|
static void
|
197 |
|
|
fr550_reset_acc_flags (SIM_CPU *cpu, INT acc)
|
198 |
|
|
{
|
199 |
|
|
set_use_not_acc_mmac (cpu, acc);
|
200 |
|
|
}
|
201 |
|
|
|
202 |
|
|
/* Detect overlap between two register ranges. Works if one of the registers
|
203 |
|
|
is -1 with width 1 (i.e. undefined), but not both. */
|
204 |
|
|
#define REG_OVERLAP(r1, w1, r2, w2) ( \
|
205 |
|
|
(r1) + (w1) - 1 >= (r2) && (r2) + (w2) - 1 >= (r1) \
|
206 |
|
|
)
|
207 |
|
|
|
208 |
|
|
/* Latency of floating point registers may be less than recorded when followed
|
209 |
|
|
by another floating point insn. */
|
210 |
|
|
static void
|
211 |
|
|
adjust_float_register_busy (SIM_CPU *cpu,
|
212 |
|
|
INT in_FRi, int iwidth,
|
213 |
|
|
INT in_FRj, int jwidth,
|
214 |
|
|
INT out_FRk, int kwidth)
|
215 |
|
|
{
|
216 |
|
|
int i;
|
217 |
|
|
/* The latency of FRk may be less than previously recorded.
|
218 |
|
|
See Table 14-15 in the LSI. */
|
219 |
|
|
if (in_FRi >= 0)
|
220 |
|
|
{
|
221 |
|
|
for (i = 0; i < iwidth; ++i)
|
222 |
|
|
{
|
223 |
|
|
if (! REG_OVERLAP (in_FRi + i, 1, out_FRk, kwidth))
|
224 |
|
|
if (use_is_fr_load (cpu, in_FRi + i))
|
225 |
|
|
decrease_FR_busy (cpu, in_FRi + i, 1);
|
226 |
|
|
else
|
227 |
|
|
enforce_full_fr_latency (cpu, in_FRi + i);
|
228 |
|
|
}
|
229 |
|
|
}
|
230 |
|
|
|
231 |
|
|
if (in_FRj >= 0)
|
232 |
|
|
{
|
233 |
|
|
for (i = 0; i < jwidth; ++i)
|
234 |
|
|
{
|
235 |
|
|
if (! REG_OVERLAP (in_FRj + i, 1, in_FRi, iwidth)
|
236 |
|
|
&& ! REG_OVERLAP (in_FRj + i, 1, out_FRk, kwidth))
|
237 |
|
|
if (use_is_fr_load (cpu, in_FRj + i))
|
238 |
|
|
decrease_FR_busy (cpu, in_FRj + i, 1);
|
239 |
|
|
else
|
240 |
|
|
enforce_full_fr_latency (cpu, in_FRj + i);
|
241 |
|
|
}
|
242 |
|
|
}
|
243 |
|
|
|
244 |
|
|
if (out_FRk >= 0)
|
245 |
|
|
{
|
246 |
|
|
for (i = 0; i < kwidth; ++i)
|
247 |
|
|
{
|
248 |
|
|
if (! REG_OVERLAP (out_FRk + i, 1, in_FRi, iwidth)
|
249 |
|
|
&& ! REG_OVERLAP (out_FRk + i, 1, in_FRj, jwidth))
|
250 |
|
|
{
|
251 |
|
|
if (use_is_fr_complex_1 (cpu, out_FRk + i))
|
252 |
|
|
decrease_FR_busy (cpu, out_FRk + i, 1);
|
253 |
|
|
else if (use_is_fr_complex_2 (cpu, out_FRk + i))
|
254 |
|
|
decrease_FR_busy (cpu, out_FRk + i, 2);
|
255 |
|
|
else
|
256 |
|
|
enforce_full_fr_latency (cpu, out_FRk + i);
|
257 |
|
|
}
|
258 |
|
|
}
|
259 |
|
|
}
|
260 |
|
|
}
|
261 |
|
|
|
262 |
|
|
static void
|
263 |
|
|
restore_float_register_busy (SIM_CPU *cpu,
|
264 |
|
|
INT in_FRi, int iwidth,
|
265 |
|
|
INT in_FRj, int jwidth,
|
266 |
|
|
INT out_FRk, int kwidth)
|
267 |
|
|
{
|
268 |
|
|
int i;
|
269 |
|
|
/* The latency of FRk may be less than previously recorded.
|
270 |
|
|
See Table 14-15 in the LSI. */
|
271 |
|
|
if (in_FRi >= 0)
|
272 |
|
|
{
|
273 |
|
|
for (i = 0; i < iwidth; ++i)
|
274 |
|
|
{
|
275 |
|
|
if (! REG_OVERLAP (in_FRi + i, 1, out_FRk, kwidth))
|
276 |
|
|
if (use_is_fr_load (cpu, in_FRi + i))
|
277 |
|
|
increase_FR_busy (cpu, in_FRi + i, 1);
|
278 |
|
|
}
|
279 |
|
|
}
|
280 |
|
|
|
281 |
|
|
if (in_FRj >= 0)
|
282 |
|
|
{
|
283 |
|
|
for (i = 0; i < jwidth; ++i)
|
284 |
|
|
{
|
285 |
|
|
if (! REG_OVERLAP (in_FRj + i, 1, in_FRi, iwidth)
|
286 |
|
|
&& ! REG_OVERLAP (in_FRj + i, 1, out_FRk, kwidth))
|
287 |
|
|
if (use_is_fr_load (cpu, in_FRj + i))
|
288 |
|
|
increase_FR_busy (cpu, in_FRj + i, 1);
|
289 |
|
|
}
|
290 |
|
|
}
|
291 |
|
|
|
292 |
|
|
if (out_FRk >= 0)
|
293 |
|
|
{
|
294 |
|
|
for (i = 0; i < kwidth; ++i)
|
295 |
|
|
{
|
296 |
|
|
if (! REG_OVERLAP (out_FRk + i, 1, in_FRi, iwidth)
|
297 |
|
|
&& ! REG_OVERLAP (out_FRk + i, 1, in_FRj, jwidth))
|
298 |
|
|
{
|
299 |
|
|
if (use_is_fr_complex_1 (cpu, out_FRk + i))
|
300 |
|
|
increase_FR_busy (cpu, out_FRk + i, 1);
|
301 |
|
|
else if (use_is_fr_complex_2 (cpu, out_FRk + i))
|
302 |
|
|
increase_FR_busy (cpu, out_FRk + i, 2);
|
303 |
|
|
}
|
304 |
|
|
}
|
305 |
|
|
}
|
306 |
|
|
}
|
307 |
|
|
|
308 |
|
|
/* Latency of floating point registers may be less than recorded when used in a
|
309 |
|
|
media insns and followed by another media insn. */
|
310 |
|
|
static void
|
311 |
|
|
adjust_float_register_busy_for_media (SIM_CPU *cpu,
|
312 |
|
|
INT in_FRi, int iwidth,
|
313 |
|
|
INT in_FRj, int jwidth,
|
314 |
|
|
INT out_FRk, int kwidth)
|
315 |
|
|
{
|
316 |
|
|
int i;
|
317 |
|
|
/* The latency of FRk may be less than previously recorded.
|
318 |
|
|
See Table 14-15 in the LSI. */
|
319 |
|
|
if (out_FRk >= 0)
|
320 |
|
|
{
|
321 |
|
|
for (i = 0; i < kwidth; ++i)
|
322 |
|
|
{
|
323 |
|
|
if (! REG_OVERLAP (out_FRk + i, 1, in_FRi, iwidth)
|
324 |
|
|
&& ! REG_OVERLAP (out_FRk + i, 1, in_FRj, jwidth))
|
325 |
|
|
{
|
326 |
|
|
if (use_is_fr_complex_1 (cpu, out_FRk + i))
|
327 |
|
|
decrease_FR_busy (cpu, out_FRk + i, 1);
|
328 |
|
|
else
|
329 |
|
|
enforce_full_fr_latency (cpu, out_FRk + i);
|
330 |
|
|
}
|
331 |
|
|
}
|
332 |
|
|
}
|
333 |
|
|
}
|
334 |
|
|
|
335 |
|
|
static void
|
336 |
|
|
restore_float_register_busy_for_media (SIM_CPU *cpu,
|
337 |
|
|
INT in_FRi, int iwidth,
|
338 |
|
|
INT in_FRj, int jwidth,
|
339 |
|
|
INT out_FRk, int kwidth)
|
340 |
|
|
{
|
341 |
|
|
int i;
|
342 |
|
|
if (out_FRk >= 0)
|
343 |
|
|
{
|
344 |
|
|
for (i = 0; i < kwidth; ++i)
|
345 |
|
|
{
|
346 |
|
|
if (! REG_OVERLAP (out_FRk + i, 1, in_FRi, iwidth)
|
347 |
|
|
&& ! REG_OVERLAP (out_FRk + i, 1, in_FRj, jwidth))
|
348 |
|
|
{
|
349 |
|
|
if (use_is_fr_complex_1 (cpu, out_FRk + i))
|
350 |
|
|
increase_FR_busy (cpu, out_FRk + i, 1);
|
351 |
|
|
}
|
352 |
|
|
}
|
353 |
|
|
}
|
354 |
|
|
}
|
355 |
|
|
|
356 |
|
|
/* Latency of accumulator registers may be less than recorded when used in a
|
357 |
|
|
media insns and followed by another media insn. */
|
358 |
|
|
static void
|
359 |
|
|
adjust_acc_busy_for_mmac (SIM_CPU *cpu,
|
360 |
|
|
INT in_ACC, int inwidth,
|
361 |
|
|
INT out_ACC, int outwidth)
|
362 |
|
|
{
|
363 |
|
|
int i;
|
364 |
|
|
/* The latency of an accumulator may be less than previously recorded.
|
365 |
|
|
See Table 14-15 in the LSI. */
|
366 |
|
|
if (in_ACC >= 0)
|
367 |
|
|
{
|
368 |
|
|
for (i = 0; i < inwidth; ++i)
|
369 |
|
|
{
|
370 |
|
|
if (use_is_acc_mmac (cpu, in_ACC + i))
|
371 |
|
|
decrease_ACC_busy (cpu, in_ACC + i, 1);
|
372 |
|
|
else
|
373 |
|
|
enforce_full_acc_latency (cpu, in_ACC + i);
|
374 |
|
|
}
|
375 |
|
|
}
|
376 |
|
|
if (out_ACC >= 0)
|
377 |
|
|
{
|
378 |
|
|
for (i = 0; i < outwidth; ++i)
|
379 |
|
|
{
|
380 |
|
|
if (! REG_OVERLAP (out_ACC + i, 1, in_ACC, inwidth))
|
381 |
|
|
{
|
382 |
|
|
if (use_is_acc_mmac (cpu, out_ACC + i))
|
383 |
|
|
decrease_ACC_busy (cpu, out_ACC + i, 1);
|
384 |
|
|
else
|
385 |
|
|
enforce_full_acc_latency (cpu, out_ACC + i);
|
386 |
|
|
}
|
387 |
|
|
}
|
388 |
|
|
}
|
389 |
|
|
}
|
390 |
|
|
|
391 |
|
|
static void
|
392 |
|
|
restore_acc_busy_for_mmac (SIM_CPU *cpu,
|
393 |
|
|
INT in_ACC, int inwidth,
|
394 |
|
|
INT out_ACC, int outwidth)
|
395 |
|
|
{
|
396 |
|
|
int i;
|
397 |
|
|
if (in_ACC >= 0)
|
398 |
|
|
{
|
399 |
|
|
for (i = 0; i < inwidth; ++i)
|
400 |
|
|
{
|
401 |
|
|
if (use_is_acc_mmac (cpu, in_ACC + i))
|
402 |
|
|
increase_ACC_busy (cpu, in_ACC + i, 1);
|
403 |
|
|
}
|
404 |
|
|
}
|
405 |
|
|
if (out_ACC >= 0)
|
406 |
|
|
{
|
407 |
|
|
for (i = 0; i < outwidth; ++i)
|
408 |
|
|
{
|
409 |
|
|
if (! REG_OVERLAP (out_ACC + i, 1, in_ACC, inwidth))
|
410 |
|
|
{
|
411 |
|
|
if (use_is_acc_mmac (cpu, out_ACC + i))
|
412 |
|
|
increase_ACC_busy (cpu, out_ACC + i, 1);
|
413 |
|
|
}
|
414 |
|
|
}
|
415 |
|
|
}
|
416 |
|
|
}
|
417 |
|
|
|
418 |
|
|
int
|
419 |
|
|
frvbf_model_fr550_u_exec (SIM_CPU *cpu, const IDESC *idesc,
|
420 |
|
|
int unit_num, int referenced)
|
421 |
|
|
{
|
422 |
|
|
return idesc->timing->units[unit_num].done;
|
423 |
|
|
}
|
424 |
|
|
|
425 |
|
|
int
|
426 |
|
|
frvbf_model_fr550_u_integer (SIM_CPU *cpu, const IDESC *idesc,
|
427 |
|
|
int unit_num, int referenced,
|
428 |
|
|
INT in_GRi, INT in_GRj, INT out_GRk,
|
429 |
|
|
INT out_ICCi_1)
|
430 |
|
|
{
|
431 |
|
|
int cycles;
|
432 |
|
|
|
433 |
|
|
/* icc0-icc4 are the upper 4 fields of the CCR. */
|
434 |
|
|
if (out_ICCi_1 >= 0)
|
435 |
|
|
out_ICCi_1 += 4;
|
436 |
|
|
|
437 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
438 |
|
|
{
|
439 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
440 |
|
|
which is not ready yet. */
|
441 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
442 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
443 |
|
|
vliw_wait_for_GR (cpu, out_GRk);
|
444 |
|
|
vliw_wait_for_CCR (cpu, out_ICCi_1);
|
445 |
|
|
handle_resource_wait (cpu);
|
446 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
447 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
448 |
|
|
load_wait_for_GR (cpu, out_GRk);
|
449 |
|
|
trace_vliw_wait_cycles (cpu);
|
450 |
|
|
return 0;
|
451 |
|
|
}
|
452 |
|
|
|
453 |
|
|
fr550_reset_ccr_flags (cpu, out_ICCi_1);
|
454 |
|
|
|
455 |
|
|
/* GRk is available immediately to the next VLIW insn as is ICCi_1. */
|
456 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
457 |
|
|
return cycles;
|
458 |
|
|
}
|
459 |
|
|
|
460 |
|
|
int
|
461 |
|
|
frvbf_model_fr550_u_imul (SIM_CPU *cpu, const IDESC *idesc,
|
462 |
|
|
int unit_num, int referenced,
|
463 |
|
|
INT in_GRi, INT in_GRj, INT out_GRk, INT out_ICCi_1)
|
464 |
|
|
{
|
465 |
|
|
int cycles;
|
466 |
|
|
/* icc0-icc4 are the upper 4 fields of the CCR. */
|
467 |
|
|
if (out_ICCi_1 >= 0)
|
468 |
|
|
out_ICCi_1 += 4;
|
469 |
|
|
|
470 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
471 |
|
|
{
|
472 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
473 |
|
|
which is not ready yet. */
|
474 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
475 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
476 |
|
|
vliw_wait_for_GRdouble (cpu, out_GRk);
|
477 |
|
|
vliw_wait_for_CCR (cpu, out_ICCi_1);
|
478 |
|
|
handle_resource_wait (cpu);
|
479 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
480 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
481 |
|
|
load_wait_for_GRdouble (cpu, out_GRk);
|
482 |
|
|
trace_vliw_wait_cycles (cpu);
|
483 |
|
|
return 0;
|
484 |
|
|
}
|
485 |
|
|
|
486 |
|
|
/* GRk has a latency of 1 cycles. */
|
487 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
488 |
|
|
update_GRdouble_latency (cpu, out_GRk, cycles + 1);
|
489 |
|
|
|
490 |
|
|
/* ICCi_1 has a latency of 1 cycle. */
|
491 |
|
|
update_CCR_latency (cpu, out_ICCi_1, cycles + 1);
|
492 |
|
|
|
493 |
|
|
fr550_reset_ccr_flags (cpu, out_ICCi_1);
|
494 |
|
|
|
495 |
|
|
return cycles;
|
496 |
|
|
}
|
497 |
|
|
|
498 |
|
|
int
|
499 |
|
|
frvbf_model_fr550_u_idiv (SIM_CPU *cpu, const IDESC *idesc,
|
500 |
|
|
int unit_num, int referenced,
|
501 |
|
|
INT in_GRi, INT in_GRj, INT out_GRk, INT out_ICCi_1)
|
502 |
|
|
{
|
503 |
|
|
int cycles;
|
504 |
|
|
FRV_VLIW *vliw;
|
505 |
|
|
int slot;
|
506 |
|
|
|
507 |
|
|
/* icc0-icc4 are the upper 4 fields of the CCR. */
|
508 |
|
|
if (out_ICCi_1 >= 0)
|
509 |
|
|
out_ICCi_1 += 4;
|
510 |
|
|
|
511 |
|
|
vliw = CPU_VLIW (cpu);
|
512 |
|
|
slot = vliw->next_slot - 1;
|
513 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_I0;
|
514 |
|
|
|
515 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
516 |
|
|
{
|
517 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
518 |
|
|
which is not ready yet. */
|
519 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
520 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
521 |
|
|
vliw_wait_for_GR (cpu, out_GRk);
|
522 |
|
|
vliw_wait_for_CCR (cpu, out_ICCi_1);
|
523 |
|
|
vliw_wait_for_idiv_resource (cpu, slot);
|
524 |
|
|
handle_resource_wait (cpu);
|
525 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
526 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
527 |
|
|
load_wait_for_GR (cpu, out_GRk);
|
528 |
|
|
trace_vliw_wait_cycles (cpu);
|
529 |
|
|
return 0;
|
530 |
|
|
}
|
531 |
|
|
|
532 |
|
|
/* GRk has a latency of 18 cycles! */
|
533 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
534 |
|
|
update_GR_latency (cpu, out_GRk, cycles + 18);
|
535 |
|
|
|
536 |
|
|
/* ICCi_1 has a latency of 18 cycles. */
|
537 |
|
|
update_CCR_latency (cpu, out_ICCi_1, cycles + 18);
|
538 |
|
|
|
539 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
540 |
|
|
{
|
541 |
|
|
/* GNER has a latency of 18 cycles. */
|
542 |
|
|
update_SPR_latency (cpu, GNER_FOR_GR (out_GRk), cycles + 18);
|
543 |
|
|
}
|
544 |
|
|
|
545 |
|
|
/* the idiv resource has a latency of 18 cycles! */
|
546 |
|
|
update_idiv_resource_latency (cpu, slot, cycles + 18);
|
547 |
|
|
|
548 |
|
|
fr550_reset_ccr_flags (cpu, out_ICCi_1);
|
549 |
|
|
|
550 |
|
|
return cycles;
|
551 |
|
|
}
|
552 |
|
|
|
553 |
|
|
int
|
554 |
|
|
frvbf_model_fr550_u_branch (SIM_CPU *cpu, const IDESC *idesc,
|
555 |
|
|
int unit_num, int referenced,
|
556 |
|
|
INT in_GRi, INT in_GRj,
|
557 |
|
|
INT in_ICCi_2, INT in_FCCi_2)
|
558 |
|
|
{
|
559 |
|
|
int cycles;
|
560 |
|
|
FRV_PROFILE_STATE *ps;
|
561 |
|
|
|
562 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
563 |
|
|
{
|
564 |
|
|
/* icc0-icc4 are the upper 4 fields of the CCR. */
|
565 |
|
|
if (in_ICCi_2 >= 0)
|
566 |
|
|
in_ICCi_2 += 4;
|
567 |
|
|
|
568 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
569 |
|
|
which is not ready yet. */
|
570 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
571 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
572 |
|
|
vliw_wait_for_CCR (cpu, in_ICCi_2);
|
573 |
|
|
vliw_wait_for_CCR (cpu, in_FCCi_2);
|
574 |
|
|
handle_resource_wait (cpu);
|
575 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
576 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
577 |
|
|
trace_vliw_wait_cycles (cpu);
|
578 |
|
|
return 0;
|
579 |
|
|
}
|
580 |
|
|
|
581 |
|
|
/* When counting branches taken or not taken, don't consider branches after
|
582 |
|
|
the first taken branch in a vliw insn. */
|
583 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
584 |
|
|
if (! ps->vliw_branch_taken)
|
585 |
|
|
{
|
586 |
|
|
/* (1 << 4): The pc is the 5th element in inputs, outputs.
