Commit | Line | Data |
---|---|---|
b34f6357 DB |
1 | /* frv simulator machine independent profiling code. |
2 | ||
dc3cf14f | 3 | Copyright (C) 1998, 1999, 2000, 2001, 2003, 2007, 2008, 2009, 2010 |
6aba47ca | 4 | Free Software Foundation, Inc. |
b34f6357 DB |
5 | Contributed by Red Hat |
6 | ||
7 | This file is part of the GNU simulators. | |
8 | ||
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
4744ac1b JB |
11 | the Free Software Foundation; either version 3 of the License, or |
12 | (at your option) any later version. | |
b34f6357 DB |
13 | |
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
18 | ||
4744ac1b JB |
19 | You should have received a copy of the GNU General Public License |
20 | along with this program. If not, see <http://www.gnu.org/licenses/>. | |
b34f6357 DB |
21 | |
22 | */ | |
23 | #define WANT_CPU | |
24 | #define WANT_CPU_FRVBF | |
25 | ||
26 | #include "sim-main.h" | |
27 | #include "bfd.h" | |
28 | ||
29 | #if WITH_PROFILE_MODEL_P | |
30 | ||
31 | #include "profile.h" | |
32 | #include "profile-fr400.h" | |
33 | #include "profile-fr500.h" | |
e930b1f5 | 34 | #include "profile-fr550.h" |
b34f6357 DB |
35 | |
36 | static void | |
37 | reset_gr_flags (SIM_CPU *cpu, INT gr) | |
38 | { | |
39 | SIM_DESC sd = CPU_STATE (cpu); | |
676a64f4 RS |
40 | if (STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr400 |
41 | || STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr450) | |
b34f6357 DB |
42 | fr400_reset_gr_flags (cpu, gr); |
43 | /* Other machines have no gr flags right now. */ | |
44 | } | |
45 | ||
46 | static void | |
47 | reset_fr_flags (SIM_CPU *cpu, INT fr) | |
48 | { | |
49 | SIM_DESC sd = CPU_STATE (cpu); | |
676a64f4 RS |
50 | if (STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr400 |
51 | || STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr450) | |
b34f6357 DB |
52 | fr400_reset_fr_flags (cpu, fr); |
53 | else if (STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr500) | |
54 | fr500_reset_fr_flags (cpu, fr); | |
55 | } | |
56 | ||
57 | static void | |
58 | reset_acc_flags (SIM_CPU *cpu, INT acc) | |
59 | { | |
60 | SIM_DESC sd = CPU_STATE (cpu); | |
676a64f4 RS |
61 | if (STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr400 |
62 | || STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr450) | |
b34f6357 DB |
63 | fr400_reset_acc_flags (cpu, acc); |
64 | /* Other machines have no acc flags right now. */ | |
65 | } | |
66 | ||
67 | static void | |
68 | reset_cc_flags (SIM_CPU *cpu, INT cc) | |
69 | { | |
70 | SIM_DESC sd = CPU_STATE (cpu); | |
71 | if (STATE_ARCHITECTURE (sd)->mach == bfd_mach_fr500) | |
72 | fr500_reset_cc_flags (cpu, cc); | |
73 | /* Other machines have no cc flags. */ | |
74 | } | |
75 | ||
76 | void | |
77 | set_use_is_gr_complex (SIM_CPU *cpu, INT gr) | |
78 | { | |
79 | if (gr != -1) | |
80 | { | |
81 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
82 | reset_gr_flags (cpu, gr); | |
83 | ps->cur_gr_complex |= (((DI)1) << gr); | |
84 | } | |
85 | } | |
86 | ||
87 | void | |
88 | set_use_not_gr_complex (SIM_CPU *cpu, INT gr) | |
89 | { | |
90 | if (gr != -1) | |
91 | { | |
92 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
93 | ps->cur_gr_complex &= ~(((DI)1) << gr); | |
94 | } | |
95 | } | |
96 | ||
97 | int | |
98 | use_is_gr_complex (SIM_CPU *cpu, INT gr) | |
99 | { | |
100 | if (gr != -1) | |
101 | { | |
102 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
103 | return ps->cur_gr_complex & (((DI)1) << gr); | |
104 | } | |
105 | return 0; | |
106 | } | |
107 | ||
108 | /* Globals flag indicates whether this insn is being modeled. */ | |
109 | enum FRV_INSN_MODELING model_insn = FRV_INSN_NO_MODELING; | |
110 | ||
111 | /* static buffer for the name of the currently most restrictive hazard. */ | |
112 | static char hazard_name[100] = ""; | |
113 | ||
114 | /* Print information about the wait applied to an entire VLIW insn. */ | |
115 | FRV_INSN_FETCH_BUFFER frv_insn_fetch_buffer[] | |
116 | = { | |
117 | {1, NO_REQNO}, {1, NO_REQNO} /* init with impossible address. */ | |
118 | }; | |
119 | ||
120 | enum cache_request | |
121 | { | |
122 | cache_load, | |
123 | cache_invalidate, | |
124 | cache_flush, | |
125 | cache_preload, | |
126 | cache_unlock | |
127 | }; | |
128 | ||
129 | /* A queue of load requests from the data cache. Use to keep track of loads | |
130 | which are still pending. */ | |
131 | /* TODO -- some of these are mutually exclusive and can use a union. */ | |
132 | typedef struct | |
133 | { | |
134 | FRV_CACHE *cache; | |
135 | unsigned reqno; | |
136 | SI address; | |
137 | int length; | |
138 | int is_signed; | |
139 | int regnum; | |
140 | int cycles; | |
141 | int regtype; | |
142 | int lock; | |
143 | int all; | |
144 | int slot; | |
145 | int active; | |
146 | enum cache_request request; | |
147 | } CACHE_QUEUE_ELEMENT; | |
148 | ||
149 | #define CACHE_QUEUE_SIZE 64 /* TODO -- make queue dynamic */ | |
150 | struct | |
151 | { | |
152 | unsigned reqno; | |
153 | int ix; | |
154 | CACHE_QUEUE_ELEMENT q[CACHE_QUEUE_SIZE]; | |
155 | } cache_queue = {0, 0}; | |
156 | ||
157 | /* Queue a request for a load from the cache. The load will be queued as | |
158 | 'inactive' and will be requested after the given number | |
159 | of cycles have passed from the point the load is activated. */ | |
160 | void | |
161 | request_cache_load (SIM_CPU *cpu, INT regnum, int regtype, int cycles) | |
162 | { | |
163 | CACHE_QUEUE_ELEMENT *q; | |
164 | FRV_VLIW *vliw; | |
165 | int slot; | |
166 | ||
167 | /* For a conditional load which was not executed, CPU_LOAD_LENGTH will be | |
168 | zero. */ | |
169 | if (CPU_LOAD_LENGTH (cpu) == 0) | |
170 | return; | |
171 | ||
172 | if (cache_queue.ix >= CACHE_QUEUE_SIZE) | |
173 | abort (); /* TODO: Make the queue dynamic */ | |
174 | ||
175 | q = & cache_queue.q[cache_queue.ix]; | |
176 | ++cache_queue.ix; | |
177 | ||
178 | q->reqno = cache_queue.reqno++; | |
179 | q->request = cache_load; | |
180 | q->cache = CPU_DATA_CACHE (cpu); | |
181 | q->address = CPU_LOAD_ADDRESS (cpu); | |
182 | q->length = CPU_LOAD_LENGTH (cpu); | |
183 | q->is_signed = CPU_LOAD_SIGNED (cpu); | |
184 | q->regnum = regnum; | |
185 | q->regtype = regtype; | |
186 | q->cycles = cycles; | |
187 | q->active = 0; | |
188 | ||
189 | vliw = CPU_VLIW (cpu); | |
190 | slot = vliw->next_slot - 1; | |
191 | q->slot = (*vliw->current_vliw)[slot]; | |
192 | ||
193 | CPU_LOAD_LENGTH (cpu) = 0; | |
194 | } | |
195 | ||
196 | /* Queue a request to flush the cache. The request will be queued as | |
197 | 'inactive' and will be requested after the given number | |
198 | of cycles have passed from the point the request is activated. */ | |
199 | void | |
200 | request_cache_flush (SIM_CPU *cpu, FRV_CACHE *cache, int cycles) | |
201 | { | |
202 | CACHE_QUEUE_ELEMENT *q; | |
203 | FRV_VLIW *vliw; | |
204 | int slot; | |
205 | ||
206 | if (cache_queue.ix >= CACHE_QUEUE_SIZE) | |
207 | abort (); /* TODO: Make the queue dynamic */ | |
208 | ||
209 | q = & cache_queue.q[cache_queue.ix]; | |
210 | ++cache_queue.ix; | |
211 | ||
212 | q->reqno = cache_queue.reqno++; | |
213 | q->request = cache_flush; | |
214 | q->cache = cache; | |
215 | q->address = CPU_LOAD_ADDRESS (cpu); | |
216 | q->all = CPU_PROFILE_STATE (cpu)->all_cache_entries; | |
217 | q->cycles = cycles; | |
218 | q->active = 0; | |
219 | ||
220 | vliw = CPU_VLIW (cpu); | |
221 | slot = vliw->next_slot - 1; | |
222 | q->slot = (*vliw->current_vliw)[slot]; | |
223 | } | |
224 | ||
225 | /* Queue a request to invalidate the cache. The request will be queued as | |
226 | 'inactive' and will be requested after the given number | |
227 | of cycles have passed from the point the request is activated. */ | |
228 | void | |
229 | request_cache_invalidate (SIM_CPU *cpu, FRV_CACHE *cache, int cycles) | |
230 | { | |
231 | CACHE_QUEUE_ELEMENT *q; | |
232 | FRV_VLIW *vliw; | |
233 | int slot; | |
234 | ||
235 | if (cache_queue.ix >= CACHE_QUEUE_SIZE) | |
236 | abort (); /* TODO: Make the queue dynamic */ | |
237 | ||
238 | q = & cache_queue.q[cache_queue.ix]; | |
239 | ++cache_queue.ix; | |
240 | ||
241 | q->reqno = cache_queue.reqno++; | |
242 | q->request = cache_invalidate; | |
243 | q->cache = cache; | |
244 | q->address = CPU_LOAD_ADDRESS (cpu); | |
245 | q->all = CPU_PROFILE_STATE (cpu)->all_cache_entries; | |
246 | q->cycles = cycles; | |
247 | q->active = 0; | |
248 | ||
249 | vliw = CPU_VLIW (cpu); | |
250 | slot = vliw->next_slot - 1; | |
251 | q->slot = (*vliw->current_vliw)[slot]; | |
252 | } | |
253 | ||
254 | /* Queue a request to preload the cache. The request will be queued as | |
255 | 'inactive' and will be requested after the given number | |
256 | of cycles have passed from the point the request is activated. */ | |
257 | void | |
258 | request_cache_preload (SIM_CPU *cpu, FRV_CACHE *cache, int cycles) | |
259 | { | |
260 | CACHE_QUEUE_ELEMENT *q; | |
261 | FRV_VLIW *vliw; | |
262 | int slot; | |
