]>
Commit | Line | Data |
---|---|---|
1 | /* Instruction scheduling pass. | |
2 | Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, | |
3 | 2000, 2001, 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc. | |
4 | Contributed by Michael Tiemann (tiemann@cygnus.com) Enhanced by, | |
5 | and currently maintained by, Jim Wilson (wilson@cygnus.com) | |
6 | ||
7 | This file is part of GCC. | |
8 | ||
9 | GCC is free software; you can redistribute it and/or modify it under | |
10 | the terms of the GNU General Public License as published by the Free | |
11 | Software Foundation; either version 2, or (at your option) any later | |
12 | version. | |
13 | ||
14 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
15 | WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
16 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
17 | for more details. | |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with GCC; see the file COPYING. If not, write to the Free | |
21 | Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA | |
22 | 02110-1301, USA. */ | |
23 | ||
24 | /* Instruction scheduling pass. This file, along with sched-deps.c, | |
25 | contains the generic parts. The actual entry point is found for | |
26 | the normal instruction scheduling pass is found in sched-rgn.c. | |
27 | ||
28 | We compute insn priorities based on data dependencies. Flow | |
29 | analysis only creates a fraction of the data-dependencies we must | |
30 | observe: namely, only those dependencies which the combiner can be | |
31 | expected to use. For this pass, we must therefore create the | |
32 | remaining dependencies we need to observe: register dependencies, | |
33 | memory dependencies, dependencies to keep function calls in order, | |
34 | and the dependence between a conditional branch and the setting of | |
35 | condition codes are all dealt with here. | |
36 | ||
37 | The scheduler first traverses the data flow graph, starting with | |
38 | the last instruction, and proceeding to the first, assigning values | |
39 | to insn_priority as it goes. This sorts the instructions | |
40 | topologically by data dependence. | |
41 | ||
42 | Once priorities have been established, we order the insns using | |
43 | list scheduling. This works as follows: starting with a list of | |
44 | all the ready insns, and sorted according to priority number, we | |
45 | schedule the insn from the end of the list by placing its | |
46 | predecessors in the list according to their priority order. We | |
47 | consider this insn scheduled by setting the pointer to the "end" of | |
48 | the list to point to the previous insn. When an insn has no | |
49 | predecessors, we either queue it until sufficient time has elapsed | |
50 | or add it to the ready list. As the instructions are scheduled or | |
51 | when stalls are introduced, the queue advances and dumps insns into | |
52 | the ready list. When all insns down to the lowest priority have | |
53 | been scheduled, the critical path of the basic block has been made | |
54 | as short as possible. The remaining insns are then scheduled in | |
55 | remaining slots. | |
56 | ||
57 | The following list shows the order in which we want to break ties | |
58 | among insns in the ready list: | |
59 | ||
60 | 1. choose insn with the longest path to end of bb, ties | |
61 | broken by | |
62 | 2. choose insn with least contribution to register pressure, | |
63 | ties broken by | |
64 | 3. prefer in-block upon interblock motion, ties broken by | |
65 | 4. prefer useful upon speculative motion, ties broken by | |
66 | 5. choose insn with largest control flow probability, ties | |
67 | broken by | |
68 | 6. choose insn with the least dependences upon the previously | |
69 | scheduled insn, or finally | |
70 | 7 choose the insn which has the most insns dependent on it. | |
71 | 8. choose insn with lowest UID. | |
72 | ||
73 | Memory references complicate matters. Only if we can be certain | |
74 | that memory references are not part of the data dependency graph | |
75 | (via true, anti, or output dependence), can we move operations past | |
76 | memory references. To first approximation, reads can be done | |
77 | independently, while writes introduce dependencies. Better | |
78 | approximations will yield fewer dependencies. | |
79 | ||
80 | Before reload, an extended analysis of interblock data dependences | |
81 | is required for interblock scheduling. This is performed in | |
82 | compute_block_backward_dependences (). | |
83 | ||
84 | Dependencies set up by memory references are treated in exactly the | |
85 | same way as other dependencies, by using LOG_LINKS backward | |
86 | dependences. LOG_LINKS are translated into INSN_DEPEND forward | |
87 | dependences for the purpose of forward list scheduling. | |
88 | ||
89 | Having optimized the critical path, we may have also unduly | |
90 | extended the lifetimes of some registers. If an operation requires | |
91 | that constants be loaded into registers, it is certainly desirable | |
92 | to load those constants as early as necessary, but no earlier. | |
93 | I.e., it will not do to load up a bunch of registers at the | |
94 | beginning of a basic block only to use them at the end, if they | |
95 | could be loaded later, since this may result in excessive register | |
96 | utilization. | |
97 | ||
98 | Note that since branches are never in basic blocks, but only end | |
99 | basic blocks, this pass will not move branches. But that is ok, | |
100 | since we can use GNU's delayed branch scheduling pass to take care | |
101 | of this case. | |
102 | ||
103 | Also note that no further optimizations based on algebraic | |
104 | identities are performed, so this pass would be a good one to | |
105 | perform instruction splitting, such as breaking up a multiply | |
106 | instruction into shifts and adds where that is profitable. | |
107 | ||
108 | Given the memory aliasing analysis that this pass should perform, | |
109 | it should be possible to remove redundant stores to memory, and to | |
110 | load values from registers instead of hitting memory. | |
111 | ||
112 | Before reload, speculative insns are moved only if a 'proof' exists | |
113 | that no exception will be caused by this, and if no live registers | |
114 | exist that inhibit the motion (live registers constraints are not | |
115 | represented by data dependence edges). | |
116 | ||
117 | This pass must update information that subsequent passes expect to | |
118 | be correct. Namely: reg_n_refs, reg_n_sets, reg_n_deaths, | |
119 | reg_n_calls_crossed, and reg_live_length. Also, BB_HEAD, BB_END. | |
120 | ||
121 | The information in the line number notes is carefully retained by | |
122 | this pass. Notes that refer to the starting and ending of | |
123 | exception regions are also carefully retained by this pass. All | |
124 | other NOTE insns are grouped in their same relative order at the | |
125 | beginning of basic blocks and regions that have been scheduled. */ | |
126 | \f | |
127 | #include "config.h" | |
128 | #include "system.h" | |
129 | #include "coretypes.h" | |
130 | #include "tm.h" | |
131 | #include "toplev.h" | |
132 | #include "rtl.h" | |
133 | #include "tm_p.h" | |
134 | #include "hard-reg-set.h" | |
135 | #include "regs.h" | |
136 | #include "function.h" | |
137 | #include "flags.h" | |
138 | #include "insn-config.h" | |
139 | #include "insn-attr.h" | |
140 | #include "except.h" | |
141 | #include "toplev.h" | |
142 | #include "recog.h" | |
143 | #include "sched-int.h" | |
144 | #include "target.h" | |
145 | #include "output.h" | |
146 | #include "params.h" | |
147 | ||
148 | #ifdef INSN_SCHEDULING | |
149 | ||
150 | /* issue_rate is the number of insns that can be scheduled in the same | |
151 | machine cycle. It can be defined in the config/mach/mach.h file, | |
152 | otherwise we set it to 1. */ | |
153 | ||
154 | static int issue_rate; | |
155 | ||
156 | /* sched-verbose controls the amount of debugging output the | |
157 | scheduler prints. It is controlled by -fsched-verbose=N: | |
158 | N>0 and no -DSR : the output is directed to stderr. | |
159 | N>=10 will direct the printouts to stderr (regardless of -dSR). | |
160 | N=1: same as -dSR. | |
161 | N=2: bb's probabilities, detailed ready list info, unit/insn info. | |
162 | N=3: rtl at abort point, control-flow, regions info. | |
163 | N=5: dependences info. */ | |
164 | ||
165 | static int sched_verbose_param = 0; | |
166 | int sched_verbose = 0; | |
167 | ||
168 | /* Debugging file. All printouts are sent to dump, which is always set, | |
169 | either to stderr, or to the dump listing file (-dRS). */ | |
170 | FILE *sched_dump = 0; | |
171 | ||
172 | /* Highest uid before scheduling. */ | |
173 | static int old_max_uid; | |
174 | ||
175 | /* fix_sched_param() is called from toplev.c upon detection | |
176 | of the -fsched-verbose=N option. */ | |
177 | ||
178 | void | |
179 | fix_sched_param (const char *param, const char *val) | |
180 | { | |
181 | if (!strcmp (param, "verbose")) | |
182 | sched_verbose_param = atoi (val); | |
183 | else | |
184 | warning (0, "fix_sched_param: unknown param: %s", param); | |
185 | } | |
186 | ||
187 | struct haifa_insn_data *h_i_d; | |
188 | ||
189 | #define LINE_NOTE(INSN) (h_i_d[INSN_UID (INSN)].line_note) | |
190 | #define INSN_TICK(INSN) (h_i_d[INSN_UID (INSN)].tick) | |
191 | #define INTER_TICK(INSN) (h_i_d[INSN_UID (INSN)].inter_tick) | |
192 | ||
193 | /* If INSN_TICK of an instruction is equal to INVALID_TICK, | |
194 | then it should be recalculated from scratch. */ | |
195 | #define INVALID_TICK (-(max_insn_queue_index + 1)) | |
196 | /* The minimal value of the INSN_TICK of an instruction. */ | |
197 | #define MIN_TICK (-max_insn_queue_index) | |
198 | ||
199 | /* Issue points are used to distinguish between instructions in max_issue (). | |
200 | For now, all instructions are equally good. */ | |
201 | #define ISSUE_POINTS(INSN) 1 | |
202 | ||
203 | /* Vector indexed by basic block number giving the starting line-number | |
204 | for each basic block. */ | |
205 | static rtx *line_note_head; | |
206 | ||
207 | /* List of important notes we must keep around. This is a pointer to the | |
208 | last element in the list. */ | |
209 | static rtx note_list; | |
210 | ||
211 | static struct spec_info_def spec_info_var; | |
212 | /* Description of the speculative part of the scheduling. | |
213 | If NULL - no speculation. */ | |
214 | static spec_info_t spec_info; | |
215 | ||
216 | /* True, if recovery block was added during scheduling of current block. | |
217 | Used to determine, if we need to fix INSN_TICKs. */ | |
218 | static bool added_recovery_block_p; | |
219 | ||
220 | /* Counters of different types of speculative instructions. */ | |
221 | static int nr_begin_data, nr_be_in_data, nr_begin_control, nr_be_in_control; | |
222 | ||
223 | /* Pointers to GLAT data. See init_glat for more information. */ | |
224 | regset *glat_start, *glat_end; | |
225 | ||
226 | /* Array used in {unlink, restore}_bb_notes. */ | |
227 | static rtx *bb_header = 0; | |
228 | ||
229 | /* Number of basic_blocks. */ | |
230 | static int old_last_basic_block; | |
231 | ||
232 | /* Basic block after which recovery blocks will be created. */ | |
233 | static basic_block before_recovery; | |
234 | ||
235 | /* Queues, etc. */ | |
236 | ||
237 | /* An instruction is ready to be scheduled when all insns preceding it | |
238 | have already been scheduled. It is important to ensure that all | |
239 | insns which use its result will not be executed until its result | |
240 | has been computed. An insn is maintained in one of four structures: | |
241 | ||
242 | (P) the "Pending" set of insns which cannot be scheduled until | |
243 | their dependencies have been satisfied. | |
244 | (Q) the "Queued" set of insns that can be scheduled when sufficient | |
245 | time has passed. | |
246 | (R) the "Ready" list of unscheduled, uncommitted insns. | |
247 | (S) the "Scheduled" list of insns. | |
248 | ||
249 | Initially, all insns are either "Pending" or "Ready" depending on | |
250 | whether their dependencies are satisfied. | |
251 | ||
252 | Insns move from the "Ready" list to the "Scheduled" list as they | |
253 | are committed to the schedule. As this occurs, the insns in the | |
254 | "Pending" list have their dependencies satisfied and move to either | |
255 | the "Ready" list or the "Queued" set depending on whether | |
256 | sufficient time has passed to make them ready. As time passes, | |
257 | insns move from the "Queued" set to the "Ready" list. | |
258 | ||
259 | The "Pending" list (P) are the insns in the INSN_DEPEND of the unscheduled | |
260 | insns, i.e., those that are ready, queued, and pending. | |
261 | The "Queued" set (Q) is implemented by the variable `insn_queue'. | |
262 | The "Ready" list (R) is implemented by the variables `ready' and | |
263 | `n_ready'. | |
264 | The "Scheduled" list (S) is the new insn chain built by this pass. | |
265 | ||
266 | The transition (R->S) is implemented in the scheduling loop in | |
267 | `schedule_block' when the best insn to schedule is chosen. | |
268 | The transitions (P->R and P->Q) are implemented in `schedule_insn' as | |
269 | insns move from the ready list to the scheduled list. | |
270 | The transition (Q->R) is implemented in 'queue_to_insn' as time | |
271 | passes or stalls are introduced. */ | |
272 | ||
273 | /* Implement a circular buffer to delay instructions until sufficient | |
274 | time has passed. For the new pipeline description interface, | |
275 | MAX_INSN_QUEUE_INDEX is a power of two minus one which is not less | |
276 | than maximal time of instruction execution computed by genattr.c on | |
277 | the base maximal time of functional unit reservations and getting a | |
278 | result. This is the longest time an insn may be queued. */ | |
279 | ||
280 | static rtx *insn_queue; | |
281 | static int q_ptr = 0; | |
282 | static int q_size = 0; | |
283 | #define NEXT_Q(X) (((X)+1) & max_insn_queue_index) | |
284 | #define NEXT_Q_AFTER(X, C) (((X)+C) & max_insn_queue_index) | |
285 | ||
286 | #define QUEUE_SCHEDULED (-3) | |
287 | #define QUEUE_NOWHERE (-2) | |
288 | #define QUEUE_READY (-1) | |
289 | /* QUEUE_SCHEDULED - INSN is scheduled. | |
290 | QUEUE_NOWHERE - INSN isn't scheduled yet and is neither in | |
291 | queue or ready list. | |
292 | QUEUE_READY - INSN is in ready list. | |
293 | N >= 0 - INSN queued for X [where NEXT_Q_AFTER (q_ptr, X) == N] cycles. */ | |
294 | ||
295 | #define QUEUE_INDEX(INSN) (h_i_d[INSN_UID (INSN)].queue_index) | |
296 | ||
297 | /* The following variable value refers for all current and future | |
298 | reservations of the processor units. */ | |
299 | state_t curr_state; | |
300 | ||
301 | /* The following variable value is size of memory representing all | |
302 | current and future reservations of the processor units. */ | |
303 | static size_t dfa_state_size; | |
304 | ||
305 | /* The following array is used to find the best insn from ready when | |
306 | the automaton pipeline interface is used. */ | |
307 | static char *ready_try; | |
308 | ||
309 | /* Describe the ready list of the scheduler. | |
310 | VEC holds space enough for all insns in the current region. VECLEN | |
311 | says how many exactly. | |
312 | FIRST is the index of the element with the highest priority; i.e. the | |
313 | last one in the ready list, since elements are ordered by ascending | |
314 | priority. | |
315 | N_READY determines how many insns are on the ready list. */ | |
316 | ||
317 | struct ready_list | |
318 | { | |
319 | rtx *vec; | |
320 | int veclen; | |
321 | int first; | |
322 | int n_ready; | |
323 | }; | |
324 | ||
325 | /* The pointer to the ready list. */ | |
326 | static struct ready_list *readyp; | |
327 | ||
328 | /* Scheduling clock. */ | |
329 | static int clock_var; | |
330 | ||
331 | /* Number of instructions in current scheduling region. */ | |
332 | static int rgn_n_insns; | |
333 | ||
334 | static int may_trap_exp (rtx, int); | |
335 | ||
336 | /* Nonzero iff the address is comprised from at most 1 register. */ | |
337 | #define CONST_BASED_ADDRESS_P(x) \ | |
338 | (REG_P (x) \ | |
339 | || ((GET_CODE (x) == PLUS || GET_CODE (x) == MINUS \ | |
340 | || (GET_CODE (x) == LO_SUM)) \ | |
341 | && (CONSTANT_P (XEXP (x, 0)) \ | |
342 | || CONSTANT_P (XEXP (x, 1))))) | |
343 | ||
344 | /* Returns a class that insn with GET_DEST(insn)=x may belong to, | |
345 | as found by analyzing insn's expression. */ | |
346 | ||
347 | static int | |
348 | may_trap_exp (rtx x, int is_store) | |
349 | { | |
350 | enum rtx_code code; | |
351 | ||
352 | if (x == 0) | |
353 | return TRAP_FREE; | |
354 | code = GET_CODE (x); | |
355 | if (is_store) | |
356 | { | |
357 | if (code == MEM && may_trap_p (x)) | |
358 | return TRAP_RISKY; | |
359 | else | |
360 | return TRAP_FREE; | |
361 | } | |
362 | if (code == MEM) | |
363 | { | |
364 | /* The insn uses memory: a volatile load. */ | |
365 | if (MEM_VOLATILE_P (x)) | |
366 | return IRISKY; | |
367 | /* An exception-free load. */ | |
368 | if (!may_trap_p (x)) | |
369 | return IFREE; | |
370 | /* A load with 1 base register, to be further checked. */ | |
371 | if (CONST_BASED_ADDRESS_P (XEXP (x, 0))) | |
372 | return PFREE_CANDIDATE; | |
373 | /* No info on the load, to be further checked. */ | |
374 | return PRISKY_CANDIDATE; | |
375 | } | |
376 | else | |
377 | { | |
378 | const char *fmt; | |
379 | int i, insn_class = TRAP_FREE; | |
380 | ||
381 | /* Neither store nor load, check if it may cause a trap. */ | |
382 | if (may_trap_p (x)) | |
383 | return TRAP_RISKY; | |
384 | /* Recursive step: walk the insn... */ | |
385 | fmt = GET_RTX_FORMAT (code); | |
386 | for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--) | |
387 | { | |
388 | if (fmt[i] == 'e') | |
389 | { | |
390 | int tmp_class = may_trap_exp (XEXP (x, i), is_store); | |
391 | insn_class = WORST_CLASS (insn_class, tmp_class); | |
392 | } | |
393 | else if (fmt[i] == 'E') | |
394 | { | |
395 | int j; | |
396 | for (j = 0; j < XVECLEN (x, i); j++) | |
397 | { | |
398 | int tmp_class = may_trap_exp (XVECEXP (x, i, j), is_store); | |
399 | insn_class = WORST_CLASS (insn_class, tmp_class); | |
400 | if (insn_class == TRAP_RISKY || insn_class == IRISKY) | |
401 | break; | |
402 | } | |
403 | } | |
404 | if (insn_class == TRAP_RISKY || insn_class == IRISKY) | |
405 | break; | |
406 | } | |
407 | return insn_class; | |
408 | } | |
409 | } | |
410 | ||
411 | /* Classifies insn for the purpose of verifying that it can be | |
412 | moved speculatively, by examining it's patterns, returning: | |
413 | TRAP_RISKY: store, or risky non-load insn (e.g. division by variable). | |
414 | TRAP_FREE: non-load insn. | |
415 | IFREE: load from a globally safe location. | |
416 | IRISKY: volatile load. | |
417 | PFREE_CANDIDATE, PRISKY_CANDIDATE: load that need to be checked for | |
418 | being either PFREE or PRISKY. */ | |
419 | ||
420 | int | |
421 | haifa_classify_insn (rtx insn) | |
422 | { | |
423 | rtx pat = PATTERN (insn); | |
424 | int tmp_class = TRAP_FREE; | |
425 | int insn_class = TRAP_FREE; | |
426 | enum rtx_code code; | |
427 | ||
428 | if (GET_CODE (pat) == PARALLEL) | |
429 | { | |
430 | int i, len = XVECLEN (pat, 0); | |
431 | ||
432 | for (i = len - 1; i >= 0; i--) | |
433 | { | |
434 | code = GET_CODE (XVECEXP (pat, 0, i)); | |
435 | switch (code) | |
436 | { | |
437 | case CLOBBER: | |
438 | /* Test if it is a 'store'. */ | |
439 | tmp_class = may_trap_exp (XEXP (XVECEXP (pat, 0, i), 0), 1); | |
440 | break; | |
441 | case SET: | |
442 | /* Test if it is a store. */ | |
443 | tmp_class = may_trap_exp (SET_DEST (XVECEXP (pat, 0, i)), 1); | |
444 | if (tmp_class == TRAP_RISKY) | |
445 | break; | |
446 | /* Test if it is a load. */ | |
447 | tmp_class | |
448 | = WORST_CLASS (tmp_class, | |
449 | may_trap_exp (SET_SRC (XVECEXP (pat, 0, i)), | |
450 | 0)); | |
451 | break; | |
452 | case COND_EXEC: | |
453 | case TRAP_IF: | |
454 | tmp_class = TRAP_RISKY; | |
455 | break; | |
456 | default: | |
457 | ; | |
458 | } | |
459 | insn_class = WORST_CLASS (insn_class, tmp_class); | |
460 | if (insn_class == TRAP_RISKY || insn_class == IRISKY) | |
461 | break; | |
462 | } | |
463 | } | |
464 | else | |
465 | { | |
466 | code = GET_CODE (pat); | |
467 | switch (code) | |
468 | { | |
469 | case CLOBBER: | |
470 | /* Test if it is a 'store'. */ | |
471 | tmp_class = may_trap_exp (XEXP (pat, 0), 1); | |
472 | break; | |
473 | case SET: | |
474 | /* Test if it is a store. */ | |
475 | tmp_class = may_trap_exp (SET_DEST (pat), 1); | |
476 | if (tmp_class == TRAP_RISKY) | |
477 | break; | |
478 | /* Test if it is a load. */ | |
479 | tmp_class = | |
480 | WORST_CLASS (tmp_class, | |
481 | may_trap_exp (SET_SRC (pat), 0)); | |
482 | break; | |
483 | case COND_EXEC: | |
484 | case TRAP_IF: | |
485 | tmp_class = TRAP_RISKY; | |
486 | break; | |