|
587 |
|
|
??? can be cleaned up */
|
588 |
|
|
PROFILE_DATA *p = CPU_PROFILE_DATA (cpu);
|
589 |
|
|
int taken = (referenced & (1 << 4)) != 0;
|
590 |
|
|
if (taken)
|
591 |
|
|
{
|
592 |
|
|
++PROFILE_MODEL_TAKEN_COUNT (p);
|
593 |
|
|
ps->vliw_branch_taken = 1;
|
594 |
|
|
}
|
595 |
|
|
else
|
596 |
|
|
++PROFILE_MODEL_UNTAKEN_COUNT (p);
|
597 |
|
|
}
|
598 |
|
|
|
599 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
600 |
|
|
return cycles;
|
601 |
|
|
}
|
602 |
|
|
|
603 |
|
|
int
|
604 |
|
|
frvbf_model_fr550_u_trap (SIM_CPU *cpu, const IDESC *idesc,
|
605 |
|
|
int unit_num, int referenced,
|
606 |
|
|
INT in_GRi, INT in_GRj,
|
607 |
|
|
INT in_ICCi_2, INT in_FCCi_2)
|
608 |
|
|
{
|
609 |
|
|
int cycles;
|
610 |
|
|
|
611 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
612 |
|
|
{
|
613 |
|
|
/* icc0-icc4 are the upper 4 fields of the CCR. */
|
614 |
|
|
if (in_ICCi_2 >= 0)
|
615 |
|
|
in_ICCi_2 += 4;
|
616 |
|
|
|
617 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
618 |
|
|
which is not ready yet. */
|
619 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
620 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
621 |
|
|
vliw_wait_for_CCR (cpu, in_ICCi_2);
|
622 |
|
|
vliw_wait_for_CCR (cpu, in_FCCi_2);
|
623 |
|
|
handle_resource_wait (cpu);
|
624 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
625 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
626 |
|
|
trace_vliw_wait_cycles (cpu);
|
627 |
|
|
return 0;
|
628 |
|
|
}
|
629 |
|
|
|
630 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
631 |
|
|
return cycles;
|
632 |
|
|
}
|
633 |
|
|
|
634 |
|
|
int
|
635 |
|
|
frvbf_model_fr550_u_check (SIM_CPU *cpu, const IDESC *idesc,
|
636 |
|
|
int unit_num, int referenced,
|
637 |
|
|
INT in_ICCi_3, INT in_FCCi_3)
|
638 |
|
|
{
|
639 |
|
|
/* Modelling for this unit is the same as for fr500. */
|
640 |
|
|
return frvbf_model_fr500_u_check (cpu, idesc, unit_num, referenced,
|
641 |
|
|
in_ICCi_3, in_FCCi_3);
|
642 |
|
|
}
|
643 |
|
|
|
644 |
|
|
int
|
645 |
|
|
frvbf_model_fr550_u_set_hilo (SIM_CPU *cpu, const IDESC *idesc,
|
646 |
|
|
int unit_num, int referenced,
|
647 |
|
|
INT out_GRkhi, INT out_GRklo)
|
648 |
|
|
{
|
649 |
|
|
int cycles;
|
650 |
|
|
|
651 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
652 |
|
|
{
|
653 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a GR
|
654 |
|
|
which is not ready yet. */
|
655 |
|
|
vliw_wait_for_GR (cpu, out_GRkhi);
|
656 |
|
|
vliw_wait_for_GR (cpu, out_GRklo);
|
657 |
|
|
handle_resource_wait (cpu);
|
658 |
|
|
load_wait_for_GR (cpu, out_GRkhi);
|
659 |
|
|
load_wait_for_GR (cpu, out_GRklo);
|
660 |
|
|
trace_vliw_wait_cycles (cpu);
|
661 |
|
|
return 0;
|
662 |
|
|
}
|
663 |
|
|
|
664 |
|
|
/* GRk is available immediately to the next VLIW insn. */
|
665 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
666 |
|
|
|
667 |
|
|
return cycles;
|
668 |
|
|
}
|
669 |
|
|
|
670 |
|
|
int
|
671 |
|
|
frvbf_model_fr550_u_gr_load (SIM_CPU *cpu, const IDESC *idesc,
|
672 |
|
|
int unit_num, int referenced,
|
673 |
|
|
INT in_GRi, INT in_GRj,
|
674 |
|
|
INT out_GRk, INT out_GRdoublek)
|
675 |
|
|
{
|
676 |
|
|
int cycles;
|
677 |
|
|
|
678 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
679 |
|
|
{
|
680 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
681 |
|
|
which is not ready yet. */
|
682 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
683 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
684 |
|
|
vliw_wait_for_GR (cpu, out_GRk);
|
685 |
|
|
vliw_wait_for_GRdouble (cpu, out_GRdoublek);
|
686 |
|
|
handle_resource_wait (cpu);
|
687 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
688 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
689 |
|
|
load_wait_for_GR (cpu, out_GRk);
|
690 |
|
|
load_wait_for_GRdouble (cpu, out_GRdoublek);
|
691 |
|
|
trace_vliw_wait_cycles (cpu);
|
692 |
|
|
return 0;
|
693 |
|
|
}
|
694 |
|
|
|
695 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
696 |
|
|
|
697 |
|
|
/* The latency of GRk for a load will depend on how long it takes to retrieve
|
698 |
|
|
the the data from the cache or memory. */
|
699 |
|
|
update_GR_latency_for_load (cpu, out_GRk, cycles);
|
700 |
|
|
update_GRdouble_latency_for_load (cpu, out_GRdoublek, cycles);
|
701 |
|
|
|
702 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
703 |
|
|
{
|
704 |
|
|
/* GNER has a latency of 2 cycles. */
|
705 |
|
|
update_SPR_latency (cpu, GNER_FOR_GR (out_GRk), cycles + 2);
|
706 |
|
|
update_SPR_latency (cpu, GNER_FOR_GR (out_GRdoublek), cycles + 2);
|
707 |
|
|
}
|
708 |
|
|
|
709 |
|
|
return cycles;
|
710 |
|
|
}
|
711 |
|
|
|
712 |
|
|
int
|
713 |
|
|
frvbf_model_fr550_u_gr_store (SIM_CPU *cpu, const IDESC *idesc,
|
714 |
|
|
int unit_num, int referenced,
|
715 |
|
|
INT in_GRi, INT in_GRj,
|
716 |
|
|
INT in_GRk, INT in_GRdoublek)
|
717 |
|
|
{
|
718 |
|
|
int cycles;
|
719 |
|
|
|
720 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
721 |
|
|
{
|
722 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
723 |
|
|
which is not ready yet. */
|
724 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
725 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
726 |
|
|
vliw_wait_for_GR (cpu, in_GRk);
|
727 |
|
|
vliw_wait_for_GRdouble (cpu, in_GRdoublek);
|
728 |
|
|
handle_resource_wait (cpu);
|
729 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
730 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
731 |
|
|
load_wait_for_GR (cpu, in_GRk);
|
732 |
|
|
load_wait_for_GRdouble (cpu, in_GRdoublek);
|
733 |
|
|
trace_vliw_wait_cycles (cpu);
|
734 |
|
|
return 0;
|
735 |
|
|
}
|
736 |
|
|
|
737 |
|
|
/* The target register is available immediately. */
|
738 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
739 |
|
|
|
740 |
|
|
return cycles;
|
741 |
|
|
}
|
742 |
|
|
|
743 |
|
|
int
|
744 |
|
|
frvbf_model_fr550_u_fr_load (SIM_CPU *cpu, const IDESC *idesc,
|
745 |
|
|
int unit_num, int referenced,
|
746 |
|
|
INT in_GRi, INT in_GRj,
|
747 |
|
|
INT out_FRk, INT out_FRdoublek)
|
748 |
|
|
{
|
749 |
|
|
int cycles;
|
750 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
751 |
|
|
{
|
752 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
753 |
|
|
which is not ready yet.
|
754 |
|
|
The latency of the registers may be less than previously recorded,
|
755 |
|
|
depending on how they were used previously.