263 | ||
264 | if (cache_queue.ix >= CACHE_QUEUE_SIZE) | |
265 | abort (); /* TODO: Make the queue dynamic */ | |
266 | ||
267 | q = & cache_queue.q[cache_queue.ix]; | |
268 | ++cache_queue.ix; | |
269 | ||
270 | q->reqno = cache_queue.reqno++; | |
271 | q->request = cache_preload; | |
272 | q->cache = cache; | |
273 | q->address = CPU_LOAD_ADDRESS (cpu); | |
274 | q->length = CPU_LOAD_LENGTH (cpu); | |
275 | q->lock = CPU_LOAD_LOCK (cpu); | |
276 | q->cycles = cycles; | |
277 | q->active = 0; | |
278 | ||
279 | vliw = CPU_VLIW (cpu); | |
280 | slot = vliw->next_slot - 1; | |
281 | q->slot = (*vliw->current_vliw)[slot]; | |
282 | ||
283 | CPU_LOAD_LENGTH (cpu) = 0; | |
284 | } | |
285 | ||
286 | /* Queue a request to unlock the cache. The request will be queued as | |
287 | 'inactive' and will be requested after the given number | |
288 | of cycles have passed from the point the request is activated. */ | |
289 | void | |
290 | request_cache_unlock (SIM_CPU *cpu, FRV_CACHE *cache, int cycles) | |
291 | { | |
292 | CACHE_QUEUE_ELEMENT *q; | |
293 | FRV_VLIW *vliw; | |
294 | int slot; | |
295 | ||
296 | if (cache_queue.ix >= CACHE_QUEUE_SIZE) | |
297 | abort (); /* TODO: Make the queue dynamic */ | |
298 | ||
299 | q = & cache_queue.q[cache_queue.ix]; | |
300 | ++cache_queue.ix; | |
301 | ||
302 | q->reqno = cache_queue.reqno++; | |
303 | q->request = cache_unlock; | |
304 | q->cache = cache; | |
305 | q->address = CPU_LOAD_ADDRESS (cpu); | |
306 | q->cycles = cycles; | |
307 | q->active = 0; | |
308 | ||
309 | vliw = CPU_VLIW (cpu); | |
310 | slot = vliw->next_slot - 1; | |
311 | q->slot = (*vliw->current_vliw)[slot]; | |
312 | } | |
313 | ||
314 | static void | |
315 | submit_cache_request (CACHE_QUEUE_ELEMENT *q) | |
316 | { | |
317 | switch (q->request) | |
318 | { | |
319 | case cache_load: | |
320 | frv_cache_request_load (q->cache, q->reqno, q->address, q->slot); | |
321 | break; | |
322 | case cache_flush: | |
323 | frv_cache_request_invalidate (q->cache, q->reqno, q->address, q->slot, | |
324 | q->all, 1/*flush*/); | |
325 | break; | |
326 | case cache_invalidate: | |
327 | frv_cache_request_invalidate (q->cache, q->reqno, q->address, q->slot, | |
328 | q->all, 0/*flush*/); | |
329 | break; | |
330 | case cache_preload: | |
331 | frv_cache_request_preload (q->cache, q->address, q->slot, | |
332 | q->length, q->lock); | |
333 | break; | |
334 | case cache_unlock: | |
335 | frv_cache_request_unlock (q->cache, q->address, q->slot); | |
336 | break; | |
337 | default: | |
338 | abort (); | |
339 | } | |
340 | } | |
341 | ||
342 | /* Activate all inactive load requests. */ | |
343 | static void | |
344 | activate_cache_requests (SIM_CPU *cpu) | |
345 | { | |
346 | int i; | |
347 | for (i = 0; i < cache_queue.ix; ++i) | |
348 | { | |
349 | CACHE_QUEUE_ELEMENT *q = & cache_queue.q[i]; | |
350 | if (! q->active) | |
351 | { | |
352 | q->active = 1; | |
353 | /* Submit the request now if the cycle count is zero. */ | |
354 | if (q->cycles == 0) | |
355 | submit_cache_request (q); | |
356 | } | |
357 | } | |
358 | } | |
359 | ||
360 | /* Check to see if a load is pending which affects the given register(s). | |
361 | */ | |
362 | int | |
363 | load_pending_for_register (SIM_CPU *cpu, int regnum, int words, int regtype) | |
364 | { | |
365 | int i; | |
366 | for (i = 0; i < cache_queue.ix; ++i) | |
367 | { | |
368 | CACHE_QUEUE_ELEMENT *q = & cache_queue.q[i]; | |
369 | ||
370 | /* Must be the same kind of register. */ | |
371 | if (! q->active || q->request != cache_load || q->regtype != regtype) | |
372 | continue; | |
373 | ||
374 | /* If the registers numbers are equal, then we have a match. */ | |
375 | if (q->regnum == regnum) | |
376 | return 1; /* load pending */ | |
377 | ||
378 | /* Check for overlap of a load with a multi-word register. */ | |
379 | if (regnum < q->regnum) | |
380 | { | |
381 | if (regnum + words > q->regnum) | |
382 | return 1; | |
383 | } | |
384 | /* Check for overlap of a multi-word load with the register. */ | |
385 | else | |
386 | { | |
387 | int data_words = (q->length + sizeof (SI) - 1) / sizeof (SI); | |
388 | if (q->regnum + data_words > regnum) | |
389 | return 1; | |
390 | } | |
391 | } | |
392 | ||
393 | return 0; /* no load pending */ | |
394 | } | |
395 | ||
396 | /* Check to see if a cache flush pending which affects the given address. */ | |
397 | static int | |
398 | flush_pending_for_address (SIM_CPU *cpu, SI address) | |
399 | { | |
400 | int line_mask = ~(CPU_DATA_CACHE (cpu)->line_size - 1); | |
401 | int i; | |
402 | for (i = 0; i < cache_queue.ix; ++i) | |
403 | { | |
404 | CACHE_QUEUE_ELEMENT *q = & cache_queue.q[i]; | |
405 | ||
406 | /* Must be the same kind of request and active. */ | |
407 | if (! q->active || q->request != cache_flush) | |
408 | continue; | |
409 | ||
410 | /* If the addresses are equal, then we have a match. */ | |
411 | if ((q->address & line_mask) == (address & line_mask)) | |
412 | return 1; /* flush pending */ | |
413 | } | |
414 | ||
415 | return 0; /* no flush pending */ | |
416 | } | |
417 | ||
418 | static void | |
419 | remove_cache_queue_element (SIM_CPU *cpu, int i) | |
420 | { | |
421 | /* If we are removing the load of a FR register, then remember which one(s). | |
422 | */ | |
423 | CACHE_QUEUE_ELEMENT q = cache_queue.q[i]; | |
424 | ||
425 | for (--cache_queue.ix; i < cache_queue.ix; ++i) | |
426 | cache_queue.q[i] = cache_queue.q[i + 1]; | |
427 | ||
428 | /* If we removed a load of a FR register, check to see if any other loads | |
429 | of that register is still queued. If not, then apply the queued post | |
430 | processing time of that register to its latency. Also apply | |
431 | 1 extra cycle of latency to the register since it was a floating point | |
432 | load. */ | |
433 | if (q.request == cache_load && q.regtype != REGTYPE_NONE) | |
434 | { | |
435 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
436 | int data_words = (q.length + sizeof (SI) - 1) / sizeof (SI); | |
437 | int j; | |
438 | for (j = 0; j < data_words; ++j) | |
439 | { | |
440 | int regnum = q.regnum + j; | |
441 | if (! load_pending_for_register (cpu, regnum, 1, q.regtype)) | |
442 | { | |
443 | if (q.regtype == REGTYPE_FR) | |
444 | { | |
445 | int *fr = ps->fr_busy; | |
446 | fr[regnum] += 1 + ps->fr_ptime[regnum]; | |
447 | ps->fr_ptime[regnum] = 0; | |
448 | } | |
449 | } | |
450 | } | |
451 | } | |
452 | } | |
453 | ||
454 | /* Copy data from the cache buffer to the target register(s). */ | |
455 | static void | |
456 | copy_load_data (SIM_CPU *current_cpu, FRV_CACHE *cache, int slot, | |
457 | CACHE_QUEUE_ELEMENT *q) | |
458 | { | |
459 | switch (q->length) | |
460 | { | |
461 | case 1: | |
462 | if (q->regtype == REGTYPE_FR) | |
463 | { | |
464 | if (q->is_signed) | |
465 | { | |
466 | QI value = CACHE_RETURN_DATA (cache, slot, q->address, QI, 1); | |
467 | SET_H_FR (q->regnum, value); | |
468 | } | |
469 | else | |
470 | { | |
471 | UQI value = CACHE_RETURN_DATA (cache, slot, q->address, UQI, 1); | |
472 | SET_H_FR (q->regnum, value); | |
473 | } | |
474 | } | |
475 | else | |
476 | { | |
477 | if (q->is_signed) | |
478 | { | |
479 | QI value = CACHE_RETURN_DATA (cache, slot, q->address, QI, 1); | |
480 | SET_H_GR (q->regnum, value); | |
481 | } | |
482 | else | |
483 | { | |
484 | UQI value = CACHE_RETURN_DATA (cache, slot, q->address, UQI, 1); | |
485 | SET_H_GR (q->regnum, value); | |
486 | } | |
487 | } | |
488 | break; | |
489 | case 2: | |
490 | if (q->regtype == REGTYPE_FR) | |
491 | { | |
492 | if (q->is_signed) | |
493 | { | |
494 | HI value = CACHE_RETURN_DATA (cache, slot, q->address, HI, 2); | |
495 | SET_H_FR (q->regnum, value); | |
496 | } | |
497 | else | |
498 | { | |
499 | UHI value = CACHE_RETURN_DATA (cache, slot, q->address, UHI, 2); | |
500 | SET_H_FR (q->regnum, value); | |
501 | } | |
502 | } | |
503 | else | |
504 | { | |
505 | if (q->is_signed) | |
506 | { | |
507 | HI value = CACHE_RETURN_DATA (cache, slot, q->address, HI, 2); | |
508 | SET_H_GR (q->regnum, value); | |
509 | } | |
510 | else | |
511 | { | |
512 | UHI value = CACHE_RETURN_DATA (cache, slot, q->address, UHI, 2); | |
513 | SET_H_GR (q->regnum, value); | |
514 | } | |
515 | } | |
516 | break; | |
517 | case 4: | |
518 | if (q->regtype == REGTYPE_FR) | |
519 | { | |
520 | SET_H_FR (q->regnum, | |
521 | CACHE_RETURN_DATA (cache, slot, q->address, SF, 4)); | |
522 | } | |
523 | else | |
524 | { | |
525 | SET_H_GR (q->regnum, | |
526 | CACHE_RETURN_DATA (cache, slot, q->address, SI, 4)); | |
527 | } | |
528 | break; | |
529 | case 8: | |
530 | if (q->regtype == REGTYPE_FR) | |
531 | { | |
532 | SET_H_FR_DOUBLE (q->regnum, | |
533 | CACHE_RETURN_DATA (cache, slot, q->address, DF, 8)); | |
534 | } | |
535 | else | |
536 | { | |
537 | SET_H_GR_DOUBLE (q->regnum, | |
538 | CACHE_RETURN_DATA (cache, slot, q->address, DI, 8)); | |
539 | } | |
540 | break; | |
541 | case 16: | |
542 | if (q->regtype == REGTYPE_FR) | |