487 | default:; | |
488 | } | |
489 | insn_class = tmp_class; | |
490 | } | |
491 | ||
492 | return insn_class; | |
493 | } | |
494 | ||
495 | /* Forward declarations. */ | |
496 | ||
497 | HAIFA_INLINE static int insn_cost1 (rtx, enum reg_note, rtx, rtx); | |
498 | static int priority (rtx); | |
499 | static int rank_for_schedule (const void *, const void *); | |
500 | static void swap_sort (rtx *, int); | |
501 | static void queue_insn (rtx, int); | |
502 | static int schedule_insn (rtx); | |
503 | static int find_set_reg_weight (rtx); | |
504 | static void find_insn_reg_weight (basic_block); | |
505 | static void find_insn_reg_weight1 (rtx); | |
506 | static void adjust_priority (rtx); | |
507 | static void advance_one_cycle (void); | |
508 | ||
509 | /* Notes handling mechanism: | |
510 | ========================= | |
511 | Generally, NOTES are saved before scheduling and restored after scheduling. | |
512 | The scheduler distinguishes between three types of notes: | |
513 | ||
514 | (1) LINE_NUMBER notes, generated and used for debugging. Here, | |
515 | before scheduling a region, a pointer to the LINE_NUMBER note is | |
516 | added to the insn following it (in save_line_notes()), and the note | |
517 | is removed (in rm_line_notes() and unlink_line_notes()). After | |
518 | scheduling the region, this pointer is used for regeneration of | |
519 | the LINE_NUMBER note (in restore_line_notes()). | |
520 | ||
521 | (2) LOOP_BEGIN, LOOP_END, SETJMP, EHREGION_BEG, EHREGION_END notes: | |
522 | Before scheduling a region, a pointer to the note is added to the insn | |
523 | that follows or precedes it. (This happens as part of the data dependence | |
524 | computation). After scheduling an insn, the pointer contained in it is | |
525 | used for regenerating the corresponding note (in reemit_notes). | |
526 | ||
527 | (3) All other notes (e.g. INSN_DELETED): Before scheduling a block, | |
528 | these notes are put in a list (in rm_other_notes() and | |
529 | unlink_other_notes ()). After scheduling the block, these notes are | |
530 | inserted at the beginning of the block (in schedule_block()). */ | |
531 | ||
532 | static rtx unlink_other_notes (rtx, rtx); | |
533 | static rtx unlink_line_notes (rtx, rtx); | |
534 | static void reemit_notes (rtx); | |
535 | ||
536 | static rtx *ready_lastpos (struct ready_list *); | |
537 | static void ready_add (struct ready_list *, rtx, bool); | |
538 | static void ready_sort (struct ready_list *); | |
539 | static rtx ready_remove_first (struct ready_list *); | |
540 | ||
541 | static void queue_to_ready (struct ready_list *); | |
542 | static int early_queue_to_ready (state_t, struct ready_list *); | |
543 | ||
544 | static void debug_ready_list (struct ready_list *); | |
545 | ||
546 | static void move_insn (rtx); | |
547 | ||
548 | /* The following functions are used to implement multi-pass scheduling | |
549 | on the first cycle. */ | |
550 | static rtx ready_element (struct ready_list *, int); | |
551 | static rtx ready_remove (struct ready_list *, int); | |
552 | static void ready_remove_insn (rtx); | |
553 | static int max_issue (struct ready_list *, int *, int); | |
554 | ||
555 | static rtx choose_ready (struct ready_list *); | |
556 | ||
557 | static void fix_inter_tick (rtx, rtx); | |
558 | static int fix_tick_ready (rtx); | |
559 | static void change_queue_index (rtx, int); | |
560 | static void resolve_dep (rtx, rtx); | |
561 | ||
562 | /* The following functions are used to implement scheduling of data/control | |
563 | speculative instructions. */ | |
564 | ||
565 | static void extend_h_i_d (void); | |
566 | static void extend_ready (int); | |
567 | static void extend_global (rtx); | |
568 | static void extend_all (rtx); | |
569 | static void init_h_i_d (rtx); | |
570 | static void generate_recovery_code (rtx); | |
571 | static void process_insn_depend_be_in_spec (rtx, rtx, ds_t); | |
572 | static void begin_speculative_block (rtx); | |
573 | static void add_to_speculative_block (rtx); | |
574 | static dw_t dep_weak (ds_t); | |
575 | static edge find_fallthru_edge (basic_block); | |
576 | static void init_before_recovery (void); | |
577 | static basic_block create_recovery_block (void); | |
578 | static void create_check_block_twin (rtx, bool); | |
579 | static void fix_recovery_deps (basic_block); | |
580 | static void associate_line_notes_with_blocks (basic_block); | |
581 | static void change_pattern (rtx, rtx); | |
582 | static int speculate_insn (rtx, ds_t, rtx *); | |
583 | static void dump_new_block_header (int, basic_block, rtx, rtx); | |
584 | static void restore_bb_notes (basic_block); | |
585 | static void extend_bb (basic_block); | |
586 | static void fix_jump_move (rtx); | |
587 | static void move_block_after_check (rtx); | |
588 | static void move_succs (VEC(edge,gc) **, basic_block); | |
589 | static void init_glat (void); | |
590 | static void init_glat1 (basic_block); | |
591 | static void attach_life_info1 (basic_block); | |
592 | static void free_glat (void); | |
593 | static void sched_remove_insn (rtx); | |
594 | static void clear_priorities (rtx); | |
595 | static void add_jump_dependencies (rtx, rtx); | |
596 | static rtx bb_note (basic_block); | |
597 | static void calc_priorities (rtx); | |
598 | #ifdef ENABLE_CHECKING | |
599 | static int has_edge_p (VEC(edge,gc) *, int); | |
600 | static void check_cfg (rtx, rtx); | |
601 | static void check_sched_flags (void); | |
602 | #endif | |
603 | ||
604 | #endif /* INSN_SCHEDULING */ | |
605 | \f | |
606 | /* Point to state used for the current scheduling pass. */ | |
607 | struct sched_info *current_sched_info; | |
608 | \f | |
609 | #ifndef INSN_SCHEDULING | |
610 | void | |
611 | schedule_insns (void) | |
612 | { | |
613 | } | |
614 | #else | |
615 | ||
616 | /* Working copy of frontend's sched_info variable. */ | |
617 | static struct sched_info current_sched_info_var; | |
618 | ||
619 | /* Pointer to the last instruction scheduled. Used by rank_for_schedule, | |
620 | so that insns independent of the last scheduled insn will be preferred | |
621 | over dependent instructions. */ | |
622 | ||
623 | static rtx last_scheduled_insn; | |
624 | ||
625 | /* Compute cost of executing INSN given the dependence LINK on the insn USED. | |
626 | This is the number of cycles between instruction issue and | |
627 | instruction results. */ | |
628 | ||
629 | HAIFA_INLINE int | |
630 | insn_cost (rtx insn, rtx link, rtx used) | |
631 | { | |
632 | return insn_cost1 (insn, used ? REG_NOTE_KIND (link) : REG_NOTE_MAX, | |
633 | link, used); | |
634 | } | |
635 | ||
636 | /* Compute cost of executing INSN given the dependence on the insn USED. | |
637 | If LINK is not NULL, then its REG_NOTE_KIND is used as a dependence type. | |
638 | Otherwise, dependence between INSN and USED is assumed to be of type | |
639 | DEP_TYPE. This function was introduced as a workaround for | |
640 | targetm.adjust_cost hook. | |
641 | This is the number of cycles between instruction issue and | |
642 | instruction results. */ | |
643 | ||
644 | HAIFA_INLINE static int | |
645 | insn_cost1 (rtx insn, enum reg_note dep_type, rtx link, rtx used) | |
646 | { | |
647 | int cost = INSN_COST (insn); | |
648 | ||
649 | if (cost < 0) | |
650 | { | |
651 | /* A USE insn, or something else we don't need to | |
652 | understand. We can't pass these directly to | |
653 | result_ready_cost or insn_default_latency because it will | |
654 | trigger a fatal error for unrecognizable insns. */ | |
655 | if (recog_memoized (insn) < 0) | |
656 | { | |
657 | INSN_COST (insn) = 0; | |
658 | return 0; | |
659 | } | |
660 | else | |
661 | { | |
662 | cost = insn_default_latency (insn); | |
663 | if (cost < 0) | |
664 | cost = 0; | |
665 | ||
666 | INSN_COST (insn) = cost; | |
667 | } | |
668 | } | |
669 | ||
670 | /* In this case estimate cost without caring how insn is used. */ | |
671 | if (used == 0) | |
672 | return cost; | |
673 | ||
674 | /* A USE insn should never require the value used to be computed. | |
675 | This allows the computation of a function's result and parameter | |
676 | values to overlap the return and call. */ | |
677 | if (recog_memoized (used) < 0) | |
678 | cost = 0; | |
679 | else | |
680 | { | |
681 | gcc_assert (!link || dep_type == REG_NOTE_KIND (link)); | |
682 | ||
683 | if (INSN_CODE (insn) >= 0) | |
684 | { | |
685 | if (dep_type == REG_DEP_ANTI) | |
686 | cost = 0; | |
687 | else if (dep_type == REG_DEP_OUTPUT) | |
688 | { | |
689 | cost = (insn_default_latency (insn) | |
690 | - insn_default_latency (used)); | |
691 | if (cost <= 0) | |
692 | cost = 1; | |
693 | } | |
694 | else if (bypass_p (insn)) | |
695 | cost = insn_latency (insn, used); | |
696 | } | |
697 | ||
698 | if (targetm.sched.adjust_cost_2) | |
699 | cost = targetm.sched.adjust_cost_2 (used, (int) dep_type, insn, cost); | |
700 | else | |
701 | { | |
702 | gcc_assert (link); | |
703 | if (targetm.sched.adjust_cost) | |
704 | cost = targetm.sched.adjust_cost (used, link, insn, cost); | |
705 | } | |
706 | ||
707 | if (cost < 0) | |
708 | cost = 0; | |
709 | } | |
710 | ||
711 | return cost; | |
712 | } | |
713 | ||
714 | /* Compute the priority number for INSN. */ | |
715 | ||
716 | static int | |
717 | priority (rtx insn) | |
718 | { | |
719 | rtx link; | |
720 | ||
721 | if (! INSN_P (insn)) | |
722 | return 0; | |
723 | ||
724 | if (! INSN_PRIORITY_KNOWN (insn)) | |
725 | { | |
726 | int this_priority = 0; | |
727 | ||
728 | if (INSN_DEPEND (insn) == 0) | |
729 | this_priority = insn_cost (insn, 0, 0); | |
730 | else | |
731 | { | |
732 | rtx prev_first, twin; | |
733 | basic_block rec; | |
734 | ||
735 | /* For recovery check instructions we calculate priority slightly | |
736 | different than that of normal instructions. Instead of walking | |
737 | through INSN_DEPEND (check) list, we walk through INSN_DEPEND list | |
738 | of each instruction in the corresponding recovery block. */ | |
739 | ||
740 | rec = RECOVERY_BLOCK (insn); | |
741 | if (!rec || rec == EXIT_BLOCK_PTR) | |
742 | { | |
743 | prev_first = PREV_INSN (insn); | |
744 | twin = insn; | |
745 | } | |
746 | else | |
747 | { | |
748 | prev_first = NEXT_INSN (BB_HEAD (rec)); | |
749 | twin = PREV_INSN (BB_END (rec)); | |
750 | } | |
751 | ||
752 | do | |
753 | { | |
754 | for (link = INSN_DEPEND (twin); link; link = XEXP (link, 1)) | |
755 | { | |
756 | rtx next; | |
757 | int next_priority; | |
758 | ||
759 | next = XEXP (link, 0); | |
760 | ||
761 | if (BLOCK_FOR_INSN (next) != rec) | |
762 | { | |
763 | /* Critical path is meaningful in block boundaries | |
764 | only. */ | |
765 | if (! (*current_sched_info->contributes_to_priority) | |
766 | (next, insn) | |
767 | /* If flag COUNT_SPEC_IN_CRITICAL_PATH is set, | |
768 | then speculative instructions will less likely be | |
769 | scheduled. That is because the priority of | |
770 | their producers will increase, and, thus, the | |
771 | producers will more likely be scheduled, thus, | |
772 | resolving the dependence. */ | |
773 | || ((current_sched_info->flags & DO_SPECULATION) | |
774 | && (DEP_STATUS (link) & SPECULATIVE) | |
775 | && !(spec_info->flags | |
776 | & COUNT_SPEC_IN_CRITICAL_PATH))) | |
777 | continue; | |
778 | ||
779 | next_priority = insn_cost1 (insn, | |
780 | twin == insn ? | |
781 | REG_NOTE_KIND (link) : | |
782 | REG_DEP_ANTI, | |
783 | twin == insn ? link : 0, | |
784 | next) + priority (next); | |
785 | ||
786 | if (next_priority > this_priority) | |
787 | this_priority = next_priority; | |
788 | } | |
789 | } | |
790 | ||
791 | twin = PREV_INSN (twin); | |
792 | } | |
793 | while (twin != prev_first); | |
794 | } | |
795 | INSN_PRIORITY (insn) = this_priority; | |
796 | INSN_PRIORITY_KNOWN (insn) = 1; | |
797 | } | |
798 | ||
799 | return INSN_PRIORITY (insn); | |
800 | } | |
801 | \f | |
802 | /* Macros and functions for keeping the priority queue sorted, and | |
803 | dealing with queuing and dequeuing of instructions. */ | |
804 | ||
805 | #define SCHED_SORT(READY, N_READY) \ | |
806 | do { if ((N_READY) == 2) \ | |
807 | swap_sort (READY, N_READY); \ | |
808 | else if ((N_READY) > 2) \ | |
809 | qsort (READY, N_READY, sizeof (rtx), rank_for_schedule); } \ | |
810 | while (0) | |
811 | ||
812 | /* Returns a positive value if x is preferred; returns a negative value if | |
813 | y is preferred. Should never return 0, since that will make the sort | |
814 | unstable. */ | |
815 | ||
816 | static int | |
817 | rank_for_schedule (const void *x, const void *y) | |
818 | { | |
819 | rtx tmp = *(const rtx *) y; | |
820 | rtx tmp2 = *(const rtx *) x; | |
821 | rtx link; | |
822 | int tmp_class, tmp2_class, depend_count1, depend_count2; | |
823 | int val, priority_val, weight_val, info_val; | |
824 | ||
825 | /* The insn in a schedule group should be issued the first. */ | |
826 | if (SCHED_GROUP_P (tmp) != SCHED_GROUP_P (tmp2)) | |
827 | return SCHED_GROUP_P (tmp2) ? 1 : -1; | |
828 | ||
829 | /* Prefer insn with higher priority. */ | |
830 | priority_val = INSN_PRIORITY (tmp2) - INSN_PRIORITY (tmp); | |
831 | ||
832 | if (priority_val) | |
833 | return priority_val; | |
834 | ||
835 | /* Prefer speculative insn with greater dependencies weakness. */ | |
836 | if (spec_info) | |
837 | { | |
838 | ds_t ds1, ds2; | |
839 | dw_t dw1, dw2; | |
840 | int dw; | |
841 | ||
842 | ds1 = TODO_SPEC (tmp) & SPECULATIVE; | |
843 | if (ds1) | |
844 | dw1 = dep_weak (ds1); | |
845 | else | |
846 | dw1 = NO_DEP_WEAK; | |
847 | ||
848 | ds2 = TODO_SPEC (tmp2) & SPECULATIVE; | |
849 | if (ds2) | |
850 | dw2 = dep_weak (ds2); | |
851 | else | |
852 | dw2 = NO_DEP_WEAK; | |
853 | ||
854 | dw = dw2 - dw1; | |
855 | if (dw > (NO_DEP_WEAK / 8) || dw < -(NO_DEP_WEAK / 8)) | |
856 | return dw; | |
857 | } | |
858 | ||
859 | /* Prefer an insn with smaller contribution to registers-pressure. */ | |
860 | if (!reload_completed && | |
861 | (weight_val = INSN_REG_WEIGHT (tmp) - INSN_REG_WEIGHT (tmp2))) | |
862 | return weight_val; | |
863 | ||
864 | info_val = (*current_sched_info->rank) (tmp, tmp2); | |
865 | if (info_val) | |
866 | return info_val; | |
867 | ||
868 | /* Compare insns based on their relation to the last-scheduled-insn. */ | |
869 | if (INSN_P (last_scheduled_insn)) | |
870 | { | |
871 | /* Classify the instructions into three classes: | |
872 | 1) Data dependent on last schedule insn. | |
873 | 2) Anti/Output dependent on last scheduled insn. | |
874 | 3) Independent of last scheduled insn, or has latency of one. | |
875 | Choose the insn from the highest numbered class if different. */ | |
876 | link = find_insn_list (tmp, INSN_DEPEND (last_scheduled_insn)); | |
877 | if (link == 0 || insn_cost (last_scheduled_insn, link, tmp) == 1) | |
878 | tmp_class = 3; | |
879 | else if (REG_NOTE_KIND (link) == 0) /* Data dependence. */ | |
880 | tmp_class = 1; | |
881 | else | |
882 | tmp_class = 2; | |
883 | ||
884 | link = find_insn_list (tmp2, INSN_DEPEND (last_scheduled_insn)); | |
885 | if (link == 0 || insn_cost (last_scheduled_insn, link, tmp2) == 1) | |
886 | tmp2_class = 3; | |
887 | else if (REG_NOTE_KIND (link) == 0) /* Data dependence. */ | |
888 | tmp2_class = 1; | |
889 | else | |
890 | tmp2_class = 2; | |
891 | ||
892 | if ((val = tmp2_class - tmp_class)) | |
893 | return val; | |
894 | } | |
895 | ||
896 | /* Prefer the insn which has more later insns that depend on it. | |
897 | This gives the scheduler more freedom when scheduling later | |
898 | instructions at the expense of added register pressure. */ | |
899 | depend_count1 = 0; | |
900 | for (link = INSN_DEPEND (tmp); link; link = XEXP (link, 1)) | |
901 | depend_count1++; | |
902 | ||
903 | depend_count2 = 0; | |
904 | for (link = INSN_DEPEND (tmp2); link; link = XEXP (link, 1)) | |
905 | depend_count2++; | |
906 | ||
907 | val = depend_count2 - depend_count1; | |
908 | if (val) | |
909 | return val; | |
910 | ||
911 | /* If insns are equally good, sort by INSN_LUID (original insn order), | |
912 | so that we make the sort stable. This minimizes instruction movement, | |
913 | thus minimizing sched's effect on debugging and cross-jumping. */ | |
914 | return INSN_LUID (tmp) - INSN_LUID (tmp2); | |
915 | } | |
916 | ||
917 | /* Resort the array A in which only element at index N may be out of order. */ | |
918 | ||
919 | HAIFA_INLINE static void | |
920 | swap_sort (rtx *a, int n) | |
921 | { | |
922 | rtx insn = a[n - 1]; | |
923 | int i = n - 2; | |
924 | ||
925 | while (i >= 0 && rank_for_schedule (a + i, &insn) >= 0) | |
926 | { | |
927 | a[i + 1] = a[i]; | |
928 | i -= 1; | |
929 | } | |
930 | a[i + 1] = insn; | |
931 | } | |
932 | ||
933 | /* Add INSN to the insn queue so that it can be executed at least | |
934 | N_CYCLES after the currently executing insn. Preserve insns | |
935 | chain for debugging purposes. */ | |
936 | ||
937 | HAIFA_INLINE static void | |
938 | queue_insn (rtx insn, int n_cycles) | |
939 | { | |
940 | int next_q = NEXT_Q_AFTER (q_ptr, n_cycles); | |
941 | rtx link = alloc_INSN_LIST (insn, insn_queue[next_q]); | |
942 | ||
943 | gcc_assert (n_cycles <= max_insn_queue_index); | |
944 | ||
945 | insn_queue[next_q] = link; | |
946 | q_size += 1; | |
947 | ||
948 | if (sched_verbose >= 2) | |
949 | { | |
950 | fprintf (sched_dump, ";;\t\tReady-->Q: insn %s: ", | |
951 | (*current_sched_info->print_insn) (insn, 0)); | |
952 | ||
953 | fprintf (sched_dump, "queued for %d cycles.\n", n_cycles); | |
954 | } | |
955 | ||
956 | QUEUE_INDEX (insn) = next_q; | |
957 | } | |
958 | ||
959 | /* Remove INSN from queue. */ | |
960 | static void | |
961 | queue_remove (rtx insn) | |
962 | { | |
963 | gcc_assert (QUEUE_INDEX (insn) >= 0); | |
964 | remove_free_INSN_LIST_elem (insn, &insn_queue[QUEUE_INDEX (insn)]); | |
965 | q_size--; | |
966 | QUEUE_INDEX (insn) = QUEUE_NOWHERE; | |
967 | } | |
968 | ||
969 | /* Return a pointer to the bottom of the ready list, i.e. the insn | |
970 | with the lowest priority. */ | |
971 | ||
972 | HAIFA_INLINE static rtx * | |
973 | ready_lastpos (struct ready_list *ready) | |
974 | { | |
975 | gcc_assert (ready->n_ready >= 1); | |
976 | return ready->vec + ready->first - ready->n_ready + 1; | |
977 | } | |
978 | ||
979 | /* Add an element INSN to the ready list so that it ends up with the | |
980 | lowest/highest priority depending on FIRST_P. */ | |
981 | ||
982 | HAIFA_INLINE static void | |
983 | ready_add (struct ready_list *ready, rtx insn, bool first_p) | |
984 | { | |
985 | if (!first_p) | |
986 | { | |
987 | if (ready->first == ready->n_ready) | |
988 | { | |
989 | memmove (ready->vec + ready->veclen - ready->n_ready, | |
990 | ready_lastpos (ready), | |
991 | ready->n_ready * sizeof (rtx)); | |
992 | ready->first = ready->veclen - 1; | |
993 | } | |
994 | ready->vec[ready->first - ready->n_ready] = insn; | |
995 | } | |
996 | else | |
997 | { | |
998 | if (ready->first == ready->veclen - 1) | |
999 | { | |
1000 | if (ready->n_ready) | |
1001 | /* ready_lastpos() fails when called with (ready->n_ready == 0). */ | |
1002 | memmove (ready->vec + ready->veclen - ready->n_ready - 1, | |
1003 | ready_lastpos (ready), | |
1004 | ready->n_ready * sizeof (rtx)); | |
1005 | ready->first = ready->veclen - 2; | |
1006 | } | |
1007 | ready->vec[++(ready->first)] = insn; | |
1008 | } | |
1009 | ||
1010 | ready->n_ready++; | |
1011 | ||
1012 | gcc_assert (QUEUE_INDEX (insn) != QUEUE_READY); | |
1013 | QUEUE_INDEX (insn) = QUEUE_READY; | |
1014 | } | |
1015 | ||
1016 | /* Remove the element with the highest priority from the ready list and | |
1017 | return it. */ | |
1018 | ||
1019 | HAIFA_INLINE static rtx | |
1020 | ready_remove_first (struct ready_list *ready) | |
1021 | { | |
1022 | rtx t; | |
1023 | ||
1024 | gcc_assert (ready->n_ready); | |
1025 | t = ready->vec[ready->first--]; | |
1026 | ready->n_ready--; | |
1027 | /* If the queue becomes empty, reset it. */ | |
1028 | if (ready->n_ready == 0) | |
1029 | ready->first = ready->veclen - 1; | |
1030 | ||
1031 | gcc_assert (QUEUE_INDEX (t) == QUEUE_READY); | |
1032 | QUEUE_INDEX (t) = QUEUE_NOWHERE; | |
1033 | ||
1034 | return t; | |
1035 | } | |
1036 | ||
1037 | /* The following code implements multi-pass scheduling for the first | |
1038 | cycle. In other words, we will try to choose ready insn which | |
1039 | permits to start maximum number of insns on the same cycle. */ | |
1040 | ||
1041 | /* Return a pointer to the element INDEX from the ready. INDEX for | |
1042 | insn with the highest priority is 0, and the lowest priority has | |
1043 | N_READY - 1. */ | |
1044 | ||
1045 | HAIFA_INLINE static rtx | |
1046 | ready_element (struct ready_list *ready, int index) | |
1047 | { | |
1048 | gcc_assert (ready->n_ready && index < ready->n_ready); | |
1049 | ||
1050 | return ready->vec[ready->first - index]; | |
1051 | } | |
1052 | ||
1053 | /* Remove the element INDEX from the ready list and return it. INDEX | |
1054 | for insn with the highest priority is 0, and the lowest priority | |