|
756 |
|
|
See Table 13-8 in the LSI. */
|
757 |
|
|
adjust_float_register_busy (cpu, -1, 1, -1, 1, out_FRk, 1);
|
758 |
|
|
adjust_float_register_busy (cpu, -1, 1, -1, 1, out_FRdoublek, 2);
|
759 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
760 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
761 |
|
|
vliw_wait_for_FR (cpu, out_FRk);
|
762 |
|
|
vliw_wait_for_FRdouble (cpu, out_FRdoublek);
|
763 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
764 |
|
|
{
|
765 |
|
|
vliw_wait_for_SPR (cpu, FNER_FOR_FR (out_FRk));
|
766 |
|
|
vliw_wait_for_SPR (cpu, FNER_FOR_FR (out_FRdoublek));
|
767 |
|
|
}
|
768 |
|
|
handle_resource_wait (cpu);
|
769 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
770 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
771 |
|
|
load_wait_for_FR (cpu, out_FRk);
|
772 |
|
|
load_wait_for_FRdouble (cpu, out_FRdoublek);
|
773 |
|
|
trace_vliw_wait_cycles (cpu);
|
774 |
|
|
return 0;
|
775 |
|
|
}
|
776 |
|
|
|
777 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
778 |
|
|
|
779 |
|
|
/* The latency of FRk for a load will depend on how long it takes to retrieve
|
780 |
|
|
the the data from the cache or memory. */
|
781 |
|
|
update_FR_latency_for_load (cpu, out_FRk, cycles);
|
782 |
|
|
update_FRdouble_latency_for_load (cpu, out_FRdoublek, cycles);
|
783 |
|
|
|
784 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
785 |
|
|
{
|
786 |
|
|
/* FNER has a latency of 3 cycles. */
|
787 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRk), cycles + 3);
|
788 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRdoublek), cycles + 3);
|
789 |
|
|
}
|
790 |
|
|
|
791 |
|
|
if (out_FRk >= 0)
|
792 |
|
|
set_use_is_fr_load (cpu, out_FRk);
|
793 |
|
|
if (out_FRdoublek >= 0)
|
794 |
|
|
{
|
795 |
|
|
set_use_is_fr_load (cpu, out_FRdoublek);
|
796 |
|
|
set_use_is_fr_load (cpu, out_FRdoublek + 1);
|
797 |
|
|
}
|
798 |
|
|
|
799 |
|
|
return cycles;
|
800 |
|
|
}
|
801 |
|
|
|
802 |
|
|
int
|
803 |
|
|
frvbf_model_fr550_u_fr_store (SIM_CPU *cpu, const IDESC *idesc,
|
804 |
|
|
int unit_num, int referenced,
|
805 |
|
|
INT in_GRi, INT in_GRj,
|
806 |
|
|
INT in_FRk, INT in_FRdoublek)
|
807 |
|
|
{
|
808 |
|
|
int cycles;
|
809 |
|
|
|
810 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
811 |
|
|
{
|
812 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
813 |
|
|
which is not ready yet. */
|
814 |
|
|
adjust_float_register_busy (cpu, in_FRk, 1, -1, 1, -1, 1);
|
815 |
|
|
adjust_float_register_busy (cpu, in_FRdoublek, 2, -1, 1, -1, 1);
|
816 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
817 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
818 |
|
|
vliw_wait_for_FR (cpu, in_FRk);
|
819 |
|
|
vliw_wait_for_FRdouble (cpu, in_FRdoublek);
|
820 |
|
|
handle_resource_wait (cpu);
|
821 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
822 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
823 |
|
|
load_wait_for_FR (cpu, in_FRk);
|
824 |
|
|
load_wait_for_FRdouble (cpu, in_FRdoublek);
|
825 |
|
|
trace_vliw_wait_cycles (cpu);
|
826 |
|
|
return 0;
|
827 |
|
|
}
|
828 |
|
|
|
829 |
|
|
/* The target register is available immediately. */
|
830 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
831 |
|
|
|
832 |
|
|
return cycles;
|
833 |
|
|
}
|
834 |
|
|
|
835 |
|
|
int
|
836 |
|
|
frvbf_model_fr550_u_ici (SIM_CPU *cpu, const IDESC *idesc,
|
837 |
|
|
int unit_num, int referenced,
|
838 |
|
|
INT in_GRi, INT in_GRj)
|
839 |
|
|
{
|
840 |
|
|
int cycles;
|
841 |
|
|
|
842 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
843 |
|
|
{
|
844 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
845 |
|
|
which is not ready yet. */
|
846 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
847 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
848 |
|
|
handle_resource_wait (cpu);
|
849 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
850 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
851 |
|
|
trace_vliw_wait_cycles (cpu);
|
852 |
|
|
return 0;
|
853 |
|
|
}
|
854 |
|
|
|
855 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
856 |
|
|
request_cache_invalidate (cpu, CPU_INSN_CACHE (cpu), cycles);
|
857 |
|
|
return cycles;
|
858 |
|
|
}
|
859 |
|
|
|
860 |
|
|
int
|
861 |
|
|
frvbf_model_fr550_u_dci (SIM_CPU *cpu, const IDESC *idesc,
|
862 |
|
|
int unit_num, int referenced,
|
863 |
|
|
INT in_GRi, INT in_GRj)
|
864 |
|
|
{
|
865 |
|
|
int cycles;
|
866 |
|
|
|
867 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
868 |
|
|
{
|
869 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
870 |
|
|
which is not ready yet. */
|
871 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
872 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
873 |
|
|
handle_resource_wait (cpu);
|
874 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
875 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
876 |
|
|
trace_vliw_wait_cycles (cpu);
|
877 |
|
|
return 0;
|
878 |
|
|
}
|
879 |
|
|
|
880 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
881 |
|
|
request_cache_invalidate (cpu, CPU_DATA_CACHE (cpu), cycles);
|
882 |
|
|
return cycles;
|
883 |
|
|
}
|
884 |
|
|
|
885 |
|
|
int
|
886 |
|
|
frvbf_model_fr550_u_dcf (SIM_CPU *cpu, const IDESC *idesc,
|
887 |
|
|
int unit_num, int referenced,
|
888 |
|
|
INT in_GRi, INT in_GRj)
|
889 |
|
|
{
|
890 |
|
|
int cycles;
|
891 |
|
|
|
892 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
893 |
|
|
{
|
894 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
895 |
|
|
which is not ready yet. */
|
896 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
897 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
898 |
|
|
handle_resource_wait (cpu);
|
899 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
900 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
901 |
|
|
trace_vliw_wait_cycles (cpu);
|
902 |
|
|
return 0;
|
903 |
|
|
}
|
904 |
|
|
|
905 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
906 |
|
|
request_cache_flush (cpu, CPU_DATA_CACHE (cpu), cycles);
|
907 |
|
|
return cycles;
|
908 |
|
|
}
|
909 |
|
|
|
910 |
|
|
int
|
911 |
|
|
frvbf_model_fr550_u_icpl (SIM_CPU *cpu, const IDESC *idesc,
|
912 |
|
|
int unit_num, int referenced,
|
913 |
|
|
INT in_GRi, INT in_GRj)
|
914 |
|
|
{
|
915 |
|
|
int cycles;
|
916 |
|
|
|
917 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
918 |
|
|
{
|
919 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
920 |
|
|
which is not ready yet. */
|
921 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
922 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
923 |
|
|
handle_resource_wait (cpu);
|
924 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
925 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
926 |
|
|
trace_vliw_wait_cycles (cpu);
|
927 |
|
|
return 0;
|
928 |
|
|
}
|
929 |
|
|
|
930 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
931 |
|
|
request_cache_preload (cpu, CPU_INSN_CACHE (cpu), cycles);
|
932 |
|
|
return cycles;
|
933 |
|
|
}
|
934 |
|
|
|
935 |
|
|
int
|
936 |
|
|
frvbf_model_fr550_u_dcpl (SIM_CPU *cpu, const IDESC *idesc,
|
937 |
|
|
int unit_num, int referenced,
|
938 |
|
|
INT in_GRi, INT in_GRj)
|
939 |
|
|
{
|
940 |
|
|
int cycles;
|
941 |
|
|
|
942 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
943 |
|
|
{
|
944 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
945 |
|
|
which is not ready yet. */
|
946 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
947 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
948 |
|
|
handle_resource_wait (cpu);
|
949 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
950 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
951 |
|
|
trace_vliw_wait_cycles (cpu);
|
952 |
|
|
return 0;
|
953 |
|
|
}
|
954 |
|
|
|
955 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
956 |
|
|
request_cache_preload (cpu, CPU_DATA_CACHE (cpu), cycles);
|
957 |
|
|
return cycles;
|
958 |
|
|
}
|
959 |
|
|
|
960 |
|
|
int
|
961 |
|
|
frvbf_model_fr550_u_icul (SIM_CPU *cpu, const IDESC *idesc,
|
962 |
|
|
int unit_num, int referenced,
|
963 |
|
|
INT in_GRi, INT in_GRj)
|
964 |
|
|
{
|
965 |
|
|
int cycles;
|
966 |
|
|
|
967 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
968 |
|
|
{
|
969 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
970 |
|
|
which is not ready yet. */
|
971 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
972 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
973 |
|
|
handle_resource_wait (cpu);
|
974 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
975 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
976 |
|
|
trace_vliw_wait_cycles (cpu);
|
977 |
|
|
return 0;
|
978 |
|
|
}
|
979 |
|
|
|
980 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
981 |
|
|
request_cache_unlock (cpu, CPU_INSN_CACHE (cpu), cycles);
|
982 |
|
|
return cycles;
|
983 |
|
|
}
|
984 |
|
|
|
985 |
|
|
int
|
986 |
|
|
frvbf_model_fr550_u_dcul (SIM_CPU *cpu, const IDESC *idesc,
|
987 |
|
|
int unit_num, int referenced,
|
988 |
|
|
INT in_GRi, INT in_GRj)
|
989 |
|
|
{
|
990 |
|
|
int cycles;
|
991 |
|
|
|
992 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
993 |
|
|
{
|
994 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
995 |
|
|
which is not ready yet. */
|
996 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
997 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
998 |
|
|
handle_resource_wait (cpu);
|
999 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
1000 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
1001 |
|
|
trace_vliw_wait_cycles (cpu);
|
1002 |
|
|
return 0;
|
1003 |
|
|
}
|
1004 |
|
|
|
1005 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1006 |
|
|
request_cache_unlock (cpu, CPU_DATA_CACHE (cpu), cycles);
|
1007 |
|
|
return cycles;
|
1008 |
|
|
}
|
1009 |
|
|
|
1010 |
|
|
int
|
1011 |
|
|
frvbf_model_fr550_u_float_arith (SIM_CPU *cpu, const IDESC *idesc,
|
1012 |
|
|
int unit_num, int referenced,
|
1013 |
|
|
INT in_FRi, INT in_FRj,
|
1014 |
|
|
INT in_FRdoublei, INT in_FRdoublej,
|
1015 |
|
|
INT out_FRk, INT out_FRdoublek)
|
1016 |
|
|
{
|
1017 |
|
|
int cycles;
|
1018 |
|
|
FRV_PROFILE_STATE *ps;
|
1019 |
|
|
FRV_VLIW *vliw;
|
1020 |
|
|
int slot;
|
1021 |
|
|
|
1022 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1023 |
|
|
return 0;
|
1024 |
|
|
|
1025 |
|
|
/* The preprocessing can execute right away. */
|
1026 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1027 |
|
|
|
1028 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1029 |
|
|
which is not ready yet. */
|
1030 |
|
|
adjust_float_register_busy (cpu, in_FRi, 1, in_FRj, 1, out_FRk, 1);
|
1031 |
|
|
adjust_float_register_busy (cpu, in_FRdoublei, 2, in_FRdoublej, 2, out_FRdoublek, 2);
|
1032 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1033 |
|
|
ps->post_wait = cycles;
|
1034 |
|
|
vliw = CPU_VLIW (cpu);
|
1035 |
|
|
slot = vliw->next_slot - 1;
|
1036 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1037 |
|
|
post_wait_for_float (cpu, slot);
|
1038 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1039 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1040 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1041 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublei);
|
1042 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublej);
|
1043 |
|
|
post_wait_for_FRdouble (cpu, out_FRdoublek);
|
1044 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1045 |
|
|
{
|
1046 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRk));
|
1047 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRdoublek));
|
1048 |
|
|
}
|
1049 |
|
|
restore_float_register_busy (cpu, in_FRi, 1, in_FRj, 1, out_FRk, 1);
|
1050 |
|
|
restore_float_register_busy (cpu, in_FRdoublei, 2, in_FRdoublej, 2, out_FRdoublek, 2);
|
1051 |
|
|
|
1052 |
|
|
/* The latency of FRk will be at least the latency of the other inputs. */
|
1053 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1054 |
|
|
update_FRdouble_latency (cpu, out_FRdoublek, ps->post_wait);
|
1055 |
|
|
|
1056 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1057 |
|
|
{
|
1058 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRk), ps->post_wait);
|
1059 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRdoublek), ps->post_wait);
|
1060 |
|
|
}
|
1061 |
|
|
|
1062 |
|
|
/* Once initiated, post-processing will take 2 cycles. */
|
1063 |
|
|
update_FR_ptime (cpu, out_FRk, 2);
|
1064 |
|
|
update_FRdouble_ptime (cpu, out_FRdoublek, 2);
|
1065 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1066 |
|
|
{
|
1067 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRk), 2);
|
1068 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRdoublek), 2);
|
1069 |
|
|
}
|
1070 |
|
|
|
1071 |
|
|
/* Mark this use of the register as a floating point op. */
|
1072 |
|
|
if (out_FRk >= 0)
|
1073 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRk);
|
1074 |
|
|
if (out_FRdoublek >= 0)
|
1075 |
|
|
{
|
1076 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRdoublek);
|
1077 |
|
|
if (out_FRdoublek < 63)
|
1078 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRdoublek + 1);
|
1079 |
|
|
}
|
1080 |
|
|
|
1081 |
|
|
/* the media point unit resource has a latency of 4 cycles */
|
1082 |
|
|
update_media_resource_latency (cpu, slot, cycles + 4);
|
1083 |
|
|
|
1084 |
|
|
return cycles;
|
1085 |
|
|
}
|
1086 |
|
|
|
1087 |
|
|
int
|
1088 |
|
|
frvbf_model_fr550_u_float_dual_arith (SIM_CPU *cpu, const IDESC *idesc,
|
1089 |
|
|
int unit_num, int referenced,
|
1090 |
|
|
INT in_FRi, INT in_FRj,
|
1091 |
|
|
INT in_FRdoublei, INT in_FRdoublej,
|
1092 |
|
|
INT out_FRk, INT out_FRdoublek)
|
1093 |
|
|
{
|
1094 |
|
|
int cycles;
|
1095 |
|
|
INT dual_FRi;
|
1096 |
|
|
INT dual_FRj;
|
1097 |
|
|
INT dual_FRk;
|
1098 |
|
|
INT dual_FRdoublei;