543 | frvbf_h_fr_quad_set_handler (current_cpu, q->regnum, | |
544 | CACHE_RETURN_DATA_ADDRESS (cache, slot, | |
545 | q->address, | |
546 | 16)); | |
547 | else | |
548 | frvbf_h_gr_quad_set_handler (current_cpu, q->regnum, | |
549 | CACHE_RETURN_DATA_ADDRESS (cache, slot, | |
550 | q->address, | |
551 | 16)); | |
552 | break; | |
553 | default: | |
554 | abort (); | |
555 | } | |
556 | } | |
557 | ||
558 | static int | |
559 | request_complete (SIM_CPU *cpu, CACHE_QUEUE_ELEMENT *q) | |
560 | { | |
561 | FRV_CACHE* cache; | |
562 | if (! q->active || q->cycles > 0) | |
563 | return 0; | |
564 | ||
565 | cache = CPU_DATA_CACHE (cpu); | |
566 | switch (q->request) | |
567 | { | |
568 | case cache_load: | |
569 | /* For loads, we must wait until the data is returned from the cache. */ | |
570 | if (frv_cache_data_in_buffer (cache, 0, q->address, q->reqno)) | |
571 | { | |
572 | copy_load_data (cpu, cache, 0, q); | |
573 | return 1; | |
574 | } | |
575 | if (frv_cache_data_in_buffer (cache, 1, q->address, q->reqno)) | |
576 | { | |
577 | copy_load_data (cpu, cache, 1, q); | |
578 | return 1; | |
579 | } | |
580 | break; | |
581 | ||
582 | case cache_flush: | |
583 | /* We must wait until the data is flushed. */ | |
584 | if (frv_cache_data_flushed (cache, 0, q->address, q->reqno)) | |
585 | return 1; | |
586 | if (frv_cache_data_flushed (cache, 1, q->address, q->reqno)) | |
587 | return 1; | |
588 | break; | |
589 | ||
590 | default: | |
591 | /* All other requests are complete once they've been made. */ | |
592 | return 1; | |
593 | } | |
594 | ||
595 | return 0; | |
596 | } | |
597 | ||
598 | /* Run the insn and data caches through the given number of cycles, taking | |
599 | note of load requests which are fullfilled as a result. */ | |
600 | static void | |
601 | run_caches (SIM_CPU *cpu, int cycles) | |
602 | { | |
603 | FRV_CACHE* data_cache = CPU_DATA_CACHE (cpu); | |
604 | FRV_CACHE* insn_cache = CPU_INSN_CACHE (cpu); | |
605 | int i; | |
606 | /* For each cycle, run the caches, noting which requests have been fullfilled | |
607 | and submitting new requests on their designated cycles. */ | |
608 | for (i = 0; i < cycles; ++i) | |
609 | { | |
610 | int j; | |
611 | /* Run the caches through 1 cycle. */ | |
612 | frv_cache_run (data_cache, 1); | |
613 | frv_cache_run (insn_cache, 1); | |
614 | ||
615 | /* Note whether prefetched insn data has been loaded yet. */ | |
616 | for (j = LS; j < FRV_CACHE_PIPELINES; ++j) | |
617 | { | |
618 | if (frv_insn_fetch_buffer[j].reqno != NO_REQNO | |
619 | && frv_cache_data_in_buffer (insn_cache, j, | |
620 | frv_insn_fetch_buffer[j].address, | |
621 | frv_insn_fetch_buffer[j].reqno)) | |
622 | frv_insn_fetch_buffer[j].reqno = NO_REQNO; | |
623 | } | |
624 | ||
625 | /* Check to see which requests have been satisfied and which should | |
626 | be submitted now. */ | |
627 | for (j = 0; j < cache_queue.ix; ++j) | |
628 | { | |
629 | CACHE_QUEUE_ELEMENT *q = & cache_queue.q[j]; | |
630 | if (! q->active) | |
631 | continue; | |
632 | ||
633 | /* If a load has been satisfied, complete the operation and remove it | |
634 | from the queue. */ | |
635 | if (request_complete (cpu, q)) | |
636 | { | |
637 | remove_cache_queue_element (cpu, j); | |
638 | --j; | |
639 | continue; | |
640 | } | |
641 | ||
642 | /* Decrease the cycle count of each queued request. | |
643 | Submit a request for each queued request whose cycle count has | |
644 | become zero. */ | |
645 | --q->cycles; | |
646 | if (q->cycles == 0) | |
647 | submit_cache_request (q); | |
648 | } | |
649 | } | |
650 | } | |
651 | ||
652 | static void | |
653 | apply_latency_adjustments (SIM_CPU *cpu) | |
654 | { | |
655 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
656 | int i; | |
657 | /* update the latencies of the registers. */ | |
658 | int *fr = ps->fr_busy; | |
659 | int *acc = ps->acc_busy; | |
660 | for (i = 0; i < 64; ++i) | |
661 | { | |
662 | if (ps->fr_busy_adjust[i] > 0) | |
663 | *fr -= ps->fr_busy_adjust[i]; /* OK if it goes negative. */ | |
664 | if (ps->acc_busy_adjust[i] > 0) | |
665 | *acc -= ps->acc_busy_adjust[i]; /* OK if it goes negative. */ | |
666 | ++fr; | |
667 | ++acc; | |
668 | } | |
669 | } | |
670 | ||
671 | /* Account for the number of cycles which have just passed in the latency of | |
672 | various system elements. Works for negative cycles too so that latency | |
673 | can be extended in the case of insn fetch latency. | |
674 | If negative or zero, then no adjustment is necessary. */ | |
675 | static void | |
676 | update_latencies (SIM_CPU *cpu, int cycles) | |
677 | { | |
678 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
679 | int i; | |
680 | /* update the latencies of the registers. */ | |
681 | int *fdiv; | |
682 | int *fsqrt; | |
683 | int *idiv; | |
e930b1f5 DB |
684 | int *flt; |
685 | int *media; | |
b34f6357 DB |
686 | int *ccr; |
687 | int *gr = ps->gr_busy; | |
688 | int *fr = ps->fr_busy; | |
689 | int *acc = ps->acc_busy; | |
153431d6 | 690 | int *spr; |
b34f6357 DB |
691 | /* This loop handles GR, FR and ACC registers. */ |
692 | for (i = 0; i < 64; ++i) | |
693 | { | |
694 | if (*gr <= cycles) | |
695 | { | |
696 | *gr = 0; | |
697 | reset_gr_flags (cpu, i); | |
698 | } | |
699 | else | |
700 | *gr -= cycles; | |
701 | /* If the busy drops to 0, then mark the register as | |
702 | "not in use". */ | |
703 | if (*fr <= cycles) | |
704 | { | |
705 | int *fr_lat = ps->fr_latency + i; | |
706 | *fr = 0; | |
707 | ps->fr_busy_adjust[i] = 0; | |
708 | /* Only clear flags if this register has no target latency. */ | |
709 | if (*fr_lat == 0) | |
710 | reset_fr_flags (cpu, i); | |
711 | } | |
712 | else | |
713 | *fr -= cycles; | |
714 | /* If the busy drops to 0, then mark the register as | |
715 | "not in use". */ | |
716 | if (*acc <= cycles) | |
717 | { | |
718 | int *acc_lat = ps->acc_latency + i; | |
719 | *acc = 0; | |
720 | ps->acc_busy_adjust[i] = 0; | |
721 | /* Only clear flags if this register has no target latency. */ | |
722 | if (*acc_lat == 0) | |
723 | reset_acc_flags (cpu, i); | |
724 | } | |
725 | else | |
726 | *acc -= cycles; | |
727 | ++gr; | |
728 | ++fr; | |
729 | ++acc; | |
730 | } | |
731 | /* This loop handles CCR registers. */ | |
732 | ccr = ps->ccr_busy; | |
733 | for (i = 0; i < 8; ++i) | |
734 | { | |
735 | if (*ccr <= cycles) | |
736 | { | |
737 | *ccr = 0; | |
738 | reset_cc_flags (cpu, i); | |
739 | } | |
740 | else | |
741 | *ccr -= cycles; | |
742 | ++ccr; | |
743 | } | |
153431d6 DB |
744 | /* This loop handles SPR registers. */ |
745 | spr = ps->spr_busy; | |
746 | for (i = 0; i < 4096; ++i) | |
747 | { | |
748 | if (*spr <= cycles) | |
749 | *spr = 0; | |
750 | else | |
751 | *spr -= cycles; | |
752 | ++spr; | |
753 | } | |
b34f6357 DB |
754 | /* This loop handles resources. */ |
755 | idiv = ps->idiv_busy; | |
756 | fdiv = ps->fdiv_busy; | |
757 | fsqrt = ps->fsqrt_busy; | |
758 | for (i = 0; i < 2; ++i) | |
759 | { | |
760 | *idiv = (*idiv <= cycles) ? 0 : (*idiv - cycles); | |
761 | *fdiv = (*fdiv <= cycles) ? 0 : (*fdiv - cycles); | |
762 | *fsqrt = (*fsqrt <= cycles) ? 0 : (*fsqrt - cycles); | |
763 | ++idiv; | |
764 | ++fdiv; | |
765 | ++fsqrt; | |
766 | } | |
e930b1f5 DB |
767 | /* Float and media units can occur in 4 slots on some machines. */ |
768 | flt = ps->float_busy; | |
769 | media = ps->media_busy; | |
770 | for (i = 0; i < 4; ++i) | |
771 | { | |
772 | *flt = (*flt <= cycles) ? 0 : (*flt - cycles); | |
773 | *media = (*media <= cycles) ? 0 : (*media - cycles); | |
774 | ++flt; | |
775 | ++media; | |
776 | } | |
b34f6357 DB |
777 | } |
778 | ||
779 | /* Print information about the wait for the given number of cycles. */ | |
780 | void | |
781 | frv_model_trace_wait_cycles (SIM_CPU *cpu, int cycles, const char *hazard_name) | |
782 | { | |
783 | if (TRACE_INSN_P (cpu) && cycles > 0) | |
784 | { | |
785 | SIM_DESC sd = CPU_STATE (cpu); | |
786 | trace_printf (sd, cpu, "**** %s wait %d cycles ***\n", | |
787 | hazard_name, cycles); | |
788 | } | |
789 | } | |
790 | ||
791 | void | |
792 | trace_vliw_wait_cycles (SIM_CPU *cpu) | |
793 | { | |
794 | if (TRACE_INSN_P (cpu)) | |
795 | { | |
796 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
797 | frv_model_trace_wait_cycles (cpu, ps->vliw_wait, hazard_name); | |
798 | } | |
799 | } | |
800 | ||
801 | /* Wait for the given number of cycles. */ | |
802 | void | |
803 | frv_model_advance_cycles (SIM_CPU *cpu, int cycles) | |
804 | { | |
805 | PROFILE_DATA *p = CPU_PROFILE_DATA (cpu); | |
806 | update_latencies (cpu, cycles); | |
807 | run_caches (cpu, cycles); | |
808 | PROFILE_MODEL_TOTAL_CYCLES (p) += cycles; | |
809 | } | |
810 | ||
811 | void | |
812 | handle_resource_wait (SIM_CPU *cpu) | |
813 | { | |
814 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
815 | if (ps->vliw_wait != 0) | |
816 | frv_model_advance_cycles (cpu, ps->vliw_wait); | |
817 | if (ps->vliw_load_stall > ps->vliw_wait) | |