1055 | has N_READY - 1. */ | |
1056 | ||
1057 | HAIFA_INLINE static rtx | |
1058 | ready_remove (struct ready_list *ready, int index) | |
1059 | { | |
1060 | rtx t; | |
1061 | int i; | |
1062 | ||
1063 | if (index == 0) | |
1064 | return ready_remove_first (ready); | |
1065 | gcc_assert (ready->n_ready && index < ready->n_ready); | |
1066 | t = ready->vec[ready->first - index]; | |
1067 | ready->n_ready--; | |
1068 | for (i = index; i < ready->n_ready; i++) | |
1069 | ready->vec[ready->first - i] = ready->vec[ready->first - i - 1]; | |
1070 | QUEUE_INDEX (t) = QUEUE_NOWHERE; | |
1071 | return t; | |
1072 | } | |
1073 | ||
1074 | /* Remove INSN from the ready list. */ | |
1075 | static void | |
1076 | ready_remove_insn (rtx insn) | |
1077 | { | |
1078 | int i; | |
1079 | ||
1080 | for (i = 0; i < readyp->n_ready; i++) | |
1081 | if (ready_element (readyp, i) == insn) | |
1082 | { | |
1083 | ready_remove (readyp, i); | |
1084 | return; | |
1085 | } | |
1086 | gcc_unreachable (); | |
1087 | } | |
1088 | ||
1089 | /* Sort the ready list READY by ascending priority, using the SCHED_SORT | |
1090 | macro. */ | |
1091 | ||
1092 | HAIFA_INLINE static void | |
1093 | ready_sort (struct ready_list *ready) | |
1094 | { | |
1095 | rtx *first = ready_lastpos (ready); | |
1096 | SCHED_SORT (first, ready->n_ready); | |
1097 | } | |
1098 | ||
1099 | /* PREV is an insn that is ready to execute. Adjust its priority if that | |
1100 | will help shorten or lengthen register lifetimes as appropriate. Also | |
1101 | provide a hook for the target to tweek itself. */ | |
1102 | ||
1103 | HAIFA_INLINE static void | |
1104 | adjust_priority (rtx prev) | |
1105 | { | |
1106 | /* ??? There used to be code here to try and estimate how an insn | |
1107 | affected register lifetimes, but it did it by looking at REG_DEAD | |
1108 | notes, which we removed in schedule_region. Nor did it try to | |
1109 | take into account register pressure or anything useful like that. | |
1110 | ||
1111 | Revisit when we have a machine model to work with and not before. */ | |
1112 | ||
1113 | if (targetm.sched.adjust_priority) | |
1114 | INSN_PRIORITY (prev) = | |
1115 | targetm.sched.adjust_priority (prev, INSN_PRIORITY (prev)); | |
1116 | } | |
1117 | ||
1118 | /* Advance time on one cycle. */ | |
1119 | HAIFA_INLINE static void | |
1120 | advance_one_cycle (void) | |
1121 | { | |
1122 | if (targetm.sched.dfa_pre_cycle_insn) | |
1123 | state_transition (curr_state, | |
1124 | targetm.sched.dfa_pre_cycle_insn ()); | |
1125 | ||
1126 | state_transition (curr_state, NULL); | |
1127 | ||
1128 | if (targetm.sched.dfa_post_cycle_insn) | |
1129 | state_transition (curr_state, | |
1130 | targetm.sched.dfa_post_cycle_insn ()); | |
1131 | } | |
1132 | ||
1133 | /* Clock at which the previous instruction was issued. */ | |
1134 | static int last_clock_var; | |
1135 | ||
1136 | /* INSN is the "currently executing insn". Launch each insn which was | |
1137 | waiting on INSN. READY is the ready list which contains the insns | |
1138 | that are ready to fire. CLOCK is the current cycle. The function | |
1139 | returns necessary cycle advance after issuing the insn (it is not | |
1140 | zero for insns in a schedule group). */ | |
1141 | ||
1142 | static int | |
1143 | schedule_insn (rtx insn) | |
1144 | { | |
1145 | rtx link; | |
1146 | int advance = 0; | |
1147 | ||
1148 | if (sched_verbose >= 1) | |
1149 | { | |
1150 | char buf[2048]; | |
1151 | ||
1152 | print_insn (buf, insn, 0); | |
1153 | buf[40] = 0; | |
1154 | fprintf (sched_dump, ";;\t%3i--> %-40s:", clock_var, buf); | |
1155 | ||
1156 | if (recog_memoized (insn) < 0) | |
1157 | fprintf (sched_dump, "nothing"); | |
1158 | else | |
1159 | print_reservation (sched_dump, insn); | |
1160 | fputc ('\n', sched_dump); | |
1161 | } | |
1162 | ||
1163 | /* Scheduling instruction should have all its dependencies resolved and | |
1164 | should have been removed from the ready list. */ | |
1165 | gcc_assert (INSN_DEP_COUNT (insn) == 0); | |
1166 | gcc_assert (!LOG_LINKS (insn)); | |
1167 | gcc_assert (QUEUE_INDEX (insn) == QUEUE_NOWHERE); | |
1168 | ||
1169 | QUEUE_INDEX (insn) = QUEUE_SCHEDULED; | |
1170 | ||
1171 | /* Now we can free RESOLVED_DEPS list. */ | |
1172 | if (current_sched_info->flags & USE_DEPS_LIST) | |
1173 | free_DEPS_LIST_list (&RESOLVED_DEPS (insn)); | |
1174 | else | |
1175 | free_INSN_LIST_list (&RESOLVED_DEPS (insn)); | |
1176 | ||
1177 | gcc_assert (INSN_TICK (insn) >= MIN_TICK); | |
1178 | if (INSN_TICK (insn) > clock_var) | |
1179 | /* INSN has been prematurely moved from the queue to the ready list. | |
1180 | This is possible only if following flag is set. */ | |
1181 | gcc_assert (flag_sched_stalled_insns); | |
1182 | ||
1183 | /* ??? Probably, if INSN is scheduled prematurely, we should leave | |
1184 | INSN_TICK untouched. This is a machine-dependent issue, actually. */ | |
1185 | INSN_TICK (insn) = clock_var; | |
1186 | ||
1187 | /* Update dependent instructions. */ | |
1188 | for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1)) | |
1189 | { | |
1190 | rtx next = XEXP (link, 0); | |
1191 | ||
1192 | resolve_dep (next, insn); | |
1193 | ||
1194 | if (!RECOVERY_BLOCK (insn) | |
1195 | || RECOVERY_BLOCK (insn) == EXIT_BLOCK_PTR) | |
1196 | { | |
1197 | int effective_cost; | |
1198 | ||
1199 | effective_cost = try_ready (next); | |
1200 | ||
1201 | if (effective_cost >= 0 | |
1202 | && SCHED_GROUP_P (next) | |
1203 | && advance < effective_cost) | |
1204 | advance = effective_cost; | |
1205 | } | |
1206 | else | |
1207 | /* Check always has only one forward dependence (to the first insn in | |
1208 | the recovery block), therefore, this will be executed only once. */ | |
1209 | { | |
1210 | gcc_assert (XEXP (link, 1) == 0); | |
1211 | fix_recovery_deps (RECOVERY_BLOCK (insn)); | |
1212 | } | |
1213 | } | |
1214 | ||
1215 | /* Annotate the instruction with issue information -- TImode | |
1216 | indicates that the instruction is expected not to be able | |
1217 | to issue on the same cycle as the previous insn. A machine | |
1218 | may use this information to decide how the instruction should | |
1219 | be aligned. */ | |
1220 | if (issue_rate > 1 | |
1221 | && GET_CODE (PATTERN (insn)) != USE | |
1222 | && GET_CODE (PATTERN (insn)) != CLOBBER) | |
1223 | { | |
1224 | if (reload_completed) | |
1225 | PUT_MODE (insn, clock_var > last_clock_var ? TImode : VOIDmode); | |
1226 | last_clock_var = clock_var; | |
1227 | } | |
1228 | ||
1229 | return advance; | |
1230 | } | |
1231 | ||
1232 | /* Functions for handling of notes. */ | |
1233 | ||
1234 | /* Delete notes beginning with INSN and put them in the chain | |
1235 | of notes ended by NOTE_LIST. | |
1236 | Returns the insn following the notes. */ | |
1237 | ||
1238 | static rtx | |
1239 | unlink_other_notes (rtx insn, rtx tail) | |
1240 | { | |
1241 | rtx prev = PREV_INSN (insn); | |
1242 | ||
1243 | while (insn != tail && NOTE_NOT_BB_P (insn)) | |
1244 | { | |
1245 | rtx next = NEXT_INSN (insn); | |
1246 | /* Delete the note from its current position. */ | |
1247 | if (prev) | |
1248 | NEXT_INSN (prev) = next; | |
1249 | if (next) | |
1250 | PREV_INSN (next) = prev; | |
1251 | ||
1252 | /* See sched_analyze to see how these are handled. */ | |
1253 | if (NOTE_LINE_NUMBER (insn) != NOTE_INSN_EH_REGION_BEG | |
1254 | && NOTE_LINE_NUMBER (insn) != NOTE_INSN_EH_REGION_END) | |
1255 | { | |
1256 | /* Insert the note at the end of the notes list. */ | |
1257 | PREV_INSN (insn) = note_list; | |
1258 | if (note_list) | |
1259 | NEXT_INSN (note_list) = insn; | |
1260 | note_list = insn; | |
1261 | } | |
1262 | ||
1263 | insn = next; | |
1264 | } | |
1265 | return insn; | |
1266 | } | |
1267 | ||
1268 | /* Delete line notes beginning with INSN. Record line-number notes so | |
1269 | they can be reused. Returns the insn following the notes. */ | |
1270 | ||
1271 | static rtx | |
1272 | unlink_line_notes (rtx insn, rtx tail) | |
1273 | { | |
1274 | rtx prev = PREV_INSN (insn); | |
1275 | ||
1276 | while (insn != tail && NOTE_P (insn)) | |
1277 | { | |
1278 | rtx next = NEXT_INSN (insn); | |
1279 | ||
1280 | if (write_symbols != NO_DEBUG && NOTE_LINE_NUMBER (insn) > 0) | |
1281 | { | |
1282 | /* Delete the note from its current position. */ | |
1283 | if (prev) | |
1284 | NEXT_INSN (prev) = next; | |
1285 | if (next) | |
1286 | PREV_INSN (next) = prev; | |
1287 | ||
1288 | /* Record line-number notes so they can be reused. */ | |
1289 | LINE_NOTE (insn) = insn; | |
1290 | } | |
1291 | else | |
1292 | prev = insn; | |
1293 | ||
1294 | insn = next; | |
1295 | } | |
1296 | return insn; | |
1297 | } | |
1298 | ||
1299 | /* Return the head and tail pointers of ebb starting at BEG and ending | |
1300 | at END. */ | |
1301 | ||
1302 | void | |
1303 | get_ebb_head_tail (basic_block beg, basic_block end, rtx *headp, rtx *tailp) | |
1304 | { | |
1305 | rtx beg_head = BB_HEAD (beg); | |
1306 | rtx beg_tail = BB_END (beg); | |
1307 | rtx end_head = BB_HEAD (end); | |
1308 | rtx end_tail = BB_END (end); | |
1309 | ||
1310 | /* Don't include any notes or labels at the beginning of the BEG | |
1311 | basic block, or notes at the end of the END basic blocks. */ | |
1312 | ||
1313 | if (LABEL_P (beg_head)) | |
1314 | beg_head = NEXT_INSN (beg_head); | |
1315 | ||
1316 | while (beg_head != beg_tail) | |
1317 | if (NOTE_P (beg_head)) | |
1318 | beg_head = NEXT_INSN (beg_head); | |
1319 | else | |
1320 | break; | |
1321 | ||
1322 | *headp = beg_head; | |
1323 | ||
1324 | if (beg == end) | |
1325 | end_head = beg_head; | |
1326 | else if (LABEL_P (end_head)) | |
1327 | end_head = NEXT_INSN (end_head); | |
1328 | ||
1329 | while (end_head != end_tail) | |
1330 | if (NOTE_P (end_tail)) | |
1331 | end_tail = PREV_INSN (end_tail); | |
1332 | else | |
1333 | break; | |
1334 | ||
1335 | *tailp = end_tail; | |
1336 | } | |
1337 | ||
1338 | /* Return nonzero if there are no real insns in the range [ HEAD, TAIL ]. */ | |
1339 | ||
1340 | int | |
1341 | no_real_insns_p (rtx head, rtx tail) | |
1342 | { | |
1343 | while (head != NEXT_INSN (tail)) | |
1344 | { | |
1345 | if (!NOTE_P (head) && !LABEL_P (head)) | |
1346 | return 0; | |
1347 | head = NEXT_INSN (head); | |
1348 | } | |
1349 | return 1; | |
1350 | } | |
1351 | ||
1352 | /* Delete line notes from one block. Save them so they can be later restored | |
1353 | (in restore_line_notes). HEAD and TAIL are the boundaries of the | |
1354 | block in which notes should be processed. */ | |
1355 | ||
1356 | void | |
1357 | rm_line_notes (rtx head, rtx tail) | |
1358 | { | |
1359 | rtx next_tail; | |
1360 | rtx insn; | |
1361 | ||
1362 | next_tail = NEXT_INSN (tail); | |
1363 | for (insn = head; insn != next_tail; insn = NEXT_INSN (insn)) | |
1364 | { | |
1365 | rtx prev; | |
1366 | ||
1367 | /* Farm out notes, and maybe save them in NOTE_LIST. | |
1368 | This is needed to keep the debugger from | |
1369 | getting completely deranged. */ | |
1370 | if (NOTE_NOT_BB_P (insn)) | |
1371 | { | |
1372 | prev = insn; | |
1373 | insn = unlink_line_notes (insn, next_tail); | |
1374 | ||
1375 | gcc_assert (prev != tail && prev != head && insn != next_tail); | |
1376 | } | |
1377 | } | |
1378 | } | |
1379 | ||
1380 | /* Save line number notes for each insn in block B. HEAD and TAIL are | |
1381 | the boundaries of the block in which notes should be processed. */ | |
1382 | ||
1383 | void | |
1384 | save_line_notes (int b, rtx head, rtx tail) | |
1385 | { | |
1386 | rtx next_tail; | |
1387 | ||
1388 | /* We must use the true line number for the first insn in the block | |
1389 | that was computed and saved at the start of this pass. We can't | |
1390 | use the current line number, because scheduling of the previous | |
1391 | block may have changed the current line number. */ | |
1392 | ||
1393 | rtx line = line_note_head[b]; | |
1394 | rtx insn; | |
1395 | ||
1396 | next_tail = NEXT_INSN (tail); | |
1397 | ||
1398 | for (insn = head; insn != next_tail; insn = NEXT_INSN (insn)) | |
1399 | if (NOTE_P (insn) && NOTE_LINE_NUMBER (insn) > 0) | |
1400 | line = insn; | |
1401 | else | |
1402 | LINE_NOTE (insn) = line; | |
1403 | } | |
1404 | ||
1405 | /* After a block was scheduled, insert line notes into the insns list. | |
1406 | HEAD and TAIL are the boundaries of the block in which notes should | |
1407 | be processed. */ | |
1408 | ||
1409 | void | |
1410 | restore_line_notes (rtx head, rtx tail) | |
1411 | { | |
1412 | rtx line, note, prev, new; | |
1413 | int added_notes = 0; | |
1414 | rtx next_tail, insn; | |
1415 | ||
1416 | head = head; | |
1417 | next_tail = NEXT_INSN (tail); | |
1418 | ||
1419 | /* Determine the current line-number. We want to know the current | |
1420 | line number of the first insn of the block here, in case it is | |
1421 | different from the true line number that was saved earlier. If | |
1422 | different, then we need a line number note before the first insn | |
1423 | of this block. If it happens to be the same, then we don't want to | |
1424 | emit another line number note here. */ | |
1425 | for (line = head; line; line = PREV_INSN (line)) | |
1426 | if (NOTE_P (line) && NOTE_LINE_NUMBER (line) > 0) | |
1427 | break; | |
1428 | ||
1429 | /* Walk the insns keeping track of the current line-number and inserting | |
1430 | the line-number notes as needed. */ | |
1431 | for (insn = head; insn != next_tail; insn = NEXT_INSN (insn)) | |
1432 | if (NOTE_P (insn) && NOTE_LINE_NUMBER (insn) > 0) | |
1433 | line = insn; | |
1434 | /* This used to emit line number notes before every non-deleted note. | |
1435 | However, this confuses a debugger, because line notes not separated | |
1436 | by real instructions all end up at the same address. I can find no | |
1437 | use for line number notes before other notes, so none are emitted. */ | |
1438 | else if (!NOTE_P (insn) | |
1439 | && INSN_UID (insn) < old_max_uid | |
1440 | && (note = LINE_NOTE (insn)) != 0 | |
1441 | && note != line | |
1442 | && (line == 0 | |
1443 | #ifdef USE_MAPPED_LOCATION | |
1444 | || NOTE_SOURCE_LOCATION (note) != NOTE_SOURCE_LOCATION (line) | |
1445 | #else | |
1446 | || NOTE_LINE_NUMBER (note) != NOTE_LINE_NUMBER (line) | |
1447 | || NOTE_SOURCE_FILE (note) != NOTE_SOURCE_FILE (line) | |
1448 | #endif | |
1449 | )) | |
1450 | { | |
1451 | line = note; | |
1452 | prev = PREV_INSN (insn); | |
1453 | if (LINE_NOTE (note)) | |
1454 | { | |
1455 | /* Re-use the original line-number note. */ | |
1456 | LINE_NOTE (note) = 0; | |
1457 | PREV_INSN (note) = prev; | |
1458 | NEXT_INSN (prev) = note; | |
1459 | PREV_INSN (insn) = note; | |
1460 | NEXT_INSN (note) = insn; | |
1461 | set_block_for_insn (note, BLOCK_FOR_INSN (insn)); | |
1462 | } | |
1463 | else | |
1464 | { | |
1465 | added_notes++; | |
1466 | new = emit_note_after (NOTE_LINE_NUMBER (note), prev); | |
1467 | #ifndef USE_MAPPED_LOCATION | |
1468 | NOTE_SOURCE_FILE (new) = NOTE_SOURCE_FILE (note); | |
1469 | #endif | |
1470 | } | |
1471 | } | |
1472 | if (sched_verbose && added_notes) | |
1473 | fprintf (sched_dump, ";; added %d line-number notes\n", added_notes); | |
1474 | } | |
1475 | ||
1476 | /* After scheduling the function, delete redundant line notes from the | |
1477 | insns list. */ | |
1478 | ||
1479 | void | |
1480 | rm_redundant_line_notes (void) | |
1481 | { | |
1482 | rtx line = 0; | |
1483 | rtx insn = get_insns (); | |
1484 | int active_insn = 0; | |
1485 | int notes = 0; | |
1486 | ||
1487 | /* Walk the insns deleting redundant line-number notes. Many of these | |
1488 | are already present. The remainder tend to occur at basic | |
1489 | block boundaries. */ | |
1490 | for (insn = get_last_insn (); insn; insn = PREV_INSN (insn)) | |
1491 | if (NOTE_P (insn) && NOTE_LINE_NUMBER (insn) > 0) | |
1492 | { | |
1493 | /* If there are no active insns following, INSN is redundant. */ | |
1494 | if (active_insn == 0) | |
1495 | { | |
1496 | notes++; | |
1497 | SET_INSN_DELETED (insn); | |
1498 | } | |
1499 | /* If the line number is unchanged, LINE is redundant. */ | |
1500 | else if (line | |
1501 | #ifdef USE_MAPPED_LOCATION | |
1502 | && NOTE_SOURCE_LOCATION (line) == NOTE_SOURCE_LOCATION (insn) | |
1503 | #else | |
1504 | && NOTE_LINE_NUMBER (line) == NOTE_LINE_NUMBER (insn) | |
1505 | && NOTE_SOURCE_FILE (line) == NOTE_SOURCE_FILE (insn) | |
1506 | #endif | |
1507 | ) | |
1508 | { | |
1509 | notes++; | |
1510 | SET_INSN_DELETED (line); | |
1511 | line = insn; | |
1512 | } | |
1513 | else | |
1514 | line = insn; | |
1515 | active_insn = 0; | |
1516 | } | |
1517 | else if (!((NOTE_P (insn) | |
1518 | && NOTE_LINE_NUMBER (insn) == NOTE_INSN_DELETED) | |
1519 | || (NONJUMP_INSN_P (insn) | |
1520 | && (GET_CODE (PATTERN (insn)) == USE | |
1521 | || GET_CODE (PATTERN (insn)) == CLOBBER)))) | |
1522 | active_insn++; | |
1523 | ||
1524 | if (sched_verbose && notes) | |
1525 | fprintf (sched_dump, ";; deleted %d line-number notes\n", notes); | |
1526 | } | |
1527 | ||
1528 | /* Delete notes between HEAD and TAIL and put them in the chain | |
1529 | of notes ended by NOTE_LIST. */ | |
1530 | ||
1531 | void | |
1532 | rm_other_notes (rtx head, rtx tail) | |
1533 | { | |
1534 | rtx next_tail; | |
1535 | rtx insn; | |
1536 | ||
1537 | note_list = 0; | |
1538 | if (head == tail && (! INSN_P (head))) | |
1539 | return; | |
1540 | ||
1541 | next_tail = NEXT_INSN (tail); | |
1542 | for (insn = head; insn != next_tail; insn = NEXT_INSN (insn)) | |
1543 | { | |
1544 | rtx prev; | |
1545 | ||
1546 | /* Farm out notes, and maybe save them in NOTE_LIST. | |
1547 | This is needed to keep the debugger from | |
1548 | getting completely deranged. */ | |
1549 | if (NOTE_NOT_BB_P (insn)) | |
1550 | { | |
1551 | prev = insn; | |
1552 | ||
1553 | insn = unlink_other_notes (insn, next_tail); | |
1554 | ||
1555 | gcc_assert (prev != tail && prev != head && insn != next_tail); | |
1556 | } | |
1557 | } | |
1558 | } | |
1559 | ||
1560 | /* Functions for computation of registers live/usage info. */ | |
1561 | ||
1562 | /* This function looks for a new register being defined. | |
1563 | If the destination register is already used by the source, | |
1564 | a new register is not needed. */ | |
1565 | ||
1566 | static int | |
1567 | find_set_reg_weight (rtx x) | |
1568 | { | |
1569 | if (GET_CODE (x) == CLOBBER | |
1570 | && register_operand (SET_DEST (x), VOIDmode)) | |
1571 | return 1; | |
1572 | if (GET_CODE (x) == SET | |
1573 | && register_operand (SET_DEST (x), VOIDmode)) | |
1574 | { | |
1575 | if (REG_P (SET_DEST (x))) | |
1576 | { | |
1577 | if (!reg_mentioned_p (SET_DEST (x), SET_SRC (x))) | |
1578 | return 1; | |
1579 | else | |
1580 | return 0; | |
1581 | } | |
1582 | return 1; | |
1583 | } | |
1584 | return 0; | |
1585 | } | |
1586 | ||
1587 | /* Calculate INSN_REG_WEIGHT for all insns of a block. */ | |
1588 | ||
1589 | static void | |
1590 | find_insn_reg_weight (basic_block bb) | |
1591 | { | |
1592 | rtx insn, next_tail, head, tail; | |
1593 | ||
1594 | get_ebb_head_tail (bb, bb, &head, &tail); | |
1595 | next_tail = NEXT_INSN (tail); | |
1596 | ||
1597 | for (insn = head; insn != next_tail; insn = NEXT_INSN (insn)) | |
1598 | find_insn_reg_weight1 (insn); | |
1599 | } | |
1600 | ||
1601 | /* Calculate INSN_REG_WEIGHT for single instruction. | |
1602 | Separated from find_insn_reg_weight because of need | |
1603 | to initialize new instruction in generate_recovery_code. */ | |
1604 | static void | |
1605 | find_insn_reg_weight1 (rtx insn) | |
1606 | { | |
1607 | int reg_weight = 0; | |
1608 | rtx x; | |
1609 | ||
1610 | /* Handle register life information. */ | |
1611 | if (! INSN_P (insn)) | |
1612 | return; | |
1613 | ||
1614 | /* Increment weight for each register born here. */ | |
1615 | x = PATTERN (insn); | |
1616 | reg_weight += find_set_reg_weight (x); | |
1617 | if (GET_CODE (x) == PARALLEL) | |
1618 | { | |
1619 | int j; | |
1620 | for (j = XVECLEN (x, 0) - 1; j >= 0; j--) | |
1621 | { | |
1622 | x = XVECEXP (PATTERN (insn), 0, j); | |
1623 | reg_weight += find_set_reg_weight (x); | |
1624 | } | |
1625 | } | |
1626 | /* Decrement weight for each register that dies here. */ | |
1627 | for (x = REG_NOTES (insn); x; x = XEXP (x, 1)) | |
1628 | { | |
1629 | if (REG_NOTE_KIND (x) == REG_DEAD | |
1630 | || REG_NOTE_KIND (x) == REG_UNUSED) | |
1631 | reg_weight--; | |
1632 | } | |
1633 | ||
1634 | INSN_REG_WEIGHT (insn) = reg_weight; | |
1635 | } | |
1636 | ||
1637 | /* Move insns that became ready to fire from queue to ready list. */ | |
1638 | ||
1639 | static void | |
1640 | queue_to_ready (struct ready_list *ready) | |
1641 | { | |
1642 | rtx insn; | |
1643 | rtx link; | |
1644 | ||
1645 | q_ptr = NEXT_Q (q_ptr); | |
1646 | ||
1647 | /* Add all pending insns that can be scheduled without stalls to the | |
1648 | ready list. */ | |
1649 | for (link = insn_queue[q_ptr]; link; link = XEXP (link, 1)) | |
1650 | { | |
1651 | insn = XEXP (link, 0); | |
1652 | q_size -= 1; | |
1653 | ||