|
1099 |
|
|
INT dual_FRdoublej;
|
1100 |
|
|
INT dual_FRdoublek;
|
1101 |
|
|
FRV_PROFILE_STATE *ps;
|
1102 |
|
|
FRV_VLIW *vliw;
|
1103 |
|
|
int slot;
|
1104 |
|
|
|
1105 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1106 |
|
|
return 0;
|
1107 |
|
|
|
1108 |
|
|
/* The preprocessing can execute right away. */
|
1109 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1110 |
|
|
|
1111 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1112 |
|
|
which is not ready yet. */
|
1113 |
|
|
dual_FRi = DUAL_REG (in_FRi);
|
1114 |
|
|
dual_FRj = DUAL_REG (in_FRj);
|
1115 |
|
|
dual_FRk = DUAL_REG (out_FRk);
|
1116 |
|
|
dual_FRdoublei = DUAL_DOUBLE (in_FRdoublei);
|
1117 |
|
|
dual_FRdoublej = DUAL_DOUBLE (in_FRdoublej);
|
1118 |
|
|
dual_FRdoublek = DUAL_DOUBLE (out_FRdoublek);
|
1119 |
|
|
|
1120 |
|
|
adjust_float_register_busy (cpu, in_FRi, 2, in_FRj, 2, out_FRk, 2);
|
1121 |
|
|
adjust_float_register_busy (cpu, in_FRdoublei, 4, in_FRdoublej, 4, out_FRdoublek, 4);
|
1122 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1123 |
|
|
ps->post_wait = cycles;
|
1124 |
|
|
vliw = CPU_VLIW (cpu);
|
1125 |
|
|
slot = vliw->next_slot - 1;
|
1126 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1127 |
|
|
post_wait_for_float (cpu, slot);
|
1128 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1129 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1130 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1131 |
|
|
post_wait_for_FR (cpu, dual_FRi);
|
1132 |
|
|
post_wait_for_FR (cpu, dual_FRj);
|
1133 |
|
|
post_wait_for_FR (cpu, dual_FRk);
|
1134 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublei);
|
1135 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublej);
|
1136 |
|
|
post_wait_for_FRdouble (cpu, out_FRdoublek);
|
1137 |
|
|
post_wait_for_FRdouble (cpu, dual_FRdoublei);
|
1138 |
|
|
post_wait_for_FRdouble (cpu, dual_FRdoublej);
|
1139 |
|
|
post_wait_for_FRdouble (cpu, dual_FRdoublek);
|
1140 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1141 |
|
|
{
|
1142 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRk));
|
1143 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (dual_FRk));
|
1144 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRdoublek));
|
1145 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (dual_FRdoublek));
|
1146 |
|
|
}
|
1147 |
|
|
restore_float_register_busy (cpu, in_FRi, 2, in_FRj, 2, out_FRk, 2);
|
1148 |
|
|
restore_float_register_busy (cpu, in_FRdoublei, 4, in_FRdoublej, 4, out_FRdoublek, 4);
|
1149 |
|
|
|
1150 |
|
|
/* The latency of FRk will be at least the latency of the other inputs. */
|
1151 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1152 |
|
|
update_FR_latency (cpu, dual_FRk, ps->post_wait);
|
1153 |
|
|
update_FRdouble_latency (cpu, out_FRdoublek, ps->post_wait);
|
1154 |
|
|
update_FRdouble_latency (cpu, dual_FRdoublek, ps->post_wait);
|
1155 |
|
|
|
1156 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1157 |
|
|
{
|
1158 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRk), ps->post_wait);
|
1159 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (dual_FRk), ps->post_wait);
|
1160 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRdoublek), ps->post_wait);
|
1161 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (dual_FRdoublek), ps->post_wait);
|
1162 |
|
|
}
|
1163 |
|
|
|
1164 |
|
|
/* Once initiated, post-processing will take 3 cycles. */
|
1165 |
|
|
update_FR_ptime (cpu, out_FRk, 3);
|
1166 |
|
|
update_FR_ptime (cpu, dual_FRk, 3);
|
1167 |
|
|
update_FRdouble_ptime (cpu, out_FRdoublek, 3);
|
1168 |
|
|
update_FRdouble_ptime (cpu, dual_FRdoublek, 3);
|
1169 |
|
|
|
1170 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1171 |
|
|
{
|
1172 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRk), 3);
|
1173 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (dual_FRk), 3);
|
1174 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRdoublek), 3);
|
1175 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (dual_FRdoublek), 3);
|
1176 |
|
|
}
|
1177 |
|
|
|
1178 |
|
|
/* Mark this use of the register as a floating point op. */
|
1179 |
|
|
if (out_FRk >= 0)
|
1180 |
|
|
fr550_reset_fr_flags (cpu, out_FRk);
|
1181 |
|
|
if (dual_FRk >= 0)
|
1182 |
|
|
fr550_reset_fr_flags (cpu, dual_FRk);
|
1183 |
|
|
if (out_FRdoublek >= 0)
|
1184 |
|
|
{
|
1185 |
|
|
fr550_reset_fr_flags (cpu, out_FRdoublek);
|
1186 |
|
|
if (out_FRdoublek < 63)
|
1187 |
|
|
fr550_reset_fr_flags (cpu, out_FRdoublek + 1);
|
1188 |
|
|
}
|
1189 |
|
|
if (dual_FRdoublek >= 0)
|
1190 |
|
|
{
|
1191 |
|
|
fr550_reset_fr_flags (cpu, dual_FRdoublek);
|
1192 |
|
|
if (dual_FRdoublek < 63)
|
1193 |
|
|
fr550_reset_fr_flags (cpu, dual_FRdoublek + 1);
|
1194 |
|
|
}
|
1195 |
|
|
|
1196 |
|
|
/* the media point unit resource has a latency of 5 cycles */
|
1197 |
|
|
update_media_resource_latency (cpu, slot, cycles + 5);
|
1198 |
|
|
|
1199 |
|
|
return cycles;
|
1200 |
|
|
}
|
1201 |
|
|
|
1202 |
|
|
int
|
1203 |
|
|
frvbf_model_fr550_u_float_div (SIM_CPU *cpu, const IDESC *idesc,
|
1204 |
|
|
int unit_num, int referenced,
|
1205 |
|
|
INT in_FRi, INT in_FRj, INT out_FRk)
|
1206 |
|
|
{
|
1207 |
|
|
int cycles;
|
1208 |
|
|
FRV_VLIW *vliw;
|
1209 |
|
|
int slot;
|
1210 |
|
|
FRV_PROFILE_STATE *ps;
|
1211 |
|
|
|
1212 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1213 |
|
|
return 0;
|
1214 |
|
|
|
1215 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1216 |
|
|
|
1217 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1218 |
|
|
which is not ready yet. */
|
1219 |
|
|
adjust_float_register_busy (cpu, in_FRi, 1, in_FRj, 1, out_FRk, 1);
|
1220 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1221 |
|
|
ps->post_wait = cycles;
|
1222 |
|
|
vliw = CPU_VLIW (cpu);
|
1223 |
|
|
slot = vliw->next_slot - 1;
|
1224 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1225 |
|
|
post_wait_for_float (cpu, slot);
|
1226 |
|
|
post_wait_for_fdiv (cpu, slot);
|
1227 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1228 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1229 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1230 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1231 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRk));
|
1232 |
|
|
restore_float_register_busy (cpu, in_FRi, 1, in_FRj, 1, out_FRk, 1);
|
1233 |
|
|
|
1234 |
|
|
/* The latency of FRk will be at least the latency of the other inputs. */
|
1235 |
|
|
/* Once initiated, post-processing will take 9 cycles. */
|
1236 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1237 |
|
|
update_FR_ptime (cpu, out_FRk, 9);
|
1238 |
|
|
|
1239 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1240 |
|
|
{
|
1241 |
|
|
/* FNER has a latency of 9 cycles. */
|
1242 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRk), ps->post_wait);
|
1243 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRk), 9);
|
1244 |
|
|
}
|
1245 |
|
|
|
1246 |
|
|
/* The latency of the fdiv unit will be at least the latency of the other
|
1247 |
|
|
inputs. Once initiated, post-processing will take 9 cycles. */
|
1248 |
|
|
update_fdiv_resource_latency (cpu, slot, ps->post_wait + 9);
|
1249 |
|
|
|
1250 |
|
|
/* the media point unit resource has a latency of 11 cycles */
|
1251 |
|
|
update_media_resource_latency (cpu, slot, cycles + 11);
|
1252 |
|
|
|
1253 |
|
|
fr550_reset_fr_flags (cpu, out_FRk);
|
1254 |
|
|
|
1255 |
|
|
return cycles;
|
1256 |
|
|
}
|
1257 |
|
|
|
1258 |
|
|
int
|
1259 |
|
|
frvbf_model_fr550_u_float_sqrt (SIM_CPU *cpu, const IDESC *idesc,
|
1260 |
|
|
int unit_num, int referenced,
|
1261 |
|
|
INT in_FRj, INT in_FRdoublej,
|
1262 |
|
|
INT out_FRk, INT out_FRdoublek)
|
1263 |
|
|
{
|
1264 |
|
|
int cycles;
|
1265 |
|
|
FRV_VLIW *vliw;
|
1266 |
|
|
int slot;
|
1267 |
|
|
FRV_PROFILE_STATE *ps;
|
1268 |
|
|
|
1269 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1270 |
|
|
return 0;
|
1271 |
|
|
|
1272 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1273 |
|
|
|
1274 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1275 |
|
|
which is not ready yet. */
|
1276 |
|
|
adjust_float_register_busy (cpu, -1, 1, in_FRj, 1, out_FRk, 1);
|
1277 |
|
|
adjust_float_register_busy (cpu, -1, 1, in_FRdoublej, 2, out_FRdoublek, 2);
|
1278 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1279 |
|
|
ps->post_wait = cycles;
|
1280 |
|
|
vliw = CPU_VLIW (cpu);
|
1281 |
|
|
slot = vliw->next_slot - 1;
|
1282 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1283 |
|
|
post_wait_for_float (cpu, slot);
|
1284 |
|
|
post_wait_for_fsqrt (cpu, slot);
|
1285 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1286 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1287 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublej);
|
1288 |
|
|
post_wait_for_FRdouble (cpu, out_FRdoublek);
|
1289 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1290 |
|
|
{
|
1291 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRk));
|
1292 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRdoublek));
|
1293 |
|
|
}
|
1294 |
|
|
restore_float_register_busy (cpu, -1, 1, in_FRj, 1, out_FRk, 1);
|
1295 |
|
|
restore_float_register_busy (cpu, -1, 1, in_FRdoublej, 2, out_FRdoublek, 2);
|
1296 |
|
|
|
1297 |
|
|
/* The latency of FRk will be at least the latency of the other inputs. */
|
1298 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1299 |
|
|
update_FRdouble_latency (cpu, out_FRdoublek, ps->post_wait);
|
1300 |
|
|
|
1301 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1302 |
|
|
{
|
1303 |
|
|
/* FNER has a latency of 14 cycles. */
|
1304 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRk), ps->post_wait);
|
1305 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRdoublek), ps->post_wait);
|
1306 |
|
|
}
|
1307 |
|
|
|
1308 |
|
|
/* Once initiated, post-processing will take 14 cycles. */
|
1309 |
|
|
update_FR_ptime (cpu, out_FRk, 14);
|
1310 |
|
|
update_FRdouble_ptime (cpu, out_FRdoublek, 14);
|
1311 |
|
|
|
1312 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1313 |
|
|
{
|
1314 |
|
|
/* FNER has a latency of 14 cycles. */
|
1315 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRk), 14);
|
1316 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRdoublek), 14);
|
1317 |
|
|
}
|
1318 |
|
|
|
1319 |
|
|
/* The latency of the sqrt unit will be the latency of the other
|
1320 |
|
|
inputs plus 14 cycles. */
|
1321 |
|
|
update_fsqrt_resource_latency (cpu, slot, ps->post_wait + 14);
|
1322 |
|
|
|
1323 |
|
|
fr550_reset_fr_flags (cpu, out_FRk);
|
1324 |
|
|
if (out_FRdoublek != -1)
|
1325 |
|
|
{
|
1326 |
|
|
fr550_reset_fr_flags (cpu, out_FRdoublek);
|
1327 |
|
|
fr550_reset_fr_flags (cpu, out_FRdoublek + 1);
|
1328 |
|
|
}
|
1329 |
|
|
|
1330 |
|
|
/* the media point unit resource has a latency of 16 cycles */
|
1331 |
|
|
update_media_resource_latency (cpu, slot, cycles + 16);
|
1332 |
|
|
|
1333 |
|
|
return cycles;
|
1334 |
|
|
}
|
1335 |
|
|
|
1336 |
|
|
int
|
1337 |
|
|
frvbf_model_fr550_u_float_compare (SIM_CPU *cpu, const IDESC *idesc,
|
1338 |
|
|
int unit_num, int referenced,
|
1339 |
|
|
INT in_FRi, INT in_FRj,
|
1340 |
|
|
INT in_FRdoublei, INT in_FRdoublej,
|
1341 |
|
|
INT out_FCCi_2)
|
1342 |
|
|
{
|
1343 |
|
|
int cycles;
|
1344 |
|
|
FRV_PROFILE_STATE *ps;
|
1345 |
|
|
FRV_VLIW *vliw;
|
1346 |
|
|
int slot;
|
1347 |
|
|
|
1348 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1349 |
|
|
return 0;
|
1350 |
|
|
|
1351 |
|
|
/* The preprocessing can execute right away. */
|
1352 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1353 |
|
|
|
1354 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1355 |
|
|
which is not ready yet. */
|
1356 |
|
|
adjust_float_register_busy (cpu, in_FRi, 1, in_FRj, 1, -1, 1);
|
1357 |
|
|
adjust_float_register_busy (cpu, in_FRdoublei, 2, in_FRdoublej, 2, -1, 1);
|
1358 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1359 |
|
|
ps->post_wait = cycles;
|
1360 |
|
|
vliw = CPU_VLIW (cpu);
|
1361 |
|
|
slot = vliw->next_slot - 1;
|
1362 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1363 |
|
|
post_wait_for_float (cpu, slot);
|
1364 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1365 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1366 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublei);
|
1367 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublej);
|
1368 |
|
|
post_wait_for_CCR (cpu, out_FCCi_2);
|
1369 |
|
|
restore_float_register_busy (cpu, in_FRi, 1, in_FRj, 1, -1, 1);
|
1370 |
|
|
restore_float_register_busy (cpu, in_FRdoublei, 2, in_FRdoublej, 2, -1, 1);
|
1371 |
|
|
|
1372 |
|
|
/* The latency of FCCi_2 will be the latency of the other inputs plus 2
|
1373 |
|
|
cycles. */
|
1374 |
|
|
update_CCR_latency (cpu, out_FCCi_2, ps->post_wait + 2);
|
1375 |
|
|
|
1376 |
|
|
/* the media point unit resource has a latency of 4 cycles */
|
1377 |
|
|
update_media_resource_latency (cpu, slot, cycles + 4);
|
1378 |
|
|
|
1379 |
|
|
set_use_is_ccr_complex (cpu, out_FCCi_2);
|
1380 |
|
|
|
1381 |
|
|
return cycles;
|
1382 |
|
|
}
|
1383 |
|
|
|
1384 |
|
|
int
|
1385 |
|
|
frvbf_model_fr550_u_float_dual_compare (SIM_CPU *cpu, const IDESC *idesc,
|
1386 |
|
|
int unit_num, int referenced,
|
1387 |
|
|
INT in_FRi, INT in_FRj,
|
1388 |
|
|
INT out_FCCi_2)
|
1389 |
|
|
{
|
1390 |
|
|
int cycles;
|
1391 |
|
|
INT dual_FRi;
|
1392 |
|
|
INT dual_FRj;
|
1393 |
|
|
INT dual_FCCi_2;
|
1394 |
|
|
FRV_PROFILE_STATE *ps;
|
1395 |
|
|
FRV_VLIW *vliw;
|
1396 |
|
|
int slot;
|
1397 |
|
|
|
1398 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1399 |
|
|
return 0;
|
1400 |
|
|
|
1401 |
|
|
/* The preprocessing can execute right away. */
|
1402 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1403 |
|
|
|
1404 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1405 |