818 | ps->vliw_load_stall -= ps->vliw_wait; | |
819 | else | |
820 | ps->vliw_load_stall = 0; | |
821 | } | |
822 | ||
823 | /* Account for the number of cycles until these resources will be available | |
824 | again. */ | |
825 | static void | |
826 | update_target_latencies (SIM_CPU *cpu) | |
827 | { | |
828 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
829 | int i; | |
830 | /* update the latencies of the registers. */ | |
831 | int *ccr_lat; | |
832 | int *gr_lat = ps->gr_latency; | |
833 | int *fr_lat = ps->fr_latency; | |
834 | int *acc_lat = ps->acc_latency; | |
153431d6 | 835 | int *spr_lat; |
b34f6357 DB |
836 | int *ccr; |
837 | int *gr = ps->gr_busy; | |
838 | int *fr = ps->fr_busy; | |
839 | int *acc = ps->acc_busy; | |
153431d6 | 840 | int *spr; |
b34f6357 DB |
841 | /* This loop handles GR, FR and ACC registers. */ |
842 | for (i = 0; i < 64; ++i) | |
843 | { | |
844 | if (*gr_lat) | |
845 | { | |
846 | *gr = *gr_lat; | |
847 | *gr_lat = 0; | |
848 | } | |
849 | if (*fr_lat) | |
850 | { | |
851 | *fr = *fr_lat; | |
852 | *fr_lat = 0; | |
853 | } | |
854 | if (*acc_lat) | |
855 | { | |
856 | *acc = *acc_lat; | |
857 | *acc_lat = 0; | |
858 | } | |
859 | ++gr; ++gr_lat; | |
860 | ++fr; ++fr_lat; | |
861 | ++acc; ++acc_lat; | |
862 | } | |
863 | /* This loop handles CCR registers. */ | |
864 | ccr = ps->ccr_busy; | |
865 | ccr_lat = ps->ccr_latency; | |
866 | for (i = 0; i < 8; ++i) | |
867 | { | |
868 | if (*ccr_lat) | |
869 | { | |
870 | *ccr = *ccr_lat; | |
871 | *ccr_lat = 0; | |
872 | } | |
873 | ++ccr; ++ccr_lat; | |
874 | } | |
153431d6 DB |
875 | /* This loop handles SPR registers. */ |
876 | spr = ps->spr_busy; | |
877 | spr_lat = ps->spr_latency; | |
878 | for (i = 0; i < 4096; ++i) | |
879 | { | |
880 | if (*spr_lat) | |
881 | { | |
882 | *spr = *spr_lat; | |
883 | *spr_lat = 0; | |
884 | } | |
885 | ++spr; ++spr_lat; | |
886 | } | |
b34f6357 DB |
887 | } |
888 | ||
889 | /* Run the caches until all pending cache flushes are complete. */ | |
890 | static void | |
891 | wait_for_flush (SIM_CPU *cpu) | |
892 | { | |
893 | SI address = CPU_LOAD_ADDRESS (cpu); | |
894 | int wait = 0; | |
895 | while (flush_pending_for_address (cpu, address)) | |
896 | { | |
897 | frv_model_advance_cycles (cpu, 1); | |
898 | ++wait; | |
899 | } | |
900 | if (TRACE_INSN_P (cpu) && wait) | |
901 | { | |
902 | sprintf (hazard_name, "Data cache flush address %p:", address); | |
903 | frv_model_trace_wait_cycles (cpu, wait, hazard_name); | |
904 | } | |
905 | } | |
906 | ||
907 | /* Initialize cycle counting for an insn. | |
908 | FIRST_P is non-zero if this is the first insn in a set of parallel | |
909 | insns. */ | |
910 | void | |
911 | frvbf_model_insn_before (SIM_CPU *cpu, int first_p) | |
912 | { | |
913 | SIM_DESC sd = CPU_STATE (cpu); | |
914 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
915 | ||
916 | ps->vliw_wait = 0; | |
917 | ps->post_wait = 0; | |
918 | memset (ps->fr_busy_adjust, 0, sizeof (ps->fr_busy_adjust)); | |
919 | memset (ps->acc_busy_adjust, 0, sizeof (ps->acc_busy_adjust)); | |
920 | ||
921 | if (first_p) | |
922 | { | |
923 | ps->vliw_insns++; | |
924 | ps->vliw_cycles = 0; | |
925 | ps->vliw_branch_taken = 0; | |
926 | ps->vliw_load_stall = 0; | |
927 | } | |
928 | ||
929 | switch (STATE_ARCHITECTURE (sd)->mach) | |
930 | { | |
931 | case bfd_mach_fr400: | |
676a64f4 | 932 | case bfd_mach_fr450: |
b34f6357 DB |
933 | fr400_model_insn_before (cpu, first_p); |
934 | break; | |
935 | case bfd_mach_fr500: | |
936 | fr500_model_insn_before (cpu, first_p); | |
937 | break; | |
e930b1f5 DB |
938 | case bfd_mach_fr550: |
939 | fr550_model_insn_before (cpu, first_p); | |
940 | break; | |
b34f6357 DB |
941 | default: |
942 | break; | |
943 | } | |
944 | ||
945 | if (first_p) | |
946 | wait_for_flush (cpu); | |
947 | } | |
948 | ||
949 | /* Record the cycles computed for an insn. | |
950 | LAST_P is non-zero if this is the last insn in a set of parallel insns, | |
951 | and we update the total cycle count. | |
952 | CYCLES is the cycle count of the insn. */ | |
953 | ||
954 | void | |
955 | frvbf_model_insn_after (SIM_CPU *cpu, int last_p, int cycles) | |
956 | { | |
957 | PROFILE_DATA *p = CPU_PROFILE_DATA (cpu); | |
958 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
959 | SIM_DESC sd = CPU_STATE (cpu); | |
960 | ||
961 | PROFILE_MODEL_CUR_INSN_CYCLES (p) = cycles; | |
962 | ||
963 | /* The number of cycles for a VLIW insn is the maximum number of cycles | |
964 | used by any individual insn within it. */ | |
965 | if (cycles > ps->vliw_cycles) | |
966 | ps->vliw_cycles = cycles; | |
967 | ||
968 | if (last_p) | |
969 | { | |
970 | /* This is the last insn in a VLIW insn. */ | |
971 | struct frv_interrupt_timer *timer = & frv_interrupt_state.timer; | |
972 | ||
973 | activate_cache_requests (cpu); /* before advancing cycles. */ | |
974 | apply_latency_adjustments (cpu); /* must go first. */ | |
975 | update_target_latencies (cpu); /* must go next. */ | |
976 | frv_model_advance_cycles (cpu, ps->vliw_cycles); | |
977 | ||
978 | PROFILE_MODEL_LOAD_STALL_CYCLES (p) += ps->vliw_load_stall; | |
979 | ||
980 | /* Check the interrupt timer. cycles contains the total cycle count. */ | |
981 | if (timer->enabled) | |
982 | { | |
983 | cycles = PROFILE_MODEL_TOTAL_CYCLES (p); | |
984 | if (timer->current % timer->value | |
985 | + (cycles - timer->current) >= timer->value) | |
986 | frv_queue_external_interrupt (cpu, timer->interrupt); | |
987 | timer->current = cycles; | |
988 | } | |
989 | ||
990 | ps->past_first_p = 0; /* Next one will be the first in a new VLIW. */ | |
991 | ps->branch_address = -1; | |
992 | } | |
993 | else | |
994 | ps->past_first_p = 1; | |
995 | ||
996 | switch (STATE_ARCHITECTURE (sd)->mach) | |
997 | { | |
998 | case bfd_mach_fr400: | |
676a64f4 | 999 | case bfd_mach_fr450: |
b34f6357 DB |
1000 | fr400_model_insn_after (cpu, last_p, cycles); |
1001 | break; | |
1002 | case bfd_mach_fr500: | |
1003 | fr500_model_insn_after (cpu, last_p, cycles); | |
1004 | break; | |
e930b1f5 DB |
1005 | case bfd_mach_fr550: |
1006 | fr550_model_insn_after (cpu, last_p, cycles); | |
1007 | break; | |
b34f6357 DB |
1008 | default: |
1009 | break; | |
1010 | } | |
1011 | } | |
1012 | ||
1013 | USI | |
1014 | frvbf_model_branch (SIM_CPU *current_cpu, PCADDR target, int hint) | |
1015 | { | |
1016 | /* Record the hint and branch address for use in profiling. */ | |
1017 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (current_cpu); | |
1018 | ps->branch_hint = hint; | |
1019 | ps->branch_address = target; | |
1020 | } | |
1021 | ||
1022 | /* Top up the latency of the given GR by the given number of cycles. */ | |
1023 | void | |
1024 | update_GR_latency (SIM_CPU *cpu, INT out_GR, int cycles) | |
1025 | { | |
1026 | if (out_GR >= 0) | |
1027 | { | |
1028 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1029 | int *gr = ps->gr_latency; | |
1030 | if (gr[out_GR] < cycles) | |
1031 | gr[out_GR] = cycles; | |
1032 | } | |
1033 | } | |
1034 | ||
1035 | void | |
1036 | decrease_GR_busy (SIM_CPU *cpu, INT in_GR, int cycles) | |
1037 | { | |
1038 | if (in_GR >= 0) | |
1039 | { | |
1040 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1041 | int *gr = ps->gr_busy; | |
1042 | gr[in_GR] -= cycles; | |
1043 | } | |
1044 | } | |
1045 | ||
1046 | /* Top up the latency of the given double GR by the number of cycles. */ | |
1047 | void | |
1048 | update_GRdouble_latency (SIM_CPU *cpu, INT out_GR, int cycles) | |
1049 | { | |
1050 | if (out_GR >= 0) | |
1051 | { | |
1052 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1053 | int *gr = ps->gr_latency; | |
1054 | if (gr[out_GR] < cycles) | |
1055 | gr[out_GR] = cycles; | |
1056 | if (out_GR < 63 && gr[out_GR + 1] < cycles) | |
1057 | gr[out_GR + 1] = cycles; | |
1058 | } | |
1059 | } | |
1060 | ||
1061 | void | |
1062 | update_GR_latency_for_load (SIM_CPU *cpu, INT out_GR, int cycles) | |
1063 | { | |
1064 | if (out_GR >= 0) | |
1065 | { | |
1066 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1067 | int *gr = ps->gr_latency; | |
1068 | ||
1069 | /* The latency of the GR will be at least the number of cycles used | |
1070 | by the insn. */ | |
1071 | if (gr[out_GR] < cycles) | |
1072 | gr[out_GR] = cycles; | |
1073 | ||
1074 | /* The latency will also depend on how long it takes to retrieve the | |
1075 | data from the cache or memory. Assume that the load is issued | |
1076 | after the last cycle of the insn. */ | |
1077 | request_cache_load (cpu, out_GR, REGTYPE_NONE, cycles); | |
1078 | } | |
1079 | } | |
1080 | ||
1081 | void | |
1082 | update_GRdouble_latency_for_load (SIM_CPU *cpu, INT out_GR, int cycles) | |
1083 | { | |
1084 | if (out_GR >= 0) | |
1085 | { | |
1086 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1087 | int *gr = ps->gr_latency; | |
1088 | ||
1089 | /* The latency of the GR will be at least the number of cycles used | |
1090 | by the insn. */ | |
1091 | if (gr[out_GR] < cycles) | |