1654 | if (sched_verbose >= 2) | |
1655 | fprintf (sched_dump, ";;\t\tQ-->Ready: insn %s: ", | |
1656 | (*current_sched_info->print_insn) (insn, 0)); | |
1657 | ||
1658 | /* If the ready list is full, delay the insn for 1 cycle. | |
1659 | See the comment in schedule_block for the rationale. */ | |
1660 | if (!reload_completed | |
1661 | && ready->n_ready > MAX_SCHED_READY_INSNS | |
1662 | && !SCHED_GROUP_P (insn)) | |
1663 | { | |
1664 | if (sched_verbose >= 2) | |
1665 | fprintf (sched_dump, "requeued because ready full\n"); | |
1666 | queue_insn (insn, 1); | |
1667 | } | |
1668 | else | |
1669 | { | |
1670 | ready_add (ready, insn, false); | |
1671 | if (sched_verbose >= 2) | |
1672 | fprintf (sched_dump, "moving to ready without stalls\n"); | |
1673 | } | |
1674 | } | |
1675 | free_INSN_LIST_list (&insn_queue[q_ptr]); | |
1676 | ||
1677 | /* If there are no ready insns, stall until one is ready and add all | |
1678 | of the pending insns at that point to the ready list. */ | |
1679 | if (ready->n_ready == 0) | |
1680 | { | |
1681 | int stalls; | |
1682 | ||
1683 | for (stalls = 1; stalls <= max_insn_queue_index; stalls++) | |
1684 | { | |
1685 | if ((link = insn_queue[NEXT_Q_AFTER (q_ptr, stalls)])) | |
1686 | { | |
1687 | for (; link; link = XEXP (link, 1)) | |
1688 | { | |
1689 | insn = XEXP (link, 0); | |
1690 | q_size -= 1; | |
1691 | ||
1692 | if (sched_verbose >= 2) | |
1693 | fprintf (sched_dump, ";;\t\tQ-->Ready: insn %s: ", | |
1694 | (*current_sched_info->print_insn) (insn, 0)); | |
1695 | ||
1696 | ready_add (ready, insn, false); | |
1697 | if (sched_verbose >= 2) | |
1698 | fprintf (sched_dump, "moving to ready with %d stalls\n", stalls); | |
1699 | } | |
1700 | free_INSN_LIST_list (&insn_queue[NEXT_Q_AFTER (q_ptr, stalls)]); | |
1701 | ||
1702 | advance_one_cycle (); | |
1703 | ||
1704 | break; | |
1705 | } | |
1706 | ||
1707 | advance_one_cycle (); | |
1708 | } | |
1709 | ||
1710 | q_ptr = NEXT_Q_AFTER (q_ptr, stalls); | |
1711 | clock_var += stalls; | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | /* Used by early_queue_to_ready. Determines whether it is "ok" to | |
1716 | prematurely move INSN from the queue to the ready list. Currently, | |
1717 | if a target defines the hook 'is_costly_dependence', this function | |
1718 | uses the hook to check whether there exist any dependences which are | |
1719 | considered costly by the target, between INSN and other insns that | |
1720 | have already been scheduled. Dependences are checked up to Y cycles | |
1721 | back, with default Y=1; The flag -fsched-stalled-insns-dep=Y allows | |
1722 | controlling this value. | |
1723 | (Other considerations could be taken into account instead (or in | |
1724 | addition) depending on user flags and target hooks. */ | |
1725 | ||
1726 | static bool | |
1727 | ok_for_early_queue_removal (rtx insn) | |
1728 | { | |
1729 | int n_cycles; | |
1730 | rtx prev_insn = last_scheduled_insn; | |
1731 | ||
1732 | if (targetm.sched.is_costly_dependence) | |
1733 | { | |
1734 | for (n_cycles = flag_sched_stalled_insns_dep; n_cycles; n_cycles--) | |
1735 | { | |
1736 | for ( ; prev_insn; prev_insn = PREV_INSN (prev_insn)) | |
1737 | { | |
1738 | rtx dep_link = 0; | |
1739 | int dep_cost; | |
1740 | ||
1741 | if (!NOTE_P (prev_insn)) | |
1742 | { | |
1743 | dep_link = find_insn_list (insn, INSN_DEPEND (prev_insn)); | |
1744 | if (dep_link) | |
1745 | { | |
1746 | dep_cost = insn_cost (prev_insn, dep_link, insn) ; | |
1747 | if (targetm.sched.is_costly_dependence (prev_insn, insn, | |
1748 | dep_link, dep_cost, | |
1749 | flag_sched_stalled_insns_dep - n_cycles)) | |
1750 | return false; | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | if (GET_MODE (prev_insn) == TImode) /* end of dispatch group */ | |
1755 | break; | |
1756 | } | |
1757 | ||
1758 | if (!prev_insn) | |
1759 | break; | |
1760 | prev_insn = PREV_INSN (prev_insn); | |
1761 | } | |
1762 | } | |
1763 | ||
1764 | return true; | |
1765 | } | |
1766 | ||
1767 | ||
1768 | /* Remove insns from the queue, before they become "ready" with respect | |
1769 | to FU latency considerations. */ | |
1770 | ||
1771 | static int | |
1772 | early_queue_to_ready (state_t state, struct ready_list *ready) | |
1773 | { | |
1774 | rtx insn; | |
1775 | rtx link; | |
1776 | rtx next_link; | |
1777 | rtx prev_link; | |
1778 | bool move_to_ready; | |
1779 | int cost; | |
1780 | state_t temp_state = alloca (dfa_state_size); | |
1781 | int stalls; | |
1782 | int insns_removed = 0; | |
1783 | ||
1784 | /* | |
1785 | Flag '-fsched-stalled-insns=X' determines the aggressiveness of this | |
1786 | function: | |
1787 | ||
1788 | X == 0: There is no limit on how many queued insns can be removed | |
1789 | prematurely. (flag_sched_stalled_insns = -1). | |
1790 | ||
1791 | X >= 1: Only X queued insns can be removed prematurely in each | |
1792 | invocation. (flag_sched_stalled_insns = X). | |
1793 | ||
1794 | Otherwise: Early queue removal is disabled. | |
1795 | (flag_sched_stalled_insns = 0) | |
1796 | */ | |
1797 | ||
1798 | if (! flag_sched_stalled_insns) | |
1799 | return 0; | |
1800 | ||
1801 | for (stalls = 0; stalls <= max_insn_queue_index; stalls++) | |
1802 | { | |
1803 | if ((link = insn_queue[NEXT_Q_AFTER (q_ptr, stalls)])) | |
1804 | { | |
1805 | if (sched_verbose > 6) | |
1806 | fprintf (sched_dump, ";; look at index %d + %d\n", q_ptr, stalls); | |
1807 | ||
1808 | prev_link = 0; | |
1809 | while (link) | |
1810 | { | |
1811 | next_link = XEXP (link, 1); | |
1812 | insn = XEXP (link, 0); | |
1813 | if (insn && sched_verbose > 6) | |
1814 | print_rtl_single (sched_dump, insn); | |
1815 | ||
1816 | memcpy (temp_state, state, dfa_state_size); | |
1817 | if (recog_memoized (insn) < 0) | |
1818 | /* non-negative to indicate that it's not ready | |
1819 | to avoid infinite Q->R->Q->R... */ | |
1820 | cost = 0; | |
1821 | else | |
1822 | cost = state_transition (temp_state, insn); | |
1823 | ||
1824 | if (sched_verbose >= 6) | |
1825 | fprintf (sched_dump, "transition cost = %d\n", cost); | |
1826 | ||
1827 | move_to_ready = false; | |
1828 | if (cost < 0) | |
1829 | { | |
1830 | move_to_ready = ok_for_early_queue_removal (insn); | |
1831 | if (move_to_ready == true) | |
1832 | { | |
1833 | /* move from Q to R */ | |
1834 | q_size -= 1; | |
1835 | ready_add (ready, insn, false); | |
1836 | ||
1837 | if (prev_link) | |
1838 | XEXP (prev_link, 1) = next_link; | |
1839 | else | |
1840 | insn_queue[NEXT_Q_AFTER (q_ptr, stalls)] = next_link; | |
1841 | ||
1842 | free_INSN_LIST_node (link); | |
1843 | ||
1844 | if (sched_verbose >= 2) | |
1845 | fprintf (sched_dump, ";;\t\tEarly Q-->Ready: insn %s\n", | |
1846 | (*current_sched_info->print_insn) (insn, 0)); | |
1847 | ||
1848 | insns_removed++; | |
1849 | if (insns_removed == flag_sched_stalled_insns) | |
1850 | /* Remove no more than flag_sched_stalled_insns insns | |
1851 | from Q at a time. */ | |
1852 | return insns_removed; | |
1853 | } | |
1854 | } | |
1855 | ||
1856 | if (move_to_ready == false) | |
1857 | prev_link = link; | |
1858 | ||
1859 | link = next_link; | |
1860 | } /* while link */ | |
1861 | } /* if link */ | |
1862 | ||
1863 | } /* for stalls.. */ | |
1864 | ||
1865 | return insns_removed; | |
1866 | } | |
1867 | ||
1868 | ||
1869 | /* Print the ready list for debugging purposes. Callable from debugger. */ | |
1870 | ||
1871 | static void | |
1872 | debug_ready_list (struct ready_list *ready) | |
1873 | { | |
1874 | rtx *p; | |
1875 | int i; | |
1876 | ||
1877 | if (ready->n_ready == 0) | |
1878 | { | |
1879 | fprintf (sched_dump, "\n"); | |
1880 | return; | |
1881 | } | |
1882 | ||
1883 | p = ready_lastpos (ready); | |
1884 | for (i = 0; i < ready->n_ready; i++) | |
1885 | fprintf (sched_dump, " %s", (*current_sched_info->print_insn) (p[i], 0)); | |
1886 | fprintf (sched_dump, "\n"); | |
1887 | } | |
1888 | ||
1889 | /* Search INSN for REG_SAVE_NOTE note pairs for | |
1890 | NOTE_INSN_EHREGION_{BEG,END}; and convert them back into | |
1891 | NOTEs. The REG_SAVE_NOTE note following first one is contains the | |
1892 | saved value for NOTE_BLOCK_NUMBER which is useful for | |
1893 | NOTE_INSN_EH_REGION_{BEG,END} NOTEs. */ | |
1894 | ||
1895 | static void | |
1896 | reemit_notes (rtx insn) | |
1897 | { | |
1898 | rtx note, last = insn; | |
1899 | ||
1900 | for (note = REG_NOTES (insn); note; note = XEXP (note, 1)) | |
1901 | { | |
1902 | if (REG_NOTE_KIND (note) == REG_SAVE_NOTE) | |
1903 | { | |
1904 | enum insn_note note_type = INTVAL (XEXP (note, 0)); | |
1905 | ||
1906 | last = emit_note_before (note_type, last); | |
1907 | remove_note (insn, note); | |
1908 | } | |
1909 | } | |
1910 | } | |
1911 | ||
1912 | /* Move INSN. Reemit notes if needed. Update CFG, if needed. */ | |
1913 | static void | |
1914 | move_insn (rtx insn) | |
1915 | { | |
1916 | rtx last = last_scheduled_insn; | |
1917 | ||
1918 | if (PREV_INSN (insn) != last) | |
1919 | { | |
1920 | basic_block bb; | |
1921 | rtx note; | |
1922 | int jump_p = 0; | |
1923 | ||
1924 | bb = BLOCK_FOR_INSN (insn); | |
1925 | ||
1926 | /* BB_HEAD is either LABEL or NOTE. */ | |
1927 | gcc_assert (BB_HEAD (bb) != insn); | |
1928 | ||
1929 | if (BB_END (bb) == insn) | |
1930 | /* If this is last instruction in BB, move end marker one | |
1931 | instruction up. */ | |
1932 | { | |
1933 | /* Jumps are always placed at the end of basic block. */ | |
1934 | jump_p = control_flow_insn_p (insn); | |
1935 | ||
1936 | gcc_assert (!jump_p | |
1937 | || ((current_sched_info->flags & SCHED_RGN) | |
1938 | && RECOVERY_BLOCK (insn) | |
1939 | && RECOVERY_BLOCK (insn) != EXIT_BLOCK_PTR) | |
1940 | || (current_sched_info->flags & SCHED_EBB)); | |
1941 | ||
1942 | gcc_assert (BLOCK_FOR_INSN (PREV_INSN (insn)) == bb); | |
1943 | ||
1944 | BB_END (bb) = PREV_INSN (insn); | |
1945 | } | |
1946 | ||
1947 | gcc_assert (BB_END (bb) != last); | |
1948 | ||
1949 | if (jump_p) | |
1950 | /* We move the block note along with jump. */ | |
1951 | { | |
1952 | /* NT is needed for assertion below. */ | |
1953 | rtx nt = current_sched_info->next_tail; | |
1954 | ||
1955 | note = NEXT_INSN (insn); | |
1956 | while (NOTE_NOT_BB_P (note) && note != nt) | |
1957 | note = NEXT_INSN (note); | |
1958 | ||
1959 | if (note != nt | |
1960 | && (LABEL_P (note) | |
1961 | || BARRIER_P (note))) | |
1962 | note = NEXT_INSN (note); | |
1963 | ||
1964 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (note)); | |
1965 | } | |
1966 | else | |
1967 | note = insn; | |
1968 | ||
1969 | NEXT_INSN (PREV_INSN (insn)) = NEXT_INSN (note); | |
1970 | PREV_INSN (NEXT_INSN (note)) = PREV_INSN (insn); | |
1971 | ||
1972 | NEXT_INSN (note) = NEXT_INSN (last); | |
1973 | PREV_INSN (NEXT_INSN (last)) = note; | |
1974 | ||
1975 | NEXT_INSN (last) = insn; | |
1976 | PREV_INSN (insn) = last; | |
1977 | ||
1978 | bb = BLOCK_FOR_INSN (last); | |
1979 | ||
1980 | if (jump_p) | |
1981 | { | |
1982 | fix_jump_move (insn); | |
1983 | ||
1984 | if (BLOCK_FOR_INSN (insn) != bb) | |
1985 | move_block_after_check (insn); | |
1986 | ||
1987 | gcc_assert (BB_END (bb) == last); | |
1988 | } | |
1989 | ||
1990 | set_block_for_insn (insn, bb); | |
1991 | ||
1992 | /* Update BB_END, if needed. */ | |
1993 | if (BB_END (bb) == last) | |
1994 | BB_END (bb) = insn; | |
1995 | } | |
1996 | ||
1997 | reemit_notes (insn); | |
1998 | ||
1999 | SCHED_GROUP_P (insn) = 0; | |
2000 | } | |
2001 | ||
2002 | /* The following structure describe an entry of the stack of choices. */ | |
2003 | struct choice_entry | |
2004 | { | |
2005 | /* Ordinal number of the issued insn in the ready queue. */ | |
2006 | int index; | |
2007 | /* The number of the rest insns whose issues we should try. */ | |
2008 | int rest; | |
2009 | /* The number of issued essential insns. */ | |
2010 | int n; | |
2011 | /* State after issuing the insn. */ | |
2012 | state_t state; | |
2013 | }; | |
2014 | ||
2015 | /* The following array is used to implement a stack of choices used in | |
2016 | function max_issue. */ | |
2017 | static struct choice_entry *choice_stack; | |
2018 | ||
2019 | /* The following variable value is number of essential insns issued on | |
2020 | the current cycle. An insn is essential one if it changes the | |
2021 | processors state. */ | |
2022 | static int cycle_issued_insns; | |
2023 | ||
2024 | /* The following variable value is maximal number of tries of issuing | |
2025 | insns for the first cycle multipass insn scheduling. We define | |
2026 | this value as constant*(DFA_LOOKAHEAD**ISSUE_RATE). We would not | |
2027 | need this constraint if all real insns (with non-negative codes) | |
2028 | had reservations because in this case the algorithm complexity is | |
2029 | O(DFA_LOOKAHEAD**ISSUE_RATE). Unfortunately, the dfa descriptions | |
2030 | might be incomplete and such insn might occur. For such | |
2031 | descriptions, the complexity of algorithm (without the constraint) | |
2032 | could achieve DFA_LOOKAHEAD ** N , where N is the queue length. */ | |
2033 | static int max_lookahead_tries; | |
2034 | ||
2035 | /* The following value is value of hook | |
2036 | `first_cycle_multipass_dfa_lookahead' at the last call of | |
2037 | `max_issue'. */ | |
2038 | static int cached_first_cycle_multipass_dfa_lookahead = 0; | |
2039 | ||
2040 | /* The following value is value of `issue_rate' at the last call of | |
2041 | `sched_init'. */ | |
2042 | static int cached_issue_rate = 0; | |
2043 | ||
2044 | /* The following function returns maximal (or close to maximal) number | |
2045 | of insns which can be issued on the same cycle and one of which | |
2046 | insns is insns with the best rank (the first insn in READY). To | |
2047 | make this function tries different samples of ready insns. READY | |
2048 | is current queue `ready'. Global array READY_TRY reflects what | |
2049 | insns are already issued in this try. MAX_POINTS is the sum of points | |
2050 | of all instructions in READY. The function stops immediately, | |
2051 | if it reached the such a solution, that all instruction can be issued. | |
2052 | INDEX will contain index of the best insn in READY. The following | |
2053 | function is used only for first cycle multipass scheduling. */ | |
2054 | static int | |
2055 | max_issue (struct ready_list *ready, int *index, int max_points) | |
2056 | { | |
2057 | int n, i, all, n_ready, best, delay, tries_num, points = -1; | |
2058 | struct choice_entry *top; | |
2059 | rtx insn; | |
2060 | ||
2061 | best = 0; | |
2062 | memcpy (choice_stack->state, curr_state, dfa_state_size); | |
2063 | top = choice_stack; | |
2064 | top->rest = cached_first_cycle_multipass_dfa_lookahead; | |
2065 | top->n = 0; | |
2066 | n_ready = ready->n_ready; | |
2067 | for (all = i = 0; i < n_ready; i++) | |
2068 | if (!ready_try [i]) | |
2069 | all++; | |
2070 | i = 0; | |
2071 | tries_num = 0; | |
2072 | for (;;) | |
2073 | { | |
2074 | if (top->rest == 0 || i >= n_ready) | |
2075 | { | |
2076 | if (top == choice_stack) | |
2077 | break; | |
2078 | if (best < top - choice_stack && ready_try [0]) | |
2079 | { | |
2080 | best = top - choice_stack; | |
2081 | *index = choice_stack [1].index; | |
2082 | points = top->n; | |
2083 | if (top->n == max_points || best == all) | |
2084 | break; | |
2085 | } | |
2086 | i = top->index; | |
2087 | ready_try [i] = 0; | |
2088 | top--; | |
2089 | memcpy (curr_state, top->state, dfa_state_size); | |
2090 | } | |
2091 | else if (!ready_try [i]) | |
2092 | { | |
2093 | tries_num++; | |
2094 | if (tries_num > max_lookahead_tries) | |
2095 | break; | |
2096 | insn = ready_element (ready, i); | |
2097 | delay = state_transition (curr_state, insn); | |
2098 | if (delay < 0) | |
2099 | { | |
2100 | if (state_dead_lock_p (curr_state)) | |
2101 | top->rest = 0; | |
2102 | else | |
2103 | top->rest--; | |
2104 | n = top->n; | |
2105 | if (memcmp (top->state, curr_state, dfa_state_size) != 0) | |
2106 | n += ISSUE_POINTS (insn); | |
2107 | top++; | |
2108 | top->rest = cached_first_cycle_multipass_dfa_lookahead; | |
2109 | top->index = i; | |
2110 | top->n = n; | |
2111 | memcpy (top->state, curr_state, dfa_state_size); | |
2112 | ready_try [i] = 1; | |
2113 | i = -1; | |
2114 | } | |
2115 | } | |
2116 | i++; | |
2117 | } | |
2118 | while (top != choice_stack) | |
2119 | { | |
2120 | ready_try [top->index] = 0; | |
2121 | top--; | |
2122 | } | |
2123 | memcpy (curr_state, choice_stack->state, dfa_state_size); | |
2124 | ||
2125 | if (sched_verbose >= 4) | |
2126 | fprintf (sched_dump, ";;\t\tChoosed insn : %s; points: %d/%d\n", | |
2127 | (*current_sched_info->print_insn) (ready_element (ready, *index), | |
2128 | 0), | |
2129 | points, max_points); | |
2130 | ||
2131 | return best; | |
2132 | } | |
2133 | ||
2134 | /* The following function chooses insn from READY and modifies | |
2135 | *N_READY and READY. The following function is used only for first | |
2136 | cycle multipass scheduling. */ | |
2137 | ||
2138 | static rtx | |
2139 | choose_ready (struct ready_list *ready) | |
2140 | { | |
2141 | int lookahead = 0; | |
2142 | ||
2143 | if (targetm.sched.first_cycle_multipass_dfa_lookahead) | |
2144 | lookahead = targetm.sched.first_cycle_multipass_dfa_lookahead (); | |
2145 | if (lookahead <= 0 || SCHED_GROUP_P (ready_element (ready, 0))) | |
2146 | return ready_remove_first (ready); | |
2147 | else | |
2148 | { | |
2149 | /* Try to choose the better insn. */ | |
2150 | int index = 0, i, n; | |
2151 | rtx insn; | |
2152 | int more_issue, max_points, try_data = 1, try_control = 1; | |
2153 | ||
2154 | if (cached_first_cycle_multipass_dfa_lookahead != lookahead) | |
2155 | { | |
2156 | cached_first_cycle_multipass_dfa_lookahead = lookahead; | |
2157 | max_lookahead_tries = 100; | |
2158 | for (i = 0; i < issue_rate; i++) | |
2159 | max_lookahead_tries *= lookahead; | |
2160 | } | |
2161 | insn = ready_element (ready, 0); | |
2162 | if (INSN_CODE (insn) < 0) | |
2163 | return ready_remove_first (ready); | |
2164 | ||
2165 | if (spec_info | |
2166 | && spec_info->flags & (PREFER_NON_DATA_SPEC | |
2167 | | PREFER_NON_CONTROL_SPEC)) | |
2168 | { | |
2169 | for (i = 0, n = ready->n_ready; i < n; i++) | |
2170 | { | |
2171 | rtx x; | |
2172 | ds_t s; | |
2173 | ||
2174 | x = ready_element (ready, i); | |
2175 | s = TODO_SPEC (x); | |
2176 | ||
2177 | if (spec_info->flags & PREFER_NON_DATA_SPEC | |
2178 | && !(s & DATA_SPEC)) | |
2179 | { | |
2180 | try_data = 0; | |
2181 | if (!(spec_info->flags & PREFER_NON_CONTROL_SPEC) | |
2182 | || !try_control) | |
2183 | break; | |
2184 | } | |
2185 | ||
2186 | if (spec_info->flags & PREFER_NON_CONTROL_SPEC | |
2187 | && !(s & CONTROL_SPEC)) | |
2188 | { | |
2189 | try_control = 0; | |
2190 | if (!(spec_info->flags & PREFER_NON_DATA_SPEC) || !try_data) | |
2191 | break; | |
2192 | } | |
2193 | } | |
2194 | } | |
2195 | ||
2196 | if ((!try_data && (TODO_SPEC (insn) & DATA_SPEC)) | |
2197 | || (!try_control && (TODO_SPEC (insn) & CONTROL_SPEC)) | |
2198 | || (targetm.sched.first_cycle_multipass_dfa_lookahead_guard_spec | |
2199 | && !targetm.sched.first_cycle_multipass_dfa_lookahead_guard_spec | |
2200 | (insn))) | |
2201 | /* Discard speculative instruction that stands first in the ready | |
2202 | list. */ | |
2203 | { | |
2204 | change_queue_index (insn, 1); | |
2205 | return 0; | |
2206 | } | |
2207 | ||
2208 | max_points = ISSUE_POINTS (insn); | |
2209 | more_issue = issue_rate - cycle_issued_insns - 1; | |
2210 | ||
2211 | for (i = 1; i < ready->n_ready; i++) | |
2212 | { | |
2213 | insn = ready_element (ready, i); | |
2214 | ready_try [i] | |
2215 | = (INSN_CODE (insn) < 0 | |
2216 | || (!try_data && (TODO_SPEC (insn) & DATA_SPEC)) | |
2217 | || (!try_control && (TODO_SPEC (insn) & CONTROL_SPEC)) | |
2218 | || (targetm.sched.first_cycle_multipass_dfa_lookahead_guard | |
2219 | && !targetm.sched.first_cycle_multipass_dfa_lookahead_guard | |
2220 | (insn))); | |
2221 | ||
2222 | if (!ready_try [i] && more_issue-- > 0) | |
2223 | max_points += ISSUE_POINTS (insn); | |
2224 | } | |
2225 | ||
2226 | if (max_issue (ready, &index, max_points) == 0) | |
2227 | return ready_remove_first (ready); | |
2228 | else | |
2229 | return ready_remove (ready, index); | |
2230 | } | |
2231 | } | |
2232 | ||
2233 | /* Use forward list scheduling to rearrange insns of block pointed to by | |
2234 | TARGET_BB, possibly bringing insns from subsequent blocks in the same | |
2235 | region. */ | |
2236 | ||
2237 | void | |
2238 | schedule_block (basic_block *target_bb, int rgn_n_insns1) | |
2239 | { | |
2240 | struct ready_list ready; | |
2241 | int i, first_cycle_insn_p; | |
2242 | int can_issue_more; | |
2243 | state_t temp_state = NULL; /* It is used for multipass scheduling. */ | |
2244 | int sort_p, advance, start_clock_var; | |
2245 | ||
2246 | /* Head/tail info for this block. */ | |
2247 | rtx prev_head = current_sched_info->prev_head; | |
2248 | rtx next_tail = current_sched_info->next_tail; | |
2249 | rtx head = NEXT_INSN (prev_head); | |