|
|
which is not ready yet. */
|
1406 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1407 |
|
|
ps->post_wait = cycles;
|
1408 |
|
|
dual_FRi = DUAL_REG (in_FRi);
|
1409 |
|
|
dual_FRj = DUAL_REG (in_FRj);
|
1410 |
|
|
dual_FCCi_2 = out_FCCi_2 + 1;
|
1411 |
|
|
adjust_float_register_busy (cpu, in_FRi, 2, in_FRj, 2, -1, 1);
|
1412 |
|
|
vliw = CPU_VLIW (cpu);
|
1413 |
|
|
slot = vliw->next_slot - 1;
|
1414 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1415 |
|
|
post_wait_for_float (cpu, slot);
|
1416 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1417 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1418 |
|
|
post_wait_for_FR (cpu, dual_FRi);
|
1419 |
|
|
post_wait_for_FR (cpu, dual_FRj);
|
1420 |
|
|
post_wait_for_CCR (cpu, out_FCCi_2);
|
1421 |
|
|
post_wait_for_CCR (cpu, dual_FCCi_2);
|
1422 |
|
|
restore_float_register_busy (cpu, in_FRi, 2, in_FRj, 2, -1, 1);
|
1423 |
|
|
|
1424 |
|
|
/* The latency of FCCi_2 will be the latency of the other inputs plus 3
|
1425 |
|
|
cycles. */
|
1426 |
|
|
update_CCR_latency (cpu, out_FCCi_2, ps->post_wait + 3);
|
1427 |
|
|
update_CCR_latency (cpu, dual_FCCi_2, ps->post_wait + 3);
|
1428 |
|
|
|
1429 |
|
|
set_use_is_ccr_complex (cpu, out_FCCi_2);
|
1430 |
|
|
if (dual_FCCi_2 >= 0)
|
1431 |
|
|
set_use_is_ccr_complex (cpu, dual_FCCi_2);
|
1432 |
|
|
|
1433 |
|
|
/* the media point unit resource has a latency of 5 cycles */
|
1434 |
|
|
update_media_resource_latency (cpu, slot, cycles + 5);
|
1435 |
|
|
|
1436 |
|
|
return cycles;
|
1437 |
|
|
}
|
1438 |
|
|
|
1439 |
|
|
int
|
1440 |
|
|
frvbf_model_fr550_u_float_convert (SIM_CPU *cpu, const IDESC *idesc,
|
1441 |
|
|
int unit_num, int referenced,
|
1442 |
|
|
INT in_FRj, INT in_FRintj, INT in_FRdoublej,
|
1443 |
|
|
INT out_FRk, INT out_FRintk,
|
1444 |
|
|
INT out_FRdoublek)
|
1445 |
|
|
{
|
1446 |
|
|
int cycles;
|
1447 |
|
|
FRV_PROFILE_STATE *ps;
|
1448 |
|
|
FRV_VLIW *vliw;
|
1449 |
|
|
int slot;
|
1450 |
|
|
|
1451 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1452 |
|
|
return 0;
|
1453 |
|
|
|
1454 |
|
|
/* The preprocessing can execute right away. */
|
1455 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1456 |
|
|
|
1457 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1458 |
|
|
which is not ready yet. */
|
1459 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1460 |
|
|
ps->post_wait = cycles;
|
1461 |
|
|
adjust_float_register_busy (cpu, -1, 1, in_FRj, 1, out_FRk, 1);
|
1462 |
|
|
adjust_float_register_busy (cpu, -1, 1, in_FRintj, 1, out_FRintk, 1);
|
1463 |
|
|
adjust_float_register_busy (cpu, -1, 1, in_FRdoublej, 2, out_FRdoublek, 2);
|
1464 |
|
|
vliw = CPU_VLIW (cpu);
|
1465 |
|
|
slot = vliw->next_slot - 1;
|
1466 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1467 |
|
|
post_wait_for_float (cpu, slot);
|
1468 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1469 |
|
|
post_wait_for_FR (cpu, in_FRintj);
|
1470 |
|
|
post_wait_for_FRdouble (cpu, in_FRdoublej);
|
1471 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1472 |
|
|
post_wait_for_FR (cpu, out_FRintk);
|
1473 |
|
|
post_wait_for_FRdouble (cpu, out_FRdoublek);
|
1474 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1475 |
|
|
{
|
1476 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRk));
|
1477 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRintk));
|
1478 |
|
|
post_wait_for_SPR (cpu, FNER_FOR_FR (out_FRdoublek));
|
1479 |
|
|
}
|
1480 |
|
|
restore_float_register_busy (cpu, -1, 1, in_FRj, 1, out_FRk, 1);
|
1481 |
|
|
restore_float_register_busy (cpu, -1, 1, in_FRintj, 1, out_FRintk, 1);
|
1482 |
|
|
restore_float_register_busy (cpu, -1, 1, in_FRdoublej, 2, out_FRdoublek, 2);
|
1483 |
|
|
|
1484 |
|
|
/* The latency of FRk will be at least the latency of the other inputs. */
|
1485 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1486 |
|
|
update_FR_latency (cpu, out_FRintk, ps->post_wait);
|
1487 |
|
|
update_FRdouble_latency (cpu, out_FRdoublek, ps->post_wait);
|
1488 |
|
|
|
1489 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1490 |
|
|
{
|
1491 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRk), ps->post_wait);
|
1492 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRintk), ps->post_wait);
|
1493 |
|
|
update_SPR_latency (cpu, FNER_FOR_FR (out_FRdoublek), ps->post_wait);
|
1494 |
|
|
}
|
1495 |
|
|
|
1496 |
|
|
/* Once initiated, post-processing will take 2 cycles. */
|
1497 |
|
|
update_FR_ptime (cpu, out_FRk, 2);
|
1498 |
|
|
update_FR_ptime (cpu, out_FRintk, 2);
|
1499 |
|
|
update_FRdouble_ptime (cpu, out_FRdoublek, 2);
|
1500 |
|
|
|
1501 |
|
|
if (CGEN_ATTR_VALUE(idesc, idesc->attrs, CGEN_INSN_NON_EXCEPTING))
|
1502 |
|
|
{
|
1503 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRk), 2);
|
1504 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRintk), 2);
|
1505 |
|
|
update_SPR_ptime (cpu, FNER_FOR_FR (out_FRdoublek), 2);
|
1506 |
|
|
}
|
1507 |
|
|
|
1508 |
|
|
/* Mark this use of the register as a floating point op. */
|
1509 |
|
|
if (out_FRk >= 0)
|
1510 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRk);
|
1511 |
|
|
if (out_FRintk >= 0)
|
1512 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRintk);
|
1513 |
|
|
if (out_FRdoublek >= 0)
|
1514 |
|
|
{
|
1515 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRdoublek);
|
1516 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRdoublek + 1);
|
1517 |
|
|
}
|
1518 |
|
|
|
1519 |
|
|
/* the media point unit resource has a latency of 4 cycles */
|
1520 |
|
|
update_media_resource_latency (cpu, slot, cycles + 4);
|
1521 |
|
|
|
1522 |
|
|
return cycles;
|
1523 |
|
|
}
|
1524 |
|
|
|
1525 |
|
|
int
|
1526 |
|
|
frvbf_model_fr550_u_spr2gr (SIM_CPU *cpu, const IDESC *idesc,
|
1527 |
|
|
int unit_num, int referenced,
|
1528 |
|
|
INT in_spr, INT out_GRj)
|
1529 |
|
|
{
|
1530 |
|
|
/* Modelling for this unit is the same as for fr500. */
|
1531 |
|
|
return frvbf_model_fr500_u_spr2gr (cpu, idesc, unit_num, referenced,
|
1532 |
|
|
in_spr, out_GRj);
|
1533 |
|
|
}
|
1534 |
|
|
|
1535 |
|
|
int
|
1536 |
|
|
frvbf_model_fr550_u_gr2spr (SIM_CPU *cpu, const IDESC *idesc,
|
1537 |
|
|
int unit_num, int referenced,
|
1538 |
|
|
INT in_GRj, INT out_spr)
|
1539 |
|
|
{
|
1540 |
|
|
int cycles;
|
1541 |
|
|
|
1542 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1543 |
|
|
{
|
1544 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
1545 |
|
|
which is not ready yet. */
|
1546 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
1547 |
|
|
vliw_wait_for_SPR (cpu, out_spr);
|
1548 |
|
|
handle_resource_wait (cpu);
|
1549 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
1550 |
|
|
trace_vliw_wait_cycles (cpu);
|
1551 |
|
|
return 0;
|
1552 |
|
|
}
|
1553 |
|
|
|
1554 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1555 |
|
|
|
1556 |
|
|
#if 0
|
1557 |
|
|
/* The latency of spr is ? cycles. */
|
1558 |
|
|
update_SPR_latency (cpu, out_spr, cycles + ?);
|
1559 |
|
|
#endif
|
1560 |
|
|
|
1561 |
|
|
return cycles;
|
1562 |
|
|
}
|
1563 |
|
|
|
1564 |
|
|
int
|
1565 |
|
|
frvbf_model_fr550_u_gr2fr (SIM_CPU *cpu, const IDESC *idesc,
|
1566 |
|
|
int unit_num, int referenced,
|
1567 |
|
|
INT in_GRj, INT out_FRk)
|
1568 |
|
|
{
|
1569 |
|
|
int cycles;
|
1570 |
|
|
|
1571 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1572 |
|
|
{
|
1573 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
1574 |
|
|
which is not ready yet.
|
1575 |
|
|
The latency of the registers may be less than previously recorded,
|
1576 |
|
|
depending on how they were used previously.
|
1577 |
|
|
See Table 14-15 in the LSI. */
|
1578 |
|
|
adjust_float_register_busy (cpu, -1, 1, -1, 1, out_FRk, 1);
|
1579 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
1580 |
|
|
vliw_wait_for_FR (cpu, out_FRk);
|
1581 |
|
|
handle_resource_wait (cpu);
|
1582 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
1583 |
|
|
load_wait_for_FR (cpu, out_FRk);
|
1584 |
|
|
trace_vliw_wait_cycles (cpu);
|
1585 |
|
|
return 0;
|
1586 |
|
|
}
|
1587 |
|
|
|
1588 |
|
|
/* The latency of FRk is 1 cycles. */
|
1589 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1590 |
|
|
update_FR_latency (cpu, out_FRk, cycles + 1);
|
1591 |
|
|
|
1592 |
|
|
set_use_is_fr_complex_1 (cpu, out_FRk);
|
1593 |
|
|
|
1594 |
|
|
return cycles;
|
1595 |
|
|
}
|
1596 |
|
|
|
1597 |
|
|
int
|
1598 |
|
|
frvbf_model_fr550_u_swap (SIM_CPU *cpu, const IDESC *idesc,
|
1599 |
|
|
int unit_num, int referenced,
|
1600 |
|
|
INT in_GRi, INT in_GRj, INT out_GRk)
|
1601 |
|
|
{
|
1602 |
|
|
int cycles;
|
1603 |
|
|
|
1604 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1605 |
|
|
{
|
1606 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
1607 |
|
|
which is not ready yet. */
|
1608 |
|
|
vliw_wait_for_GR (cpu, in_GRi);
|
1609 |
|
|
vliw_wait_for_GR (cpu, in_GRj);
|
1610 |
|
|
vliw_wait_for_GR (cpu, out_GRk);
|
1611 |
|
|
handle_resource_wait (cpu);
|
1612 |
|
|
load_wait_for_GR (cpu, in_GRi);
|
1613 |
|
|
load_wait_for_GR (cpu, in_GRj);
|
1614 |
|
|
load_wait_for_GR (cpu, out_GRk);
|
1615 |
|
|
trace_vliw_wait_cycles (cpu);
|
1616 |
|
|
return 0;
|
1617 |
|
|
}
|
1618 |
|
|
|
1619 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1620 |
|
|
|
1621 |
|
|
/* The latency of GRk will depend on how long it takes to swap
|
1622 |
|
|
the the data from the cache or memory. */
|
1623 |
|
|
update_GR_latency_for_swap (cpu, out_GRk, cycles);
|
1624 |
|
|
|
1625 |
|
|
return cycles;
|
1626 |
|
|
}
|
1627 |
|
|
|
1628 |
|
|
int
|
1629 |
|
|
frvbf_model_fr550_u_fr2fr (SIM_CPU *cpu, const IDESC *idesc,
|
1630 |
|
|
int unit_num, int referenced,
|
1631 |
|
|
INT in_FRj, INT out_FRk)
|
1632 |
|
|
{
|
1633 |
|
|
int cycles;
|
1634 |
|
|
|
1635 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1636 |
|
|
{
|
1637 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
1638 |
|
|
which is not ready yet.
|
1639 |
|
|
The latency of the registers may be less than previously recorded,
|
1640 |
|
|
depending on how they were used previously.
|
1641 |
|
|
See Table 14-15 in the LSI. */
|
1642 |
|
|
adjust_float_register_busy (cpu, -1, 1, in_FRj, 1, out_FRk, 1);
|
1643 |
|
|
vliw_wait_for_FR (cpu, in_FRj);
|
1644 |
|
|
vliw_wait_for_FR (cpu, out_FRk);
|
1645 |
|
|
handle_resource_wait (cpu);
|
1646 |
|
|
load_wait_for_FR (cpu, in_FRj);
|
1647 |
|
|
load_wait_for_FR (cpu, out_FRk);
|
1648 |
|
|
trace_vliw_wait_cycles (cpu);
|
1649 |
|
|
return 0;
|
1650 |
|
|
}
|
1651 |
|
|
|
1652 |
|
|
/* The latency of FRj is 2 cycles. */
|
1653 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1654 |
|
|
update_FR_latency (cpu, out_FRk, cycles + 2);
|
1655 |
|
|
|
1656 |
|
|
set_use_is_fr_complex_2 (cpu, out_FRk);
|
1657 |
|
|
|
1658 |
|
|
return cycles;
|
1659 |
|
|
}
|
1660 |
|
|
|
1661 |
|
|
int
|
1662 |
|
|
frvbf_model_fr550_u_fr2gr (SIM_CPU *cpu, const IDESC *idesc,
|
1663 |
|
|
int unit_num, int referenced,
|
1664 |
|
|
INT in_FRk, INT out_GRj)
|
1665 |
|
|
{
|
1666 |
|
|
int cycles;
|
1667 |
|
|
|
1668 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1669 |
|
|
{
|
1670 |
|
|
/* The entire VLIW insn must wait if there is a dependency on a register
|
1671 |
|
|
which is not ready yet.
|
1672 |
|
|
The latency of the registers may be less than previously recorded,
|
1673 |
|
|
depending on how they were used previously.
|
1674 |
|
|
See Table 14-15 in the LSI. */
|
1675 |
|
|
adjust_float_register_busy (cpu, in_FRk, 1, -1, 1, -1, 1);
|
1676 |
|
|
vliw_wait_for_FR (cpu, in_FRk);
|
1677 |
|
|
vliw_wait_for_GR (cpu, out_GRj);
|
1678 |
|
|
handle_resource_wait (cpu);
|
1679 |
|
|
load_wait_for_FR (cpu, in_FRk);
|
1680 |
|
|
load_wait_for_GR (cpu, out_GRj);
|
1681 |
|
|
trace_vliw_wait_cycles (cpu);
|
1682 |
|
|
return 0;
|
1683 |
|
|
}
|
1684 |
|
|
|
1685 |
|
|
/* The latency of GRj is 1 cycle. */
|
1686 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1687 |
|
|
update_GR_latency (cpu, out_GRj, cycles + 1);
|
1688 |
|
|
|
1689 |
|
|
return cycles;
|
1690 |
|
|
}
|
1691 |
|
|
|
1692 |
|
|
int
|
1693 |
|
|
frvbf_model_fr550_u_clrgr (SIM_CPU *cpu, const IDESC *idesc,
|
1694 |
|
|
int unit_num, int referenced,
|
1695 |
|
|
INT in_GRk)
|
1696 |
|
|
{
|
1697 |
|
|
/* Modelling for this unit is the same as for fr500. */
|
1698 |
|
|
return frvbf_model_fr500_u_clrgr (cpu, idesc, unit_num, referenced, in_GRk);
|
1699 |
|
|
}
|
1700 |
|
|
|
1701 |
|
|
int
|
1702 |
|
|
frvbf_model_fr550_u_clrfr (SIM_CPU *cpu, const IDESC *idesc,
|
1703 |
|
|
int unit_num, int referenced,
|
1704 |
|
|
INT in_FRk)
|
1705 |
|
|
{
|
1706 |
|
|
/* Modelling for this unit is the same as for fr500. */
|
1707 |
|
|
return frvbf_model_fr500_u_clrfr (cpu, idesc, unit_num, referenced, in_FRk);
|
1708 |
|
|
}
|
1709 |
|
|
|
1710 |
|
|
int
|
1711 |
|
|
frvbf_model_fr550_u_commit (SIM_CPU *cpu, const IDESC *idesc,
|
1712 |
|
|
int unit_num, int referenced,
|
1713 |
|
|
INT in_GRk, INT in_FRk)
|
1714 |
|
|
{
|
1715 |
|
|
/* Modelling for this unit is the same as for fr500. */
|
1716 |
|
|
return frvbf_model_fr500_u_commit (cpu, idesc, unit_num, referenced,
|
1717 |
|
|
in_GRk, in_FRk);
|
1718 |
|
|
}
|
1719 |
|
|
|
1720 |
|
|
int
|
1721 |
|
|
frvbf_model_fr550_u_media (SIM_CPU *cpu, const IDESC *idesc,
|
1722 |
|
|
int unit_num, int referenced,
|
1723 |
|
|
INT in_FRi, INT in_FRj, INT out_FRk)
|
1724 |
|
|
{
|
1725 |
|
|
int cycles;
|
1726 |
|
|
FRV_PROFILE_STATE *ps;
|
1727 |
|
|
FRV_VLIW *vliw;
|
1728 |
|
|
int slot;
|
1729 |
|
|
|
1730 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1731 |
|
|
return 0;
|
1732 |
|
|
|
1733 |
|
|
/* The preprocessing can execute right away. */
|
1734 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1735 |
|
|
|
1736 |
|
|
/* If the previous use of the registers was a media op,
|
1737 |
|
|
then their latency may be less than previously recorded.