1092 | gr[out_GR] = cycles; | |
1093 | if (out_GR < 63 && gr[out_GR + 1] < cycles) | |
1094 | gr[out_GR + 1] = cycles; | |
1095 | ||
1096 | /* The latency will also depend on how long it takes to retrieve the | |
1097 | data from the cache or memory. Assume that the load is issued | |
1098 | after the last cycle of the insn. */ | |
1099 | request_cache_load (cpu, out_GR, REGTYPE_NONE, cycles); | |
1100 | } | |
1101 | } | |
1102 | ||
1103 | void | |
1104 | update_GR_latency_for_swap (SIM_CPU *cpu, INT out_GR, int cycles) | |
1105 | { | |
1106 | update_GR_latency_for_load (cpu, out_GR, cycles); | |
1107 | } | |
1108 | ||
1109 | /* Top up the latency of the given FR by the given number of cycles. */ | |
1110 | void | |
1111 | update_FR_latency (SIM_CPU *cpu, INT out_FR, int cycles) | |
1112 | { | |
1113 | if (out_FR >= 0) | |
1114 | { | |
1115 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1116 | int *fr = ps->fr_latency; | |
1117 | if (fr[out_FR] < cycles) | |
1118 | fr[out_FR] = cycles; | |
1119 | } | |
1120 | } | |
1121 | ||
1122 | /* Top up the latency of the given double FR by the number of cycles. */ | |
1123 | void | |
1124 | update_FRdouble_latency (SIM_CPU *cpu, INT out_FR, int cycles) | |
1125 | { | |
1126 | if (out_FR >= 0) | |
1127 | { | |
1128 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1129 | int *fr = ps->fr_latency; | |
1130 | if (fr[out_FR] < cycles) | |
1131 | fr[out_FR] = cycles; | |
1132 | if (out_FR < 63 && fr[out_FR + 1] < cycles) | |
1133 | fr[out_FR + 1] = cycles; | |
1134 | } | |
1135 | } | |
1136 | ||
1137 | void | |
1138 | update_FR_latency_for_load (SIM_CPU *cpu, INT out_FR, int cycles) | |
1139 | { | |
1140 | if (out_FR >= 0) | |
1141 | { | |
1142 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1143 | int *fr = ps->fr_latency; | |
1144 | ||
1145 | /* The latency of the FR will be at least the number of cycles used | |
1146 | by the insn. */ | |
1147 | if (fr[out_FR] < cycles) | |
1148 | fr[out_FR] = cycles; | |
1149 | ||
1150 | /* The latency will also depend on how long it takes to retrieve the | |
1151 | data from the cache or memory. Assume that the load is issued | |
1152 | after the last cycle of the insn. */ | |
1153 | request_cache_load (cpu, out_FR, REGTYPE_FR, cycles); | |
1154 | } | |
1155 | } | |
1156 | ||
1157 | void | |
1158 | update_FRdouble_latency_for_load (SIM_CPU *cpu, INT out_FR, int cycles) | |
1159 | { | |
1160 | if (out_FR >= 0) | |
1161 | { | |
1162 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1163 | int *fr = ps->fr_latency; | |
1164 | ||
1165 | /* The latency of the FR will be at least the number of cycles used | |
1166 | by the insn. */ | |
1167 | if (fr[out_FR] < cycles) | |
1168 | fr[out_FR] = cycles; | |
1169 | if (out_FR < 63 && fr[out_FR + 1] < cycles) | |
1170 | fr[out_FR + 1] = cycles; | |
1171 | ||
1172 | /* The latency will also depend on how long it takes to retrieve the | |
1173 | data from the cache or memory. Assume that the load is issued | |
1174 | after the last cycle of the insn. */ | |
1175 | request_cache_load (cpu, out_FR, REGTYPE_FR, cycles); | |
1176 | } | |
1177 | } | |
1178 | ||
1179 | /* Top up the post-processing time of the given FR by the given number of | |
1180 | cycles. */ | |
1181 | void | |
1c453cd6 DB |
1182 | update_FR_ptime (SIM_CPU *cpu, INT out_FR, int cycles) |
1183 | { | |
1184 | if (out_FR >= 0) | |
1185 | { | |
1186 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1187 | /* If a load is pending on this register, then add the cycles to | |
1188 | the post processing time for this register. Otherwise apply it | |
1189 | directly to the latency of the register. */ | |
1190 | if (! load_pending_for_register (cpu, out_FR, 1, REGTYPE_FR)) | |
1191 | { | |
1192 | int *fr = ps->fr_latency; | |
1193 | fr[out_FR] += cycles; | |
1194 | } | |
1195 | else | |
1196 | ps->fr_ptime[out_FR] += cycles; | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | void | |
1201 | update_FRdouble_ptime (SIM_CPU *cpu, INT out_FR, int cycles) | |
1202 | { | |
1203 | if (out_FR >= 0) | |
1204 | { | |
1205 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1206 | /* If a load is pending on this register, then add the cycles to | |
1207 | the post processing time for this register. Otherwise apply it | |
1208 | directly to the latency of the register. */ | |
1209 | if (! load_pending_for_register (cpu, out_FR, 2, REGTYPE_FR)) | |
1210 | { | |
1211 | int *fr = ps->fr_latency; | |
1212 | fr[out_FR] += cycles; | |
1213 | if (out_FR < 63) | |
1214 | fr[out_FR + 1] += cycles; | |
1215 | } | |
1216 | else | |
1217 | { | |
1218 | ps->fr_ptime[out_FR] += cycles; | |
1219 | if (out_FR < 63) | |
1220 | ps->fr_ptime[out_FR + 1] += cycles; | |
1221 | } | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | /* Top up the post-processing time of the given ACC by the given number of | |
1226 | cycles. */ | |
1227 | void | |
b34f6357 DB |
1228 | update_ACC_ptime (SIM_CPU *cpu, INT out_ACC, int cycles) |
1229 | { | |
1230 | if (out_ACC >= 0) | |
1231 | { | |
1232 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1233 | /* No load can be pending on this register. Apply the cycles | |
1234 | directly to the latency of the register. */ | |
1235 | int *acc = ps->acc_latency; | |
1236 | acc[out_ACC] += cycles; | |
1237 | } | |
1238 | } | |
1239 | ||
1c453cd6 DB |
1240 | /* Top up the post-processing time of the given SPR by the given number of |
1241 | cycles. */ | |
1242 | void | |
1243 | update_SPR_ptime (SIM_CPU *cpu, INT out_SPR, int cycles) | |
1244 | { | |
1245 | if (out_SPR >= 0) | |
1246 | { | |
1247 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1248 | /* No load can be pending on this register. Apply the cycles | |
1249 | directly to the latency of the register. */ | |
1250 | int *spr = ps->spr_latency; | |
1251 | spr[out_SPR] += cycles; | |
1252 | } | |
1253 | } | |
1254 | ||
b34f6357 DB |
1255 | void |
1256 | decrease_ACC_busy (SIM_CPU *cpu, INT out_ACC, int cycles) | |
1257 | { | |
1258 | if (out_ACC >= 0) | |
1259 | { | |
1260 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1261 | int *acc = ps->acc_busy; | |
1262 | acc[out_ACC] -= cycles; | |
1263 | if (ps->acc_busy_adjust[out_ACC] >= 0 | |
1264 | && cycles > ps->acc_busy_adjust[out_ACC]) | |
1265 | ps->acc_busy_adjust[out_ACC] = cycles; | |
1266 | } | |
1267 | } | |
1268 | ||
1c453cd6 DB |
1269 | void |
1270 | increase_ACC_busy (SIM_CPU *cpu, INT out_ACC, int cycles) | |
1271 | { | |
1272 | if (out_ACC >= 0) | |
1273 | { | |
1274 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1275 | int *acc = ps->acc_busy; | |
1276 | acc[out_ACC] += cycles; | |
1277 | } | |
1278 | } | |
1279 | ||
1280 | void | |
1281 | enforce_full_acc_latency (SIM_CPU *cpu, INT in_ACC) | |
1282 | { | |
1283 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1284 | ps->acc_busy_adjust [in_ACC] = -1; | |
1285 | } | |
1286 | ||
b34f6357 DB |
1287 | void |
1288 | decrease_FR_busy (SIM_CPU *cpu, INT out_FR, int cycles) | |
1289 | { | |
1290 | if (out_FR >= 0) | |
1291 | { | |
1292 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1293 | int *fr = ps->fr_busy; | |
1294 | fr[out_FR] -= cycles; | |
1295 | if (ps->fr_busy_adjust[out_FR] >= 0 | |
1296 | && cycles > ps->fr_busy_adjust[out_FR]) | |
1297 | ps->fr_busy_adjust[out_FR] = cycles; | |
1298 | } | |
1299 | } | |
1300 | ||
1301 | void | |
1302 | increase_FR_busy (SIM_CPU *cpu, INT out_FR, int cycles) | |
1303 | { | |
1304 | if (out_FR >= 0) | |
1305 | { | |
1306 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1307 | int *fr = ps->fr_busy; | |
1308 | fr[out_FR] += cycles; | |
1309 | } | |
1310 | } | |
1311 | ||
1312 | /* Top up the latency of the given ACC by the given number of cycles. */ | |
1313 | void | |
1314 | update_ACC_latency (SIM_CPU *cpu, INT out_ACC, int cycles) | |
1315 | { | |
1316 | if (out_ACC >= 0) | |
1317 | { | |
1318 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1319 | int *acc = ps->acc_latency; | |
1320 | if (acc[out_ACC] < cycles) | |
1321 | acc[out_ACC] = cycles; | |
1322 | } | |
1323 | } | |
1324 | ||
1325 | /* Top up the latency of the given CCR by the given number of cycles. */ | |
1326 | void | |
1327 | update_CCR_latency (SIM_CPU *cpu, INT out_CCR, int cycles) | |
1328 | { | |
1329 | if (out_CCR >= 0) | |
1330 | { | |
1331 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1332 | int *ccr = ps->ccr_latency; | |
1333 | if (ccr[out_CCR] < cycles) | |
1334 | ccr[out_CCR] = cycles; | |
1335 | } | |
1336 | } | |
1337 | ||
153431d6 DB |
1338 | /* Top up the latency of the given SPR by the given number of cycles. */ |
1339 | void | |
1340 | update_SPR_latency (SIM_CPU *cpu, INT out_SPR, int cycles) | |
1341 | { | |
1342 | if (out_SPR >= 0) | |
1343 | { | |
1344 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1345 | int *spr = ps->spr_latency; | |
1346 | if (spr[out_SPR] < cycles) | |
1347 | spr[out_SPR] = cycles; | |
1348 | } | |