2250 | rtx tail = PREV_INSN (next_tail); | |
2251 | ||
2252 | /* We used to have code to avoid getting parameters moved from hard | |
2253 | argument registers into pseudos. | |
2254 | ||
2255 | However, it was removed when it proved to be of marginal benefit | |
2256 | and caused problems because schedule_block and compute_forward_dependences | |
2257 | had different notions of what the "head" insn was. */ | |
2258 | ||
2259 | gcc_assert (head != tail || INSN_P (head)); | |
2260 | ||
2261 | added_recovery_block_p = false; | |
2262 | ||
2263 | /* Debug info. */ | |
2264 | if (sched_verbose) | |
2265 | dump_new_block_header (0, *target_bb, head, tail); | |
2266 | ||
2267 | state_reset (curr_state); | |
2268 | ||
2269 | /* Allocate the ready list. */ | |
2270 | readyp = &ready; | |
2271 | ready.vec = NULL; | |
2272 | ready_try = NULL; | |
2273 | choice_stack = NULL; | |
2274 | ||
2275 | rgn_n_insns = -1; | |
2276 | extend_ready (rgn_n_insns1 + 1); | |
2277 | ||
2278 | ready.first = ready.veclen - 1; | |
2279 | ready.n_ready = 0; | |
2280 | ||
2281 | /* It is used for first cycle multipass scheduling. */ | |
2282 | temp_state = alloca (dfa_state_size); | |
2283 | ||
2284 | if (targetm.sched.md_init) | |
2285 | targetm.sched.md_init (sched_dump, sched_verbose, ready.veclen); | |
2286 | ||
2287 | /* We start inserting insns after PREV_HEAD. */ | |
2288 | last_scheduled_insn = prev_head; | |
2289 | ||
2290 | gcc_assert (NOTE_P (last_scheduled_insn) | |
2291 | && BLOCK_FOR_INSN (last_scheduled_insn) == *target_bb); | |
2292 | ||
2293 | /* Initialize INSN_QUEUE. Q_SIZE is the total number of insns in the | |
2294 | queue. */ | |
2295 | q_ptr = 0; | |
2296 | q_size = 0; | |
2297 | ||
2298 | insn_queue = alloca ((max_insn_queue_index + 1) * sizeof (rtx)); | |
2299 | memset (insn_queue, 0, (max_insn_queue_index + 1) * sizeof (rtx)); | |
2300 | ||
2301 | /* Start just before the beginning of time. */ | |
2302 | clock_var = -1; | |
2303 | ||
2304 | /* We need queue and ready lists and clock_var be initialized | |
2305 | in try_ready () (which is called through init_ready_list ()). */ | |
2306 | (*current_sched_info->init_ready_list) (); | |
2307 | ||
2308 | /* The algorithm is O(n^2) in the number of ready insns at any given | |
2309 | time in the worst case. Before reload we are more likely to have | |
2310 | big lists so truncate them to a reasonable size. */ | |
2311 | if (!reload_completed && ready.n_ready > MAX_SCHED_READY_INSNS) | |
2312 | { | |
2313 | ready_sort (&ready); | |
2314 | ||
2315 | /* Find first free-standing insn past MAX_SCHED_READY_INSNS. */ | |
2316 | for (i = MAX_SCHED_READY_INSNS; i < ready.n_ready; i++) | |
2317 | if (!SCHED_GROUP_P (ready_element (&ready, i))) | |
2318 | break; | |
2319 | ||
2320 | if (sched_verbose >= 2) | |
2321 | { | |
2322 | fprintf (sched_dump, | |
2323 | ";;\t\tReady list on entry: %d insns\n", ready.n_ready); | |
2324 | fprintf (sched_dump, | |
2325 | ";;\t\t before reload => truncated to %d insns\n", i); | |
2326 | } | |
2327 | ||
2328 | /* Delay all insns past it for 1 cycle. */ | |
2329 | while (i < ready.n_ready) | |
2330 | queue_insn (ready_remove (&ready, i), 1); | |
2331 | } | |
2332 | ||
2333 | /* Now we can restore basic block notes and maintain precise cfg. */ | |
2334 | restore_bb_notes (*target_bb); | |
2335 | ||
2336 | last_clock_var = -1; | |
2337 | ||
2338 | advance = 0; | |
2339 | ||
2340 | sort_p = TRUE; | |
2341 | /* Loop until all the insns in BB are scheduled. */ | |
2342 | while ((*current_sched_info->schedule_more_p) ()) | |
2343 | { | |
2344 | do | |
2345 | { | |
2346 | start_clock_var = clock_var; | |
2347 | ||
2348 | clock_var++; | |
2349 | ||
2350 | advance_one_cycle (); | |
2351 | ||
2352 | /* Add to the ready list all pending insns that can be issued now. | |
2353 | If there are no ready insns, increment clock until one | |
2354 | is ready and add all pending insns at that point to the ready | |
2355 | list. */ | |
2356 | queue_to_ready (&ready); | |
2357 | ||
2358 | gcc_assert (ready.n_ready); | |
2359 | ||
2360 | if (sched_verbose >= 2) | |
2361 | { | |
2362 | fprintf (sched_dump, ";;\t\tReady list after queue_to_ready: "); | |
2363 | debug_ready_list (&ready); | |
2364 | } | |
2365 | advance -= clock_var - start_clock_var; | |
2366 | } | |
2367 | while (advance > 0); | |
2368 | ||
2369 | if (sort_p) | |
2370 | { | |
2371 | /* Sort the ready list based on priority. */ | |
2372 | ready_sort (&ready); | |
2373 | ||
2374 | if (sched_verbose >= 2) | |
2375 | { | |
2376 | fprintf (sched_dump, ";;\t\tReady list after ready_sort: "); | |
2377 | debug_ready_list (&ready); | |
2378 | } | |
2379 | } | |
2380 | ||
2381 | /* Allow the target to reorder the list, typically for | |
2382 | better instruction bundling. */ | |
2383 | if (sort_p && targetm.sched.reorder | |
2384 | && (ready.n_ready == 0 | |
2385 | || !SCHED_GROUP_P (ready_element (&ready, 0)))) | |
2386 | can_issue_more = | |
2387 | targetm.sched.reorder (sched_dump, sched_verbose, | |
2388 | ready_lastpos (&ready), | |
2389 | &ready.n_ready, clock_var); | |
2390 | else | |
2391 | can_issue_more = issue_rate; | |
2392 | ||
2393 | first_cycle_insn_p = 1; | |
2394 | cycle_issued_insns = 0; | |
2395 | for (;;) | |
2396 | { | |
2397 | rtx insn; | |
2398 | int cost; | |
2399 | bool asm_p = false; | |
2400 | ||
2401 | if (sched_verbose >= 2) | |
2402 | { | |
2403 | fprintf (sched_dump, ";;\tReady list (t = %3d): ", | |
2404 | clock_var); | |
2405 | debug_ready_list (&ready); | |
2406 | } | |
2407 | ||
2408 | if (ready.n_ready == 0 | |
2409 | && can_issue_more | |
2410 | && reload_completed) | |
2411 | { | |
2412 | /* Allow scheduling insns directly from the queue in case | |
2413 | there's nothing better to do (ready list is empty) but | |
2414 | there are still vacant dispatch slots in the current cycle. */ | |
2415 | if (sched_verbose >= 6) | |
2416 | fprintf(sched_dump,";;\t\tSecond chance\n"); | |
2417 | memcpy (temp_state, curr_state, dfa_state_size); | |
2418 | if (early_queue_to_ready (temp_state, &ready)) | |
2419 | ready_sort (&ready); | |
2420 | } | |
2421 | ||
2422 | if (ready.n_ready == 0 || !can_issue_more | |
2423 | || state_dead_lock_p (curr_state) | |
2424 | || !(*current_sched_info->schedule_more_p) ()) | |
2425 | break; | |
2426 | ||
2427 | /* Select and remove the insn from the ready list. */ | |
2428 | if (sort_p) | |
2429 | { | |
2430 | insn = choose_ready (&ready); | |
2431 | if (!insn) | |
2432 | continue; | |
2433 | } | |
2434 | else | |
2435 | insn = ready_remove_first (&ready); | |
2436 | ||
2437 | if (targetm.sched.dfa_new_cycle | |
2438 | && targetm.sched.dfa_new_cycle (sched_dump, sched_verbose, | |
2439 | insn, last_clock_var, | |
2440 | clock_var, &sort_p)) | |
2441 | /* SORT_P is used by the target to override sorting | |
2442 | of the ready list. This is needed when the target | |
2443 | has modified its internal structures expecting that | |
2444 | the insn will be issued next. As we need the insn | |
2445 | to have the highest priority (so it will be returned by | |
2446 | the ready_remove_first call above), we invoke | |
2447 | ready_add (&ready, insn, true). | |
2448 | But, still, there is one issue: INSN can be later | |
2449 | discarded by scheduler's front end through | |
2450 | current_sched_info->can_schedule_ready_p, hence, won't | |
2451 | be issued next. */ | |
2452 | { | |
2453 | ready_add (&ready, insn, true); | |
2454 | break; | |
2455 | } | |
2456 | ||
2457 | sort_p = TRUE; | |
2458 | memcpy (temp_state, curr_state, dfa_state_size); | |
2459 | if (recog_memoized (insn) < 0) | |
2460 | { | |
2461 | asm_p = (GET_CODE (PATTERN (insn)) == ASM_INPUT | |
2462 | || asm_noperands (PATTERN (insn)) >= 0); | |
2463 | if (!first_cycle_insn_p && asm_p) | |
2464 | /* This is asm insn which is tryed to be issued on the | |
2465 | cycle not first. Issue it on the next cycle. */ | |
2466 | cost = 1; | |
2467 | else | |
2468 | /* A USE insn, or something else we don't need to | |
2469 | understand. We can't pass these directly to | |
2470 | state_transition because it will trigger a | |
2471 | fatal error for unrecognizable insns. */ | |
2472 | cost = 0; | |
2473 | } | |
2474 | else | |
2475 | { | |
2476 | cost = state_transition (temp_state, insn); | |
2477 | if (cost < 0) | |
2478 | cost = 0; | |
2479 | else if (cost == 0) | |
2480 | cost = 1; | |
2481 | } | |
2482 | ||
2483 | if (cost >= 1) | |
2484 | { | |
2485 | queue_insn (insn, cost); | |
2486 | if (SCHED_GROUP_P (insn)) | |
2487 | { | |
2488 | advance = cost; | |
2489 | break; | |
2490 | } | |
2491 | ||
2492 | continue; | |
2493 | } | |
2494 | ||
2495 | if (current_sched_info->can_schedule_ready_p | |
2496 | && ! (*current_sched_info->can_schedule_ready_p) (insn)) | |
2497 | /* We normally get here only if we don't want to move | |
2498 | insn from the split block. */ | |
2499 | { | |
2500 | TODO_SPEC (insn) = (TODO_SPEC (insn) & ~SPECULATIVE) | HARD_DEP; | |
2501 | continue; | |
2502 | } | |
2503 | ||
2504 | /* DECISION is made. */ | |
2505 | ||
2506 | if (TODO_SPEC (insn) & SPECULATIVE) | |
2507 | generate_recovery_code (insn); | |
2508 | ||
2509 | if (control_flow_insn_p (last_scheduled_insn) | |
2510 | /* This is used to to switch basic blocks by request | |
2511 | from scheduler front-end (actually, sched-ebb.c only). | |
2512 | This is used to process blocks with single fallthru | |
2513 | edge. If succeeding block has jump, it [jump] will try | |
2514 | move at the end of current bb, thus corrupting CFG. */ | |
2515 | || current_sched_info->advance_target_bb (*target_bb, insn)) | |
2516 | { | |
2517 | *target_bb = current_sched_info->advance_target_bb | |
2518 | (*target_bb, 0); | |
2519 | ||
2520 | if (sched_verbose) | |
2521 | { | |
2522 | rtx x; | |
2523 | ||
2524 | x = next_real_insn (last_scheduled_insn); | |
2525 | gcc_assert (x); | |
2526 | dump_new_block_header (1, *target_bb, x, tail); | |
2527 | } | |
2528 | ||
2529 | last_scheduled_insn = bb_note (*target_bb); | |
2530 | } | |
2531 | ||
2532 | /* Update counters, etc in the scheduler's front end. */ | |
2533 | (*current_sched_info->begin_schedule_ready) (insn, | |
2534 | last_scheduled_insn); | |
2535 | ||
2536 | move_insn (insn); | |
2537 | last_scheduled_insn = insn; | |
2538 | ||
2539 | if (memcmp (curr_state, temp_state, dfa_state_size) != 0) | |
2540 | { | |
2541 | cycle_issued_insns++; | |
2542 | memcpy (curr_state, temp_state, dfa_state_size); | |
2543 | } | |
2544 | ||
2545 | if (targetm.sched.variable_issue) | |
2546 | can_issue_more = | |
2547 | targetm.sched.variable_issue (sched_dump, sched_verbose, | |
2548 | insn, can_issue_more); | |
2549 | /* A naked CLOBBER or USE generates no instruction, so do | |
2550 | not count them against the issue rate. */ | |
2551 | else if (GET_CODE (PATTERN (insn)) != USE | |
2552 | && GET_CODE (PATTERN (insn)) != CLOBBER) | |
2553 | can_issue_more--; | |
2554 | ||
2555 | advance = schedule_insn (insn); | |
2556 | ||
2557 | /* After issuing an asm insn we should start a new cycle. */ | |
2558 | if (advance == 0 && asm_p) | |
2559 | advance = 1; | |
2560 | if (advance != 0) | |
2561 | break; | |
2562 | ||
2563 | first_cycle_insn_p = 0; | |
2564 | ||
2565 | /* Sort the ready list based on priority. This must be | |
2566 | redone here, as schedule_insn may have readied additional | |
2567 | insns that will not be sorted correctly. */ | |
2568 | if (ready.n_ready > 0) | |
2569 | ready_sort (&ready); | |
2570 | ||
2571 | if (targetm.sched.reorder2 | |
2572 | && (ready.n_ready == 0 | |
2573 | || !SCHED_GROUP_P (ready_element (&ready, 0)))) | |
2574 | { | |
2575 | can_issue_more = | |
2576 | targetm.sched.reorder2 (sched_dump, sched_verbose, | |
2577 | ready.n_ready | |
2578 | ? ready_lastpos (&ready) : NULL, | |
2579 | &ready.n_ready, clock_var); | |
2580 | } | |
2581 | } | |
2582 | } | |
2583 | ||
2584 | /* Debug info. */ | |
2585 | if (sched_verbose) | |
2586 | { | |
2587 | fprintf (sched_dump, ";;\tReady list (final): "); | |
2588 | debug_ready_list (&ready); | |
2589 | } | |
2590 | ||
2591 | if (current_sched_info->queue_must_finish_empty) | |
2592 | /* Sanity check -- queue must be empty now. Meaningless if region has | |
2593 | multiple bbs. */ | |
2594 | gcc_assert (!q_size && !ready.n_ready); | |
2595 | else | |
2596 | { | |
2597 | /* We must maintain QUEUE_INDEX between blocks in region. */ | |
2598 | for (i = ready.n_ready - 1; i >= 0; i--) | |
2599 | { | |
2600 | rtx x; | |
2601 | ||
2602 | x = ready_element (&ready, i); | |
2603 | QUEUE_INDEX (x) = QUEUE_NOWHERE; | |
2604 | TODO_SPEC (x) = (TODO_SPEC (x) & ~SPECULATIVE) | HARD_DEP; | |
2605 | } | |
2606 | ||
2607 | if (q_size) | |
2608 | for (i = 0; i <= max_insn_queue_index; i++) | |
2609 | { | |
2610 | rtx link; | |
2611 | for (link = insn_queue[i]; link; link = XEXP (link, 1)) | |
2612 | { | |
2613 | rtx x; | |
2614 | ||
2615 | x = XEXP (link, 0); | |
2616 | QUEUE_INDEX (x) = QUEUE_NOWHERE; | |
2617 | TODO_SPEC (x) = (TODO_SPEC (x) & ~SPECULATIVE) | HARD_DEP; | |
2618 | } | |
2619 | free_INSN_LIST_list (&insn_queue[i]); | |
2620 | } | |
2621 | } | |
2622 | ||
2623 | if (!current_sched_info->queue_must_finish_empty | |
2624 | || added_recovery_block_p) | |
2625 | { | |
2626 | /* INSN_TICK (minimum clock tick at which the insn becomes | |
2627 | ready) may be not correct for the insn in the subsequent | |
2628 | blocks of the region. We should use a correct value of | |
2629 | `clock_var' or modify INSN_TICK. It is better to keep | |
2630 | clock_var value equal to 0 at the start of a basic block. | |
2631 | Therefore we modify INSN_TICK here. */ | |
2632 | fix_inter_tick (NEXT_INSN (prev_head), last_scheduled_insn); | |
2633 | } | |
2634 | ||
2635 | #ifdef ENABLE_CHECKING | |
2636 | /* After the reload the ia64 backend doesn't maintain BB_END, so | |
2637 | if we want to check anything, better do it now. | |
2638 | And it already clobbered previously scheduled code. */ | |
2639 | if (reload_completed) | |
2640 | check_cfg (BB_HEAD (BLOCK_FOR_INSN (prev_head)), 0); | |
2641 | #endif | |
2642 | ||
2643 | if (targetm.sched.md_finish) | |
2644 | targetm.sched.md_finish (sched_dump, sched_verbose); | |
2645 | ||
2646 | /* Update head/tail boundaries. */ | |
2647 | head = NEXT_INSN (prev_head); | |
2648 | tail = last_scheduled_insn; | |
2649 | ||
2650 | /* Restore-other-notes: NOTE_LIST is the end of a chain of notes | |
2651 | previously found among the insns. Insert them at the beginning | |
2652 | of the insns. */ | |
2653 | if (note_list != 0) | |
2654 | { | |
2655 | basic_block head_bb = BLOCK_FOR_INSN (head); | |
2656 | rtx note_head = note_list; | |
2657 | ||
2658 | while (PREV_INSN (note_head)) | |
2659 | { | |
2660 | set_block_for_insn (note_head, head_bb); | |
2661 | note_head = PREV_INSN (note_head); | |
2662 | } | |
2663 | /* In the above cycle we've missed this note: */ | |
2664 | set_block_for_insn (note_head, head_bb); | |
2665 | ||
2666 | PREV_INSN (note_head) = PREV_INSN (head); | |
2667 | NEXT_INSN (PREV_INSN (head)) = note_head; | |
2668 | PREV_INSN (head) = note_list; | |
2669 | NEXT_INSN (note_list) = head; | |
2670 | head = note_head; | |
2671 | } | |
2672 | ||
2673 | /* Debugging. */ | |
2674 | if (sched_verbose) | |
2675 | { | |
2676 | fprintf (sched_dump, ";; total time = %d\n;; new head = %d\n", | |
2677 | clock_var, INSN_UID (head)); | |
2678 | fprintf (sched_dump, ";; new tail = %d\n\n", | |
2679 | INSN_UID (tail)); | |
2680 | } | |
2681 | ||
2682 | current_sched_info->head = head; | |
2683 | current_sched_info->tail = tail; | |
2684 | ||
2685 | free (ready.vec); | |
2686 | ||
2687 | free (ready_try); | |
2688 | for (i = 0; i <= rgn_n_insns; i++) | |
2689 | free (choice_stack [i].state); | |
2690 | free (choice_stack); | |
2691 | } | |
2692 | \f | |
2693 | /* Set_priorities: compute priority of each insn in the block. */ | |
2694 | ||
2695 | int | |
2696 | set_priorities (rtx head, rtx tail) | |
2697 | { | |
2698 | rtx insn; | |
2699 | int n_insn; | |
2700 | int sched_max_insns_priority = | |
2701 | current_sched_info->sched_max_insns_priority; | |
2702 | rtx prev_head; | |
2703 | ||
2704 | if (head == tail && (! INSN_P (head))) | |
2705 | return 0; | |
2706 | ||
2707 | n_insn = 0; | |
2708 | ||
2709 | prev_head = PREV_INSN (head); | |
2710 | for (insn = tail; insn != prev_head; insn = PREV_INSN (insn)) | |
2711 | { | |
2712 | if (!INSN_P (insn)) | |
2713 | continue; | |
2714 | ||
2715 | n_insn++; | |
2716 | (void) priority (insn); | |
2717 | ||
2718 | if (INSN_PRIORITY_KNOWN (insn)) | |
2719 | sched_max_insns_priority = | |
2720 | MAX (sched_max_insns_priority, INSN_PRIORITY (insn)); | |
2721 | } | |
2722 | ||
2723 | current_sched_info->sched_max_insns_priority = sched_max_insns_priority; | |
2724 | ||
2725 | return n_insn; | |
2726 | } | |
2727 | ||
2728 | /* Next LUID to assign to an instruction. */ | |
2729 | static int luid; | |
2730 | ||
2731 | /* Initialize some global state for the scheduler. */ | |
2732 | ||
2733 | void | |
2734 | sched_init (void) | |
2735 | { | |
2736 | basic_block b; | |
2737 | rtx insn; | |
2738 | int i; | |
2739 | ||
2740 | /* Switch to working copy of sched_info. */ | |
2741 | memcpy (¤t_sched_info_var, current_sched_info, | |
2742 | sizeof (current_sched_info_var)); | |
2743 | current_sched_info = ¤t_sched_info_var; | |
2744 | ||
2745 | /* Disable speculative loads in their presence if cc0 defined. */ | |
2746 | #ifdef HAVE_cc0 | |
2747 | flag_schedule_speculative_load = 0; | |
2748 | #endif | |
2749 | ||
2750 | /* Set dump and sched_verbose for the desired debugging output. If no | |
2751 | dump-file was specified, but -fsched-verbose=N (any N), print to stderr. | |
2752 | For -fsched-verbose=N, N>=10, print everything to stderr. */ | |
2753 | sched_verbose = sched_verbose_param; | |
2754 | if (sched_verbose_param == 0 && dump_file) | |
2755 | sched_verbose = 1; | |
2756 | sched_dump = ((sched_verbose_param >= 10 || !dump_file) | |
2757 | ? stderr : dump_file); | |
2758 | ||
2759 | /* Initialize SPEC_INFO. */ | |
2760 | if (targetm.sched.set_sched_flags) | |
2761 | { | |
2762 | spec_info = &spec_info_var; | |
2763 | targetm.sched.set_sched_flags (spec_info); | |
2764 | if (current_sched_info->flags & DO_SPECULATION) | |
2765 | spec_info->weakness_cutoff = | |
2766 | (PARAM_VALUE (PARAM_SCHED_SPEC_PROB_CUTOFF) * MAX_DEP_WEAK) / 100; | |
2767 | else | |
2768 | /* So we won't read anything accidently. */ | |
2769 | spec_info = 0; | |
2770 | #ifdef ENABLE_CHECKING | |
2771 | check_sched_flags (); | |
2772 | #endif | |
2773 | } | |
2774 | else | |
2775 | /* So we won't read anything accidently. */ | |
2776 | spec_info = 0; | |
2777 | ||
2778 | /* Initialize issue_rate. */ | |
2779 | if (targetm.sched.issue_rate) | |
2780 | issue_rate = targetm.sched.issue_rate (); | |
2781 | else | |
2782 | issue_rate = 1; | |
2783 | ||
2784 | if (cached_issue_rate != issue_rate) | |
2785 | { | |
2786 | cached_issue_rate = issue_rate; | |
2787 | /* To invalidate max_lookahead_tries: */ | |
2788 | cached_first_cycle_multipass_dfa_lookahead = 0; | |
2789 | } | |
2790 | ||
2791 | old_max_uid = 0; | |
2792 | h_i_d = 0; | |
2793 | extend_h_i_d (); | |
2794 | ||
2795 | for (i = 0; i < old_max_uid; i++) | |
2796 | { | |
2797 | h_i_d[i].cost = -1; | |
2798 | h_i_d[i].todo_spec = HARD_DEP; | |
2799 | h_i_d[i].queue_index = QUEUE_NOWHERE; | |
2800 | h_i_d[i].tick = INVALID_TICK; | |
2801 | h_i_d[i].inter_tick = INVALID_TICK; | |
2802 | } | |
2803 | ||
2804 | if (targetm.sched.init_dfa_pre_cycle_insn) | |
2805 | targetm.sched.init_dfa_pre_cycle_insn (); | |
2806 | ||
2807 | if (targetm.sched.init_dfa_post_cycle_insn) | |
2808 | targetm.sched.init_dfa_post_cycle_insn (); | |
2809 | ||
2810 | dfa_start (); | |
2811 | dfa_state_size = state_size (); | |
2812 | curr_state = xmalloc (dfa_state_size); | |
2813 | ||
2814 | h_i_d[0].luid = 0; | |
2815 | luid = 1; | |
2816 | FOR_EACH_BB (b) | |
2817 | for (insn = BB_HEAD (b); ; insn = NEXT_INSN (insn)) | |
2818 | { | |
2819 | INSN_LUID (insn) = luid; | |
2820 | ||
2821 | /* Increment the next luid, unless this is a note. We don't | |
2822 | really need separate IDs for notes and we don't want to | |
2823 | schedule differently depending on whether or not there are | |
2824 | line-number notes, i.e., depending on whether or not we're | |
2825 | generating debugging information. */ | |
2826 | if (!NOTE_P (insn)) | |
2827 | ++luid; | |
2828 | ||
2829 | if (insn == BB_END (b)) | |
2830 | break; | |
2831 | } | |
2832 | ||
2833 | init_dependency_caches (luid); | |
2834 | ||
2835 | init_alias_analysis (); | |
2836 | ||
2837 | line_note_head = 0; | |
2838 | old_last_basic_block = 0; | |
2839 | glat_start = 0; | |
2840 | glat_end = 0; | |
2841 | extend_bb (0); | |
2842 | ||
2843 | if (current_sched_info->flags & USE_GLAT) | |
2844 | init_glat (); | |
2845 | ||
2846 | /* Compute INSN_REG_WEIGHT for all blocks. We must do this before | |
2847 | removing death notes. */ | |
2848 | FOR_EACH_BB_REVERSE (b) | |