|
1738 |
|
|
See Table 14-15 in the LSI. */
|
1739 |
|
|
adjust_float_register_busy_for_media (cpu, in_FRi, 1, in_FRj, 1, out_FRk, 1);
|
1740 |
|
|
|
1741 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1742 |
|
|
which is not ready yet. */
|
1743 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1744 |
|
|
ps->post_wait = cycles;
|
1745 |
|
|
vliw = CPU_VLIW (cpu);
|
1746 |
|
|
slot = vliw->next_slot - 1;
|
1747 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1748 |
|
|
post_wait_for_media (cpu, slot);
|
1749 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1750 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1751 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1752 |
|
|
|
1753 |
|
|
/* Restore the busy cycles of the registers we used. */
|
1754 |
|
|
restore_float_register_busy_for_media (cpu, in_FRi, 1, in_FRj, 1, out_FRk, 1);
|
1755 |
|
|
|
1756 |
|
|
/* The latency of tht output register will be at least the latency of the
|
1757 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
1758 |
|
|
if (out_FRk >= 0)
|
1759 |
|
|
{
|
1760 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1761 |
|
|
update_FR_ptime (cpu, out_FRk, 1);
|
1762 |
|
|
/* Mark this use of the register as a media op. */
|
1763 |
|
|
set_use_is_fr_complex_1 (cpu, out_FRk);
|
1764 |
|
|
}
|
1765 |
|
|
|
1766 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
1767 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
1768 |
|
|
|
1769 |
|
|
return cycles;
|
1770 |
|
|
}
|
1771 |
|
|
|
1772 |
|
|
int
|
1773 |
|
|
frvbf_model_fr550_u_media_quad (SIM_CPU *cpu, const IDESC *idesc,
|
1774 |
|
|
int unit_num, int referenced,
|
1775 |
|
|
INT in_FRi, INT in_FRj,
|
1776 |
|
|
INT out_FRk)
|
1777 |
|
|
{
|
1778 |
|
|
int cycles;
|
1779 |
|
|
INT dual_FRi;
|
1780 |
|
|
INT dual_FRj;
|
1781 |
|
|
INT dual_FRk;
|
1782 |
|
|
FRV_PROFILE_STATE *ps;
|
1783 |
|
|
FRV_VLIW *vliw;
|
1784 |
|
|
int slot;
|
1785 |
|
|
|
1786 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1787 |
|
|
return 0;
|
1788 |
|
|
|
1789 |
|
|
/* The preprocessing can execute right away. */
|
1790 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1791 |
|
|
|
1792 |
|
|
dual_FRi = DUAL_REG (in_FRi);
|
1793 |
|
|
dual_FRj = DUAL_REG (in_FRj);
|
1794 |
|
|
dual_FRk = DUAL_REG (out_FRk);
|
1795 |
|
|
|
1796 |
|
|
/* The latency of the registers may be less than previously recorded,
|
1797 |
|
|
depending on how they were used previously.
|
1798 |
|
|
See Table 14-15 in the LSI. */
|
1799 |
|
|
adjust_float_register_busy_for_media (cpu, in_FRi, 2, in_FRj, 2, out_FRk, 2);
|
1800 |
|
|
|
1801 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1802 |
|
|
which is not ready yet. */
|
1803 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1804 |
|
|
ps->post_wait = cycles;
|
1805 |
|
|
vliw = CPU_VLIW (cpu);
|
1806 |
|
|
slot = vliw->next_slot - 1;
|
1807 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1808 |
|
|
post_wait_for_media (cpu, slot);
|
1809 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1810 |
|
|
post_wait_for_FR (cpu, dual_FRi);
|
1811 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1812 |
|
|
post_wait_for_FR (cpu, dual_FRj);
|
1813 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1814 |
|
|
post_wait_for_FR (cpu, dual_FRk);
|
1815 |
|
|
|
1816 |
|
|
/* Restore the busy cycles of the registers we used. */
|
1817 |
|
|
restore_float_register_busy_for_media (cpu, in_FRi, 2, in_FRj, 2, out_FRk, 2);
|
1818 |
|
|
|
1819 |
|
|
/* The latency of the output register will be at least the latency of the
|
1820 |
|
|
other inputs. Once initiated, post-processing take 1 cycle. */
|
1821 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1822 |
|
|
update_FR_ptime (cpu, out_FRk, 1);
|
1823 |
|
|
set_use_is_fr_complex_1 (cpu, out_FRk);
|
1824 |
|
|
|
1825 |
|
|
if (dual_FRk >= 0)
|
1826 |
|
|
{
|
1827 |
|
|
update_FR_latency (cpu, dual_FRk, ps->post_wait);
|
1828 |
|
|
update_FR_ptime (cpu, dual_FRk, 1);
|
1829 |
|
|
set_use_is_fr_complex_1 (cpu, dual_FRk);
|
1830 |
|
|
}
|
1831 |
|
|
|
1832 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
1833 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
1834 |
|
|
|
1835 |
|
|
return cycles;
|
1836 |
|
|
}
|
1837 |
|
|
|
1838 |
|
|
int
|
1839 |
|
|
frvbf_model_fr550_u_media_dual_expand (SIM_CPU *cpu, const IDESC *idesc,
|
1840 |
|
|
int unit_num, int referenced,
|
1841 |
|
|
INT in_FRi, INT out_FRk)
|
1842 |
|
|
{
|
1843 |
|
|
int cycles;
|
1844 |
|
|
INT dual_FRk;
|
1845 |
|
|
FRV_PROFILE_STATE *ps;
|
1846 |
|
|
FRV_VLIW *vliw;
|
1847 |
|
|
int slot;
|
1848 |
|
|
|
1849 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1850 |
|
|
return 0;
|
1851 |
|
|
|
1852 |
|
|
/* The preprocessing can execute right away. */
|
1853 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1854 |
|
|
|
1855 |
|
|
/* If the previous use of the registers was a media op,
|
1856 |
|
|
then their latency will be less than previously recorded.
|
1857 |
|
|
See Table 14-15 in the LSI. */
|
1858 |
|
|
dual_FRk = DUAL_REG (out_FRk);
|
1859 |
|
|
adjust_float_register_busy_for_media (cpu, in_FRi, 1, -1, 1, out_FRk, 2);
|
1860 |
|
|
|
1861 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1862 |
|
|
which is not ready yet. */
|
1863 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1864 |
|
|
ps->post_wait = cycles;
|
1865 |
|
|
vliw = CPU_VLIW (cpu);
|
1866 |
|
|
slot = vliw->next_slot - 1;
|
1867 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1868 |
|
|
post_wait_for_media (cpu, slot);
|
1869 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1870 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1871 |
|
|
post_wait_for_FR (cpu, dual_FRk);
|
1872 |
|
|
|
1873 |
|
|
/* Restore the busy cycles of the registers we used. */
|
1874 |
|
|
restore_float_register_busy_for_media (cpu, in_FRi, 1, -1, 1, out_FRk, 2);
|
1875 |
|
|
|
1876 |
|
|
/* The latency of the output register will be at least the latency of the
|
1877 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
1878 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1879 |
|
|
update_FR_ptime (cpu, out_FRk, 1);
|
1880 |
|
|
set_use_is_fr_complex_1 (cpu, out_FRk);
|
1881 |
|
|
|
1882 |
|
|
if (dual_FRk >= 0)
|
1883 |
|
|
{
|
1884 |
|
|
update_FR_latency (cpu, dual_FRk, ps->post_wait);
|
1885 |
|
|
update_FR_ptime (cpu, dual_FRk, 1);
|
1886 |
|
|
set_use_is_fr_complex_1 (cpu, dual_FRk);
|
1887 |
|
|
}
|
1888 |
|
|
|
1889 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
1890 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
1891 |
|
|
|
1892 |
|
|
return cycles;
|
1893 |
|
|
}
|
1894 |
|
|
|
1895 |
|
|
int
|
1896 |
|
|
frvbf_model_fr550_u_media_3_dual (SIM_CPU *cpu, const IDESC *idesc,
|
1897 |
|
|
int unit_num, int referenced,
|
1898 |
|
|
INT in_FRi, INT out_FRk)
|
1899 |
|
|
{
|
1900 |
|
|
int cycles;
|
1901 |
|
|
INT dual_FRi;
|
1902 |
|
|
FRV_PROFILE_STATE *ps;
|
1903 |
|
|
FRV_VLIW *vliw;
|
1904 |
|
|
int slot;
|
1905 |
|
|
|
1906 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1907 |
|
|
return 0;
|
1908 |
|
|
|
1909 |
|
|
/* The preprocessing can execute right away. */
|
1910 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1911 |
|
|
|
1912 |
|
|
dual_FRi = DUAL_REG (in_FRi);
|
1913 |
|
|
|
1914 |
|
|
/* The latency of the registers may be less than previously recorded,
|
1915 |
|
|
depending on how they were used previously.
|
1916 |
|
|
See Table 14-15 in the LSI. */
|
1917 |
|
|
adjust_float_register_busy_for_media (cpu, in_FRi, 2, -1, 1, out_FRk, 1);
|
1918 |
|
|
|
1919 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1920 |
|
|
which is not ready yet. */
|
1921 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1922 |
|
|
ps->post_wait = cycles;
|
1923 |
|
|
vliw = CPU_VLIW (cpu);
|
1924 |
|
|
slot = vliw->next_slot - 1;
|
1925 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1926 |
|
|
post_wait_for_media (cpu, slot);
|
1927 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
1928 |
|
|
post_wait_for_FR (cpu, dual_FRi);
|
1929 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1930 |
|
|
|
1931 |
|
|
/* Restore the busy cycles of the registers we used. */
|
1932 |
|
|
restore_float_register_busy_for_media (cpu, in_FRi, 2, -1, 1, out_FRk, 1);
|
1933 |
|
|
|
1934 |
|
|
/* The latency of the output register will be at least the latency of the
|
1935 |
|
|
other inputs. Once initiated, post-processing takes 1 cycle. */
|
1936 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1937 |
|
|
update_FR_ptime (cpu, out_FRk, 1);
|
1938 |
|
|
|
1939 |
|
|
set_use_is_fr_complex_1 (cpu, out_FRk);
|
1940 |
|
|
|
1941 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
1942 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
1943 |
|
|
|
1944 |
|
|
return cycles;
|
1945 |
|
|
}
|
1946 |
|
|
|
1947 |
|
|
int
|
1948 |
|
|
frvbf_model_fr550_u_media_3_acc (SIM_CPU *cpu, const IDESC *idesc,
|
1949 |
|
|
int unit_num, int referenced,
|
1950 |
|
|
INT in_FRj, INT in_ACC40Si,
|
1951 |
|
|
INT out_FRk)
|
1952 |
|
|
{
|
1953 |
|
|
int cycles;
|
1954 |
|
|
FRV_PROFILE_STATE *ps;
|
1955 |
|
|
FRV_VLIW *vliw;
|
1956 |
|
|
int slot;
|
1957 |
|
|
|
1958 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
1959 |
|
|
return 0;
|
1960 |
|
|
|
1961 |
|
|
/* The preprocessing can execute right away. */
|
1962 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
1963 |
|
|
|
1964 |
|
|
/* If the previous use of the registers was a media op,
|
1965 |
|
|
then their latency will be less than previously recorded.
|
1966 |
|
|
See Table 14-15 in the LSI. */
|
1967 |
|
|
adjust_float_register_busy_for_media (cpu, -1, 1, in_FRj, 1, out_FRk, 1);
|
1968 |
|
|
|
1969 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
1970 |
|
|
which is not ready yet. */
|
1971 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
1972 |
|
|
ps->post_wait = cycles;
|
1973 |
|
|
vliw = CPU_VLIW (cpu);
|
1974 |
|
|
slot = vliw->next_slot - 1;
|
1975 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
1976 |
|
|
post_wait_for_media (cpu, slot);
|
1977 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
1978 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
1979 |
|
|
post_wait_for_ACC (cpu, in_ACC40Si);
|
1980 |
|
|
|
1981 |
|
|
/* Restore the busy cycles of the registers we used. */
|
1982 |
|
|
restore_float_register_busy_for_media (cpu, -1, 1, in_FRj, 1, out_FRk, 1);
|
1983 |
|
|
|
1984 |
|
|
/* The latency of tht output register will be at least the latency of the
|
1985 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
1986 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
1987 |
|
|
update_FR_ptime (cpu, out_FRk, 1);
|
1988 |
|
|
|
1989 |
|
|
set_use_is_fr_complex_1 (cpu, out_FRk);
|
1990 |
|
|
|
1991 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
1992 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
1993 |
|
|
|
1994 |
|
|
return cycles;
|
1995 |
|
|
}
|
1996 |
|
|
|
1997 |
|
|
int
|
1998 |
|
|
frvbf_model_fr550_u_media_3_acc_dual (SIM_CPU *cpu, const IDESC *idesc,
|
1999 |
|
|
int unit_num, int referenced,
|
2000 |
|
|
INT in_ACC40Si, INT out_FRk)
|
2001 |
|
|
{
|
2002 |
|
|
int cycles;
|
2003 |
|
|
FRV_PROFILE_STATE *ps;
|
2004 |
|
|
INT ACC40Si_1;
|
2005 |
|
|
INT dual_FRk;
|
2006 |
|
|
FRV_VLIW *vliw;
|
2007 |
|
|
int slot;
|
2008 |
|
|
|
2009 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2010 |
|
|
return 0;
|
2011 |
|
|
|
2012 |
|
|
/* The preprocessing can execute right away. */
|
2013 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2014 |
|
|
|
2015 |
|
|
ACC40Si_1 = DUAL_REG (in_ACC40Si);
|
2016 |
|
|
dual_FRk = DUAL_REG (out_FRk);
|
2017 |
|
|
|
2018 |
|
|
/* If the previous use of the registers was a media op,
|
2019 |
|
|
then their latency will be less than previously recorded.
|
2020 |
|
|
See Table 14-15 in the LSI. */
|
2021 |
|
|
adjust_float_register_busy_for_media (cpu, -1, 1, -1, 1, out_FRk, 2);
|
2022 |
|
|
|
2023 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
2024 |
|
|
which is not ready yet. */
|
2025 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2026 |
|
|
ps->post_wait = cycles;
|
2027 |
|
|
vliw = CPU_VLIW (cpu);
|
2028 |
|
|
slot = vliw->next_slot - 1;
|
2029 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2030 |
|
|
post_wait_for_media (cpu, slot);
|
2031 |
|
|
post_wait_for_ACC (cpu, in_ACC40Si);
|
2032 |
|
|
post_wait_for_ACC (cpu, ACC40Si_1);
|
2033 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
2034 |
|
|
post_wait_for_FR (cpu, dual_FRk);
|
2035 |
|
|
|
2036 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2037 |
|
|
restore_float_register_busy_for_media (cpu, -1, 1, -1, 1, out_FRk, 2);
|
2038 |
|
|
|
2039 |
|
|
/* The latency of the output register will be at least the latency of the
|
2040 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2041 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
2042 |
|
|
update_FR_ptime (cpu, out_FRk, 1);
|
2043 |
|
|
set_use_is_fr_complex_1 (cpu, out_FRk);
|
2044 |
|
|
if (dual_FRk >= 0)
|
2045 |
|
|
{
|
2046 |
|
|
update_FR_latency (cpu, dual_FRk, ps->post_wait);
|
2047 |
|
|
update_FR_ptime (cpu, dual_FRk, 1);
|
2048 |
|
|
set_use_is_fr_complex_1 (cpu, dual_FRk);
|
2049 |
|
|
}
|
2050 |
|
|
|
2051 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2052 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2053 |
|
|
|
2054 |
|
|
return cycles;
|
2055 |
|
|
}
|
2056 |
|
|
|
2057 |
|
|
int
|
2058 |
|
|
frvbf_model_fr550_u_media_3_wtacc (SIM_CPU *cpu, const IDESC *idesc,
|
2059 |
|
|
int unit_num, int referenced,
|
2060 |
|
|
INT in_FRi, INT out_ACC40Sk)
|
2061 |
|
|
{
|
2062 |
|
|
int cycles;
|
2063 |
|
|
FRV_PROFILE_STATE *ps;
|
2064 |
|
|
FRV_VLIW *vliw;
|
2065 |
|
|
int slot;
|
2066 |
|
|
|
2067 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2068 |
|
|
return 0;
|
2069 |
|
|
|
2070 |
|
|
/* The preprocessing can execute right away. */
|
2071 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2072 |
|
|
|
2073 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2074 |
|
|
|
2075 |
|
|
/* The latency of the registers may be less than previously recorded,
|
2076 |
|
|
depending on how they were used previously.