1349 | } | |
1350 | ||
b34f6357 DB |
1351 | /* Top up the latency of the given integer division resource by the given |
1352 | number of cycles. */ | |
1353 | void | |
1354 | update_idiv_resource_latency (SIM_CPU *cpu, INT in_resource, int cycles) | |
1355 | { | |
1356 | /* operate directly on the busy cycles since each resource can only | |
1357 | be used once in a VLIW insn. */ | |
1358 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1359 | int *r = ps->idiv_busy; | |
1360 | r[in_resource] = cycles; | |
1361 | } | |
1362 | ||
1363 | /* Set the latency of the given resource to the given number of cycles. */ | |
1364 | void | |
1365 | update_fdiv_resource_latency (SIM_CPU *cpu, INT in_resource, int cycles) | |
1366 | { | |
1367 | /* operate directly on the busy cycles since each resource can only | |
1368 | be used once in a VLIW insn. */ | |
1369 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1370 | int *r = ps->fdiv_busy; | |
1371 | r[in_resource] = cycles; | |
1372 | } | |
1373 | ||
1374 | /* Set the latency of the given resource to the given number of cycles. */ | |
1375 | void | |
1376 | update_fsqrt_resource_latency (SIM_CPU *cpu, INT in_resource, int cycles) | |
1377 | { | |
1378 | /* operate directly on the busy cycles since each resource can only | |
1379 | be used once in a VLIW insn. */ | |
1380 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1381 | int *r = ps->fsqrt_busy; | |
1382 | r[in_resource] = cycles; | |
1383 | } | |
1384 | ||
e930b1f5 DB |
1385 | /* Set the latency of the given resource to the given number of cycles. */ |
1386 | void | |
1387 | update_float_resource_latency (SIM_CPU *cpu, INT in_resource, int cycles) | |
1388 | { | |
1389 | /* operate directly on the busy cycles since each resource can only | |
1390 | be used once in a VLIW insn. */ | |
1391 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1392 | int *r = ps->float_busy; | |
1393 | r[in_resource] = cycles; | |
1394 | } | |
1395 | ||
1396 | void | |
1397 | update_media_resource_latency (SIM_CPU *cpu, INT in_resource, int cycles) | |
1398 | { | |
1399 | /* operate directly on the busy cycles since each resource can only | |
1400 | be used once in a VLIW insn. */ | |
1401 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1402 | int *r = ps->media_busy; | |
1403 | r[in_resource] = cycles; | |
1404 | } | |
1405 | ||
b34f6357 DB |
1406 | /* Set the branch penalty to the given number of cycles. */ |
1407 | void | |
1408 | update_branch_penalty (SIM_CPU *cpu, int cycles) | |
1409 | { | |
1410 | /* operate directly on the busy cycles since only one branch can occur | |
1411 | in a VLIW insn. */ | |
1412 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1413 | ps->branch_penalty = cycles; | |
1414 | } | |
1415 | ||
1416 | /* Check the availability of the given GR register and update the number | |
1417 | of cycles the current VLIW insn must wait until it is available. */ | |
1418 | void | |
1419 | vliw_wait_for_GR (SIM_CPU *cpu, INT in_GR) | |
1420 | { | |
1421 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1422 | int *gr = ps->gr_busy; | |
1423 | /* If the latency of the register is greater than the current wait | |
1424 | then update the current wait. */ | |
1425 | if (in_GR >= 0 && gr[in_GR] > ps->vliw_wait) | |
1426 | { | |
1427 | if (TRACE_INSN_P (cpu)) | |
1428 | sprintf (hazard_name, "Data hazard for gr%d:", in_GR); | |
1429 | ps->vliw_wait = gr[in_GR]; | |
1430 | } | |
1431 | } | |
1432 | ||
1433 | /* Check the availability of the given GR register and update the number | |
1434 | of cycles the current VLIW insn must wait until it is available. */ | |
1435 | void | |
1436 | vliw_wait_for_GRdouble (SIM_CPU *cpu, INT in_GR) | |
1437 | { | |
1438 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1439 | int *gr = ps->gr_busy; | |
1440 | /* If the latency of the register is greater than the current wait | |
1441 | then update the current wait. */ | |
1442 | if (in_GR >= 0) | |
1443 | { | |
1444 | if (gr[in_GR] > ps->vliw_wait) | |
1445 | { | |
1446 | if (TRACE_INSN_P (cpu)) | |
1447 | sprintf (hazard_name, "Data hazard for gr%d:", in_GR); | |
1448 | ps->vliw_wait = gr[in_GR]; | |
1449 | } | |
1450 | if (in_GR < 63 && gr[in_GR + 1] > ps->vliw_wait) | |
1451 | { | |
1452 | if (TRACE_INSN_P (cpu)) | |
1453 | sprintf (hazard_name, "Data hazard for gr%d:", in_GR + 1); | |
1454 | ps->vliw_wait = gr[in_GR + 1]; | |
1455 | } | |
1456 | } | |
1457 | } | |
1458 | ||
1459 | /* Check the availability of the given FR register and update the number | |
1460 | of cycles the current VLIW insn must wait until it is available. */ | |
1461 | void | |
1462 | vliw_wait_for_FR (SIM_CPU *cpu, INT in_FR) | |
1463 | { | |
1464 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1465 | int *fr = ps->fr_busy; | |
1466 | /* If the latency of the register is greater than the current wait | |
1467 | then update the current wait. */ | |
1468 | if (in_FR >= 0 && fr[in_FR] > ps->vliw_wait) | |
1469 | { | |
1470 | if (TRACE_INSN_P (cpu)) | |
1471 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR); | |
1472 | ps->vliw_wait = fr[in_FR]; | |
1473 | } | |
1474 | } | |
1475 | ||
1476 | /* Check the availability of the given GR register and update the number | |
1477 | of cycles the current VLIW insn must wait until it is available. */ | |
1478 | void | |
1479 | vliw_wait_for_FRdouble (SIM_CPU *cpu, INT in_FR) | |
1480 | { | |
1481 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1482 | int *fr = ps->fr_busy; | |
1483 | /* If the latency of the register is greater than the current wait | |
1484 | then update the current wait. */ | |
1485 | if (in_FR >= 0) | |
1486 | { | |
1487 | if (fr[in_FR] > ps->vliw_wait) | |
1488 | { | |
1489 | if (TRACE_INSN_P (cpu)) | |
1490 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR); | |
1491 | ps->vliw_wait = fr[in_FR]; | |
1492 | } | |
1493 | if (in_FR < 63 && fr[in_FR + 1] > ps->vliw_wait) | |
1494 | { | |
1495 | if (TRACE_INSN_P (cpu)) | |
1496 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR + 1); | |
1497 | ps->vliw_wait = fr[in_FR + 1]; | |
1498 | } | |
1499 | } | |
1500 | } | |
1501 | ||
1502 | /* Check the availability of the given CCR register and update the number | |
1503 | of cycles the current VLIW insn must wait until it is available. */ | |
1504 | void | |
1505 | vliw_wait_for_CCR (SIM_CPU *cpu, INT in_CCR) | |
1506 | { | |
1507 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1508 | int *ccr = ps->ccr_busy; | |
1509 | /* If the latency of the register is greater than the current wait | |
1510 | then update the current wait. */ | |
1511 | if (in_CCR >= 0 && ccr[in_CCR] > ps->vliw_wait) | |
1512 | { | |
1513 | if (TRACE_INSN_P (cpu)) | |
1514 | { | |
1515 | if (in_CCR > 3) | |
1516 | sprintf (hazard_name, "Data hazard for icc%d:", in_CCR-4); | |
1517 | else | |
1518 | sprintf (hazard_name, "Data hazard for fcc%d:", in_CCR); | |
1519 | } | |
1520 | ps->vliw_wait = ccr[in_CCR]; | |
1521 | } | |
1522 | } | |
1523 | ||
1524 | /* Check the availability of the given ACC register and update the number | |
1525 | of cycles the current VLIW insn must wait until it is available. */ | |
1526 | void | |
1527 | vliw_wait_for_ACC (SIM_CPU *cpu, INT in_ACC) | |
1528 | { | |
1529 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1530 | int *acc = ps->acc_busy; | |
1531 | /* If the latency of the register is greater than the current wait | |
1532 | then update the current wait. */ | |
1533 | if (in_ACC >= 0 && acc[in_ACC] > ps->vliw_wait) | |
1534 | { | |
1535 | if (TRACE_INSN_P (cpu)) | |
1536 | sprintf (hazard_name, "Data hazard for acc%d:", in_ACC); | |
1537 | ps->vliw_wait = acc[in_ACC]; | |
1538 | } | |
1539 | } | |
1540 | ||
153431d6 DB |
1541 | /* Check the availability of the given SPR register and update the number |
1542 | of cycles the current VLIW insn must wait until it is available. */ | |
1543 | void | |
1544 | vliw_wait_for_SPR (SIM_CPU *cpu, INT in_SPR) | |
1545 | { | |
1546 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1547 | int *spr = ps->spr_busy; | |
1548 | /* If the latency of the register is greater than the current wait | |
1549 | then update the current wait. */ | |
1550 | if (in_SPR >= 0 && spr[in_SPR] > ps->vliw_wait) | |
1551 | { | |
1552 | if (TRACE_INSN_P (cpu)) | |
1553 | sprintf (hazard_name, "Data hazard for spr %d:", in_SPR); | |
1554 | ps->vliw_wait = spr[in_SPR]; | |
1555 | } | |
1556 | } | |
1557 | ||
b34f6357 DB |
1558 | /* Check the availability of the given integer division resource and update |
1559 | the number of cycles the current VLIW insn must wait until it is available. | |
1560 | */ | |
1561 | void | |
1562 | vliw_wait_for_idiv_resource (SIM_CPU *cpu, INT in_resource) | |
1563 | { | |
1564 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1565 | int *r = ps->idiv_busy; | |
1566 | /* If the latency of the resource is greater than the current wait | |
1567 | then update the current wait. */ | |