2849 | find_insn_reg_weight (b); | |
2850 | ||
2851 | if (targetm.sched.md_init_global) | |
2852 | targetm.sched.md_init_global (sched_dump, sched_verbose, old_max_uid); | |
2853 | ||
2854 | nr_begin_data = nr_begin_control = nr_be_in_data = nr_be_in_control = 0; | |
2855 | before_recovery = 0; | |
2856 | ||
2857 | #ifdef ENABLE_CHECKING | |
2858 | /* This is used preferably for finding bugs in check_cfg () itself. */ | |
2859 | check_cfg (0, 0); | |
2860 | #endif | |
2861 | } | |
2862 | ||
2863 | /* Free global data used during insn scheduling. */ | |
2864 | ||
2865 | void | |
2866 | sched_finish (void) | |
2867 | { | |
2868 | free (h_i_d); | |
2869 | free (curr_state); | |
2870 | dfa_finish (); | |
2871 | free_dependency_caches (); | |
2872 | end_alias_analysis (); | |
2873 | free (line_note_head); | |
2874 | free_glat (); | |
2875 | ||
2876 | if (targetm.sched.md_finish_global) | |
2877 | targetm.sched.md_finish_global (sched_dump, sched_verbose); | |
2878 | ||
2879 | if (spec_info && spec_info->dump) | |
2880 | { | |
2881 | char c = reload_completed ? 'a' : 'b'; | |
2882 | ||
2883 | fprintf (spec_info->dump, | |
2884 | ";; %s:\n", current_function_name ()); | |
2885 | ||
2886 | fprintf (spec_info->dump, | |
2887 | ";; Procedure %cr-begin-data-spec motions == %d\n", | |
2888 | c, nr_begin_data); | |
2889 | fprintf (spec_info->dump, | |
2890 | ";; Procedure %cr-be-in-data-spec motions == %d\n", | |
2891 | c, nr_be_in_data); | |
2892 | fprintf (spec_info->dump, | |
2893 | ";; Procedure %cr-begin-control-spec motions == %d\n", | |
2894 | c, nr_begin_control); | |
2895 | fprintf (spec_info->dump, | |
2896 | ";; Procedure %cr-be-in-control-spec motions == %d\n", | |
2897 | c, nr_be_in_control); | |
2898 | } | |
2899 | ||
2900 | #ifdef ENABLE_CHECKING | |
2901 | /* After reload ia64 backend clobbers CFG, so can't check anything. */ | |
2902 | if (!reload_completed) | |
2903 | check_cfg (0, 0); | |
2904 | #endif | |
2905 | ||
2906 | current_sched_info = NULL; | |
2907 | } | |
2908 | ||
2909 | /* Fix INSN_TICKs of the instructions in the current block as well as | |
2910 | INSN_TICKs of their dependents. | |
2911 | HEAD and TAIL are the begin and the end of the current scheduled block. */ | |
2912 | static void | |
2913 | fix_inter_tick (rtx head, rtx tail) | |
2914 | { | |
2915 | /* Set of instructions with corrected INSN_TICK. */ | |
2916 | bitmap_head processed; | |
2917 | int next_clock = clock_var + 1; | |
2918 | ||
2919 | bitmap_initialize (&processed, 0); | |
2920 | ||
2921 | /* Iterates over scheduled instructions and fix their INSN_TICKs and | |
2922 | INSN_TICKs of dependent instructions, so that INSN_TICKs are consistent | |
2923 | across different blocks. */ | |
2924 | for (tail = NEXT_INSN (tail); head != tail; head = NEXT_INSN (head)) | |
2925 | { | |
2926 | if (INSN_P (head)) | |
2927 | { | |
2928 | int tick; | |
2929 | rtx link; | |
2930 | ||
2931 | tick = INSN_TICK (head); | |
2932 | gcc_assert (tick >= MIN_TICK); | |
2933 | ||
2934 | /* Fix INSN_TICK of instruction from just scheduled block. */ | |
2935 | if (!bitmap_bit_p (&processed, INSN_LUID (head))) | |
2936 | { | |
2937 | bitmap_set_bit (&processed, INSN_LUID (head)); | |
2938 | tick -= next_clock; | |
2939 | ||
2940 | if (tick < MIN_TICK) | |
2941 | tick = MIN_TICK; | |
2942 | ||
2943 | INSN_TICK (head) = tick; | |
2944 | } | |
2945 | ||
2946 | for (link = INSN_DEPEND (head); link; link = XEXP (link, 1)) | |
2947 | { | |
2948 | rtx next; | |
2949 | ||
2950 | next = XEXP (link, 0); | |
2951 | tick = INSN_TICK (next); | |
2952 | ||
2953 | if (tick != INVALID_TICK | |
2954 | /* If NEXT has its INSN_TICK calculated, fix it. | |
2955 | If not - it will be properly calculated from | |
2956 | scratch later in fix_tick_ready. */ | |
2957 | && !bitmap_bit_p (&processed, INSN_LUID (next))) | |
2958 | { | |
2959 | bitmap_set_bit (&processed, INSN_LUID (next)); | |
2960 | tick -= next_clock; | |
2961 | ||
2962 | if (tick < MIN_TICK) | |
2963 | tick = MIN_TICK; | |
2964 | ||
2965 | if (tick > INTER_TICK (next)) | |
2966 | INTER_TICK (next) = tick; | |
2967 | else | |
2968 | tick = INTER_TICK (next); | |
2969 | ||
2970 | INSN_TICK (next) = tick; | |
2971 | } | |
2972 | } | |
2973 | } | |
2974 | } | |
2975 | bitmap_clear (&processed); | |
2976 | } | |
2977 | ||
2978 | /* Check if NEXT is ready to be added to the ready or queue list. | |
2979 | If "yes", add it to the proper list. | |
2980 | Returns: | |
2981 | -1 - is not ready yet, | |
2982 | 0 - added to the ready list, | |
2983 | 0 < N - queued for N cycles. */ | |
2984 | int | |
2985 | try_ready (rtx next) | |
2986 | { | |
2987 | ds_t old_ts, *ts; | |
2988 | rtx link; | |
2989 | ||
2990 | ts = &TODO_SPEC (next); | |
2991 | old_ts = *ts; | |
2992 | ||
2993 | gcc_assert (!(old_ts & ~(SPECULATIVE | HARD_DEP)) | |
2994 | && ((old_ts & HARD_DEP) | |
2995 | || (old_ts & SPECULATIVE))); | |
2996 | ||
2997 | if (!(current_sched_info->flags & DO_SPECULATION)) | |
2998 | { | |
2999 | if (!LOG_LINKS (next)) | |
3000 | *ts &= ~HARD_DEP; | |
3001 | } | |
3002 | else | |
3003 | { | |
3004 | *ts &= ~SPECULATIVE & ~HARD_DEP; | |
3005 | ||
3006 | link = LOG_LINKS (next); | |
3007 | if (link) | |
3008 | { | |
3009 | /* LOG_LINKS are maintained sorted. | |
3010 | So if DEP_STATUS of the first dep is SPECULATIVE, | |
3011 | than all other deps are speculative too. */ | |
3012 | if (DEP_STATUS (link) & SPECULATIVE) | |
3013 | { | |
3014 | /* Now we've got NEXT with speculative deps only. | |
3015 | 1. Look at the deps to see what we have to do. | |
3016 | 2. Check if we can do 'todo'. */ | |
3017 | *ts = DEP_STATUS (link) & SPECULATIVE; | |
3018 | while ((link = XEXP (link, 1))) | |
3019 | *ts = ds_merge (*ts, DEP_STATUS (link) & SPECULATIVE); | |
3020 | ||
3021 | if (dep_weak (*ts) < spec_info->weakness_cutoff) | |
3022 | /* Too few points. */ | |
3023 | *ts = (*ts & ~SPECULATIVE) | HARD_DEP; | |
3024 | } | |
3025 | else | |
3026 | *ts |= HARD_DEP; | |
3027 | } | |
3028 | } | |
3029 | ||
3030 | if (*ts & HARD_DEP) | |
3031 | gcc_assert (*ts == old_ts | |
3032 | && QUEUE_INDEX (next) == QUEUE_NOWHERE); | |
3033 | else if (current_sched_info->new_ready) | |
3034 | *ts = current_sched_info->new_ready (next, *ts); | |
3035 | ||
3036 | /* * if !(old_ts & SPECULATIVE) (e.g. HARD_DEP or 0), then insn might | |
3037 | have its original pattern or changed (speculative) one. This is due | |
3038 | to changing ebb in region scheduling. | |
3039 | * But if (old_ts & SPECULATIVE), then we are pretty sure that insn | |
3040 | has speculative pattern. | |
3041 | ||
3042 | We can't assert (!(*ts & HARD_DEP) || *ts == old_ts) here because | |
3043 | control-speculative NEXT could have been discarded by sched-rgn.c | |
3044 | (the same case as when discarded by can_schedule_ready_p ()). */ | |
3045 | ||
3046 | if ((*ts & SPECULATIVE) | |
3047 | /* If (old_ts == *ts), then (old_ts & SPECULATIVE) and we don't | |
3048 | need to change anything. */ | |
3049 | && *ts != old_ts) | |
3050 | { | |
3051 | int res; | |
3052 | rtx new_pat; | |
3053 | ||
3054 | gcc_assert ((*ts & SPECULATIVE) && !(*ts & ~SPECULATIVE)); | |
3055 | ||
3056 | res = speculate_insn (next, *ts, &new_pat); | |
3057 | ||
3058 | switch (res) | |
3059 | { | |
3060 | case -1: | |
3061 | /* It would be nice to change DEP_STATUS of all dependences, | |
3062 | which have ((DEP_STATUS & SPECULATIVE) == *ts) to HARD_DEP, | |
3063 | so we won't reanalyze anything. */ | |
3064 | *ts = (*ts & ~SPECULATIVE) | HARD_DEP; | |
3065 | break; | |
3066 | ||
3067 | case 0: | |
3068 | /* We follow the rule, that every speculative insn | |
3069 | has non-null ORIG_PAT. */ | |
3070 | if (!ORIG_PAT (next)) | |
3071 | ORIG_PAT (next) = PATTERN (next); | |
3072 | break; | |
3073 | ||
3074 | case 1: | |
3075 | if (!ORIG_PAT (next)) | |
3076 | /* If we gonna to overwrite the original pattern of insn, | |
3077 | save it. */ | |
3078 | ORIG_PAT (next) = PATTERN (next); | |
3079 | ||
3080 | change_pattern (next, new_pat); | |
3081 | break; | |
3082 | ||
3083 | default: | |
3084 | gcc_unreachable (); | |
3085 | } | |
3086 | } | |
3087 | ||
3088 | /* We need to restore pattern only if (*ts == 0), because otherwise it is | |
3089 | either correct (*ts & SPECULATIVE), | |
3090 | or we simply don't care (*ts & HARD_DEP). */ | |
3091 | ||
3092 | gcc_assert (!ORIG_PAT (next) | |
3093 | || !RECOVERY_BLOCK (next) | |
3094 | || RECOVERY_BLOCK (next) == EXIT_BLOCK_PTR); | |
3095 | ||
3096 | if (*ts & HARD_DEP) | |
3097 | { | |
3098 | /* We can't assert (QUEUE_INDEX (next) == QUEUE_NOWHERE) here because | |
3099 | control-speculative NEXT could have been discarded by sched-rgn.c | |
3100 | (the same case as when discarded by can_schedule_ready_p ()). */ | |
3101 | /*gcc_assert (QUEUE_INDEX (next) == QUEUE_NOWHERE);*/ | |
3102 | ||
3103 | change_queue_index (next, QUEUE_NOWHERE); | |
3104 | return -1; | |
3105 | } | |
3106 | else if (!(*ts & BEGIN_SPEC) && ORIG_PAT (next) && !RECOVERY_BLOCK (next)) | |
3107 | /* We should change pattern of every previously speculative | |
3108 | instruction - and we determine if NEXT was speculative by using | |
3109 | ORIG_PAT field. Except one case - simple checks have ORIG_PAT | |
3110 | pat too, hence we also check for the RECOVERY_BLOCK. */ | |
3111 | { | |
3112 | change_pattern (next, ORIG_PAT (next)); | |
3113 | ORIG_PAT (next) = 0; | |
3114 | } | |
3115 | ||
3116 | if (sched_verbose >= 2) | |
3117 | { | |
3118 | int s = TODO_SPEC (next); | |
3119 | ||
3120 | fprintf (sched_dump, ";;\t\tdependencies resolved: insn %s", | |
3121 | (*current_sched_info->print_insn) (next, 0)); | |
3122 | ||
3123 | if (spec_info && spec_info->dump) | |
3124 | { | |
3125 | if (s & BEGIN_DATA) | |
3126 | fprintf (spec_info->dump, "; data-spec;"); | |
3127 | if (s & BEGIN_CONTROL) | |
3128 | fprintf (spec_info->dump, "; control-spec;"); | |
3129 | if (s & BE_IN_CONTROL) | |
3130 | fprintf (spec_info->dump, "; in-control-spec;"); | |
3131 | } | |
3132 | ||
3133 | fprintf (sched_dump, "\n"); | |
3134 | } | |
3135 | ||
3136 | adjust_priority (next); | |
3137 | ||
3138 | return fix_tick_ready (next); | |
3139 | } | |
3140 | ||
3141 | /* Calculate INSN_TICK of NEXT and add it to either ready or queue list. */ | |
3142 | static int | |
3143 | fix_tick_ready (rtx next) | |
3144 | { | |
3145 | rtx link; | |
3146 | int tick, delay; | |
3147 | ||
3148 | link = RESOLVED_DEPS (next); | |
3149 | ||
3150 | if (link) | |
3151 | { | |
3152 | int full_p; | |
3153 | ||
3154 | tick = INSN_TICK (next); | |
3155 | /* if tick is not equal to INVALID_TICK, then update | |
3156 | INSN_TICK of NEXT with the most recent resolved dependence | |
3157 | cost. Otherwise, recalculate from scratch. */ | |
3158 | full_p = tick == INVALID_TICK; | |
3159 | do | |
3160 | { | |
3161 | rtx pro; | |
3162 | int tick1; | |
3163 | ||
3164 | pro = XEXP (link, 0); | |
3165 | gcc_assert (INSN_TICK (pro) >= MIN_TICK); | |
3166 | ||
3167 | tick1 = INSN_TICK (pro) + insn_cost (pro, link, next); | |
3168 | if (tick1 > tick) | |
3169 | tick = tick1; | |
3170 | } | |
3171 | while ((link = XEXP (link, 1)) && full_p); | |
3172 | } | |
3173 | else | |
3174 | tick = -1; | |
3175 | ||
3176 | INSN_TICK (next) = tick; | |
3177 | ||
3178 | delay = tick - clock_var; | |
3179 | if (delay <= 0) | |
3180 | delay = QUEUE_READY; | |
3181 | ||
3182 | change_queue_index (next, delay); | |
3183 | ||
3184 | return delay; | |
3185 | } | |
3186 | ||
3187 | /* Move NEXT to the proper queue list with (DELAY >= 1), | |
3188 | or add it to the ready list (DELAY == QUEUE_READY), | |
3189 | or remove it from ready and queue lists at all (DELAY == QUEUE_NOWHERE). */ | |
3190 | static void | |
3191 | change_queue_index (rtx next, int delay) | |
3192 | { | |
3193 | int i = QUEUE_INDEX (next); | |
3194 | ||
3195 | gcc_assert (QUEUE_NOWHERE <= delay && delay <= max_insn_queue_index | |
3196 | && delay != 0); | |
3197 | gcc_assert (i != QUEUE_SCHEDULED); | |
3198 | ||
3199 | if ((delay > 0 && NEXT_Q_AFTER (q_ptr, delay) == i) | |
3200 | || (delay < 0 && delay == i)) | |
3201 | /* We have nothing to do. */ | |
3202 | return; | |
3203 | ||
3204 | /* Remove NEXT from wherever it is now. */ | |
3205 | if (i == QUEUE_READY) | |
3206 | ready_remove_insn (next); | |
3207 | else if (i >= 0) | |
3208 | queue_remove (next); | |
3209 | ||
3210 | /* Add it to the proper place. */ | |
3211 | if (delay == QUEUE_READY) | |
3212 | ready_add (readyp, next, false); | |
3213 | else if (delay >= 1) | |
3214 | queue_insn (next, delay); | |
3215 | ||
3216 | if (sched_verbose >= 2) | |
3217 | { | |
3218 | fprintf (sched_dump, ";;\t\ttick updated: insn %s", | |
3219 | (*current_sched_info->print_insn) (next, 0)); | |
3220 | ||
3221 | if (delay == QUEUE_READY) | |
3222 | fprintf (sched_dump, " into ready\n"); | |
3223 | else if (delay >= 1) | |
3224 | fprintf (sched_dump, " into queue with cost=%d\n", delay); | |
3225 | else | |
3226 | fprintf (sched_dump, " removed from ready or queue lists\n"); | |
3227 | } | |
3228 | } | |
3229 | ||
3230 | /* INSN is being scheduled. Resolve the dependence between INSN and NEXT. */ | |
3231 | static void | |
3232 | resolve_dep (rtx next, rtx insn) | |
3233 | { | |
3234 | rtx dep; | |
3235 | ||
3236 | INSN_DEP_COUNT (next)--; | |
3237 | ||
3238 | dep = remove_list_elem (insn, &LOG_LINKS (next)); | |
3239 | XEXP (dep, 1) = RESOLVED_DEPS (next); | |
3240 | RESOLVED_DEPS (next) = dep; | |
3241 | ||
3242 | gcc_assert ((INSN_DEP_COUNT (next) != 0 || !LOG_LINKS (next)) | |
3243 | && (LOG_LINKS (next) || INSN_DEP_COUNT (next) == 0)); | |
3244 | } | |
3245 | ||
3246 | /* Extend H_I_D data. */ | |
3247 | static void | |
3248 | extend_h_i_d (void) | |
3249 | { | |
3250 | /* We use LUID 0 for the fake insn (UID 0) which holds dependencies for | |
3251 | pseudos which do not cross calls. */ | |
3252 | int new_max_uid = get_max_uid() + 1; | |
3253 | ||
3254 | h_i_d = xrecalloc (h_i_d, new_max_uid, old_max_uid, sizeof (*h_i_d)); | |
3255 | old_max_uid = new_max_uid; | |
3256 | ||
3257 | if (targetm.sched.h_i_d_extended) | |
3258 | targetm.sched.h_i_d_extended (); | |
3259 | } | |
3260 | ||
3261 | /* Extend READY, READY_TRY and CHOICE_STACK arrays. | |
3262 | N_NEW_INSNS is the number of additional elements to allocate. */ | |
3263 | static void | |
3264 | extend_ready (int n_new_insns) | |
3265 | { | |
3266 | int i; | |
3267 | ||
3268 | readyp->veclen = rgn_n_insns + n_new_insns + 1 + issue_rate; | |
3269 | readyp->vec = XRESIZEVEC (rtx, readyp->vec, readyp->veclen); | |
3270 | ||
3271 | ready_try = xrecalloc (ready_try, rgn_n_insns + n_new_insns + 1, | |
3272 | rgn_n_insns + 1, sizeof (char)); | |
3273 | ||
3274 | rgn_n_insns += n_new_insns; | |
3275 | ||
3276 | choice_stack = XRESIZEVEC (struct choice_entry, choice_stack, | |
3277 | rgn_n_insns + 1); | |
3278 | ||
3279 | for (i = rgn_n_insns; n_new_insns--; i--) | |
3280 | choice_stack[i].state = xmalloc (dfa_state_size); | |
3281 | } | |
3282 | ||
3283 | /* Extend global scheduler structures (those, that live across calls to | |
3284 | schedule_block) to include information about just emitted INSN. */ | |
3285 | static void | |
3286 | extend_global (rtx insn) | |
3287 | { | |
3288 | gcc_assert (INSN_P (insn)); | |
3289 | /* These structures have scheduler scope. */ | |
3290 | extend_h_i_d (); | |
3291 | init_h_i_d (insn); | |
3292 | ||
3293 | extend_dependency_caches (1, 0); | |
3294 | } | |
3295 | ||
3296 | /* Extends global and local scheduler structures to include information | |
3297 | about just emitted INSN. */ | |
3298 | static void | |
3299 | extend_all (rtx insn) | |
3300 | { | |
3301 | extend_global (insn); | |
3302 | ||
3303 | /* These structures have block scope. */ | |
3304 | extend_ready (1); | |
3305 | ||
3306 | (*current_sched_info->add_remove_insn) (insn, 0); | |
3307 | } | |
3308 | ||
3309 | /* Initialize h_i_d entry of the new INSN with default values. | |
3310 | Values, that are not explicitly initialized here, hold zero. */ | |
3311 | static void | |
3312 | init_h_i_d (rtx insn) | |
3313 | { | |
3314 | INSN_LUID (insn) = luid++; | |
3315 | INSN_COST (insn) = -1; | |
3316 | TODO_SPEC (insn) = HARD_DEP; | |
3317 | QUEUE_INDEX (insn) = QUEUE_NOWHERE; | |
3318 | INSN_TICK (insn) = INVALID_TICK; | |
3319 | INTER_TICK (insn) = INVALID_TICK; | |
3320 | find_insn_reg_weight1 (insn); | |
3321 | } | |
3322 | ||
3323 | /* Generates recovery code for INSN. */ | |
3324 | static void | |
3325 | generate_recovery_code (rtx insn) | |
3326 | { | |
3327 | if (TODO_SPEC (insn) & BEGIN_SPEC) | |
3328 | begin_speculative_block (insn); | |
3329 | ||
3330 | /* Here we have insn with no dependencies to | |
3331 | instructions other then CHECK_SPEC ones. */ | |
3332 | ||
3333 | if (TODO_SPEC (insn) & BE_IN_SPEC) | |
3334 | add_to_speculative_block (insn); | |
3335 | } | |
3336 | ||
3337 | /* Helper function. | |
3338 | Tries to add speculative dependencies of type FS between instructions | |
3339 | in LINK list and TWIN. */ | |
3340 | static void | |
3341 | process_insn_depend_be_in_spec (rtx link, rtx twin, ds_t fs) | |
3342 | { | |
3343 | for (; link; link = XEXP (link, 1)) | |
3344 | { | |
3345 | ds_t ds; | |
3346 | rtx consumer; | |
3347 | ||
3348 | consumer = XEXP (link, 0); | |
3349 | ||
3350 | ds = DEP_STATUS (link); | |
3351 | ||
3352 | if (/* If we want to create speculative dep. */ | |
3353 | fs | |
3354 | /* And we can do that because this is a true dep. */ | |
3355 | && (ds & DEP_TYPES) == DEP_TRUE) | |
3356 | { | |
3357 | gcc_assert (!(ds & BE_IN_SPEC)); | |
3358 | ||
3359 | if (/* If this dep can be overcome with 'begin speculation'. */ | |
3360 | ds & BEGIN_SPEC) | |
3361 | /* Then we have a choice: keep the dep 'begin speculative' | |
3362 | or transform it into 'be in speculative'. */ | |
3363 | { | |
3364 | if (/* In try_ready we assert that if insn once became ready | |
3365 | it can be removed from the ready (or queue) list only | |
3366 | due to backend decision. Hence we can't let the | |
3367 | probability of the speculative dep to decrease. */ | |
3368 | dep_weak (ds) <= dep_weak (fs)) | |
3369 | /* Transform it to be in speculative. */ | |
3370 | ds = (ds & ~BEGIN_SPEC) | fs; | |
3371 | } | |
3372 | else | |
3373 | /* Mark the dep as 'be in speculative'. */ | |
3374 | ds |= fs; | |
3375 | } | |
3376 | ||
3377 | add_back_forw_dep (consumer, twin, REG_NOTE_KIND (link), ds); | |
3378 | } | |
3379 | } | |
3380 | ||
3381 | /* Generates recovery code for BEGIN speculative INSN. */ | |
3382 | static void | |
3383 | begin_speculative_block (rtx insn) | |
3384 | { | |
3385 | if (TODO_SPEC (insn) & BEGIN_DATA) | |
3386 | nr_begin_data++; | |
3387 | if (TODO_SPEC (insn) & BEGIN_CONTROL) | |
3388 | nr_begin_control++; | |
3389 | ||
3390 | create_check_block_twin (insn, false); | |
3391 | ||
3392 | TODO_SPEC (insn) &= ~BEGIN_SPEC; | |
3393 | } | |
3394 | ||
3395 | /* Generates recovery code for BE_IN speculative INSN. */ | |
3396 | static void | |
3397 | add_to_speculative_block (rtx insn) | |
3398 | { | |
3399 | ds_t ts; | |
3400 | rtx link, twins = NULL; | |
3401 | ||
3402 | ts = TODO_SPEC (insn); | |
3403 | gcc_assert (!(ts & ~BE_IN_SPEC)); | |
3404 | ||
3405 | if (ts & BE_IN_DATA) | |
3406 | nr_be_in_data++; | |
3407 | if (ts & BE_IN_CONTROL) | |
3408 | nr_be_in_control++; | |
3409 | ||
3410 | TODO_SPEC (insn) &= ~BE_IN_SPEC; | |
3411 | gcc_assert (!TODO_SPEC (insn)); | |
3412 | ||
3413 | DONE_SPEC (insn) |= ts; | |
3414 | ||
3415 | /* First we convert all simple checks to branchy. */ | |
3416 | for (link = LOG_LINKS (insn); link;) | |
3417 | { | |
3418 | rtx check; | |
3419 | ||
3420 | check = XEXP (link, 0); | |
3421 | ||
3422 | if (RECOVERY_BLOCK (check)) | |
3423 | { | |
3424 | create_check_block_twin (check, true); | |
3425 | link = LOG_LINKS (insn); | |
3426 | } | |
3427 | else | |
3428 | link = XEXP (link, 1); | |
3429 | } | |
3430 | ||
3431 | clear_priorities (insn); | |
3432 | ||
3433 | do | |
3434 | { | |
3435 | rtx link, check, twin; | |
3436 | basic_block rec; | |
3437 | ||
3438 | link = LOG_LINKS (insn); | |
3439 | gcc_assert (!(DEP_STATUS (link) & BEGIN_SPEC) | |
3440 | && (DEP_STATUS (link) & BE_IN_SPEC) | |
3441 | && (DEP_STATUS (link) & DEP_TYPES) == DEP_TRUE); | |
3442 | ||
3443 | check = XEXP (link, 0); | |
3444 | gcc_assert (!RECOVERY_BLOCK (check) && !ORIG_PAT (check) | |
3445 | && QUEUE_INDEX (check) == QUEUE_NOWHERE); | |
3446 | ||
3447 | rec = BLOCK_FOR_INSN (check); | |
3448 | ||
3449 | twin = emit_insn_before (copy_rtx (PATTERN (insn)), BB_END (rec)); | |
3450 | extend_global (twin); | |
3451 | ||
3452 | RESOLVED_DEPS (twin) = copy_DEPS_LIST_list (RESOLVED_DEPS (insn)); | |
3453 | ||
3454 | if (sched_verbose && spec_info->dump) | |
3455 | /* INSN_BB (insn) isn't determined for twin insns yet. | |
3456 | So we can't use current_sched_info->print_insn. */ | |
3457 | fprintf (spec_info->dump, ";;\t\tGenerated twin insn : %d/rec%d\n", | |
3458 | INSN_UID (twin), rec->index); | |
3459 | ||
3460 | twins = alloc_INSN_LIST (twin, twins); | |