|
2077 |
|
|
See Table 14-15 in the LSI. */
|
2078 |
|
|
adjust_float_register_busy_for_media (cpu, in_FRi, 1, -1, 1, -1, 1);
|
2079 |
|
|
|
2080 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
2081 |
|
|
which is not ready yet. */
|
2082 |
|
|
ps->post_wait = cycles;
|
2083 |
|
|
vliw = CPU_VLIW (cpu);
|
2084 |
|
|
slot = vliw->next_slot - 1;
|
2085 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2086 |
|
|
post_wait_for_media (cpu, slot);
|
2087 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
2088 |
|
|
post_wait_for_ACC (cpu, out_ACC40Sk);
|
2089 |
|
|
|
2090 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2091 |
|
|
restore_float_register_busy_for_media (cpu, in_FRi, 1, -1, 1, -1, 1);
|
2092 |
|
|
|
2093 |
|
|
/* The latency of the output register will be at least the latency of the
|
2094 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2095 |
|
|
update_ACC_latency (cpu, out_ACC40Sk, ps->post_wait);
|
2096 |
|
|
update_ACC_ptime (cpu, out_ACC40Sk, 1);
|
2097 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Sk);
|
2098 |
|
|
|
2099 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2100 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2101 |
|
|
|
2102 |
|
|
return cycles;
|
2103 |
|
|
}
|
2104 |
|
|
|
2105 |
|
|
int
|
2106 |
|
|
frvbf_model_fr550_u_media_3_mclracc (SIM_CPU *cpu, const IDESC *idesc,
|
2107 |
|
|
int unit_num, int referenced)
|
2108 |
|
|
{
|
2109 |
|
|
int cycles;
|
2110 |
|
|
FRV_PROFILE_STATE *ps;
|
2111 |
|
|
FRV_VLIW *vliw;
|
2112 |
|
|
int slot;
|
2113 |
|
|
int i;
|
2114 |
|
|
|
2115 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2116 |
|
|
return 0;
|
2117 |
|
|
|
2118 |
|
|
/* The preprocessing can execute right away. */
|
2119 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2120 |
|
|
|
2121 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2122 |
|
|
|
2123 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
2124 |
|
|
which is not ready yet. */
|
2125 |
|
|
ps->post_wait = cycles;
|
2126 |
|
|
vliw = CPU_VLIW (cpu);
|
2127 |
|
|
slot = vliw->next_slot - 1;
|
2128 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2129 |
|
|
post_wait_for_media (cpu, slot);
|
2130 |
|
|
|
2131 |
|
|
/* If A was 1 and the accumulator was ACC0, then we must check all
|
2132 |
|
|
accumulators. Otherwise just wait for the specified accumulator. */
|
2133 |
|
|
if (ps->mclracc_A && ps->mclracc_acc == 0)
|
2134 |
|
|
{
|
2135 |
|
|
for (i = 0; i < 8; ++i)
|
2136 |
|
|
post_wait_for_ACC (cpu, i);
|
2137 |
|
|
}
|
2138 |
|
|
else
|
2139 |
|
|
post_wait_for_ACC (cpu, ps->mclracc_acc);
|
2140 |
|
|
|
2141 |
|
|
/* The latency of the output registers will be at least the latency of the
|
2142 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2143 |
|
|
if (ps->mclracc_A && ps->mclracc_acc == 0)
|
2144 |
|
|
{
|
2145 |
|
|
for (i = 0; i < 8; ++i)
|
2146 |
|
|
{
|
2147 |
|
|
update_ACC_latency (cpu, i, ps->post_wait);
|
2148 |
|
|
update_ACC_ptime (cpu, i, 1);
|
2149 |
|
|
set_use_is_acc_mmac (cpu, i);
|
2150 |
|
|
}
|
2151 |
|
|
}
|
2152 |
|
|
else
|
2153 |
|
|
{
|
2154 |
|
|
update_ACC_latency (cpu, ps->mclracc_acc, ps->post_wait);
|
2155 |
|
|
update_ACC_ptime (cpu, ps->mclracc_acc, 1);
|
2156 |
|
|
set_use_is_acc_mmac (cpu, ps->mclracc_acc);
|
2157 |
|
|
}
|
2158 |
|
|
|
2159 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2160 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2161 |
|
|
|
2162 |
|
|
return cycles;
|
2163 |
|
|
}
|
2164 |
|
|
|
2165 |
|
|
int
|
2166 |
|
|
frvbf_model_fr550_u_media_set (SIM_CPU *cpu, const IDESC *idesc,
|
2167 |
|
|
int unit_num, int referenced,
|
2168 |
|
|
INT out_FRk)
|
2169 |
|
|
{
|
2170 |
|
|
int cycles;
|
2171 |
|
|
FRV_PROFILE_STATE *ps;
|
2172 |
|
|
FRV_VLIW *vliw;
|
2173 |
|
|
int slot;
|
2174 |
|
|
|
2175 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2176 |
|
|
return 0;
|
2177 |
|
|
|
2178 |
|
|
/* The preprocessing can execute right away. */
|
2179 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2180 |
|
|
|
2181 |
|
|
/* If the previous use of the registers was a media op,
|
2182 |
|
|
then their latency will be less than previously recorded.
|
2183 |
|
|
See Table 14-15 in the LSI. */
|
2184 |
|
|
adjust_float_register_busy_for_media (cpu, -1, 1, -1, 1, out_FRk, 1);
|
2185 |
|
|
|
2186 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
2187 |
|
|
which is not ready yet. */
|
2188 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2189 |
|
|
ps->post_wait = cycles;
|
2190 |
|
|
vliw = CPU_VLIW (cpu);
|
2191 |
|
|
slot = vliw->next_slot - 1;
|
2192 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2193 |
|
|
post_wait_for_media (cpu, slot);
|
2194 |
|
|
post_wait_for_FR (cpu, out_FRk);
|
2195 |
|
|
|
2196 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2197 |
|
|
restore_float_register_busy_for_media (cpu, -1, 1, -1, 1, out_FRk, 1);
|
2198 |
|
|
|
2199 |
|
|
/* The latency of the output register will be at least the latency of the
|
2200 |
|
|
other inputs. Once initiated, post-processing takes 1 cycle. */
|
2201 |
|
|
update_FR_latency (cpu, out_FRk, ps->post_wait);
|
2202 |
|
|
update_FR_ptime (cpu, out_FRk, 1);
|
2203 |
|
|
fr550_reset_acc_flags (cpu, out_FRk);
|
2204 |
|
|
|
2205 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2206 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2207 |
|
|
|
2208 |
|
|
return cycles;
|
2209 |
|
|
}
|
2210 |
|
|
|
2211 |
|
|
int
|
2212 |
|
|
frvbf_model_fr550_u_media_4 (SIM_CPU *cpu, const IDESC *idesc,
|
2213 |
|
|
int unit_num, int referenced,
|
2214 |
|
|
INT in_FRi, INT in_FRj,
|
2215 |
|
|
INT out_ACC40Sk, INT out_ACC40Uk)
|
2216 |
|
|
{
|
2217 |
|
|
int cycles;
|
2218 |
|
|
INT dual_ACC40Sk;
|
2219 |
|
|
INT dual_ACC40Uk;
|
2220 |
|
|
FRV_PROFILE_STATE *ps;
|
2221 |
|
|
FRV_VLIW *vliw;
|
2222 |
|
|
int slot;
|
2223 |
|
|
|
2224 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2225 |
|
|
return 0;
|
2226 |
|
|
|
2227 |
|
|
/* The preprocessing can execute right away. */
|
2228 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2229 |
|
|
|
2230 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2231 |
|
|
dual_ACC40Sk = DUAL_REG (out_ACC40Sk);
|
2232 |
|
|
dual_ACC40Uk = DUAL_REG (out_ACC40Uk);
|
2233 |
|
|
|
2234 |
|
|
/* The latency of the registers may be less than previously recorded,
|
2235 |
|
|
depending on how they were used previously.
|
2236 |
|
|
See Table 14-15 in the LSI. */
|
2237 |
|
|
adjust_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Sk, 2);
|
2238 |
|
|
adjust_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Uk, 2);
|
2239 |
|
|
|
2240 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
2241 |
|
|
which is not ready yet. */
|
2242 |
|
|
ps->post_wait = cycles;
|
2243 |
|
|
vliw = CPU_VLIW (cpu);
|
2244 |
|
|
slot = vliw->next_slot - 1;
|
2245 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2246 |
|
|
post_wait_for_media (cpu, slot);
|
2247 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
2248 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
2249 |
|
|
post_wait_for_ACC (cpu, out_ACC40Sk);
|
2250 |
|
|
post_wait_for_ACC (cpu, dual_ACC40Sk);
|
2251 |
|
|
post_wait_for_ACC (cpu, out_ACC40Uk);
|
2252 |
|
|
post_wait_for_ACC (cpu, dual_ACC40Uk);
|
2253 |
|
|
|
2254 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2255 |
|
|
restore_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Sk, 2);
|
2256 |
|
|
restore_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Uk, 2);
|
2257 |
|
|
|
2258 |
|
|
/* The latency of the output register will be at least the latency of the
|
2259 |
|
|
other inputs. Once initiated, post-processing will take 1 cycles. */
|
2260 |
|
|
if (out_ACC40Sk >= 0)
|
2261 |
|
|
{
|
2262 |
|
|
update_ACC_latency (cpu, out_ACC40Sk, ps->post_wait + 1);
|
2263 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Sk);
|
2264 |
|
|
}
|
2265 |
|
|
if (dual_ACC40Sk >= 0)
|
2266 |
|
|
{
|
2267 |
|
|
update_ACC_latency (cpu, dual_ACC40Sk, ps->post_wait + 1);
|
2268 |
|
|
set_use_is_acc_mmac (cpu, dual_ACC40Sk);
|
2269 |
|
|
}
|
2270 |
|
|
if (out_ACC40Uk >= 0)
|
2271 |
|
|
{
|
2272 |
|
|
update_ACC_latency (cpu, out_ACC40Uk, ps->post_wait + 1);
|
2273 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Uk);
|
2274 |
|
|
}
|
2275 |
|
|
if (dual_ACC40Uk >= 0)
|
2276 |
|
|
{
|
2277 |
|
|
update_ACC_latency (cpu, dual_ACC40Uk, ps->post_wait + 1);
|
2278 |
|
|
set_use_is_acc_mmac (cpu, dual_ACC40Uk);
|
2279 |
|
|
}
|
2280 |
|
|
|
2281 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2282 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2283 |
|
|
|
2284 |
|
|
return cycles;
|
2285 |
|
|
}
|
2286 |
|
|
|
2287 |
|
|
int
|
2288 |
|
|
frvbf_model_fr550_u_media_4_acc (SIM_CPU *cpu, const IDESC *idesc,
|
2289 |
|
|
int unit_num, int referenced,
|
2290 |
|
|
INT in_ACC40Si, INT out_ACC40Sk)
|
2291 |
|
|
{
|
2292 |
|
|
int cycles;
|
2293 |
|
|
INT ACC40Si_1;
|
2294 |
|
|
FRV_PROFILE_STATE *ps;
|
2295 |
|
|
FRV_VLIW *vliw;
|
2296 |
|
|
int slot;
|
2297 |
|
|
|
2298 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2299 |
|
|
return 0;
|
2300 |
|
|
|
2301 |
|
|
/* The preprocessing can execute right away. */
|
2302 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2303 |
|
|
|
2304 |
|
|
ACC40Si_1 = DUAL_REG (in_ACC40Si);
|
2305 |
|
|
|
2306 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2307 |
|
|
/* The latency of the registers may be less than previously recorded,
|
2308 |
|
|
depending on how they were used previously.
|
2309 |
|
|
See Table 14-15 in the LSI. */
|
2310 |
|
|
adjust_acc_busy_for_mmac (cpu, in_ACC40Si, 2, out_ACC40Sk, 1);
|
2311 |
|
|
|
2312 |
|
|
/* The post processing must wait if there is a dependency on a register
|
2313 |
|
|
which is not ready yet. */
|
2314 |
|
|
ps->post_wait = cycles;
|
2315 |
|
|
vliw = CPU_VLIW (cpu);
|
2316 |
|
|
slot = vliw->next_slot - 1;
|
2317 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2318 |
|
|
post_wait_for_media (cpu, slot);
|
2319 |
|
|
post_wait_for_ACC (cpu, in_ACC40Si);
|
2320 |
|
|
post_wait_for_ACC (cpu, ACC40Si_1);
|
2321 |
|
|
post_wait_for_ACC (cpu, out_ACC40Sk);
|
2322 |
|
|
|
2323 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2324 |
|
|
restore_acc_busy_for_mmac (cpu, in_ACC40Si, 2, out_ACC40Sk, 1);
|
2325 |
|
|
|
2326 |
|
|
/* The latency of the output register will be at least the latency of the
|
2327 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2328 |
|
|
update_ACC_latency (cpu, out_ACC40Sk, ps->post_wait + 1);
|
2329 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Sk);
|
2330 |
|
|
|
2331 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2332 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2333 |
|
|
|
2334 |
|
|
return cycles;
|
2335 |
|
|
}
|
2336 |
|
|
|
2337 |
|
|
int
|
2338 |
|
|
frvbf_model_fr550_u_media_4_acc_dual (SIM_CPU *cpu, const IDESC *idesc,
|
2339 |
|
|
int unit_num, int referenced,
|
2340 |
|
|
INT in_ACC40Si, INT out_ACC40Sk)
|
2341 |
|
|
{
|
2342 |
|
|
int cycles;
|
2343 |
|
|
INT ACC40Si_1;
|
2344 |
|
|
INT ACC40Si_2;
|
2345 |
|
|
INT ACC40Si_3;
|
2346 |
|
|
INT ACC40Sk_1;
|
2347 |
|
|
FRV_PROFILE_STATE *ps;
|
2348 |
|
|
FRV_VLIW *vliw;
|
2349 |
|
|
int slot;
|
2350 |
|
|
|
2351 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2352 |
|
|
return 0;
|
2353 |
|
|
|
2354 |
|
|
/* The preprocessing can execute right away. */
|
2355 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2356 |
|
|
|
2357 |
|
|
ACC40Si_1 = DUAL_REG (in_ACC40Si);
|
2358 |
|
|
ACC40Si_2 = DUAL_REG (ACC40Si_1);
|
2359 |
|
|
ACC40Si_3 = DUAL_REG (ACC40Si_2);
|
2360 |
|
|
ACC40Sk_1 = DUAL_REG (out_ACC40Sk);
|
2361 |
|
|
|
2362 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2363 |
|
|
/* The latency of the registers may be less than previously recorded,
|
2364 |
|
|
depending on how they were used previously.