1568 | if (r[in_resource] > ps->vliw_wait) | |
1569 | { | |
1570 | if (TRACE_INSN_P (cpu)) | |
1571 | { | |
1572 | sprintf (hazard_name, "Resource hazard for integer division in slot I%d:", in_resource); | |
1573 | } | |
1574 | ps->vliw_wait = r[in_resource]; | |
1575 | } | |
1576 | } | |
1577 | ||
1578 | /* Check the availability of the given float division resource and update | |
1579 | the number of cycles the current VLIW insn must wait until it is available. | |
1580 | */ | |
1581 | void | |
1582 | vliw_wait_for_fdiv_resource (SIM_CPU *cpu, INT in_resource) | |
1583 | { | |
1584 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1585 | int *r = ps->fdiv_busy; | |
1586 | /* If the latency of the resource is greater than the current wait | |
1587 | then update the current wait. */ | |
1588 | if (r[in_resource] > ps->vliw_wait) | |
1589 | { | |
1590 | if (TRACE_INSN_P (cpu)) | |
1591 | { | |
1c453cd6 | 1592 | sprintf (hazard_name, "Resource hazard for floating point division in slot F%d:", in_resource); |
b34f6357 DB |
1593 | } |
1594 | ps->vliw_wait = r[in_resource]; | |
1595 | } | |
1596 | } | |
1597 | ||
1598 | /* Check the availability of the given float square root resource and update | |
1599 | the number of cycles the current VLIW insn must wait until it is available. | |
1600 | */ | |
1601 | void | |
1602 | vliw_wait_for_fsqrt_resource (SIM_CPU *cpu, INT in_resource) | |
1603 | { | |
1604 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1605 | int *r = ps->fsqrt_busy; | |
1606 | /* If the latency of the resource is greater than the current wait | |
1607 | then update the current wait. */ | |
1608 | if (r[in_resource] > ps->vliw_wait) | |
1609 | { | |
1610 | if (TRACE_INSN_P (cpu)) | |
1611 | { | |
1c453cd6 | 1612 | sprintf (hazard_name, "Resource hazard for square root in slot F%d:", in_resource); |
b34f6357 DB |
1613 | } |
1614 | ps->vliw_wait = r[in_resource]; | |
1615 | } | |
1616 | } | |
1617 | ||
e930b1f5 DB |
1618 | /* Check the availability of the given float unit resource and update |
1619 | the number of cycles the current VLIW insn must wait until it is available. | |
1620 | */ | |
1621 | void | |
1622 | vliw_wait_for_float_resource (SIM_CPU *cpu, INT in_resource) | |
1623 | { | |
1624 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1625 | int *r = ps->float_busy; | |
1626 | /* If the latency of the resource is greater than the current wait | |
1627 | then update the current wait. */ | |
1628 | if (r[in_resource] > ps->vliw_wait) | |
1629 | { | |
1630 | if (TRACE_INSN_P (cpu)) | |
1631 | { | |
1632 | sprintf (hazard_name, "Resource hazard for floating point unit in slot F%d:", in_resource); | |
1633 | } | |
1634 | ps->vliw_wait = r[in_resource]; | |
1635 | } | |
1636 | } | |
1637 | ||
1638 | /* Check the availability of the given media unit resource and update | |
1639 | the number of cycles the current VLIW insn must wait until it is available. | |
1640 | */ | |
1641 | void | |
1642 | vliw_wait_for_media_resource (SIM_CPU *cpu, INT in_resource) | |
1643 | { | |
1644 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1645 | int *r = ps->media_busy; | |
1646 | /* If the latency of the resource is greater than the current wait | |
1647 | then update the current wait. */ | |
1648 | if (r[in_resource] > ps->vliw_wait) | |
1649 | { | |
1650 | if (TRACE_INSN_P (cpu)) | |
1651 | { | |
1652 | sprintf (hazard_name, "Resource hazard for media unit in slot M%d:", in_resource); | |
1653 | } | |
1654 | ps->vliw_wait = r[in_resource]; | |
1655 | } | |
1656 | } | |
1657 | ||
b34f6357 DB |
1658 | /* Run the caches until all requests for the given register(s) are satisfied. */ |
1659 | void | |
1660 | load_wait_for_GR (SIM_CPU *cpu, INT in_GR) | |
1661 | { | |
1662 | if (in_GR >= 0) | |
1663 | { | |
1664 | int wait = 0; | |
1665 | while (load_pending_for_register (cpu, in_GR, 1/*words*/, REGTYPE_NONE)) | |
1666 | { | |
1667 | frv_model_advance_cycles (cpu, 1); | |
1668 | ++wait; | |
1669 | } | |
1670 | if (wait) | |
1671 | { | |
1672 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1673 | ps->vliw_wait += wait; | |
1674 | ps->vliw_load_stall += wait; | |
1675 | if (TRACE_INSN_P (cpu)) | |
1676 | sprintf (hazard_name, "Data hazard for gr%d:", in_GR); | |
1677 | } | |
1678 | } | |
1679 | } | |
1680 | ||
1681 | void | |
1682 | load_wait_for_FR (SIM_CPU *cpu, INT in_FR) | |
1683 | { | |
1684 | if (in_FR >= 0) | |
1685 | { | |
1686 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1687 | int *fr; | |
1688 | int wait = 0; | |
1689 | while (load_pending_for_register (cpu, in_FR, 1/*words*/, REGTYPE_FR)) | |
1690 | { | |
1691 | frv_model_advance_cycles (cpu, 1); | |
1692 | ++wait; | |
1693 | } | |
1694 | /* Post processing time may have been added to the register's | |
1695 | latency after the loads were processed. Account for that too. | |
1696 | */ | |
1697 | fr = ps->fr_busy; | |
1698 | if (fr[in_FR]) | |
1699 | { | |
1700 | wait += fr[in_FR]; | |
1701 | frv_model_advance_cycles (cpu, fr[in_FR]); | |
1702 | } | |
1703 | /* Update the vliw_wait with the number of cycles we waited for the | |
1704 | load and any post-processing. */ | |
1705 | if (wait) | |
1706 | { | |
1707 | ps->vliw_wait += wait; | |
1708 | ps->vliw_load_stall += wait; | |
1709 | if (TRACE_INSN_P (cpu)) | |
1710 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR); | |
1711 | } | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | void | |
1716 | load_wait_for_GRdouble (SIM_CPU *cpu, INT in_GR) | |
1717 | { | |
1718 | if (in_GR >= 0) | |
1719 | { | |
1720 | int wait = 0; | |
1721 | while (load_pending_for_register (cpu, in_GR, 2/*words*/, REGTYPE_NONE)) | |
1722 | { | |
1723 | frv_model_advance_cycles (cpu, 1); | |
1724 | ++wait; | |
1725 | } | |
1726 | if (wait) | |
1727 | { | |
1728 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1729 | ps->vliw_wait += wait; | |
1730 | ps->vliw_load_stall += wait; | |
1731 | if (TRACE_INSN_P (cpu)) | |
1732 | sprintf (hazard_name, "Data hazard for gr%d:", in_GR); | |
1733 | } | |
1734 | } | |
1735 | } | |
1736 | ||
1737 | void | |
1738 | load_wait_for_FRdouble (SIM_CPU *cpu, INT in_FR) | |
1739 | { | |
1740 | if (in_FR >= 0) | |
1741 | { | |
1742 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1743 | int *fr; | |
1744 | int wait = 0; | |
1745 | while (load_pending_for_register (cpu, in_FR, 2/*words*/, REGTYPE_FR)) | |
1746 | { | |
1747 | frv_model_advance_cycles (cpu, 1); | |
1748 | ++wait; | |
1749 | } | |
1750 | /* Post processing time may have been added to the registers' | |
1751 | latencies after the loads were processed. Account for that too. | |
1752 | */ | |
1753 | fr = ps->fr_busy; | |
1754 | if (fr[in_FR]) | |
1755 | { | |
1756 | wait += fr[in_FR]; | |
1757 | frv_model_advance_cycles (cpu, fr[in_FR]); | |
1758 | } | |
1759 | if (in_FR < 63) | |
1760 | { | |
1761 | if (fr[in_FR + 1]) | |
1762 | { | |
1763 | wait += fr[in_FR + 1]; | |
1764 | frv_model_advance_cycles (cpu, fr[in_FR + 1]); | |
1765 | } | |
1766 | } | |
1767 | /* Update the vliw_wait with the number of cycles we waited for the | |
1768 | load and any post-processing. */ | |
1769 | if (wait) | |
1770 | { | |
1771 | ps->vliw_wait += wait; | |
1772 | ps->vliw_load_stall += wait; | |
1773 | if (TRACE_INSN_P (cpu)) | |
1774 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR); | |
1775 | } | |
1776 | } | |
1777 | } | |
1778 | ||
1779 | void | |
1780 | enforce_full_fr_latency (SIM_CPU *cpu, INT in_FR) | |
1781 | { | |
1782 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1783 | ps->fr_busy_adjust [in_FR] = -1; | |
1784 | } | |
1785 | ||
1786 | /* Calculate how long the post processing for a floating point insn must | |
1787 | wait for resources to become available. */ | |
1788 | int | |
1789 | post_wait_for_FR (SIM_CPU *cpu, INT in_FR) | |
1790 | { | |
1791 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1792 | int *fr = ps->fr_busy; | |
1793 | ||
1794 | if (in_FR >= 0 && fr[in_FR] > ps->post_wait) | |
1795 | { | |
1796 | ps->post_wait = fr[in_FR]; | |
1797 | if (TRACE_INSN_P (cpu)) | |
1798 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR); | |
1799 | } | |
1800 | } | |
1801 | ||
1802 | /* Calculate how long the post processing for a floating point insn must | |
1803 | wait for resources to become available. */ | |
1804 | int | |
1805 | post_wait_for_FRdouble (SIM_CPU *cpu, INT in_FR) | |
1806 | { | |
1807 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1808 | int *fr = ps->fr_busy; | |
1809 | ||
1810 | if (in_FR >= 0) | |
1811 | { | |
1812 | if (fr[in_FR] > ps->post_wait) | |
1813 | { | |
1814 | ps->post_wait = fr[in_FR]; | |
1815 | if (TRACE_INSN_P (cpu)) | |
1816 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR); | |
1817 | } | |
1818 | if (in_FR < 63 && fr[in_FR + 1] > ps->post_wait) | |
1819 | { | |
1820 | ps->post_wait = fr[in_FR + 1]; | |
1821 | if (TRACE_INSN_P (cpu)) | |
1822 | sprintf (hazard_name, "Data hazard for fr%d:", in_FR + 1); | |