3461 | ||
3462 | /* Add dependences between TWIN and all appropriate | |
3463 | instructions from REC. */ | |
3464 | do | |
3465 | { | |
3466 | add_back_forw_dep (twin, check, REG_DEP_TRUE, DEP_TRUE); | |
3467 | ||
3468 | do | |
3469 | { | |
3470 | link = XEXP (link, 1); | |
3471 | if (link) | |
3472 | { | |
3473 | check = XEXP (link, 0); | |
3474 | if (BLOCK_FOR_INSN (check) == rec) | |
3475 | break; | |
3476 | } | |
3477 | else | |
3478 | break; | |
3479 | } | |
3480 | while (1); | |
3481 | } | |
3482 | while (link); | |
3483 | ||
3484 | process_insn_depend_be_in_spec (INSN_DEPEND (insn), twin, ts); | |
3485 | ||
3486 | for (link = LOG_LINKS (insn); link;) | |
3487 | { | |
3488 | check = XEXP (link, 0); | |
3489 | ||
3490 | if (BLOCK_FOR_INSN (check) == rec) | |
3491 | { | |
3492 | delete_back_forw_dep (insn, check); | |
3493 | link = LOG_LINKS (insn); | |
3494 | } | |
3495 | else | |
3496 | link = XEXP (link, 1); | |
3497 | } | |
3498 | } | |
3499 | while (LOG_LINKS (insn)); | |
3500 | ||
3501 | /* We can't add the dependence between insn and twin earlier because | |
3502 | that would make twin appear in the INSN_DEPEND (insn). */ | |
3503 | while (twins) | |
3504 | { | |
3505 | rtx twin; | |
3506 | ||
3507 | twin = XEXP (twins, 0); | |
3508 | calc_priorities (twin); | |
3509 | add_back_forw_dep (twin, insn, REG_DEP_OUTPUT, DEP_OUTPUT); | |
3510 | ||
3511 | twin = XEXP (twins, 1); | |
3512 | free_INSN_LIST_node (twins); | |
3513 | twins = twin; | |
3514 | } | |
3515 | } | |
3516 | ||
3517 | /* Extends and fills with zeros (only the new part) array pointed to by P. */ | |
3518 | void * | |
3519 | xrecalloc (void *p, size_t new_nmemb, size_t old_nmemb, size_t size) | |
3520 | { | |
3521 | gcc_assert (new_nmemb >= old_nmemb); | |
3522 | p = XRESIZEVAR (void, p, new_nmemb * size); | |
3523 | memset (((char *) p) + old_nmemb * size, 0, (new_nmemb - old_nmemb) * size); | |
3524 | return p; | |
3525 | } | |
3526 | ||
3527 | /* Return the probability of speculation success for the speculation | |
3528 | status DS. */ | |
3529 | static dw_t | |
3530 | dep_weak (ds_t ds) | |
3531 | { | |
3532 | ds_t res = 1, dt; | |
3533 | int n = 0; | |
3534 | ||
3535 | dt = FIRST_SPEC_TYPE; | |
3536 | do | |
3537 | { | |
3538 | if (ds & dt) | |
3539 | { | |
3540 | res *= (ds_t) get_dep_weak (ds, dt); | |
3541 | n++; | |
3542 | } | |
3543 | ||
3544 | if (dt == LAST_SPEC_TYPE) | |
3545 | break; | |
3546 | dt <<= SPEC_TYPE_SHIFT; | |
3547 | } | |
3548 | while (1); | |
3549 | ||
3550 | gcc_assert (n); | |
3551 | while (--n) | |
3552 | res /= MAX_DEP_WEAK; | |
3553 | ||
3554 | if (res < MIN_DEP_WEAK) | |
3555 | res = MIN_DEP_WEAK; | |
3556 | ||
3557 | gcc_assert (res <= MAX_DEP_WEAK); | |
3558 | ||
3559 | return (dw_t) res; | |
3560 | } | |
3561 | ||
3562 | /* Helper function. | |
3563 | Find fallthru edge from PRED. */ | |
3564 | static edge | |
3565 | find_fallthru_edge (basic_block pred) | |
3566 | { | |
3567 | edge e; | |
3568 | edge_iterator ei; | |
3569 | basic_block succ; | |
3570 | ||
3571 | succ = pred->next_bb; | |
3572 | gcc_assert (succ->prev_bb == pred); | |
3573 | ||
3574 | if (EDGE_COUNT (pred->succs) <= EDGE_COUNT (succ->preds)) | |
3575 | { | |
3576 | FOR_EACH_EDGE (e, ei, pred->succs) | |
3577 | if (e->flags & EDGE_FALLTHRU) | |
3578 | { | |
3579 | gcc_assert (e->dest == succ); | |
3580 | return e; | |
3581 | } | |
3582 | } | |
3583 | else | |
3584 | { | |
3585 | FOR_EACH_EDGE (e, ei, succ->preds) | |
3586 | if (e->flags & EDGE_FALLTHRU) | |
3587 | { | |
3588 | gcc_assert (e->src == pred); | |
3589 | return e; | |
3590 | } | |
3591 | } | |
3592 | ||
3593 | return NULL; | |
3594 | } | |
3595 | ||
3596 | /* Initialize BEFORE_RECOVERY variable. */ | |
3597 | static void | |
3598 | init_before_recovery (void) | |
3599 | { | |
3600 | basic_block last; | |
3601 | edge e; | |
3602 | ||
3603 | last = EXIT_BLOCK_PTR->prev_bb; | |
3604 | e = find_fallthru_edge (last); | |
3605 | ||
3606 | if (e) | |
3607 | { | |
3608 | /* We create two basic blocks: | |
3609 | 1. Single instruction block is inserted right after E->SRC | |
3610 | and has jump to | |
3611 | 2. Empty block right before EXIT_BLOCK. | |
3612 | Between these two blocks recovery blocks will be emitted. */ | |
3613 | ||
3614 | basic_block single, empty; | |
3615 | rtx x, label; | |
3616 | ||
3617 | single = create_empty_bb (last); | |
3618 | empty = create_empty_bb (single); | |
3619 | ||
3620 | single->count = last->count; | |
3621 | empty->count = last->count; | |
3622 | single->frequency = last->frequency; | |
3623 | empty->frequency = last->frequency; | |
3624 | BB_COPY_PARTITION (single, last); | |
3625 | BB_COPY_PARTITION (empty, last); | |
3626 | ||
3627 | redirect_edge_succ (e, single); | |
3628 | make_single_succ_edge (single, empty, 0); | |
3629 | make_single_succ_edge (empty, EXIT_BLOCK_PTR, | |
3630 | EDGE_FALLTHRU | EDGE_CAN_FALLTHRU); | |
3631 | ||
3632 | label = block_label (empty); | |
3633 | x = emit_jump_insn_after (gen_jump (label), BB_END (single)); | |
3634 | JUMP_LABEL (x) = label; | |
3635 | LABEL_NUSES (label)++; | |
3636 | extend_global (x); | |
3637 | ||
3638 | emit_barrier_after (x); | |
3639 | ||
3640 | add_block (empty, 0); | |
3641 | add_block (single, 0); | |
3642 | ||
3643 | before_recovery = single; | |
3644 | ||
3645 | if (sched_verbose >= 2 && spec_info->dump) | |
3646 | fprintf (spec_info->dump, | |
3647 | ";;\t\tFixed fallthru to EXIT : %d->>%d->%d->>EXIT\n", | |
3648 | last->index, single->index, empty->index); | |
3649 | } | |
3650 | else | |
3651 | before_recovery = last; | |
3652 | } | |
3653 | ||
3654 | /* Returns new recovery block. */ | |
3655 | static basic_block | |
3656 | create_recovery_block (void) | |
3657 | { | |
3658 | rtx label; | |
3659 | basic_block rec; | |
3660 | ||
3661 | added_recovery_block_p = true; | |
3662 | ||
3663 | if (!before_recovery) | |
3664 | init_before_recovery (); | |
3665 | ||
3666 | label = gen_label_rtx (); | |
3667 | gcc_assert (BARRIER_P (NEXT_INSN (BB_END (before_recovery)))); | |
3668 | label = emit_label_after (label, NEXT_INSN (BB_END (before_recovery))); | |
3669 | ||
3670 | rec = create_basic_block (label, label, before_recovery); | |
3671 | emit_barrier_after (BB_END (rec)); | |
3672 | ||
3673 | if (BB_PARTITION (before_recovery) != BB_UNPARTITIONED) | |
3674 | BB_SET_PARTITION (rec, BB_COLD_PARTITION); | |
3675 | ||
3676 | if (sched_verbose && spec_info->dump) | |
3677 | fprintf (spec_info->dump, ";;\t\tGenerated recovery block rec%d\n", | |
3678 | rec->index); | |
3679 | ||
3680 | before_recovery = rec; | |
3681 | ||
3682 | return rec; | |
3683 | } | |
3684 | ||
3685 | /* This function creates recovery code for INSN. If MUTATE_P is nonzero, | |
3686 | INSN is a simple check, that should be converted to branchy one. */ | |
3687 | static void | |
3688 | create_check_block_twin (rtx insn, bool mutate_p) | |
3689 | { | |
3690 | basic_block rec; | |
3691 | rtx label, check, twin, link; | |
3692 | ds_t fs; | |
3693 | ||
3694 | gcc_assert (ORIG_PAT (insn) | |
3695 | && (!mutate_p | |
3696 | || (RECOVERY_BLOCK (insn) == EXIT_BLOCK_PTR | |
3697 | && !(TODO_SPEC (insn) & SPECULATIVE)))); | |
3698 | ||
3699 | /* Create recovery block. */ | |
3700 | if (mutate_p || targetm.sched.needs_block_p (insn)) | |
3701 | { | |
3702 | rec = create_recovery_block (); | |
3703 | label = BB_HEAD (rec); | |
3704 | } | |
3705 | else | |
3706 | { | |
3707 | rec = EXIT_BLOCK_PTR; | |
3708 | label = 0; | |
3709 | } | |
3710 | ||
3711 | /* Emit CHECK. */ | |
3712 | check = targetm.sched.gen_check (insn, label, mutate_p); | |
3713 | ||
3714 | if (rec != EXIT_BLOCK_PTR) | |
3715 | { | |
3716 | /* To have mem_reg alive at the beginning of second_bb, | |
3717 | we emit check BEFORE insn, so insn after splitting | |
3718 | insn will be at the beginning of second_bb, which will | |
3719 | provide us with the correct life information. */ | |
3720 | check = emit_jump_insn_before (check, insn); | |
3721 | JUMP_LABEL (check) = label; | |
3722 | LABEL_NUSES (label)++; | |
3723 | } | |
3724 | else | |
3725 | check = emit_insn_before (check, insn); | |
3726 | ||
3727 | /* Extend data structures. */ | |
3728 | extend_all (check); | |
3729 | RECOVERY_BLOCK (check) = rec; | |
3730 | ||
3731 | if (sched_verbose && spec_info->dump) | |
3732 | fprintf (spec_info->dump, ";;\t\tGenerated check insn : %s\n", | |
3733 | (*current_sched_info->print_insn) (check, 0)); | |
3734 | ||
3735 | gcc_assert (ORIG_PAT (insn)); | |
3736 | ||
3737 | /* Initialize TWIN (twin is a duplicate of original instruction | |
3738 | in the recovery block). */ | |
3739 | if (rec != EXIT_BLOCK_PTR) | |
3740 | { | |
3741 | rtx link; | |
3742 | ||
3743 | for (link = RESOLVED_DEPS (insn); link; link = XEXP (link, 1)) | |
3744 | if (DEP_STATUS (link) & DEP_OUTPUT) | |
3745 | { | |
3746 | RESOLVED_DEPS (check) = | |
3747 | alloc_DEPS_LIST (XEXP (link, 0), RESOLVED_DEPS (check), DEP_TRUE); | |
3748 | PUT_REG_NOTE_KIND (RESOLVED_DEPS (check), REG_DEP_TRUE); | |
3749 | } | |
3750 | ||
3751 | twin = emit_insn_after (ORIG_PAT (insn), BB_END (rec)); | |
3752 | extend_global (twin); | |
3753 | ||
3754 | if (sched_verbose && spec_info->dump) | |
3755 | /* INSN_BB (insn) isn't determined for twin insns yet. | |
3756 | So we can't use current_sched_info->print_insn. */ | |
3757 | fprintf (spec_info->dump, ";;\t\tGenerated twin insn : %d/rec%d\n", | |
3758 | INSN_UID (twin), rec->index); | |
3759 | } | |
3760 | else | |
3761 | { | |
3762 | ORIG_PAT (check) = ORIG_PAT (insn); | |
3763 | HAS_INTERNAL_DEP (check) = 1; | |
3764 | twin = check; | |
3765 | /* ??? We probably should change all OUTPUT dependencies to | |
3766 | (TRUE | OUTPUT). */ | |
3767 | } | |
3768 | ||
3769 | RESOLVED_DEPS (twin) = copy_DEPS_LIST_list (RESOLVED_DEPS (insn)); | |
3770 | ||
3771 | if (rec != EXIT_BLOCK_PTR) | |
3772 | /* In case of branchy check, fix CFG. */ | |
3773 | { | |
3774 | basic_block first_bb, second_bb; | |
3775 | rtx jump; | |
3776 | edge e; | |
3777 | int edge_flags; | |
3778 | ||
3779 | first_bb = BLOCK_FOR_INSN (check); | |
3780 | e = split_block (first_bb, check); | |
3781 | /* split_block emits note if *check == BB_END. Probably it | |
3782 | is better to rip that note off. */ | |
3783 | gcc_assert (e->src == first_bb); | |
3784 | second_bb = e->dest; | |
3785 | ||
3786 | /* This is fixing of incoming edge. */ | |
3787 | /* ??? Which other flags should be specified? */ | |
3788 | if (BB_PARTITION (first_bb) != BB_PARTITION (rec)) | |
3789 | /* Partition type is the same, if it is "unpartitioned". */ | |
3790 | edge_flags = EDGE_CROSSING; | |
3791 | else | |
3792 | edge_flags = 0; | |
3793 | ||
3794 | e = make_edge (first_bb, rec, edge_flags); | |
3795 | ||
3796 | add_block (second_bb, first_bb); | |
3797 | ||
3798 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (BB_HEAD (second_bb))); | |
3799 | label = block_label (second_bb); | |
3800 | jump = emit_jump_insn_after (gen_jump (label), BB_END (rec)); | |
3801 | JUMP_LABEL (jump) = label; | |
3802 | LABEL_NUSES (label)++; | |
3803 | extend_global (jump); | |
3804 | ||
3805 | if (BB_PARTITION (second_bb) != BB_PARTITION (rec)) | |
3806 | /* Partition type is the same, if it is "unpartitioned". */ | |
3807 | { | |
3808 | /* Rewritten from cfgrtl.c. */ | |
3809 | if (flag_reorder_blocks_and_partition | |
3810 | && targetm.have_named_sections | |
3811 | /*&& !any_condjump_p (jump)*/) | |
3812 | /* any_condjump_p (jump) == false. | |
3813 | We don't need the same note for the check because | |
3814 | any_condjump_p (check) == true. */ | |
3815 | { | |
3816 | REG_NOTES (jump) = gen_rtx_EXPR_LIST (REG_CROSSING_JUMP, | |
3817 | NULL_RTX, | |
3818 | REG_NOTES (jump)); | |
3819 | } | |
3820 | edge_flags = EDGE_CROSSING; | |
3821 | } | |
3822 | else | |
3823 | edge_flags = 0; | |
3824 | ||
3825 | make_single_succ_edge (rec, second_bb, edge_flags); | |
3826 | ||
3827 | add_block (rec, EXIT_BLOCK_PTR); | |
3828 | } | |
3829 | ||
3830 | /* Move backward dependences from INSN to CHECK and | |
3831 | move forward dependences from INSN to TWIN. */ | |
3832 | for (link = LOG_LINKS (insn); link; link = XEXP (link, 1)) | |
3833 | { | |
3834 | ds_t ds; | |
3835 | ||
3836 | /* If BEGIN_DATA: [insn ~~TRUE~~> producer]: | |
3837 | check --TRUE--> producer ??? or ANTI ??? | |
3838 | twin --TRUE--> producer | |
3839 | twin --ANTI--> check | |
3840 | ||
3841 | If BEGIN_CONTROL: [insn ~~ANTI~~> producer]: | |
3842 | check --ANTI--> producer | |
3843 | twin --ANTI--> producer | |
3844 | twin --ANTI--> check | |
3845 | ||
3846 | If BE_IN_SPEC: [insn ~~TRUE~~> producer]: | |
3847 | check ~~TRUE~~> producer | |
3848 | twin ~~TRUE~~> producer | |
3849 | twin --ANTI--> check */ | |
3850 | ||
3851 | ds = DEP_STATUS (link); | |
3852 | ||
3853 | if (ds & BEGIN_SPEC) | |
3854 | { | |
3855 | gcc_assert (!mutate_p); | |
3856 | ds &= ~BEGIN_SPEC; | |
3857 | } | |
3858 | ||
3859 | if (rec != EXIT_BLOCK_PTR) | |
3860 | { | |
3861 | add_back_forw_dep (check, XEXP (link, 0), REG_NOTE_KIND (link), ds); | |
3862 | add_back_forw_dep (twin, XEXP (link, 0), REG_NOTE_KIND (link), ds); | |
3863 | } | |
3864 | else | |
3865 | add_back_forw_dep (check, XEXP (link, 0), REG_NOTE_KIND (link), ds); | |
3866 | } | |
3867 | ||
3868 | for (link = LOG_LINKS (insn); link;) | |
3869 | if ((DEP_STATUS (link) & BEGIN_SPEC) | |
3870 | || mutate_p) | |
3871 | /* We can delete this dep only if we totally overcome it with | |
3872 | BEGIN_SPECULATION. */ | |
3873 | { | |
3874 | delete_back_forw_dep (insn, XEXP (link, 0)); | |
3875 | link = LOG_LINKS (insn); | |
3876 | } | |
3877 | else | |
3878 | link = XEXP (link, 1); | |
3879 | ||
3880 | fs = 0; | |
3881 | ||
3882 | /* Fields (DONE_SPEC (x) & BEGIN_SPEC) and CHECK_SPEC (x) are set only | |
3883 | here. */ | |
3884 | ||
3885 | gcc_assert (!DONE_SPEC (insn)); | |
3886 | ||
3887 | if (!mutate_p) | |
3888 | { | |
3889 | ds_t ts = TODO_SPEC (insn); | |
3890 | ||
3891 | DONE_SPEC (insn) = ts & BEGIN_SPEC; | |
3892 | CHECK_SPEC (check) = ts & BEGIN_SPEC; | |
3893 | ||
3894 | if (ts & BEGIN_DATA) | |
3895 | fs = set_dep_weak (fs, BE_IN_DATA, get_dep_weak (ts, BEGIN_DATA)); | |
3896 | if (ts & BEGIN_CONTROL) | |
3897 | fs = set_dep_weak (fs, BE_IN_CONTROL, get_dep_weak (ts, BEGIN_CONTROL)); | |
3898 | } | |
3899 | else | |
3900 | CHECK_SPEC (check) = CHECK_SPEC (insn); | |
3901 | ||
3902 | /* Future speculations: call the helper. */ | |
3903 | process_insn_depend_be_in_spec (INSN_DEPEND (insn), twin, fs); | |
3904 | ||
3905 | if (rec != EXIT_BLOCK_PTR) | |
3906 | { | |
3907 | /* Which types of dependencies should we use here is, | |
3908 | generally, machine-dependent question... But, for now, | |
3909 | it is not. */ | |
3910 | ||
3911 | if (!mutate_p) | |
3912 | { | |
3913 | add_back_forw_dep (check, insn, REG_DEP_TRUE, DEP_TRUE); | |
3914 | add_back_forw_dep (twin, insn, REG_DEP_OUTPUT, DEP_OUTPUT); | |
3915 | } | |
3916 | else | |
3917 | { | |
3918 | if (spec_info->dump) | |
3919 | fprintf (spec_info->dump, ";;\t\tRemoved simple check : %s\n", | |
3920 | (*current_sched_info->print_insn) (insn, 0)); | |
3921 | ||
3922 | for (link = INSN_DEPEND (insn); link; link = INSN_DEPEND (insn)) | |
3923 | delete_back_forw_dep (XEXP (link, 0), insn); | |
3924 | ||
3925 | if (QUEUE_INDEX (insn) != QUEUE_NOWHERE) | |
3926 | try_ready (check); | |
3927 | ||
3928 | sched_remove_insn (insn); | |
3929 | } | |
3930 | ||
3931 | add_back_forw_dep (twin, check, REG_DEP_ANTI, DEP_ANTI); | |
3932 | } | |
3933 | else | |
3934 | add_back_forw_dep (check, insn, REG_DEP_TRUE, DEP_TRUE | DEP_OUTPUT); | |
3935 | ||
3936 | if (!mutate_p) | |
3937 | /* Fix priorities. If MUTATE_P is nonzero, this is not necessary, | |
3938 | because it'll be done later in add_to_speculative_block. */ | |
3939 | { | |
3940 | clear_priorities (twin); | |
3941 | calc_priorities (twin); | |
3942 | } | |
3943 | } | |
3944 | ||
3945 | /* Removes dependency between instructions in the recovery block REC | |
3946 | and usual region instructions. It keeps inner dependences so it | |
3947 | won't be necessary to recompute them. */ | |
3948 | static void | |
3949 | fix_recovery_deps (basic_block rec) | |
3950 | { | |
3951 | rtx note, insn, link, jump, ready_list = 0; | |
3952 | bitmap_head in_ready; | |
3953 | ||
3954 | bitmap_initialize (&in_ready, 0); | |
3955 | ||
3956 | /* NOTE - a basic block note. */ | |
3957 | note = NEXT_INSN (BB_HEAD (rec)); | |
3958 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (note)); | |
3959 | insn = BB_END (rec); | |
3960 | gcc_assert (JUMP_P (insn)); | |
3961 | insn = PREV_INSN (insn); | |
3962 | ||
3963 | do | |
3964 | { | |
3965 | for (link = INSN_DEPEND (insn); link;) | |
3966 | { | |
3967 | rtx consumer; | |
3968 | ||
3969 | consumer = XEXP (link, 0); | |
3970 | ||
3971 | if (BLOCK_FOR_INSN (consumer) != rec) | |
3972 | { | |
3973 | delete_back_forw_dep (consumer, insn); | |
3974 | ||
3975 | if (!bitmap_bit_p (&in_ready, INSN_LUID (consumer))) | |
3976 | { | |
3977 | ready_list = alloc_INSN_LIST (consumer, ready_list); | |
3978 | bitmap_set_bit (&in_ready, INSN_LUID (consumer)); | |
3979 | } | |
3980 | ||
3981 | link = INSN_DEPEND (insn); | |
3982 | } | |
3983 | else | |
3984 | { | |
3985 | gcc_assert ((DEP_STATUS (link) & DEP_TYPES) == DEP_TRUE); | |
3986 | ||
3987 | link = XEXP (link, 1); | |
3988 | } | |
3989 | } | |
3990 | ||
3991 | insn = PREV_INSN (insn); | |
3992 | } | |
3993 | while (insn != note); | |
3994 | ||
3995 | bitmap_clear (&in_ready); | |
3996 | ||
3997 | /* Try to add instructions to the ready or queue list. */ | |
3998 | for (link = ready_list; link; link = XEXP (link, 1)) | |
3999 | try_ready (XEXP (link, 0)); | |
4000 | free_INSN_LIST_list (&ready_list); | |
4001 | ||
4002 | /* Fixing jump's dependences. */ | |
4003 | insn = BB_HEAD (rec); | |
4004 | jump = BB_END (rec); | |
4005 | ||
4006 | gcc_assert (LABEL_P (insn)); | |
4007 | insn = NEXT_INSN (insn); | |
4008 | ||
4009 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (insn)); | |
4010 | add_jump_dependencies (insn, jump); | |
4011 | } | |
4012 | ||
4013 | /* The function saves line notes at the beginning of block B. */ | |
4014 | static void | |
4015 | associate_line_notes_with_blocks (basic_block b) | |
4016 | { | |
4017 | rtx line; | |
4018 | ||
4019 | for (line = BB_HEAD (b); line; line = PREV_INSN (line)) | |
4020 | if (NOTE_P (line) && NOTE_LINE_NUMBER (line) > 0) | |
4021 | { | |
4022 | line_note_head[b->index] = line; | |
4023 | break; | |
4024 | } | |
4025 | /* Do a forward search as well, since we won't get to see the first | |
4026 | notes in a basic block. */ | |
4027 | for (line = BB_HEAD (b); line; line = NEXT_INSN (line)) | |
4028 | { | |
4029 | if (INSN_P (line)) | |
4030 | break; | |
4031 | if (NOTE_P (line) && NOTE_LINE_NUMBER (line) > 0) | |
4032 | line_note_head[b->index] = line; | |
4033 | } | |
4034 | } | |
4035 | ||
4036 | /* Changes pattern of the INSN to NEW_PAT. */ | |
4037 | static void | |
4038 | change_pattern (rtx insn, rtx new_pat) | |
4039 | { | |
4040 | int t; | |
4041 | ||
4042 | t = validate_change (insn, &PATTERN (insn), new_pat, 0); | |
4043 | gcc_assert (t); | |
4044 | /* Invalidate INSN_COST, so it'll be recalculated. */ | |
4045 | INSN_COST (insn) = -1; | |
4046 | /* Invalidate INSN_TICK, so it'll be recalculated. */ | |
4047 | INSN_TICK (insn) = INVALID_TICK; | |
4048 | dfa_clear_single_insn_cache (insn); | |
4049 | } | |
4050 | ||
4051 | ||
4052 | /* -1 - can't speculate, | |
4053 | 0 - for speculation with REQUEST mode it is OK to use | |
4054 | current instruction pattern, | |
4055 | 1 - need to change pattern for *NEW_PAT to be speculative. */ | |
4056 | static int | |
4057 | speculate_insn (rtx insn, ds_t request, rtx *new_pat) | |
4058 | { | |
4059 | gcc_assert (current_sched_info->flags & DO_SPECULATION | |
4060 | && (request & SPECULATIVE)); | |
4061 | ||
4062 | if (!NONJUMP_INSN_P (insn) | |
4063 | || HAS_INTERNAL_DEP (insn) | |
4064 | || SCHED_GROUP_P (insn) | |
4065 | || side_effects_p (PATTERN (insn)) | |
4066 | || (request & spec_info->mask) != request) | |
4067 | return -1; | |
4068 | ||
4069 | gcc_assert (!RECOVERY_BLOCK (insn)); | |
4070 | ||
4071 | if (request & BE_IN_SPEC) | |
4072 | { | |
4073 | if (may_trap_p (PATTERN (insn))) | |
4074 | return -1; | |
4075 | ||
4076 | if (!(request & BEGIN_SPEC)) | |
4077 | return 0; | |
4078 | } | |
4079 | ||
4080 | return targetm.sched.speculate_insn (insn, request & BEGIN_SPEC, new_pat); | |
4081 | } | |
4082 | ||
4083 | /* Print some information about block BB, which starts with HEAD and | |
4084 | ends with TAIL, before scheduling it. | |
4085 | I is zero, if scheduler is about to start with the fresh ebb. */ | |
4086 | static void | |
4087 | dump_new_block_header (int i, basic_block bb, rtx head, rtx tail) | |
4088 | { | |
4089 | if (!i) | |
4090 | fprintf (sched_dump, | |
4091 | ";; ======================================================\n"); | |
4092 | else | |
4093 | fprintf (sched_dump, | |