|
2365 |
|
|
See Table 14-15 in the LSI. */
|
2366 |
|
|
adjust_acc_busy_for_mmac (cpu, in_ACC40Si, 4, out_ACC40Sk, 2);
|
2367 |
|
|
|
2368 |
|
|
/* The post processing must wait if there is a dependency on a register
|
2369 |
|
|
which is not ready yet. */
|
2370 |
|
|
ps->post_wait = cycles;
|
2371 |
|
|
vliw = CPU_VLIW (cpu);
|
2372 |
|
|
slot = vliw->next_slot - 1;
|
2373 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2374 |
|
|
post_wait_for_media (cpu, slot);
|
2375 |
|
|
post_wait_for_ACC (cpu, in_ACC40Si);
|
2376 |
|
|
post_wait_for_ACC (cpu, ACC40Si_1);
|
2377 |
|
|
post_wait_for_ACC (cpu, ACC40Si_2);
|
2378 |
|
|
post_wait_for_ACC (cpu, ACC40Si_3);
|
2379 |
|
|
post_wait_for_ACC (cpu, out_ACC40Sk);
|
2380 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_1);
|
2381 |
|
|
|
2382 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2383 |
|
|
restore_acc_busy_for_mmac (cpu, in_ACC40Si, 4, out_ACC40Sk, 2);
|
2384 |
|
|
|
2385 |
|
|
/* The latency of the output register will be at least the latency of the
|
2386 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2387 |
|
|
update_ACC_latency (cpu, out_ACC40Sk, ps->post_wait + 1);
|
2388 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Sk);
|
2389 |
|
|
if (ACC40Sk_1 >= 0)
|
2390 |
|
|
{
|
2391 |
|
|
update_ACC_latency (cpu, ACC40Sk_1, ps->post_wait + 1);
|
2392 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_1);
|
2393 |
|
|
}
|
2394 |
|
|
|
2395 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2396 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2397 |
|
|
|
2398 |
|
|
return cycles;
|
2399 |
|
|
}
|
2400 |
|
|
|
2401 |
|
|
int
|
2402 |
|
|
frvbf_model_fr550_u_media_4_add_sub (SIM_CPU *cpu, const IDESC *idesc,
|
2403 |
|
|
int unit_num, int referenced,
|
2404 |
|
|
INT in_ACC40Si, INT out_ACC40Sk)
|
2405 |
|
|
{
|
2406 |
|
|
int cycles;
|
2407 |
|
|
INT ACC40Si_1;
|
2408 |
|
|
INT ACC40Sk_1;
|
2409 |
|
|
FRV_PROFILE_STATE *ps;
|
2410 |
|
|
FRV_VLIW *vliw;
|
2411 |
|
|
int slot;
|
2412 |
|
|
|
2413 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2414 |
|
|
return 0;
|
2415 |
|
|
|
2416 |
|
|
/* The preprocessing can execute right away. */
|
2417 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2418 |
|
|
|
2419 |
|
|
ACC40Si_1 = DUAL_REG (in_ACC40Si);
|
2420 |
|
|
ACC40Sk_1 = DUAL_REG (out_ACC40Sk);
|
2421 |
|
|
|
2422 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2423 |
|
|
/* The latency of the registers may be less than previously recorded,
|
2424 |
|
|
depending on how they were used previously.
|
2425 |
|
|
See Table 14-15 in the LSI. */
|
2426 |
|
|
adjust_acc_busy_for_mmac (cpu, in_ACC40Si, 2, out_ACC40Sk, 2);
|
2427 |
|
|
|
2428 |
|
|
/* The post processing must wait if there is a dependency on a register
|
2429 |
|
|
which is not ready yet. */
|
2430 |
|
|
ps->post_wait = cycles;
|
2431 |
|
|
vliw = CPU_VLIW (cpu);
|
2432 |
|
|
slot = vliw->next_slot - 1;
|
2433 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2434 |
|
|
post_wait_for_media (cpu, slot);
|
2435 |
|
|
post_wait_for_ACC (cpu, in_ACC40Si);
|
2436 |
|
|
post_wait_for_ACC (cpu, ACC40Si_1);
|
2437 |
|
|
post_wait_for_ACC (cpu, out_ACC40Sk);
|
2438 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_1);
|
2439 |
|
|
|
2440 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2441 |
|
|
restore_acc_busy_for_mmac (cpu, in_ACC40Si, 2, out_ACC40Sk, 2);
|
2442 |
|
|
|
2443 |
|
|
/* The latency of the output register will be at least the latency of the
|
2444 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2445 |
|
|
update_ACC_latency (cpu, out_ACC40Sk, ps->post_wait + 1);
|
2446 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Sk);
|
2447 |
|
|
if (ACC40Sk_1 >= 0)
|
2448 |
|
|
{
|
2449 |
|
|
update_ACC_latency (cpu, ACC40Sk_1, ps->post_wait + 1);
|
2450 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_1);
|
2451 |
|
|
}
|
2452 |
|
|
|
2453 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2454 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2455 |
|
|
|
2456 |
|
|
return cycles;
|
2457 |
|
|
}
|
2458 |
|
|
|
2459 |
|
|
int
|
2460 |
|
|
frvbf_model_fr550_u_media_4_add_sub_dual (SIM_CPU *cpu, const IDESC *idesc,
|
2461 |
|
|
int unit_num, int referenced,
|
2462 |
|
|
INT in_ACC40Si, INT out_ACC40Sk)
|
2463 |
|
|
{
|
2464 |
|
|
int cycles;
|
2465 |
|
|
INT ACC40Si_1;
|
2466 |
|
|
INT ACC40Si_2;
|
2467 |
|
|
INT ACC40Si_3;
|
2468 |
|
|
INT ACC40Sk_1;
|
2469 |
|
|
INT ACC40Sk_2;
|
2470 |
|
|
INT ACC40Sk_3;
|
2471 |
|
|
FRV_PROFILE_STATE *ps;
|
2472 |
|
|
FRV_VLIW *vliw;
|
2473 |
|
|
int slot;
|
2474 |
|
|
|
2475 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2476 |
|
|
return 0;
|
2477 |
|
|
|
2478 |
|
|
/* The preprocessing can execute right away. */
|
2479 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2480 |
|
|
|
2481 |
|
|
ACC40Si_1 = DUAL_REG (in_ACC40Si);
|
2482 |
|
|
ACC40Si_2 = DUAL_REG (ACC40Si_1);
|
2483 |
|
|
ACC40Si_3 = DUAL_REG (ACC40Si_2);
|
2484 |
|
|
ACC40Sk_1 = DUAL_REG (out_ACC40Sk);
|
2485 |
|
|
ACC40Sk_2 = DUAL_REG (ACC40Sk_1);
|
2486 |
|
|
ACC40Sk_3 = DUAL_REG (ACC40Sk_2);
|
2487 |
|
|
|
2488 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2489 |
|
|
/* The latency of the registers may be less than previously recorded,
|
2490 |
|
|
depending on how they were used previously.
|
2491 |
|
|
See Table 14-15 in the LSI. */
|
2492 |
|
|
adjust_acc_busy_for_mmac (cpu, in_ACC40Si, 4, out_ACC40Sk, 4);
|
2493 |
|
|
|
2494 |
|
|
/* The post processing must wait if there is a dependency on a register
|
2495 |
|
|
which is not ready yet. */
|
2496 |
|
|
ps->post_wait = cycles;
|
2497 |
|
|
vliw = CPU_VLIW (cpu);
|
2498 |
|
|
slot = vliw->next_slot - 1;
|
2499 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2500 |
|
|
post_wait_for_media (cpu, slot);
|
2501 |
|
|
post_wait_for_ACC (cpu, in_ACC40Si);
|
2502 |
|
|
post_wait_for_ACC (cpu, ACC40Si_1);
|
2503 |
|
|
post_wait_for_ACC (cpu, ACC40Si_2);
|
2504 |
|
|
post_wait_for_ACC (cpu, ACC40Si_3);
|
2505 |
|
|
post_wait_for_ACC (cpu, out_ACC40Sk);
|
2506 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_1);
|
2507 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_2);
|
2508 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_3);
|
2509 |
|
|
|
2510 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2511 |
|
|
restore_acc_busy_for_mmac (cpu, in_ACC40Si, 4, out_ACC40Sk, 4);
|
2512 |
|
|
|
2513 |
|
|
/* The latency of the output register will be at least the latency of the
|
2514 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2515 |
|
|
update_ACC_latency (cpu, out_ACC40Sk, ps->post_wait + 1);
|
2516 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Sk);
|
2517 |
|
|
if (ACC40Sk_1 >= 0)
|
2518 |
|
|
{
|
2519 |
|
|
update_ACC_latency (cpu, ACC40Sk_1, ps->post_wait + 1);
|
2520 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_1);
|
2521 |
|
|
}
|
2522 |
|
|
if (ACC40Sk_2 >= 0)
|
2523 |
|
|
{
|
2524 |
|
|
update_ACC_latency (cpu, ACC40Sk_2, ps->post_wait + 1);
|
2525 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_2);
|
2526 |
|
|
}
|
2527 |
|
|
if (ACC40Sk_3 >= 0)
|
2528 |
|
|
{
|
2529 |
|
|
update_ACC_latency (cpu, ACC40Sk_3, ps->post_wait + 1);
|
2530 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_3);
|
2531 |
|
|
}
|
2532 |
|
|
|
2533 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2534 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2535 |
|
|
|
2536 |
|
|
return cycles;
|
2537 |
|
|
}
|
2538 |
|
|
|
2539 |
|
|
int
|
2540 |
|
|
frvbf_model_fr550_u_media_4_quad (SIM_CPU *cpu, const IDESC *idesc,
|
2541 |
|
|
int unit_num, int referenced,
|
2542 |
|
|
INT in_FRi, INT in_FRj,
|
2543 |
|
|
INT out_ACC40Sk, INT out_ACC40Uk)
|
2544 |
|
|
{
|
2545 |
|
|
int cycles;
|
2546 |
|
|
INT dual_FRi;
|
2547 |
|
|
INT dual_FRj;
|
2548 |
|
|
INT ACC40Sk_1;
|
2549 |
|
|
INT ACC40Sk_2;
|
2550 |
|
|
INT ACC40Sk_3;
|
2551 |
|
|
INT ACC40Uk_1;
|
2552 |
|
|
INT ACC40Uk_2;
|
2553 |
|
|
INT ACC40Uk_3;
|
2554 |
|
|
FRV_PROFILE_STATE *ps;
|
2555 |
|
|
FRV_VLIW *vliw;
|
2556 |
|
|
int slot;
|
2557 |
|
|
|
2558 |
|
|
if (model_insn == FRV_INSN_MODEL_PASS_1)
|
2559 |
|
|
return 0;
|
2560 |
|
|
|
2561 |
|
|
/* The preprocessing can execute right away. */
|
2562 |
|
|
cycles = idesc->timing->units[unit_num].done;
|
2563 |
|
|
|
2564 |
|
|
dual_FRi = DUAL_REG (in_FRi);
|
2565 |
|
|
dual_FRj = DUAL_REG (in_FRj);
|
2566 |
|
|
ACC40Sk_1 = DUAL_REG (out_ACC40Sk);
|
2567 |
|
|
ACC40Sk_2 = DUAL_REG (ACC40Sk_1);
|
2568 |
|
|
ACC40Sk_3 = DUAL_REG (ACC40Sk_2);
|
2569 |
|
|
ACC40Uk_1 = DUAL_REG (out_ACC40Uk);
|
2570 |
|
|
ACC40Uk_2 = DUAL_REG (ACC40Uk_1);
|
2571 |
|
|
ACC40Uk_3 = DUAL_REG (ACC40Uk_2);
|
2572 |
|
|
|
2573 |
|
|
ps = CPU_PROFILE_STATE (cpu);
|
2574 |
|
|
/* The latency of the registers may be less than previously recorded,
|
2575 |
|
|
depending on how they were used previously.
|
2576 |
|
|
See Table 14-15 in the LSI. */
|
2577 |
|
|
adjust_float_register_busy_for_media (cpu, in_FRi, 2, in_FRj, 2, -1, 1);
|
2578 |
|
|
adjust_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Sk, 4);
|
2579 |
|
|
adjust_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Uk, 4);
|
2580 |
|
|
|
2581 |
|
|
/* The post processing must wait if there is a dependency on a FR
|
2582 |
|
|
which is not ready yet. */
|
2583 |
|
|
ps->post_wait = cycles;
|
2584 |
|
|
vliw = CPU_VLIW (cpu);
|
2585 |
|
|
slot = vliw->next_slot - 1;
|
2586 |
|
|
slot = (*vliw->current_vliw)[slot] - UNIT_FM0;
|
2587 |
|
|
post_wait_for_media (cpu, slot);
|
2588 |
|
|
post_wait_for_FR (cpu, in_FRi);
|
2589 |
|
|
post_wait_for_FR (cpu, dual_FRi);
|
2590 |
|
|
post_wait_for_FR (cpu, in_FRj);
|
2591 |
|
|
post_wait_for_FR (cpu, dual_FRj);
|
2592 |
|
|
post_wait_for_ACC (cpu, out_ACC40Sk);
|
2593 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_1);
|
2594 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_2);
|
2595 |
|
|
post_wait_for_ACC (cpu, ACC40Sk_3);
|
2596 |
|
|
post_wait_for_ACC (cpu, out_ACC40Uk);
|
2597 |
|
|
post_wait_for_ACC (cpu, ACC40Uk_1);
|
2598 |
|
|
post_wait_for_ACC (cpu, ACC40Uk_2);
|
2599 |
|
|
post_wait_for_ACC (cpu, ACC40Uk_3);
|
2600 |
|
|
|
2601 |
|
|
/* Restore the busy cycles of the registers we used. */
|
2602 |
|
|
restore_float_register_busy_for_media (cpu, in_FRi, 2, in_FRj, 2, -1, 1);
|
2603 |
|
|
restore_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Sk, 4);
|
2604 |
|
|
restore_acc_busy_for_mmac (cpu, -1, 1, out_ACC40Uk, 4);
|
2605 |
|
|
|
2606 |
|
|
/* The latency of the output register will be at least the latency of the
|
2607 |
|
|
other inputs. Once initiated, post-processing will take 1 cycle. */
|
2608 |
|
|
if (out_ACC40Sk >= 0)
|
2609 |
|
|
{
|
2610 |
|
|
update_ACC_latency (cpu, out_ACC40Sk, ps->post_wait + 1);
|
2611 |
|
|
|
2612 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Sk);
|
2613 |
|
|
if (ACC40Sk_1 >= 0)
|
2614 |
|
|
{
|
2615 |
|
|
update_ACC_latency (cpu, ACC40Sk_1, ps->post_wait + 1);
|
2616 |
|
|
|
2617 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_1);
|
2618 |
|
|
}
|
2619 |
|
|
if (ACC40Sk_2 >= 0)
|
2620 |
|
|
{
|
2621 |
|
|
update_ACC_latency (cpu, ACC40Sk_2, ps->post_wait + 1);
|
2622 |
|
|
|
2623 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_2);
|
2624 |
|
|
}
|
2625 |
|
|
if (ACC40Sk_3 >= 0)
|
2626 |
|
|
{
|
2627 |
|
|
update_ACC_latency (cpu, ACC40Sk_3, ps->post_wait + 1);
|
2628 |
|
|
|
2629 |
|
|
set_use_is_acc_mmac (cpu, ACC40Sk_3);
|
2630 |
|
|
}
|
2631 |
|
|
}
|
2632 |
|
|
else if (out_ACC40Uk >= 0)
|
2633 |
|
|
{
|
2634 |
|
|
update_ACC_latency (cpu, out_ACC40Uk, ps->post_wait + 1);
|
2635 |
|
|
|
2636 |
|
|
set_use_is_acc_mmac (cpu, out_ACC40Uk);
|
2637 |
|
|
if (ACC40Uk_1 >= 0)
|
2638 |
|
|
{
|
2639 |
|
|
update_ACC_latency (cpu, ACC40Uk_1, ps->post_wait + 1);
|
2640 |
|
|
|
2641 |
|
|
set_use_is_acc_mmac (cpu, ACC40Uk_1);
|
2642 |
|
|
}
|
2643 |
|
|
if (ACC40Uk_2 >= 0)
|
2644 |
|
|
{
|
2645 |
|
|
update_ACC_latency (cpu, ACC40Uk_2, ps->post_wait + 1);
|
2646 |
|
|
|
2647 |
|
|
set_use_is_acc_mmac (cpu, ACC40Uk_2);
|
2648 |
|
|
}
|
2649 |
|
|
if (ACC40Uk_3 >= 0)
|
2650 |
|
|
{
|
2651 |
|
|
update_ACC_latency (cpu, ACC40Uk_3, ps->post_wait + 1);
|
2652 |
|
|
|
2653 |
|
|
set_use_is_acc_mmac (cpu, ACC40Uk_3);
|
2654 |
|
|
}
|
2655 |
|
|
}
|
2656 |
|
|
|
2657 |
|
|
/* the floating point unit resource has a latency of 3 cycles */
|
2658 |
|
|
update_float_resource_latency (cpu, slot, cycles + 3);
|
2659 |
|
|
|
2660 |
|
|
return cycles;
|
2661 |
|
|
}
|
2662 |
|
|
|
2663 |
|
|
#endif /* WITH_PROFILE_MODEL_P */
|