1823 | } | |
1824 | } | |
1825 | } | |
1826 | ||
1827 | int | |
1828 | post_wait_for_ACC (SIM_CPU *cpu, INT in_ACC) | |
1829 | { | |
1830 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1831 | int *acc = ps->acc_busy; | |
1832 | ||
1833 | if (in_ACC >= 0 && acc[in_ACC] > ps->post_wait) | |
1834 | { | |
1835 | ps->post_wait = acc[in_ACC]; | |
1836 | if (TRACE_INSN_P (cpu)) | |
1837 | sprintf (hazard_name, "Data hazard for acc%d:", in_ACC); | |
1838 | } | |
1839 | } | |
1840 | ||
1841 | int | |
1842 | post_wait_for_CCR (SIM_CPU *cpu, INT in_CCR) | |
1843 | { | |
1844 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1845 | int *ccr = ps->ccr_busy; | |
1846 | ||
1847 | if (in_CCR >= 0 && ccr[in_CCR] > ps->post_wait) | |
1848 | { | |
1849 | ps->post_wait = ccr[in_CCR]; | |
1850 | if (TRACE_INSN_P (cpu)) | |
1851 | { | |
1852 | if (in_CCR > 3) | |
1853 | sprintf (hazard_name, "Data hazard for icc%d:", in_CCR - 4); | |
1854 | else | |
1855 | sprintf (hazard_name, "Data hazard for fcc%d:", in_CCR); | |
1856 | } | |
1857 | } | |
1858 | } | |
1859 | ||
1c453cd6 DB |
1860 | int |
1861 | post_wait_for_SPR (SIM_CPU *cpu, INT in_SPR) | |
1862 | { | |
1863 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1864 | int *spr = ps->spr_busy; | |
1865 | ||
1866 | if (in_SPR >= 0 && spr[in_SPR] > ps->post_wait) | |
1867 | { | |
1868 | ps->post_wait = spr[in_SPR]; | |
1869 | if (TRACE_INSN_P (cpu)) | |
1870 | sprintf (hazard_name, "Data hazard for spr[%d]:", in_SPR); | |
1871 | } | |
1872 | } | |
1873 | ||
b34f6357 DB |
1874 | int |
1875 | post_wait_for_fdiv (SIM_CPU *cpu, INT slot) | |
1876 | { | |
1877 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1878 | int *fdiv = ps->fdiv_busy; | |
1879 | ||
1880 | /* Multiple floating point divisions in the same slot need only wait 1 | |
1881 | extra cycle. */ | |
1882 | if (fdiv[slot] > 0 && 1 > ps->post_wait) | |
1883 | { | |
1884 | ps->post_wait = 1; | |
1885 | if (TRACE_INSN_P (cpu)) | |
1886 | { | |
1887 | sprintf (hazard_name, "Resource hazard for floating point division in slot F%d:", slot); | |
1888 | } | |
1889 | } | |
1890 | } | |
1891 | ||
1892 | int | |
1893 | post_wait_for_fsqrt (SIM_CPU *cpu, INT slot) | |
1894 | { | |
1895 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1896 | int *fsqrt = ps->fsqrt_busy; | |
1897 | ||
1898 | /* Multiple floating point square roots in the same slot need only wait 1 | |
1899 | extra cycle. */ | |
1900 | if (fsqrt[slot] > 0 && 1 > ps->post_wait) | |
1901 | { | |
1902 | ps->post_wait = 1; | |
1903 | if (TRACE_INSN_P (cpu)) | |
1904 | { | |
1905 | sprintf (hazard_name, "Resource hazard for square root in slot F%d:", slot); | |
1906 | } | |
1907 | } | |
1908 | } | |
1909 | ||
e930b1f5 DB |
1910 | int |
1911 | post_wait_for_float (SIM_CPU *cpu, INT slot) | |
1912 | { | |
1913 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1914 | int *flt = ps->float_busy; | |
1915 | ||
1916 | /* Multiple floating point square roots in the same slot need only wait 1 | |
1917 | extra cycle. */ | |
1918 | if (flt[slot] > ps->post_wait) | |
1919 | { | |
1920 | ps->post_wait = flt[slot]; | |
1921 | if (TRACE_INSN_P (cpu)) | |
1922 | { | |
1923 | sprintf (hazard_name, "Resource hazard for floating point unit in slot F%d:", slot); | |
1924 | } | |
1925 | } | |
1926 | } | |
1927 | ||
1928 | int | |
1929 | post_wait_for_media (SIM_CPU *cpu, INT slot) | |
1930 | { | |
1931 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1932 | int *media = ps->media_busy; | |
1933 | ||
1934 | /* Multiple floating point square roots in the same slot need only wait 1 | |
1935 | extra cycle. */ | |
1936 | if (media[slot] > ps->post_wait) | |
1937 | { | |
1938 | ps->post_wait = media[slot]; | |
1939 | if (TRACE_INSN_P (cpu)) | |
1940 | { | |
1941 | sprintf (hazard_name, "Resource hazard for media unit in slot M%d:", slot); | |
1942 | } | |
1943 | } | |
1944 | } | |
1945 | ||
b34f6357 DB |
1946 | /* Print cpu-specific profile information. */ |
1947 | #define COMMAS(n) sim_add_commas (comma_buf, sizeof (comma_buf), (n)) | |
1948 | ||
1949 | static void | |
1950 | print_cache (SIM_CPU *cpu, FRV_CACHE *cache, const char *cache_name) | |
1951 | { | |
1952 | SIM_DESC sd = CPU_STATE (cpu); | |
1953 | ||
1954 | if (cache != NULL) | |
1955 | { | |
1956 | char comma_buf[20]; | |
1957 | unsigned accesses; | |
1958 | ||
1959 | sim_io_printf (sd, " %s Cache\n\n", cache_name); | |
1960 | accesses = cache->statistics.accesses; | |
1961 | sim_io_printf (sd, " Total accesses: %s\n", COMMAS (accesses)); | |
1962 | if (accesses != 0) | |
1963 | { | |
1964 | float rate; | |
1965 | unsigned hits = cache->statistics.hits; | |
1966 | sim_io_printf (sd, " Hits: %s\n", COMMAS (hits)); | |
1967 | rate = (float)hits / accesses; | |
1968 | sim_io_printf (sd, " Hit rate: %.2f%%\n", rate * 100); | |
1969 | } | |
1970 | } | |
1971 | else | |
1972 | sim_io_printf (sd, " Model %s has no %s cache\n", | |
1973 | MODEL_NAME (CPU_MODEL (cpu)), cache_name); | |
1974 | ||
1975 | sim_io_printf (sd, "\n"); | |
1976 | } | |
1977 | ||
ba9c4053 DB |
1978 | /* This table must correspond to the UNIT_ATTR table in |
1979 | opcodes/frv-desc.h. Only the units up to UNIT_C need be | |
1980 | listed since the others cannot occur after mapping. */ | |
b34f6357 DB |
1981 | static char * |
1982 | slot_names[] = | |
1983 | { | |
1984 | "none", | |
e930b1f5 DB |
1985 | "I0", "I1", "I01", "I2", "I3", "IALL", |
1986 | "FM0", "FM1", "FM01", "FM2", "FM3", "FMALL", "FMLOW", | |
b34f6357 DB |
1987 | "B0", "B1", "B01", |
1988 | "C" | |
1989 | }; | |
1990 | ||
1991 | static void | |
1992 | print_parallel (SIM_CPU *cpu, int verbose) | |
1993 | { | |
1994 | SIM_DESC sd = CPU_STATE (cpu); | |
1995 | PROFILE_DATA *p = CPU_PROFILE_DATA (cpu); | |
1996 | FRV_PROFILE_STATE *ps = CPU_PROFILE_STATE (cpu); | |
1997 | unsigned total, vliw; | |
1998 | char comma_buf[20]; | |
1999 | float average; | |
2000 | ||
2001 | sim_io_printf (sd, "Model %s Parallelization\n\n", | |
2002 | MODEL_NAME (CPU_MODEL (cpu))); | |
2003 | ||
2004 | total = PROFILE_TOTAL_INSN_COUNT (p); | |
2005 | sim_io_printf (sd, " Total instructions: %s\n", COMMAS (total)); | |
2006 | vliw = ps->vliw_insns; | |
2007 | sim_io_printf (sd, " VLIW instructions: %s\n", COMMAS (vliw)); | |
2008 | average = (float)total / vliw; | |
2009 | sim_io_printf (sd, " Average VLIW length: %.2f\n", average); | |
2010 | average = (float)PROFILE_MODEL_TOTAL_CYCLES (p) / vliw; | |
2011 | sim_io_printf (sd, " Cycles per VLIW instruction: %.2f\n", average); | |
2012 | average = (float)total / PROFILE_MODEL_TOTAL_CYCLES (p); | |
2013 | sim_io_printf (sd, " Instructions per cycle: %.2f\n", average); | |
2014 | ||
2015 | if (verbose) | |
2016 | { | |
2017 | int i; | |
2018 | int max_val = 0; | |
2019 | int max_name_len = 0; | |
2020 | for (i = UNIT_NIL + 1; i < UNIT_NUM_UNITS; ++i) | |
2021 | { | |
ba9c4053 DB |
2022 | if (INSNS_IN_SLOT (i)) |
2023 | { | |
2024 | int len; | |
2025 | if (INSNS_IN_SLOT (i) > max_val) | |
2026 | max_val = INSNS_IN_SLOT (i); | |
2027 | len = strlen (slot_names[i]); | |
2028 | if (len > max_name_len) | |
2029 | max_name_len = len; | |
2030 | } | |
b34f6357 DB |
2031 | } |
2032 | if (max_val > 0) | |
2033 | { | |
2034 | sim_io_printf (sd, "\n"); | |
2035 | sim_io_printf (sd, " Instructions per slot:\n"); | |
2036 | sim_io_printf (sd, "\n"); | |
2037 | for (i = UNIT_NIL + 1; i < UNIT_NUM_UNITS; ++i) | |
2038 | { | |
2039 | if (INSNS_IN_SLOT (i) != 0) | |
2040 | { | |
2041 | sim_io_printf (sd, " %*s: %*s: ", | |
2042 | max_name_len, slot_names[i], | |
2043 | max_val < 10000 ? 5 : 10, | |
2044 | COMMAS (INSNS_IN_SLOT (i))); | |
f9bdfd29 | 2045 | sim_profile_print_bar (sd, cpu, PROFILE_HISTOGRAM_WIDTH, |
b34f6357 DB |
2046 | INSNS_IN_SLOT (i), |
2047 | max_val); | |
2048 | sim_io_printf (sd, "\n"); | |
2049 | } | |
2050 | } | |
2051 | } /* details to print */ | |
2052 | } /* verbose */ | |
2053 | ||
2054 | sim_io_printf (sd, "\n"); | |
2055 | } | |
2056 | ||
2057 | void | |
2058 | frv_profile_info (SIM_CPU *cpu, int verbose) | |
2059 | { | |
2060 | /* FIXME: Need to add smp support. */ | |
2061 | PROFILE_DATA *p = CPU_PROFILE_DATA (cpu); | |
2062 | ||
2063 | #if WITH_PROFILE_PARALLEL_P | |
2064 | if (PROFILE_FLAGS (p) [PROFILE_PARALLEL_IDX]) | |
2065 | print_parallel (cpu, verbose); | |
2066 | #endif | |
2067 | ||
2068 | #if WITH_PROFILE_CACHE_P | |
2069 | if (PROFILE_FLAGS (p) [PROFILE_CACHE_IDX]) | |
2070 | { | |
2071 | SIM_DESC sd = CPU_STATE (cpu); | |
2072 | sim_io_printf (sd, "Model %s Cache Statistics\n\n", | |
2073 | MODEL_NAME (CPU_MODEL (cpu))); | |
2074 | print_cache (cpu, CPU_INSN_CACHE (cpu), "Instruction"); | |
2075 | print_cache (cpu, CPU_DATA_CACHE (cpu), "Data"); | |
2076 | } | |
2077 | #endif /* WITH_PROFILE_CACHE_P */ | |
2078 | } | |
2079 | ||
2080 | /* A hack to get registers referenced for profiling. */ | |
2081 | SI frv_ref_SI (SI ref) {return ref;} | |
2082 | #endif /* WITH_PROFILE_MODEL_P */ |