4094 | ";; =====================ADVANCING TO=====================\n"); | |
4095 | fprintf (sched_dump, | |
4096 | ";; -- basic block %d from %d to %d -- %s reload\n", | |
4097 | bb->index, INSN_UID (head), INSN_UID (tail), | |
4098 | (reload_completed ? "after" : "before")); | |
4099 | fprintf (sched_dump, | |
4100 | ";; ======================================================\n"); | |
4101 | fprintf (sched_dump, "\n"); | |
4102 | } | |
4103 | ||
4104 | /* Unlink basic block notes and labels and saves them, so they | |
4105 | can be easily restored. We unlink basic block notes in EBB to | |
4106 | provide back-compatibility with the previous code, as target backends | |
4107 | assume, that there'll be only instructions between | |
4108 | current_sched_info->{head and tail}. We restore these notes as soon | |
4109 | as we can. | |
4110 | FIRST (LAST) is the first (last) basic block in the ebb. | |
4111 | NB: In usual case (FIRST == LAST) nothing is really done. */ | |
4112 | void | |
4113 | unlink_bb_notes (basic_block first, basic_block last) | |
4114 | { | |
4115 | /* We DON'T unlink basic block notes of the first block in the ebb. */ | |
4116 | if (first == last) | |
4117 | return; | |
4118 | ||
4119 | bb_header = xmalloc (last_basic_block * sizeof (*bb_header)); | |
4120 | ||
4121 | /* Make a sentinel. */ | |
4122 | if (last->next_bb != EXIT_BLOCK_PTR) | |
4123 | bb_header[last->next_bb->index] = 0; | |
4124 | ||
4125 | first = first->next_bb; | |
4126 | do | |
4127 | { | |
4128 | rtx prev, label, note, next; | |
4129 | ||
4130 | label = BB_HEAD (last); | |
4131 | if (LABEL_P (label)) | |
4132 | note = NEXT_INSN (label); | |
4133 | else | |
4134 | note = label; | |
4135 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (note)); | |
4136 | ||
4137 | prev = PREV_INSN (label); | |
4138 | next = NEXT_INSN (note); | |
4139 | gcc_assert (prev && next); | |
4140 | ||
4141 | NEXT_INSN (prev) = next; | |
4142 | PREV_INSN (next) = prev; | |
4143 | ||
4144 | bb_header[last->index] = label; | |
4145 | ||
4146 | if (last == first) | |
4147 | break; | |
4148 | ||
4149 | last = last->prev_bb; | |
4150 | } | |
4151 | while (1); | |
4152 | } | |
4153 | ||
4154 | /* Restore basic block notes. | |
4155 | FIRST is the first basic block in the ebb. */ | |
4156 | static void | |
4157 | restore_bb_notes (basic_block first) | |
4158 | { | |
4159 | if (!bb_header) | |
4160 | return; | |
4161 | ||
4162 | /* We DON'T unlink basic block notes of the first block in the ebb. */ | |
4163 | first = first->next_bb; | |
4164 | /* Remember: FIRST is actually a second basic block in the ebb. */ | |
4165 | ||
4166 | while (first != EXIT_BLOCK_PTR | |
4167 | && bb_header[first->index]) | |
4168 | { | |
4169 | rtx prev, label, note, next; | |
4170 | ||
4171 | label = bb_header[first->index]; | |
4172 | prev = PREV_INSN (label); | |
4173 | next = NEXT_INSN (prev); | |
4174 | ||
4175 | if (LABEL_P (label)) | |
4176 | note = NEXT_INSN (label); | |
4177 | else | |
4178 | note = label; | |
4179 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (note)); | |
4180 | ||
4181 | bb_header[first->index] = 0; | |
4182 | ||
4183 | NEXT_INSN (prev) = label; | |
4184 | NEXT_INSN (note) = next; | |
4185 | PREV_INSN (next) = note; | |
4186 | ||
4187 | first = first->next_bb; | |
4188 | } | |
4189 | ||
4190 | free (bb_header); | |
4191 | bb_header = 0; | |
4192 | } | |
4193 | ||
4194 | /* Extend per basic block data structures of the scheduler. | |
4195 | If BB is NULL, initialize structures for the whole CFG. | |
4196 | Otherwise, initialize them for the just created BB. */ | |
4197 | static void | |
4198 | extend_bb (basic_block bb) | |
4199 | { | |
4200 | rtx insn; | |
4201 | ||
4202 | if (write_symbols != NO_DEBUG) | |
4203 | { | |
4204 | /* Save-line-note-head: | |
4205 | Determine the line-number at the start of each basic block. | |
4206 | This must be computed and saved now, because after a basic block's | |
4207 | predecessor has been scheduled, it is impossible to accurately | |
4208 | determine the correct line number for the first insn of the block. */ | |
4209 | line_note_head = xrecalloc (line_note_head, last_basic_block, | |
4210 | old_last_basic_block, | |
4211 | sizeof (*line_note_head)); | |
4212 | ||
4213 | if (bb) | |
4214 | associate_line_notes_with_blocks (bb); | |
4215 | else | |
4216 | FOR_EACH_BB (bb) | |
4217 | associate_line_notes_with_blocks (bb); | |
4218 | } | |
4219 | ||
4220 | old_last_basic_block = last_basic_block; | |
4221 | ||
4222 | if (current_sched_info->flags & USE_GLAT) | |
4223 | { | |
4224 | glat_start = xrealloc (glat_start, | |
4225 | last_basic_block * sizeof (*glat_start)); | |
4226 | glat_end = xrealloc (glat_end, last_basic_block * sizeof (*glat_end)); | |
4227 | } | |
4228 | ||
4229 | /* The following is done to keep current_sched_info->next_tail non null. */ | |
4230 | ||
4231 | insn = BB_END (EXIT_BLOCK_PTR->prev_bb); | |
4232 | if (NEXT_INSN (insn) == 0 | |
4233 | || (!NOTE_P (insn) | |
4234 | && !LABEL_P (insn) | |
4235 | /* Don't emit a NOTE if it would end up before a BARRIER. */ | |
4236 | && !BARRIER_P (NEXT_INSN (insn)))) | |
4237 | { | |
4238 | emit_note_after (NOTE_INSN_DELETED, insn); | |
4239 | /* Make insn to appear outside BB. */ | |
4240 | BB_END (EXIT_BLOCK_PTR->prev_bb) = insn; | |
4241 | } | |
4242 | } | |
4243 | ||
4244 | /* Add a basic block BB to extended basic block EBB. | |
4245 | If EBB is EXIT_BLOCK_PTR, then BB is recovery block. | |
4246 | If EBB is NULL, then BB should be a new region. */ | |
4247 | void | |
4248 | add_block (basic_block bb, basic_block ebb) | |
4249 | { | |
4250 | gcc_assert (current_sched_info->flags & DETACH_LIFE_INFO | |
4251 | && bb->il.rtl->global_live_at_start == 0 | |
4252 | && bb->il.rtl->global_live_at_end == 0); | |
4253 | ||
4254 | extend_bb (bb); | |
4255 | ||
4256 | glat_start[bb->index] = 0; | |
4257 | glat_end[bb->index] = 0; | |
4258 | ||
4259 | if (current_sched_info->add_block) | |
4260 | /* This changes only data structures of the front-end. */ | |
4261 | current_sched_info->add_block (bb, ebb); | |
4262 | } | |
4263 | ||
4264 | /* Helper function. | |
4265 | Fix CFG after both in- and inter-block movement of | |
4266 | control_flow_insn_p JUMP. */ | |
4267 | static void | |
4268 | fix_jump_move (rtx jump) | |
4269 | { | |
4270 | basic_block bb, jump_bb, jump_bb_next; | |
4271 | ||
4272 | bb = BLOCK_FOR_INSN (PREV_INSN (jump)); | |
4273 | jump_bb = BLOCK_FOR_INSN (jump); | |
4274 | jump_bb_next = jump_bb->next_bb; | |
4275 | ||
4276 | gcc_assert (current_sched_info->flags & SCHED_EBB | |
4277 | || (RECOVERY_BLOCK (jump) | |
4278 | && RECOVERY_BLOCK (jump) != EXIT_BLOCK_PTR)); | |
4279 | ||
4280 | if (!NOTE_INSN_BASIC_BLOCK_P (BB_END (jump_bb_next))) | |
4281 | /* if jump_bb_next is not empty. */ | |
4282 | BB_END (jump_bb) = BB_END (jump_bb_next); | |
4283 | ||
4284 | if (BB_END (bb) != PREV_INSN (jump)) | |
4285 | /* Then there are instruction after jump that should be placed | |
4286 | to jump_bb_next. */ | |
4287 | BB_END (jump_bb_next) = BB_END (bb); | |
4288 | else | |
4289 | /* Otherwise jump_bb_next is empty. */ | |
4290 | BB_END (jump_bb_next) = NEXT_INSN (BB_HEAD (jump_bb_next)); | |
4291 | ||
4292 | /* To make assertion in move_insn happy. */ | |
4293 | BB_END (bb) = PREV_INSN (jump); | |
4294 | ||
4295 | update_bb_for_insn (jump_bb_next); | |
4296 | } | |
4297 | ||
4298 | /* Fix CFG after interblock movement of control_flow_insn_p JUMP. */ | |
4299 | static void | |
4300 | move_block_after_check (rtx jump) | |
4301 | { | |
4302 | basic_block bb, jump_bb, jump_bb_next; | |
4303 | VEC(edge,gc) *t; | |
4304 | ||
4305 | bb = BLOCK_FOR_INSN (PREV_INSN (jump)); | |
4306 | jump_bb = BLOCK_FOR_INSN (jump); | |
4307 | jump_bb_next = jump_bb->next_bb; | |
4308 | ||
4309 | update_bb_for_insn (jump_bb); | |
4310 | ||
4311 | gcc_assert (RECOVERY_BLOCK (jump) | |
4312 | || RECOVERY_BLOCK (BB_END (jump_bb_next))); | |
4313 | ||
4314 | unlink_block (jump_bb_next); | |
4315 | link_block (jump_bb_next, bb); | |
4316 | ||
4317 | t = bb->succs; | |
4318 | bb->succs = 0; | |
4319 | move_succs (&(jump_bb->succs), bb); | |
4320 | move_succs (&(jump_bb_next->succs), jump_bb); | |
4321 | move_succs (&t, jump_bb_next); | |
4322 | ||
4323 | if (current_sched_info->fix_recovery_cfg) | |
4324 | current_sched_info->fix_recovery_cfg | |
4325 | (bb->index, jump_bb->index, jump_bb_next->index); | |
4326 | } | |
4327 | ||
4328 | /* Helper function for move_block_after_check. | |
4329 | This functions attaches edge vector pointed to by SUCCSP to | |
4330 | block TO. */ | |
4331 | static void | |
4332 | move_succs (VEC(edge,gc) **succsp, basic_block to) | |
4333 | { | |
4334 | edge e; | |
4335 | edge_iterator ei; | |
4336 | ||
4337 | gcc_assert (to->succs == 0); | |
4338 | ||
4339 | to->succs = *succsp; | |
4340 | ||
4341 | FOR_EACH_EDGE (e, ei, to->succs) | |
4342 | e->src = to; | |
4343 | ||
4344 | *succsp = 0; | |
4345 | } | |
4346 | ||
4347 | /* Initialize GLAT (global_live_at_{start, end}) structures. | |
4348 | GLAT structures are used to substitute global_live_{start, end} | |
4349 | regsets during scheduling. This is necessary to use such functions as | |
4350 | split_block (), as they assume consistency of register live information. */ | |
4351 | static void | |
4352 | init_glat (void) | |
4353 | { | |
4354 | basic_block bb; | |
4355 | ||
4356 | FOR_ALL_BB (bb) | |
4357 | init_glat1 (bb); | |
4358 | } | |
4359 | ||
4360 | /* Helper function for init_glat. */ | |
4361 | static void | |
4362 | init_glat1 (basic_block bb) | |
4363 | { | |
4364 | gcc_assert (bb->il.rtl->global_live_at_start != 0 | |
4365 | && bb->il.rtl->global_live_at_end != 0); | |
4366 | ||
4367 | glat_start[bb->index] = bb->il.rtl->global_live_at_start; | |
4368 | glat_end[bb->index] = bb->il.rtl->global_live_at_end; | |
4369 | ||
4370 | if (current_sched_info->flags & DETACH_LIFE_INFO) | |
4371 | { | |
4372 | bb->il.rtl->global_live_at_start = 0; | |
4373 | bb->il.rtl->global_live_at_end = 0; | |
4374 | } | |
4375 | } | |
4376 | ||
4377 | /* Attach reg_live_info back to basic blocks. | |
4378 | Also save regsets, that should not have been changed during scheduling, | |
4379 | for checking purposes (see check_reg_live). */ | |
4380 | void | |
4381 | attach_life_info (void) | |
4382 | { | |
4383 | basic_block bb; | |
4384 | ||
4385 | FOR_ALL_BB (bb) | |
4386 | attach_life_info1 (bb); | |
4387 | } | |
4388 | ||
4389 | /* Helper function for attach_life_info. */ | |
4390 | static void | |
4391 | attach_life_info1 (basic_block bb) | |
4392 | { | |
4393 | gcc_assert (bb->il.rtl->global_live_at_start == 0 | |
4394 | && bb->il.rtl->global_live_at_end == 0); | |
4395 | ||
4396 | if (glat_start[bb->index]) | |
4397 | { | |
4398 | gcc_assert (glat_end[bb->index]); | |
4399 | ||
4400 | bb->il.rtl->global_live_at_start = glat_start[bb->index]; | |
4401 | bb->il.rtl->global_live_at_end = glat_end[bb->index]; | |
4402 | ||
4403 | /* Make them NULL, so they won't be freed in free_glat. */ | |
4404 | glat_start[bb->index] = 0; | |
4405 | glat_end[bb->index] = 0; | |
4406 | ||
4407 | #ifdef ENABLE_CHECKING | |
4408 | if (bb->index < NUM_FIXED_BLOCKS | |
4409 | || current_sched_info->region_head_or_leaf_p (bb, 0)) | |
4410 | { | |
4411 | glat_start[bb->index] = ALLOC_REG_SET (®_obstack); | |
4412 | COPY_REG_SET (glat_start[bb->index], | |
4413 | bb->il.rtl->global_live_at_start); | |
4414 | } | |
4415 | ||
4416 | if (bb->index < NUM_FIXED_BLOCKS | |
4417 | || current_sched_info->region_head_or_leaf_p (bb, 1)) | |
4418 | { | |
4419 | glat_end[bb->index] = ALLOC_REG_SET (®_obstack); | |
4420 | COPY_REG_SET (glat_end[bb->index], bb->il.rtl->global_live_at_end); | |
4421 | } | |
4422 | #endif | |
4423 | } | |
4424 | else | |
4425 | { | |
4426 | gcc_assert (!glat_end[bb->index]); | |
4427 | ||
4428 | bb->il.rtl->global_live_at_start = ALLOC_REG_SET (®_obstack); | |
4429 | bb->il.rtl->global_live_at_end = ALLOC_REG_SET (®_obstack); | |
4430 | } | |
4431 | } | |
4432 | ||
4433 | /* Free GLAT information. */ | |
4434 | static void | |
4435 | free_glat (void) | |
4436 | { | |
4437 | #ifdef ENABLE_CHECKING | |
4438 | if (current_sched_info->flags & DETACH_LIFE_INFO) | |
4439 | { | |
4440 | basic_block bb; | |
4441 | ||
4442 | FOR_ALL_BB (bb) | |
4443 | { | |
4444 | if (glat_start[bb->index]) | |
4445 | FREE_REG_SET (glat_start[bb->index]); | |
4446 | if (glat_end[bb->index]) | |
4447 | FREE_REG_SET (glat_end[bb->index]); | |
4448 | } | |
4449 | } | |
4450 | #endif | |
4451 | ||
4452 | free (glat_start); | |
4453 | free (glat_end); | |
4454 | } | |
4455 | ||
4456 | /* Remove INSN from the instruction stream. | |
4457 | INSN should have any dependencies. */ | |
4458 | static void | |
4459 | sched_remove_insn (rtx insn) | |
4460 | { | |
4461 | change_queue_index (insn, QUEUE_NOWHERE); | |
4462 | current_sched_info->add_remove_insn (insn, 1); | |
4463 | remove_insn (insn); | |
4464 | } | |
4465 | ||
4466 | /* Clear priorities of all instructions, that are | |
4467 | forward dependent on INSN. */ | |
4468 | static void | |
4469 | clear_priorities (rtx insn) | |
4470 | { | |
4471 | rtx link; | |
4472 | ||
4473 | for (link = LOG_LINKS (insn); link; link = XEXP (link, 1)) | |
4474 | { | |
4475 | rtx pro; | |
4476 | ||
4477 | pro = XEXP (link, 0); | |
4478 | if (INSN_PRIORITY_KNOWN (pro)) | |
4479 | { | |
4480 | INSN_PRIORITY_KNOWN (pro) = 0; | |
4481 | clear_priorities (pro); | |
4482 | } | |
4483 | } | |
4484 | } | |
4485 | ||
4486 | /* Recompute priorities of instructions, whose priorities might have been | |
4487 | changed due to changes in INSN. */ | |
4488 | static void | |
4489 | calc_priorities (rtx insn) | |
4490 | { | |
4491 | rtx link; | |
4492 | ||
4493 | for (link = LOG_LINKS (insn); link; link = XEXP (link, 1)) | |
4494 | { | |
4495 | rtx pro; | |
4496 | ||
4497 | pro = XEXP (link, 0); | |
4498 | if (!INSN_PRIORITY_KNOWN (pro)) | |
4499 | { | |
4500 | priority (pro); | |
4501 | calc_priorities (pro); | |
4502 | } | |
4503 | } | |
4504 | } | |
4505 | ||
4506 | ||
4507 | /* Add dependences between JUMP and other instructions in the recovery | |
4508 | block. INSN is the first insn the recovery block. */ | |
4509 | static void | |
4510 | add_jump_dependencies (rtx insn, rtx jump) | |
4511 | { | |
4512 | do | |
4513 | { | |
4514 | insn = NEXT_INSN (insn); | |
4515 | if (insn == jump) | |
4516 | break; | |
4517 | ||
4518 | if (!INSN_DEPEND (insn)) | |
4519 | add_back_forw_dep (jump, insn, REG_DEP_ANTI, DEP_ANTI); | |
4520 | } | |
4521 | while (1); | |
4522 | gcc_assert (LOG_LINKS (jump)); | |
4523 | } | |
4524 | ||
4525 | /* Return the NOTE_INSN_BASIC_BLOCK of BB. */ | |
4526 | static rtx | |
4527 | bb_note (basic_block bb) | |
4528 | { | |
4529 | rtx note; | |
4530 | ||
4531 | note = BB_HEAD (bb); | |
4532 | if (LABEL_P (note)) | |
4533 | note = NEXT_INSN (note); | |
4534 | ||
4535 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (note)); | |
4536 | return note; | |
4537 | } | |
4538 | ||
4539 | #ifdef ENABLE_CHECKING | |
4540 | extern void debug_spec_status (ds_t); | |
4541 | ||
4542 | /* Dump information about the dependence status S. */ | |
4543 | void | |
4544 | debug_spec_status (ds_t s) | |
4545 | { | |
4546 | FILE *f = stderr; | |
4547 | ||
4548 | if (s & BEGIN_DATA) | |
4549 | fprintf (f, "BEGIN_DATA: %d; ", get_dep_weak (s, BEGIN_DATA)); | |
4550 | if (s & BE_IN_DATA) | |
4551 | fprintf (f, "BE_IN_DATA: %d; ", get_dep_weak (s, BE_IN_DATA)); | |
4552 | if (s & BEGIN_CONTROL) | |
4553 | fprintf (f, "BEGIN_CONTROL: %d; ", get_dep_weak (s, BEGIN_CONTROL)); | |
4554 | if (s & BE_IN_CONTROL) | |
4555 | fprintf (f, "BE_IN_CONTROL: %d; ", get_dep_weak (s, BE_IN_CONTROL)); | |
4556 | ||
4557 | if (s & HARD_DEP) | |
4558 | fprintf (f, "HARD_DEP; "); | |
4559 | ||
4560 | if (s & DEP_TRUE) | |
4561 | fprintf (f, "DEP_TRUE; "); | |
4562 | if (s & DEP_ANTI) | |
4563 | fprintf (f, "DEP_ANTI; "); | |
4564 | if (s & DEP_OUTPUT) | |
4565 | fprintf (f, "DEP_OUTPUT; "); | |
4566 | ||
4567 | fprintf (f, "\n"); | |
4568 | } | |
4569 | ||
4570 | /* Helper function for check_cfg. | |
4571 | Return nonzero, if edge vector pointed to by EL has edge with TYPE in | |
4572 | its flags. */ | |
4573 | static int | |
4574 | has_edge_p (VEC(edge,gc) *el, int type) | |
4575 | { | |
4576 | edge e; | |
4577 | edge_iterator ei; | |
4578 | ||
4579 | FOR_EACH_EDGE (e, ei, el) | |
4580 | if (e->flags & type) | |
4581 | return 1; | |
4582 | return 0; | |
4583 | } | |
4584 | ||
4585 | /* Check few properties of CFG between HEAD and TAIL. | |
4586 | If HEAD (TAIL) is NULL check from the beginning (till the end) of the | |
4587 | instruction stream. */ | |
4588 | static void | |
4589 | check_cfg (rtx head, rtx tail) | |
4590 | { | |
4591 | rtx next_tail; | |
4592 | basic_block bb = 0; | |
4593 | int not_first = 0, not_last; | |
4594 | ||
4595 | if (head == NULL) | |
4596 | head = get_insns (); | |
4597 | if (tail == NULL) | |
4598 | tail = get_last_insn (); | |
4599 | next_tail = NEXT_INSN (tail); | |
4600 | ||
4601 | do | |
4602 | { | |
4603 | not_last = head != tail; | |
4604 | ||
4605 | if (not_first) | |
4606 | gcc_assert (NEXT_INSN (PREV_INSN (head)) == head); | |
4607 | if (not_last) | |
4608 | gcc_assert (PREV_INSN (NEXT_INSN (head)) == head); | |
4609 | ||
4610 | if (LABEL_P (head) | |
4611 | || (NOTE_INSN_BASIC_BLOCK_P (head) | |
4612 | && (!not_first | |
4613 | || (not_first && !LABEL_P (PREV_INSN (head)))))) | |
4614 | { | |
4615 | gcc_assert (bb == 0); | |
4616 | bb = BLOCK_FOR_INSN (head); | |
4617 | if (bb != 0) | |
4618 | gcc_assert (BB_HEAD (bb) == head); | |
4619 | else | |
4620 | /* This is the case of jump table. See inside_basic_block_p (). */ | |
4621 | gcc_assert (LABEL_P (head) && !inside_basic_block_p (head)); | |
4622 | } | |
4623 | ||
4624 | if (bb == 0) | |
4625 | { | |
4626 | gcc_assert (!inside_basic_block_p (head)); | |
4627 | head = NEXT_INSN (head); | |
4628 | } | |
4629 | else | |
4630 | { | |
4631 | gcc_assert (inside_basic_block_p (head) | |
4632 | || NOTE_P (head)); | |
4633 | gcc_assert (BLOCK_FOR_INSN (head) == bb); | |
4634 | ||
4635 | if (LABEL_P (head)) | |
4636 | { | |
4637 | head = NEXT_INSN (head); | |
4638 | gcc_assert (NOTE_INSN_BASIC_BLOCK_P (head)); | |
4639 | } | |
4640 | else | |
4641 | { | |
4642 | if (control_flow_insn_p (head)) | |
4643 | { | |
4644 | gcc_assert (BB_END (bb) == head); | |
4645 | ||
4646 | if (any_uncondjump_p (head)) | |
4647 | gcc_assert (EDGE_COUNT (bb->succs) == 1 | |
4648 | && BARRIER_P (NEXT_INSN (head))); | |
4649 | else if (any_condjump_p (head)) | |
4650 | gcc_assert (EDGE_COUNT (bb->succs) > 1 | |
4651 | && !BARRIER_P (NEXT_INSN (head))); | |
4652 | } | |
4653 | if (BB_END (bb) == head) | |
4654 | { | |
4655 | if (EDGE_COUNT (bb->succs) > 1) | |
4656 | gcc_assert (control_flow_insn_p (head) | |
4657 | || has_edge_p (bb->succs, EDGE_COMPLEX)); | |
4658 | bb = 0; | |
4659 | } | |
4660 | ||
4661 | head = NEXT_INSN (head); | |
4662 | } | |
4663 | } | |
4664 | ||
4665 | not_first = 1; | |
4666 | } | |
4667 | while (head != next_tail); | |
4668 | ||
4669 | gcc_assert (bb == 0); | |
4670 | } | |
4671 | ||
4672 | /* Perform a few consistency checks of flags in different data structures. */ | |
4673 | static void | |
4674 | check_sched_flags (void) | |
4675 | { | |
4676 | unsigned int f = current_sched_info->flags; | |
4677 | ||
4678 | if (flag_sched_stalled_insns) | |
4679 | gcc_assert (!(f & DO_SPECULATION)); | |
4680 | if (f & DO_SPECULATION) | |
4681 | gcc_assert (!flag_sched_stalled_insns | |
4682 | && (f & DETACH_LIFE_INFO) | |
4683 | && spec_info | |
4684 | && spec_info->mask); | |
4685 | if (f & DETACH_LIFE_INFO) | |
4686 | gcc_assert (f & USE_GLAT); | |
4687 | } | |
4688 | ||
4689 | /* Check global_live_at_{start, end} regsets. | |
4690 | If FATAL_P is TRUE, then abort execution at the first failure. | |
4691 | Otherwise, print diagnostics to STDERR (this mode is for calling | |
4692 | from debugger). */ | |
4693 | void | |
4694 | check_reg_live (bool fatal_p) | |
4695 | { | |
4696 | basic_block bb; | |
4697 | ||
4698 | FOR_ALL_BB (bb) | |
4699 | { | |
4700 | int i; | |
4701 | ||
4702 | i = bb->index; | |
4703 | ||
4704 | if (glat_start[i]) | |
4705 | { | |
4706 | bool b = bitmap_equal_p (bb->il.rtl->global_live_at_start, | |
4707 | glat_start[i]); | |
4708 | ||
4709 | if (!b) | |
4710 | { | |
4711 | gcc_assert (!fatal_p); | |
4712 | ||
4713 | fprintf (stderr, ";; check_reg_live_at_start (%d) failed.\n", i); | |
4714 | } | |
4715 | } | |
4716 | ||
4717 | if (glat_end[i]) | |
4718 | { | |
4719 | bool b = bitmap_equal_p (bb->il.rtl->global_live_at_end, | |
4720 | glat_end[i]); | |
4721 | ||
4722 | if (!b) | |
4723 | { | |
4724 | gcc_assert (!fatal_p); | |
4725 | ||
4726 | fprintf (stderr, ";; check_reg_live_at_end (%d) failed.\n", i); | |
4727 | } | |
4728 | } | |
4729 | } | |
4730 | } | |
4731 | #endif /* ENABLE_CHECKING */ | |
4732 | ||
4733 | #endif /* INSN_SCHEDULING */ |